Search results for: convolution

Authors: Jacintha Menezes, Ramalingam Dharmalingam, Palaiahnakote Shivakumara

Abstract:

In the recent years, Omani acid lime cultivation and production has been affected by Citrus greening or Huanglongbing (HLB) disease. HLB disease is one of the most destructive diseases for citrus, with no remedies or countermeasures to stop the disease. Currently used Polymerase chain reaction (PCR) and enzyme-linked immunosorbent assay (ELISA) HLB detection tests require lengthy and labor-intensive laboratory procedures. Furthermore, the equipment and staff needed to carry out the laboratory procedures are frequently specialized hence making them a less optimal solution for the detection of the disease. The current research uses hyperspectral imaging technology for automatic detection of citrus trees with HLB disease. Omani citrus tree leaf images were captured through portable Specim IQ hyperspectral camera. The research considered healthy, nutrition deficient, and HLB infected leaf samples based on the Polymerase chain reaction (PCR) test. The highresolution image samples were sliced to into sub cubes. The sub cubes were further processed to obtain RGB images with spatial features. Similarly, RGB spectral slices were obtained through a moving window on the wavelength. The resized spectral-Spatial RGB images were given to Convolution Neural Networks for deep features extraction. The current research was able to classify a given sample to the appropriate class with 92.86% accuracy indicating the effectiveness of the proposed techniques. The significant bands with a difference in three types of leaves are found to be 560nm, 678nm, 726 nm and 750nm.

Keywords: huanglongbing (HLB), hyperspectral imaging (HSI), · omani citrus, CNN

Procedia PDF Downloads 45

Authors: Zhongmin Wang, Wudong Fan, Hengshan Zhang, Yimin Zhou

Abstract:

In data-driven prognostic methods, the prediction accuracy of the estimation for remaining useful life of bearings mainly depends on the performance of health indicators, which are usually fused some statistical features extracted from vibrating signals. However, the existing health indicators have the following two drawbacks: (1) The differnet ranges of the statistical features have the different contributions to construct the health indicators, the expert knowledge is required to extract the features. (2) When convolutional neural networks are utilized to tackle time-frequency features of signals, the time-series of signals are not considered. To overcome these drawbacks, in this study, the method combining convolutional neural network with gated recurrent unit is proposed to extract the time-frequency image features. The extracted features are utilized to construct health indicator and predict remaining useful life of bearings. First, original signals are converted into time-frequency images by using continuous wavelet transform so as to form the original feature sets. Second, with convolutional and pooling layers of convolutional neural networks, the most sensitive features of time-frequency images are selected from the original feature sets. Finally, these selected features are fed into the gated recurrent unit to construct the health indicator. The results state that the proposed method shows the enhance performance than the related studies which have used the same bearing dataset provided by PRONOSTIA.

Keywords: continuous wavelet transform, convolution neural net-work, gated recurrent unit, health indicators, remaining useful life

Procedia PDF Downloads 101

Authors: Adel Khorramrouz, Monireh Ahmadi Bani, Ehsan Norouzi, Morvarid Lalenoor

Abstract:

Background: Plantar pressure distribution measurement has been used for a long time to assess foot disorders. Plantar pressure is an important component affecting the foot and ankle function and Changes in plantar pressure distribution could indicate various foot and ankle disorders. Morphologic and mechanical properties of the foot may be important factors affecting the plantar pressure distribution. Accurate and early measurement may help to reduce the prevalence of pes planus. With recent developments in technology, new techniques such as machine learning have been used to assist clinicians in predicting patients with foot disorders. Significance of the study: This study proposes a neural network learning-based flat foot classification methodology using static foot pressure distribution. Methodologies: Data were collected from 895 patients who were referred to a foot clinic due to foot disorders. Patients with pes planus were labeled by an experienced physician based on clinical examination. Then all subjects (with and without pes planus) were evaluated for static plantar pressures distribution. Patients who were diagnosed with the flat foot in both feet were included in the study. In the next step, the leg length was normalized and the network was trained for plantar pressure mapping images. Findings: From a total of 895 image data, 581 were labeled as pes planus. A computational neural network (CNN) ran to evaluate the performance of the proposed model. The prediction accuracy of the basic CNN-based model was performed and the prediction model was derived through the proposed methodology. In the basic CNN model, the training accuracy was 79.14%, and the test accuracy was 72.09%. Conclusion: This model can be easily and simply used by patients with pes planus and doctors to predict the classification of pes planus and prescreen for possible musculoskeletal disorders related to this condition. However, more models need to be considered and compared for higher accuracy.

Keywords: foot disorder, machine learning, neural network, pes planus

Procedia PDF Downloads 325

Authors: Edward Holupka, John Rossman, Tye Morancy, Joseph Aronovitz, Irving Kaplan

Abstract:

A common modality for the treatment of early stage prostate cancer is the implantation of radioactive seeds directly into the prostate. The radioactive seeds are positioned inside the prostate to achieve optimal radiation dose coverage to the prostate. These radioactive seeds are positioned inside the prostate using Transrectal ultrasound imaging. Once all of the planned seeds have been implanted, two dimensional transaxial transrectal ultrasound images separated by 2 mm are obtained through out the prostate, beginning at the base of the prostate up to and including the apex. A common deep neural network, called DetectNet was trained to automatically determine the position of the implanted radioactive seeds within the prostate under ultrasound imaging. The results of the training using 950 training ultrasound images and 90 validation ultrasound images. The commonly used metrics for successful training were used to evaluate the efficacy and accuracy of the trained deep neural network and resulted in an loss_bbox (train) = 0.00, loss_coverage (train) = 1.89e-8, loss_bbox (validation) = 11.84, loss_coverage (validation) = 9.70, mAP (validation) = 66.87%, precision (validation) = 81.07%, and a recall (validation) = 82.29%, where train and validation refers to the training image set and validation refers to the validation training set. On the hardware platform used, the training expended 12.8 seconds per epoch. The network was trained for over 10,000 epochs. In addition, the seed locations as determined by the Deep Neural Network were compared to the seed locations as determined by a commercial software based on a one to three months after implant CT. The Deep Learning approach was within \strikeout off\uuline off\uwave off2.29\uuline default\uwave default mm of the seed locations determined by the commercial software. The Deep Learning approach to the determination of radioactive seed locations is robust, accurate, and fast and well within spatial agreement with the gold standard of CT determined seed coordinates.

Keywords: prostate, deep neural network, seed implant, ultrasound

Procedia PDF Downloads 165

Authors: Jiangnan Yang, Shuangli Liu, Jinjun Wu, Xinyu Su, Nan Hai

Abstract:

These recent years have witnessed that convolutional neural network (CNN)-based methods for detecting infrared dim small targets have achieved outstanding performance. However, these methods typically employ standard convolutions, neglecting to fully consider the spatial characteristics of the pixel distribution of infrared dim small targets. Therefore, based on interleaved group convolutions, we propose a Pinwheel-shaped convolution (PConv) to replace the first two layers of standard convolutions in the backbone network. Compared to standard convolutions, PConv better conforms to the pixel spatial distribution of dim small targets and introduces only a minimal increase in parameters while effectively enhancing the extraction of features from dim small targets. Additionally, the latest loss function, combining intersection over union (IoU) loss and distance loss, fails to adequately account for the sensitivity differences in scale and position for targets of different sizes, thereby underutilizing the detection performance of neural networks for weak, small targets. To address this, we introduce dynamic completely IoU (DCIoU) loss, dynamically adjusting the influence coefficients of IoU and distance based on the size of the target bounding box, enhancing the neural network's ability to converge on targets of varying scales, thereby significantly boosting detection accuracy. Finally, we have curated the largest and most challenging dataset of real-shot single-frame infrared dim small target detection to date: SIRST-UAVB. Integrating PConv and DCIoU into the state-of-the-art small target detection algorithm, we conduct tests on both public datasets and SIRST-UAVB, achieving significant performance improvements, thus validating the effectiveness and generalizability of our approach.

Keywords: infrared dim small target detection, deep learning, interleaved group convolutions, feature extraction, loss function, dataset

Procedia PDF Downloads 5

Authors: K. R. Roopesh Bharatwaj, Avinash Maharana, Favour Tobi Aborisade, Roger Young

Abstract:

Presently, there are few cases of complete automation of drones and its allied intelligence capabilities. In essence, the potential of the drone has not yet been fully utilized. This paper presents feasible methods to build an intelligent drone with smart capabilities such as self-driving, and obstacle avoidance. It does this through advanced Reinforcement Learning Techniques and performs object detection using latest advanced algorithms, which are capable of processing light weight models with fast training in real time instances. For the scope of this paper, after researching on the various algorithms and comparing them, we finally implemented the Deep-Q-Networks (DQN) algorithm in the AirSim Simulator. In future works, we plan to implement further advanced self-driving and object detection algorithms, we also plan to implement voice-based speech recognition for the entire drone operation which would provide an option of speech communication between users (People) and the drone in the time of unavoidable circumstances. Thus, making drones an interactive intelligent Robotic Voice Enabled Service Assistant. This proposed drone has a wide scope of usability and is applicable in scenarios such as Disaster management, Air Transport of essentials, Agriculture, Manufacturing, Monitoring people movements in public area, and Defense. Also discussed, is the entire drone communication based on the satellite broadband Internet technology for faster computation and seamless communication service for uninterrupted network during disasters and remote location operations. This paper will explain the feasible algorithms required to go about achieving this goal and is more of a reference paper for future researchers going down this path.

Keywords: convolution neural network, natural language processing, obstacle avoidance, satellite broadband technology, self-driving

Procedia PDF Downloads 213

Authors: Yaolin Tian, Shanxiong Chen, Fujia Zhao, Xiaoyu Lin, Hailing Xiong

Abstract:

Ancient books are significant culture inheritors and their background textures convey the potential history information. However, multi-style texture recovery of ancient books has received little attention. Restricted by insufficient ancient textures and complex handling process, the generation of ancient textures confronts with new challenges. For instance, training without sufficient data usually brings about overfitting or mode collapse, so some of the outputs are prone to be fake. Recently, image generation and style transfer based on deep learning are widely applied in computer vision. Breakthroughs within the field make it possible to conduct research upon multi-style texture recovery of ancient books. Under the circumstances, we proposed a network of layout analysis and image fusion system. Firstly, we trained models by using Deep Convolution Generative against Networks (DCGAN) to synthesize multi-style ancient textures; then, we analyzed layouts based on the Position Rearrangement (PR) algorithm that we proposed to adjust the layout structure of foreground content; at last, we realized our goal by fusing rearranged foreground texts and generated background. In experiments, diversified samples such as ancient Yi, Jurchen, Seal were selected as our training sets. Then, the performances of different fine-turning models were gradually improved by adjusting DCGAN model in parameters as well as structures. In order to evaluate the results scientifically, cross entropy loss function and Fréchet Inception Distance (FID) are selected to be our assessment criteria. Eventually, we got model M8 with lowest FID score. Compared with DCGAN model proposed by Radford at el., the FID score of M8 improved by 19.26%, enhancing the quality of the synthetic images profoundly.

Keywords: deep learning, image fusion, image generation, layout analysis

Procedia PDF Downloads 111

Authors: Alexander A. Karabutov, Natalia B. Podymova, Elena B. Cherepetskaya

Abstract:

The theoretical analysis is carried out to get the relation between the ultrasonic wave velocity and the value of residual stresses. The laser-ultrasonic method is developed to evaluate the residual stresses and subsurface defects in metals. The method is based on the laser thermooptical excitation of longitudinal ultrasonic wave sand their detection by a broadband piezoelectric detector. A laser pulse with the time duration of 8 ns of the full width at half of maximum and with the energy of 300 µJ is absorbed in a thin layer of the special generator that is inclined relative to the object under study. The non-uniform heating of the generator causes the formation of a broadband powerful pulse of longitudinal ultrasonic waves. It is shown that the temporal profile of this pulse is the convolution of the temporal envelope of the laser pulse and the profile of the in-depth distribution of the heat sources. The ultrasonic waves reach the surface of the object through the prism that serves as an acoustic duct. At the interface ‚laser-ultrasonic transducer-object‘ the conversion of the most part of the longitudinal wave energy takes place into the shear, subsurface longitudinal and Rayleigh waves. They spread within the subsurface layer of the studied object and are detected by the piezoelectric detector. The electrical signal that corresponds to the detected acoustic signal is acquired by an analog-to-digital converter and when is mathematically processed and visualized with a personal computer. The distance between the generator and the piezodetector as well as the spread times of acoustic waves in the acoustic ducts are the characteristic parameters of the laser-ultrasonic transducer and are determined using the calibration samples. There lative precision of the measurement of the velocity of longitudinal ultrasonic waves is 0.05% that corresponds to approximately ±3 m/s for the steels of conventional quality. This precision allows one to determine the mechanical stress in the steel samples with the minimal detection threshold of approximately 22.7 MPa. The results are presented for the measured dependencies of the velocity of longitudinal ultrasonic waves in the samples on the values of the applied compression stress in the range of 20-100 MPa.

Keywords: laser-ultrasonic method, longitudinal ultrasonic waves, metals, residual stresses

Procedia PDF Downloads 293

Authors: Carolina Zambrana, Grover Zurita

Abstract:

The food loss production that occurs in deficient agricultural production is one of the major problems worldwide. This puts the population's food security and the efficiency of farming investments at risk. It is to be expected that this food security will be achieved with the own and efficient production of each country. It will have an impact on the well-being of its population and, thus, also on food sovereignty. The production losses in quantity and quality occur due to the lack of efficient detection of diseases at an early stage. It is very difficult to solve the agriculture efficiency using traditional methods since it takes a long time to be carried out due to detection imprecision of the main diseases, especially when the production areas are extensive. Therefore, the main objective of this research study is to perform a systematic literature review, of the latest five years, of Precision Agriculture (PA) to be able to understand the state of the art of the set of new technologies, procedures, and optimization processes with Artificial Intelligence (AI). This study will focus on Corns, Potatoes, and Rice diagnostic diseases. The extensive literature review will be performed on Elsevier, Scopus, and IEEE databases. In addition, this research will focus on advanced digital imaging processing and the development of software and hardware for PA. The convolution neural network will be handling special attention due to its outstanding diagnostic results. Moreover, the studied data will be incorporated with artificial intelligence algorithms for the automatic diagnosis of crop quality. Finally, precision agriculture with technology applied to the agricultural sector allows the land to be exploited efficiently. This system requires sensors, drones, data acquisition cards, and global positioning systems. This research seeks to merge different areas of science, control engineering, electronics, digital image processing, and artificial intelligence for the development, in the near future, of a low-cost image measurement system that allows the optimization of crops with AI.

Keywords: precision agriculture, convolutional neural network, deep learning, artificial intelligence

Procedia PDF Downloads 52

Authors: Heewon Jeong, Seongpyo Kim, Joon Ha Kim

Abstract:

Over the last few tens of years, the coastal regions of Korea have experienced red tide algal blooms, which are harmful and toxic to both humans and marine organisms due to their potential threat. It was accelerated owing to eutrophication by human activities, certain oceanic processes, and climate change. Previous studies have tried to monitoring and predicting the algae concentration of the ocean with the bio-optical algorithms applied to color images of the satellite. However, the accurate estimation of algal blooms remains problems to challenges because of the complexity of coastal waters. Therefore, this study suggests a new method to identify the concentration of red tide algal bloom from images of geostationary ocean color imager (GOCI) which are representing the water environment of the sea in Korea. The method employed GOCI images, which took the water leaving radiances centered at 443nm, 490nm and 660nm respectively, as well as observed weather data (i.e., humidity, temperature and atmospheric pressure) for the database to apply optical characteristics of algae and train deep learning algorithm. Convolution neural network (CNN) was used to extract the significant features from the images. And then artificial neural network (ANN) was used to estimate the concentration of algae from the extracted features. For training of the deep learning model, backpropagation learning strategy is developed. The established methods were tested and compared with the performances of GOCI data processing system (GDPS), which is based on standard image processing algorithms and optical algorithms. The model had better performance to estimate algae concentration than the GDPS which is impossible to estimate greater than 5mg/m³. Thus, deep learning model trained successfully to assess algae concentration in spite of the complexity of water environment. Furthermore, the results of this system and methodology can be used to improve the performances of remote sensing. Acknowledgement: This work was supported by the 'Climate Technology Development and Application' research project (#K07731) through a grant provided by GIST in 2017.

Keywords: deep learning, algae concentration, remote sensing, satellite

Procedia PDF Downloads 152

Authors: Juan Francisco Reinoso Gordo, Francisco Javier Ariza-López, José Rodríguez Avi, Domingo Barrera Rosillo

Abstract:

The evaluation of the quality of a data product is based on the comparison of the product with a reference of greater accuracy. In the case of MDE data products, quality assessment usually focuses on positional accuracy and few studies consider other terrain characteristics, such as slope and orientation. The proposal that is made consists of evaluating the similarity of two DEMs (a product and a reference), through the joint analysis of the distribution functions of the variables of interest, for example, elevations, slopes and orientations. This is a multivariable approach that focuses on distribution functions, not on single parameters such as mean values or dispersions (e.g. root mean squared error or variance). This is considered to be a more holistic approach. The use of the Kolmogorov-Smirnov test is proposed due to its non-parametric nature, since the distributions of the variables of interest cannot always be adequately modeled by parametric models (e.g. the Normal distribution model). In addition, its application to the multivariate case is carried out jointly by means of a single test on the convolution of the distribution functions of the variables considered, which avoids the use of corrections such as Bonferroni when several statistics hypothesis tests are carried out together. In this work, two DEM products have been considered, DEM02 with a resolution of 2x2 meters and DEM05 with a resolution of 5x5 meters, both generated by the National Geographic Institute of Spain. DEM02 is considered as the reference and DEM05 as the product to be evaluated. In addition, the slope and aspect derived models have been calculated by GIS operations on the two DEM datasets. Through sample simulation processes, the adequate behavior of the Kolmogorov-Smirnov statistical test has been verified when the null hypothesis is true, which allows calibrating the value of the statistic for the desired significance value (e.g. 5%). Once the process has been calibrated, the same process can be applied to compare the similarity of different DEM data sets (e.g. the DEM05 versus the DEM02). In summary, an innovative alternative for the comparison of DEM data sets based on a multinomial non-parametric perspective has been proposed by means of a single Kolmogorov-Smirnov test. This new approach could be extended to other DEM features of interest (e.g. curvature, etc.) and to more than three variables

Keywords: data quality, DEM, kolmogorov-smirnov test, multivariate DEM comparison

Procedia PDF Downloads 88

Authors: Anoop T. R., Otman Basir, Robert F. Hess, Eileen E. Birch, Brooke A. Koritala, Reed M. Jost, Becky Luu, David Stager, Ben Thompson

Abstract:

Strabismus refers to a misalignment of the eyes. Early detection and treatment of strabismus in childhood can prevent the development of permanent vision loss due to abnormal development of visual brain areas. We developed a two-stage method for strabismus detection and classification based on photographs of the face. The first stage detects the presence or absence of strabismus, and the second stage classifies the type of strabismus. The first stage comprises face detection using Haar cascade, facial landmark estimation, face alignment, aligned face landmark detection, segmentation of the eye region, and detection of strabismus using VGG 16 convolution neural networks. Face alignment transforms the face to a canonical pose to ensure consistency in subsequent analysis. Using facial landmarks, the eye region is segmented from the aligned face and fed into a VGG 16 CNN model, which has been trained to classify strabismus. The CNN determines whether strabismus is present and classifies the type of strabismus (exotropia, esotropia, and vertical deviation). If stage 1 detects strabismus, the eye region image is fed into stage 2, which starts with the estimation of pupil center coordinates using mask R-CNN deep neural networks. Then, the distance between the pupil coordinates and eye landmarks is calculated along with the angle that the pupil coordinates make with the horizontal and vertical axis. The distance and angle information is used to characterize the degree and direction of the strabismic eye misalignment. This model was tested on 100 clinically labeled images of children with (n = 50) and without (n = 50) strabismus. The True Positive Rate (TPR) and False Positive Rate (FPR) of the first stage were 94% and 6% respectively. The classification stage has produced a TPR of 94.73%, 94.44%, and 100% for esotropia, exotropia, and vertical deviations, respectively. This method also had an FPR of 5.26%, 5.55%, and 0% for esotropia, exotropia, and vertical deviation, respectively. The addition of one more feature related to the location of corneal light reflections may reduce the FPR, which was primarily due to children with pseudo-strabismus (the appearance of strabismus due to a wide nasal bridge or skin folds on the nasal side of the eyes).

Keywords: strabismus, deep neural networks, face detection, facial landmarks, face alignment, segmentation, VGG 16, mask R-CNN, pupil coordinates, angle deviation, horizontal and vertical deviation

Procedia PDF Downloads 55

Authors: Jalil ur Rehman, Ramesh C, Tailor, Isa Khan, Jahanzeeb Ashraf, Muhammad Afzal, Geofferry S. Ibbott

Abstract:

The purpose of this analysis is to estimate secondary cancer risks after VMAT compared to other modalities of head and neck radiotherapy (IMRT, 3DCRT). Computer tomography (CT) scans of Radiological Physics Center (RPC) head and neck phantom were acquired with CT scanner and exported via DICOM to the treatment planning system (TPS). Treatment planning was done using four arc (182-178 and 180-184, clockwise and anticlockwise) for volumetric modulated arc therapy (VMAT) , Nine fields (200, 240, 280, 320,0,40,80,120 and 160), which has been commonly used at MD Anderson Cancer Center Houston for intensity modulated radiation therapy (IMRT) and four fields for three dimensional radiation therapy (3DCRT) were used. True beam linear accelerator of 6MV photon energy was used for dose delivery, and dose calculation was done with CC convolution algorithm with prescription dose of 6.6 Gy. Primary Target Volume (PTV) coverage, mean and maximal doses, DVHs and volumes receiving more than 2 Gy and 3.8 Gy of OARs were calculated and compared. Absolute point dose and planar dose were measured with thermoluminescent dosimeters (TLDs) and GafChromic EBT2 film, respectively. Quality Assurance of VMAT and IMRT were performed by using ArcCHECK method with gamma index criteria of 3%/3mm dose difference to distance to agreement (DD/DTA). PTV coverage was found 90.80 %, 95.80 % and 95.82 % for 3DCRT, IMRT and VMAT respectively. VMAT delivered the lowest maximal doses to esophagus (2.3 Gy), brain (4.0 Gy) and thyroid (2.3 Gy) compared to all other studied techniques. In comparison, maximal doses for 3DCRT were found higher than VMAT for all studied OARs. Whereas, IMRT delivered maximal higher doses 26%, 5% and 26% for esophagus, normal brain and thyroid, respectively, compared to VMAT. It was noted that esophagus volume receiving more than 2 Gy was 3.6 % for VMAT, 23.6 % for IMRT and up to 100 % for 3DCRT. Good agreement was observed between measured doses and those calculated with TPS. The averages relative standard errors (RSE) of three deliveries within eight TLD capsule locations were, 0.9%, 0.8% and 0.6% for 3DCRT, IMRT and VMAT, respectively. The gamma analysis for all plans met the ±5%/3 mm criteria (over 90% passed) and results of QA were greater than 98%. The calculations for maximal doses and volumes of OARs suggest that the estimated risk of secondary cancer induction after VMAT is considerably lower than IMRT and 3DCRT.

Keywords: RPC, 3DCRT, IMRT, VMAT, EBT2 film, TLD

Procedia PDF Downloads 482

Authors: Hamed Kalhori, Lin Ye

Abstract:

In this study, an inverse method was developed to reconstruct the magnitude and duration of impact forces exerted to a rectangular carbon fibre-epoxy composite honeycomb sandwich panel. The dynamic signals captured by Piezoelectric (PZT) sensors installed on the panel remotely from the impact locations were utilized to reconstruct the impact force generated by an instrumented hammer through an extended deconvolution approach. Two discretized forms of convolution integral are considered; the traditional one with an explicit transfer function and the modified one without an explicit transfer function. Deconvolution, usually applied to reconstruct the time history (e.g. magnitude) of a stochastic force at a defined location, is extended to identify both the location and magnitude of the impact force among a number of potential impact locations. It is assumed that a number of impact forces are simultaneously exerted to all potential locations, but the magnitude of all forces except one is zero, implicating that the impact occurs only at one location. The extended deconvolution is then applied to determine the magnitude as well as location (among the potential ones), incorporating the linear superposition of responses resulted from impact at each potential location. The problem can be categorized into under-determined (the number of sensors is less than that of impact locations), even-determined (the number of sensors equals that of impact locations), or over-determined (the number of sensors is greater than that of impact locations) cases. For an under-determined case, it comprises three potential impact locations and one PZT sensor for the rectangular carbon fibre-epoxy composite honeycomb sandwich panel. Assessments are conducted to evaluate the factors affecting the precision of the reconstructed force. Truncated Singular Value Decomposition (TSVD) and the Tikhonov regularization are independently chosen to regularize the problem to find the most suitable method for this system. The selection of optimal value of the regularization parameter is investigated through L-curve and Generalized Cross Validation (GCV) methods. In addition, the effect of different width of signal windows on the reconstructed force is examined. It is observed that the impact force generated by the instrumented impact hammer is sensitive to the impact locations of the structure, having a shape from a simple half-sine to a complicated one. The accuracy of the reconstructed impact force is evaluated using the correlation co-efficient between the reconstructed force and the actual one. Based on this criterion, it is concluded that the forces reconstructed by using the extended deconvolution without an explicit transfer function together with Tikhonov regularization match well with the actual forces in terms of magnitude and duration.

Keywords: honeycomb composite panel, deconvolution, impact localization, force reconstruction

Procedia PDF Downloads 508

Authors: Mohammadreza Hedyehzadeh, Mahdi Yousefi

Abstract:

Colon cancer is one of the most common cancer, which predicted to increase its prevalence due to the bad eating habits of peoples. Nowadays, due to the busyness of people, the use of fast foods is increasing, and therefore, diagnosis of this disease and its treatment are of particular importance. To determine the best treatment approach for each specific colon cancer patients, the oncologist should be known the stage of the tumor. The most common method to determine the tumor stage is TNM staging system. In this system, M indicates the presence of metastasis, N indicates the extent of spread to the lymph nodes, and T indicates the size of the tumor. It is clear that in order to determine all three of these parameters, an imaging method must be used, and the gold standard imaging protocols for this purpose are CT and PET/CT. In CT imaging, due to the use of X-rays, the risk of cancer and the absorbed dose of the patient is high, while in the PET/CT method, there is a lack of access to the device due to its high cost. Therefore, in this study, we aimed to estimate the tumor size and the extent of its spread to the lymph nodes using MR images. More than 1300 MR images collected from the TCIA portal, and in the first step (pre-processing), histogram equalization to improve image qualities and resizing to get the same image size was done. Two expert radiologists, which work more than 21 years on colon cancer cases, segmented the images and extracted the tumor region from the images. The next step is feature extraction from segmented images and then classify the data into three classes: T0N0، T3N1 و T3N2. In this article, the VGG-16 convolutional neural network has been used to perform both of the above-mentioned tasks, i.e., feature extraction and classification. This network has 13 convolution layers for feature extraction and three fully connected layers with the softmax activation function for classification. In order to validate the proposed method, the 10-fold cross validation method used in such a way that the data was randomly divided into three parts: training (70% of data), validation (10% of data) and the rest for testing. It is repeated 10 times, each time, the accuracy, sensitivity and specificity of the model are calculated and the average of ten repetitions is reported as the result. The accuracy, specificity and sensitivity of the proposed method for testing dataset was 89/09%, 95/8% and 96/4%. Compared to previous studies, using a safe imaging technique (MRI) and non-use of predefined hand-crafted imaging features to determine the stage of colon cancer patients are some of the study advantages.

Keywords: colon cancer, VGG-16, magnetic resonance imaging, tumor size, lymph node metastasis

Procedia PDF Downloads 33

Authors: Ole-Christian Galbo Engstrøm, Erik Schou Dreier, Birthe Møller Jespersen, Kim Steenstrup Pedersen

Abstract:

Grain quality analysis is a multi-parameterized problem that includes a variety of qualitative and quantitative parameters such as grain type classification, damage type classification, and nutrient regression. Currently, these parameters require human inspection, a multitude of instruments employing a variety of sensor technologies, and predictive model types or destructive and slow chemical analysis. This paper investigates the feasibility of applying near-infrared hyperspectral imaging (NIR-HSI) to grain quality analysis. For this study two datasets of NIR hyperspectral images in the wavelength range of 900 nm - 1700 nm have been used. Both datasets contain images of sparsely and densely packed grain kernels. The first dataset contains ~87,000 image crops of bulk wheat samples from 63 harvests where protein value has been determined by the FOSS Infratec NOVA which is the golden industry standard for protein content estimation in bulk samples of cereal grain. The second dataset consists of ~28,000 image crops of bulk grain kernels from seven different wheat varieties and a single rye variety. In the first dataset, protein regression analysis is the problem to solve while variety classification analysis is the problem to solve in the second dataset. Deep convolutional neural networks (CNNs) have the potential to utilize spatio-spectral correlations within a hyperspectral image to simultaneously estimate the qualitative and quantitative parameters. CNNs can autonomously derive meaningful representations of the input data reducing the need for advanced preprocessing techniques required for classical chemometric model types such as artificial neural networks (ANNs) and partial least-squares regression (PLS-R). A comparison between different CNN architectures utilizing 2D and 3D convolution is conducted. These results are compared to the performance of ANNs and PLS-R. Additionally, a variety of preprocessing techniques from image analysis and chemometrics are tested. These include centering, scaling, standard normal variate (SNV), Savitzky-Golay (SG) filtering, and detrending. The results indicate that the combination of NIR-HSI and CNNs has the potential to be the foundation for an automatic system unifying qualitative and quantitative grain quality analysis within a single sensor technology and predictive model type.

Keywords: deep learning, grain analysis, hyperspectral imaging, preprocessing techniques

Procedia PDF Downloads 68

Authors: Tayo P. Ogundunmade, Adedayo A. Adepoju

Abstract:

Terrorist attacks in liberal democracies bring about a few pessimistic results, for example, sabotaged public support in the governments they target, disturbing the peace of a protected environment underwritten by the state, and a limitation of individuals from adding to the advancement of the country, among others. Hence, seeking for techniques to understand the different factors involved in terrorism and how to deal with those factors in order to completely stop or reduce terrorist activities is the topmost priority of the government in every country. This research aim is to develop an efficient deep learning-based predictive model for the prediction of future terrorist activities in Nigeria, addressing low-quality prediction accuracy problems associated with the existing solution methods. The proposed predictive AI-based model as a counterterrorism tool will be useful by governments and law enforcement agencies to protect the lives of individuals in society and to improve the quality of life in general. A Heterogeneous Bayesian Neural Network (HETBNN) model was derived with Gaussian error normal distribution. Three primary transfer functions (HOTTFs), as well as two derived transfer functions (HETTFs) arising from the convolution of the HOTTFs, are namely; Symmetric Saturated Linear transfer function (SATLINS ), Hyperbolic Tangent transfer function (TANH), Hyperbolic Tangent sigmoid transfer function (TANSIG), Symmetric Saturated Linear and Hyperbolic Tangent transfer function (SATLINS-TANH) and Symmetric Saturated Linear and Hyperbolic Tangent Sigmoid transfer function (SATLINS-TANSIG). Data on the Terrorist activities in Nigeria gathered through questionnaires for the purpose of this study were used. Mean Square Error (MSE), Mean Absolute Error (MAE) and Test Error are the forecast prediction criteria. The results showed that the HETFs performed better in terms of prediction and factors associated with terrorist activities in Nigeria were determined. The proposed predictive deep learning-based model will be useful to governments and law enforcement agencies as an effective counterterrorism mechanism to understand the parameters of terrorism and to design strategies to deal with terrorism before an incident actually happens and potentially causes the loss of precious lives. The proposed predictive AI-based model will reduce the chances of terrorist activities and is particularly helpful for security agencies to predict future terrorist activities.

Keywords: activation functions, Bayesian neural network, mean square error, test error, terrorism

Procedia PDF Downloads 135

Authors: Mir Jalil Razavi, Tianming Liu, Xianqiao Wang

Abstract:

Cortical folding, characterized by convex gyri and concave sulci, has an intrinsic relationship to the brain’s functional organization. Understanding the mechanism of the brain’s convoluted patterns can provide useful clues into normal and pathological brain function. During the development, the cerebral cortex experiences a noticeable expansion in volume and surface area accompanied by tremendous tissue folding which may be attributed to many possible factors. Despite decades of endeavors, the fundamental mechanism and key regulators of this crucial process remain incompletely understood. Therefore, to taking even a small role in unraveling of brain folding mystery, we present a mechanical model to find mechanism of 3-hinges formation in a growing brain that it has not been addressed before. A 3-hinge is defined as a gyral region where three gyral crests (hinge-lines) join. The reasons that how and why brain prefers to develop 3-hinges have not been answered very well. Therefore, we offer a theoretical and computational explanation to mechanism of 3-hinges formation in a growing brain and validate it by experimental observations. In theoretical approach, the dynamic behavior of brain tissue is examined and described with the aid of a large strain and nonlinear constitutive model. Derived constitute model is used in the computational model to define material behavior. Since the theoretical approach cannot predict the evolution of cortical complex convolution after instability, non-linear finite element models are employed to study the 3-hinges formation and secondary morphological folds of the developing brain. Three-dimensional (3D) finite element analyses on a multi-layer soft tissue model which mimics a small piece of the brain are performed to investigate the fundamental mechanism of consistent hinge formation in the cortical folding. Results show that after certain amount growth of cortex, mechanical model starts to be unstable and then by formation of creases enters to a new configuration with lower strain energy. By further growth of the model, formed shallow creases start to form convoluted patterns and then develop 3-hinge patterns. Simulation results related to 3-hinges in models show good agreement with experimental observations from macaque, chimpanzee and human brain images. These results have great potential to reveal fundamental principles of brain architecture and to produce a unified theoretical framework that convincingly explains the intrinsic relationship between cortical folding and 3-hinges formation. This achieved fundamental understanding of the intrinsic relationship between cortical folding and 3-hinges formation would potentially shed new insights into the diagnosis of many brain disorders such as schizophrenia, autism, lissencephaly and polymicrogyria.

Keywords: brain, cortical folding, finite element, three hinge

Procedia PDF Downloads 205

Authors: Brittany Richardson, Ying Wang

Abstract:

For physical therapy, exercise prescription, athlete training, and regular fitness training, it is crucial to perform health assessments or fitness assessments periodically. An accurate assessment is propitious for tracking recovery progress, preventing potential injury and making long-range training plans. Assessments include necessary measurements, height, weight, blood pressure, heart rate, body fat, etc. and advanced evaluation, muscle group strength, stability-mobility, and movement evaluation, etc. In the current standard assessment procedures, the accuracy of assessments, especially advanced evaluations, largely depends on the experience of physicians, coaches, and personal trainers. And it is challenging to track clients’ progress in the current assessment. Unlike the tradition assessment, in this paper, we present a deep learning based face recognition algorithm for accurate, comprehensive and trackable assessment. Based on the result from our assessment, physicians, coaches, and personal trainers are able to adjust the training targets and methods. The system categorizes the difficulty levels of the current activity for the client or user, furthermore make more comprehensive assessments based on tracking muscle group over time using a designed landmark detection method. The system also includes the function of grading and correcting the form of the clients during exercise. Experienced coaches and personal trainer can tell the clients' limit based on their facial expression and muscle group movements, even during the first several sessions. Similar to this, using a convolution neural network, the system is trained with people’s facial expression to differentiate challenge levels for clients. It uses landmark detection for subtle changes in muscle groups movements. It measures the proximal mobility of the hips and thoracic spine, the proximal stability of the scapulothoracic region and distal mobility of the glenohumeral joint, as well as distal mobility, and its effect on the kinetic chain. This system integrates data from other fitness assistant devices, including but not limited to Apple Watch, Fitbit, etc. for a improved training and testing performance. The system itself doesn’t require history data for an individual client, but the history data of a client can be used to create a more effective exercise plan. In order to validate the performance of the proposed work, an experimental design is presented. The results show that the proposed work contributes towards improving the quality of exercise plan, execution, progress tracking, and performance.

Keywords: exercise prescription, facial recognition, landmark detection, fitness assessments

Procedia PDF Downloads 101

Authors: Manaranjan Pradhan, Shailaja Grover, U. Dinesh Kumar

Abstract:

Farming community in India, as well as other parts of the world, is one of the highly stressed communities due to reasons such as increasing input costs (cost of seeds, fertilizers, pesticide), droughts, reduced revenue leading to farmer suicides. Lack of integrated farm advisory system in India adds to the farmers problems. Farmers need right information during the early stages of crop’s lifecycle to prevent damage and loss in revenue. In this paper, we use deep learning techniques to develop an early warning system for detection of crop diseases using images taken by farmers using their smart phone. The research work leads to building a smart assistant using analytics and big data which could help the farmers with early diagnosis of the crop diseases and corrective actions. The classical approach for crop disease management has been to identify diseases at crop level. Recently, ImageNet Classification using the convolutional neural network (CNN) has been successfully used to identify diseases at individual plant level. Our model uses convolution filters, max pooling, dense layers and dropouts (to avoid overfitting). The models are built for binary classification (healthy or not healthy) and multi class classification (identifying which disease). Transfer learning is used to modify the weights of parameters learnt through ImageNet dataset and apply them on crop diseases, which reduces number of epochs to learn. One shot learning is used to learn from very few images, while data augmentation techniques are used to improve accuracy with images taken from farms by using techniques such as rotation, zoom, shift and blurred images. Models built using combination of these techniques are more robust for deploying in the real world. Our model is validated using tomato crop. In India, tomato is affected by 10 different diseases. Our model achieves an accuracy of more than 95% in correctly classifying the diseases. The main contribution of our research is to create a personal assistant for farmers for managing plant disease, although the model was validated using tomato crop, it can be easily extended to other crops. The advancement of technology in computing and availability of large data has made possible the success of deep learning applications in computer vision, natural language processing, image recognition, etc. With these robust models and huge smartphone penetration, feasibility of implementation of these models is high resulting in timely advise to the farmers and thus increasing the farmers' income and reducing the input costs.

Keywords: analytics in agriculture, CNN, crop disease detection, data augmentation, image recognition, one shot learning, transfer learning

Procedia PDF Downloads 95

Authors: Alexander Norbach

Abstract:

The script in this paper describes the use of advanced simulation environment using electronic systems (Microcontroller, Operational Amplifiers, and FPGA). The simulation may be used for all dynamic systems with the diffusion and the ionisation behaviour also. By additionally required observer structure, the system works with parallel real-time simulation based on diffusion model and the state-space representation for other dynamics. The proposed deposited model may be used for electrodynamic effects, including ionising effects and eddy current distribution also. With the script and proposed method, it is possible to calculate the spatial distribution of the electromagnetic fields in real-time. For further purpose, the spatial temperature distribution may be used also. With upon system, the uncertainties, unknown initial states and disturbances may be determined. This provides the estimation of the more precise system states for the required system, and additionally, the estimation of the ionising disturbances that occur due to radiation effects. The results have shown that a system can be also developed and adopted specifically for space systems with the real-time calculation of the radiation effects only. Electronic systems can take damage caused by impacts with charged particle flux in space or radiation environment. In order to be able to react to these processes, it must be calculated within a shorter time that ionising radiation and dose is present. All available sensors shall be used to observe the spatial distributions. By measured value of size and known location of the sensors, the entire distribution can be calculated retroactively or more accurately. With the formation, the type of ionisation and the direct effect to the systems and thus possible prevent processes can be activated up to the shutdown. The results show possibilities to perform more qualitative and faster simulations independent of kind of systems space-systems and radiation environment also. The paper gives additionally an overview of the diffusion effects and their mechanisms. For the modelling and derivation of equations, the extended current equation is used. The size K represents the proposed charge density drifting vector. The extended diffusion equation was derived and shows the quantising character and has similar law like the Klein-Gordon equation. These kinds of PDE's (Partial Differential Equations) are analytically solvable by giving initial distribution conditions (Cauchy problem) and boundary conditions (Dirichlet boundary condition). For a simpler structure, a transfer function for B- and E- fields was analytically calculated. With known discretised responses g₁(k·Ts) and g₂(k·Ts), the electric current or voltage may be calculated using a convolution; g₁ is the direct function and g₂ is a recursive function. The analytical results are good enough for calculation of fields with diffusion effects. Within the scope of this work, a proposed model of the consideration of the electromagnetic diffusion effects of arbitrary current 'waveforms' has been developed. The advantage of the proposed calculation of diffusion is the real-time capability, which is not really possible with the FEM programs available today. It makes sense in the further course of research to use these methods and to investigate them thoroughly.

Keywords: advanced observer, electrodynamics, systems, diffusion, partial differential equations, solver

Procedia PDF Downloads 101

Authors: Sree Bash Chandra Debnath, Didier Tonneau, Carole Fauquet, Agnes Tallet, Julien Darreon

Abstract:

Purpose: Inorganic scintillating dosimetry is the most recent promising technique to solve several dosimetric issues and provide quality assurance in radiation therapy. Despite several advantages, the major issue of using scintillating detectors is the Cerenkov effect, typically induced in the visible emission range. In this context, the purpose of this research work is to evaluate the performance of a novel infrared inorganic scintillator detector (IR-ISD) in the radiation therapy treatment to ensure Cerenkov free signal and the best matches between the delivered and prescribed doses during treatment. Methods: A simple and small-scale infrared inorganic scintillating detector of 100 µm diameter with a sensitive scintillating volume of 2x10-6 mm3 was developed. A prototype of the dose verification system has been introduced based on PTIR1470/F (provided by Phosphor Technology®) material used in the proposed novel IR-ISD. The detector was tested on an Elekta LINAC system tuned at 6 MV/15MV and a brachytherapy source (Ir-192) used in the patient treatment protocol. The associated dose rate was measured in count rate (photons/s) using a highly sensitive photon counter (sensitivity ~20ph/s). Overall measurements were performed in IBATM water tank phantoms by following international Technical Reports series recommendations (TRS 381) for radiotherapy and TG43U1 recommendations for brachytherapy. The performance of the detector was tested through several dosimetric parameters such as PDD, beam profiling, Cerenkov measurement, dose linearity, dose rate linearity repeatability, and scintillator stability. Finally, a comparative study is also shown using a reference microdiamond dosimeter, Monte-Carlo (MC) simulation, and data from recent literature. Results: This study is highlighting the complete removal of the Cerenkov effect especially for small field radiation beam characterization. The detector provides an entire linear response with the dose in the 4cGy to 800 cGy range, independently of the field size selected from 5 x 5 cm² down to 0.5 x 0.5 cm². A perfect repeatability (0.2 % variation from average) with day-to-day reproducibility (0.3% variation) was observed. Measurements demonstrated that ISD has superlinear behavior with dose rate (R2=1) varying from 50 cGy/s to 1000 cGy/s. PDD profiles obtained in water present identical behavior with a build-up maximum depth dose at 15 mm for different small fields irradiation. A low dimension of 0.5 x 0.5 cm² field profiles have been characterized, and the field cross profile presents a Gaussian-like shape. The standard deviation (1σ) of the scintillating signal remains within 0.02% while having a very low convolution effect, thanks to lower sensitive volume. Finally, during brachytherapy, a comparison with MC simulations shows that considering energy dependency, measurement agrees within 0.8% till 0.2 cm source to detector distance. Conclusion: The proposed scintillating detector in this study shows no- Cerenkov radiation and efficient performance for several radiation therapy measurement parameters. Therefore, it is anticipated that the IR-ISD system can be promoted to validate with direct clinical investigations, such as appropriate dose verification and quality control in the Treatment Planning System (TPS).

Keywords: IR-Scintillating detector, dose measurement, micro-scintillators, Cerenkov effect

Procedia PDF Downloads 153

Authors: Thilini M. Yatanwala

Abstract:

CCTV (Closed-Circuit-Television) Surveillance System is being used in public places over decades and a large variety of data is being produced every moment. However, most of the CCTV data is stored in isolation without having integrity. As a result, identification of the behavior of suspicious people along with their location has become strenuous. This research was conducted to acquire more accurate and reliable timely information from the CCTV video records. The implemented system can identify human objects in public places based on outfit colors. Inter-process communication technologies were used to implement the CCTV camera network to track people in the premises. The research was conducted in three stages and in the first stage human objects were filtered from other movable objects available in public places. In the second stage people were uniquely identified based on their outfit colors and in the third stage an individual was continuously tracked in the CCTV network. A face detection algorithm was implemented using cascade classifier based on the training model to detect human objects. HAAR feature based two-dimensional convolution operator was introduced to identify features of the human face such as region of eyes, region of nose and bridge of the nose based on darkness and lightness of facial area. In the second stage outfit colors of human objects were analyzed by dividing the area into upper left, upper right, lower left, lower right of the body. Mean color, mod color and standard deviation of each area were extracted as crucial factors to uniquely identify human object using histogram based approach. Color based measurements were written in to XML files and separate directories were maintained to store XML files related to each camera according to time stamp. As the third stage of the approach, inter-process communication techniques were used to implement an acknowledgement based CCTV camera network to continuously track individuals in a network of cameras. Real time analysis of XML files generated in each camera can determine the path of individual to monitor full activity sequence. Higher efficiency was achieved by sending and receiving acknowledgments only among adjacent cameras. Suspicious incidents such as a person staying in a sensitive area for a longer period or a person disappeared from the camera coverage can be detected in this approach. The system was tested for 150 people with the accuracy level of 82%. However, this approach was unable to produce expected results in the presence of group of people wearing similar type of outfits. This approach can be applied to any existing camera network without changing the physical arrangement of CCTV cameras. The study of human identification and suspicious incident detection using outfit color analysis can achieve higher level of accuracy and the project will be continued by integrating motion and gait feature analysis techniques to derive more information from CCTV videos.

Keywords: CCTV surveillance, human detection and identification, image processing, inter-process communication, security, suspicious detection

Procedia PDF Downloads 150

Authors: Monica Racha, Siva Chandrasekaran, Alex Stojcevski

Abstract:

The engineering programs focus on promoting students' personal and professional development by ensuring that students acquire technical and professional competencies during four-year studies. The traditional engineering laboratory provides an opportunity for students to "practice by doing," and laboratory facilities aid them in obtaining insight and understanding of their discipline. Due to rapid technological advancements and the current COVID-19 outbreak, the traditional labs were transforming into virtual learning environments. Aim: To better understand the limitations of the physical laboratory, this research study aims to use a Machine Learning (ML) algorithm that interfaces with the Augmented Reality HoloLens and predicts the image behavior to classify and detect the electronic components. The automated electronic components error classification and detection automatically detect and classify the position of all components on a breadboard by using the ML algorithm. This research will assist first-year undergraduate engineering students in conducting laboratory practices without any supervision. With the help of HoloLens, and ML algorithm, students will reduce component placement error on a breadboard and increase the efficiency of simple laboratory practices virtually. Method: The images of breadboards, resistors, capacitors, transistors, and other electrical components will be collected using HoloLens 2 and stored in a database. The collected image dataset will then be used for training a machine learning model. The raw images will be cleaned, processed, and labeled to facilitate further analysis of components error classification and detection. For instance, when students conduct laboratory experiments, the HoloLens captures images of students placing different components on a breadboard. The images are forwarded to the server for detection in the background. A hybrid Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs) algorithm will be used to train the dataset for object recognition and classification. The convolution layer extracts image features, which are then classified using Support Vector Machine (SVM). By adequately labeling the training data and classifying, the model will predict, categorize, and assess students in placing components correctly. As a result, the data acquired through HoloLens includes images of students assembling electronic components. It constantly checks to see if students appropriately position components in the breadboard and connect the components to function. When students misplace any components, the HoloLens predicts the error before the user places the components in the incorrect proportion and fosters students to correct their mistakes. This hybrid Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs) algorithm automating electronic component error classification and detection approach eliminates component connection problems and minimizes the risk of component damage. Conclusion: These augmented reality smart glasses powered by machine learning provide a wide range of benefits to supervisors, professionals, and students. It helps customize the learning experience, which is particularly beneficial in large classes with limited time. It determines the accuracy with which machine learning algorithms can forecast whether students are making the correct decisions and completing their laboratory tasks.

Keywords: augmented reality, machine learning, object recognition, virtual laboratories

Procedia PDF Downloads 109

Authors: Qianwei Cheng, A. K. M. Mahbubur Rahman, Anis Sarker, Abu Bakar Siddik Nayem, Ovi Paul, Amin Ahsan Ali, M. Ashraful Amin, Ryosuke Shibasaki, Moinul Zaber

Abstract:

Thomas Friedman, in his famous book, argued that the world in this 21st century is flat and will continue to be flatter. This is attributed to rapid globalization and the interdependence of humanity that engendered tremendous in-flow of human migration towards the urban spaces. In order to keep the urban environment sustainable, policy makers need to plan based on extensive analysis of the urban environment. With the advent of high definition satellite images, high resolution data, computational methods such as deep neural network analysis, and hardware capable of high-speed analysis; urban planning is seeing a paradigm shift. Legacy data on urban environments are now being complemented with high-volume, high-frequency data. However, the first step of understanding urban space lies in useful categorization of the space that is usable for data collection, analysis, and visualization. In this paper, we propose a pragmatic categorization method that is readily usable for machine analysis and show applicability of the methodology on a developing world setting. Categorization to plan sustainable urban spaces should encompass the buildings and their surroundings. However, the state-of-the-art is mostly dominated by classification of building structures, building types, etc. and largely represents the developed world. Hence, these methods and models are not sufficient for developing countries such as Bangladesh, where the surrounding environment is crucial for the categorization. Moreover, these categorizations propose small-scale classifications, which give limited information, have poor scalability and are slow to compute in real time. Our proposed method is divided into two steps-categorization and automation. We categorize the urban area in terms of informal and formal spaces and take the surrounding environment into account. 50 km × 50 km Google Earth image of Dhaka, Bangladesh was visually annotated and categorized by an expert and consequently a map was drawn. The categorization is based broadly on two dimensions-the state of urbanization and the architectural form of urban environment. Consequently, the urban space is divided into four categories: 1) highly informal area; 2) moderately informal area; 3) moderately formal area; and 4) highly formal area. In total, sixteen sub-categories were identified. For semantic segmentation and automatic categorization, Google’s DeeplabV3+ model was used. The model uses Atrous convolution operation to analyze different layers of texture and shape. This allows us to enlarge the field of view of the filters to incorporate larger context. Image encompassing 70% of the urban space was used to train the model, and the remaining 30% was used for testing and validation. The model is able to segment with 75% accuracy and 60% Mean Intersection over Union (mIoU). In this paper, we propose a pragmatic categorization method that is readily applicable for automatic use in both developing and developed world context. The method can be augmented for real-time socio-economic comparative analysis among cities. It can be an essential tool for the policy makers to plan future sustainable urban spaces.

Keywords: semantic segmentation, urban environment, deep learning, urban building, classification

Procedia PDF Downloads 151