Search results for: deep neural image models

Authors: Yang Yang, Dan Liu

Abstract:

Anomaly detection models for network flow based on machine learning have poor detection performance under extremely unbalanced training data conditions and also have slow detection speed and large resource consumption when deploying on network edge devices. Embedding multi-teacher knowledge distillation (mKD) in anomaly detection can transfer knowledge from multiple teacher models to a single model. Inspired by this, we proposed a state-of-the-art model, mKDNAD, to improve detection performance. mKDNAD mine and integrate the knowledge of one-dimensional sequence and two-dimensional image implicit in network flow to improve the detection accuracy of small sample classes. The multi-teacher knowledge distillation method guides the train of the student model, thus speeding up the model's detection speed and reducing the number of model parameters. Experiments in the CICIDS2017 dataset verify the improvements of our method in the detection speed and the detection accuracy in dealing with the small sample classes.

Keywords: network flow anomaly detection (NAD), multi-teacher knowledge distillation, machine learning, deep learning

Procedia PDF Downloads 101

Authors: Hooman Torabifard

Abstract:

In the last few years, with the progress of technology and computers and artificial intelligence entry into all kinds of scientific and industrial fields, the lifestyles of human life have changed and in general, the way of humans live on earth has many changes and development. Until now, some of the changes has occurred in the context of digital images and image processing and still continues. However, besides all the benefits, there have been disadvantages. One of these disadvantages is the multiplicity of images with high volume and data; the focus of this paper is on improving and developing a method for summarizing and enhancing the productivity of these images. The general method used for this purpose in this paper consists of a set of methods based on data obtained from image processing and using the PSO (Particle swarm optimization) algorithm. In the remainder of this paper, the method used is elaborated in detail.

Keywords: image summarization, particle swarm optimization, image threshold, image processing

Procedia PDF Downloads 113

Authors: M. Dhana Lakshmi, S. Sakthivel Murugan

Abstract:

As light passes from source to observer in the water medium, it is scattered by the suspended particulate matter. This scattering effect will plague the captured images with non-uniform illumination, blurring details, halo artefacts, weak edges, etc. To overcome this, pixel normalization with an Amended Unsharp Mask (AUM) filter is proposed to enhance the degraded image. To validate the robustness of the proposed technique irrespective of atmospheric light, the considered datasets are collected on dual locations. For those images, the maxima and minima pixel intensity value is computed and normalized; then the AUM filter is applied to strengthen the blurred edges. Finally, the enhanced image is obtained with good illumination and contrast. Thus, the proposed technique removes the effect of scattering called de-hazing and restores the perceptual information with enhanced edge detail. Both qualitative and quantitative analyses are done on considering the standard non-reference metric called underwater image sharpness measure (UISM), and underwater image quality measure (UIQM) is used to measure color, sharpness, and contrast for both of the location images. It is observed that the proposed technique has shown overwhelming performance compared to other deep-based enhancement networks and traditional techniques in an adaptive manner.

Keywords: underwater drone imagery, pixel normalization, thresholding, masking, unsharp mask filter

Procedia PDF Downloads 177

Authors: Vijay Pal Singh, Yogesh Chandra Kotiyal

Abstract:

Structures are a combination of various load carrying members which transfer the loads to the foundation from the superstructure safely. At the design stage, the loading of the structure is defined and appropriate material choices are made based upon their properties, mainly related to strength. The strength of materials kept on reducing with time because of many factors like environmental exposure and deformation caused by unpredictable external loads. Hence, to predict the strength of materials used in structures, various techniques are used. Among these techniques, Non-Destructive Techniques (NDT) are the one that can be used to predict the strength without damaging the structure. In the present study, the compressive strength of concrete has been predicted using Artificial Neural Network (ANN). The predicted strength was compared with the experimentally obtained actual compressive strength of concrete and equations were developed for different models. A good co-relation has been obtained between the predicted strength by these models and experimental values. Further, the co-relation has been developed using two NDT techniques for prediction of strength by regression analysis. It was found that the percentage error has been reduced between the predicted strength by using combined techniques in place of single techniques.

Keywords: rebound, ultra-sonic pulse, penetration, ANN, NDT, regression

Procedia PDF Downloads 404

Authors: Alper Taner, Yesim Benal Oztekin, Huseyin Duran

Abstract:

In this study, an Artificial Neural Network (ANN) was developed in order to classify barley varieties. For this purpose, physical properties of barley varieties were determined and ANN techniques were used. The physical properties of 8 barley varieties grown in Turkey, namely thousand kernel weight, geometric mean diameter, sphericity, kernel volume, surface area, bulk density, true density, porosity and colour parameters of grain, were determined and it was found that these properties were statistically significant with respect to varieties. As ANN model, three models, N-l, N-2 and N-3 were constructed. The performances of these models were compared. It was determined that the best-fit model was N-1. In the N-1 model, the structure of the model was designed to be 11 input layers, 2 hidden layers and 1 output layer. Thousand kernel weight, geometric mean diameter, sphericity, kernel volume, surface area, bulk density, true density, porosity and colour parameters of grain were used as input parameter; and varieties as output parameter. R2, Root Mean Square Error and Mean Error for the N-l model were found as 99.99%, 0.00074 and 0.009%, respectively. All results obtained by the N-l model were observed to have been quite consistent with real data. By this model, it would be possible to construct automation systems for classification and cleaning in flourmills.

Keywords: physical properties, artificial neural networks, barley, classification

Procedia PDF Downloads 158

Authors: Ning Xue, Srinivas Merugu, Ignacio Delgado Martinez, Tao Sun, John Tsang, Shih-Cheng Yen

Abstract:

We have developed a polyimide based neural interface electrode to record nerve signals from the sciatic nerve of a rat. The neural interface electrode has a split-ring shape, with four protruding gold electrodes for recording, and two reference gold electrodes around the split-ring. The split-ring electrode can be opened up to encircle the sciatic nerve. The four electrodes can be bent to sit on top of the nerve and hold the device in position, while the split-ring frame remains flat. In comparison, while traditional cuff electrodes can only fit certain sizes of the nerve, the developed device can fit a variety of rat sciatic nerve dimensions from 0.6 mm to 1.0 mm, and adapt to the chronic changes in the nerve as the electrode tips are bendable. The electrochemical impedance spectroscopy measurement was conducted. The gold electrode impedance is on the order of 10 kΩ, showing excellent charge injection capacity to record neural signals.

Keywords: impedance, neural interface, split-ring electrode, neural signal recording

Procedia PDF Downloads 353

Authors: Mohamed Lotfy, Ghada Soliman

Abstract:

Recognizing Arabic dot-matrix digits is a challenging problem due to the unique characteristics of dot-matrix fonts, such as irregular dot spacing and varying dot sizes. This paper presents an approach for recognizing Arabic digits printed in dot matrix format. The proposed model is based on Convolutional Neural Networks (CNN) that take the dot matrix as input and generate embeddings that are rounded to generate binary representations of the digits. The binary embeddings are then used to perform Optical Character Recognition (OCR) on the digit images. To overcome the challenge of the limited availability of dotted Arabic expiration date images, we developed a True Type Font (TTF) for generating synthetic images of Arabic dot-matrix characters. The model was trained on a synthetic dataset of 3287 images and 658 synthetic images for testing, representing realistic expiration dates from 2019 to 2027 in the format of yyyy/mm/dd. Our model achieved an accuracy of 98.94% on the expiry date recognition with Arabic dot matrix format using fewer parameters and less computational resources than traditional CNN-based models. By investigating and presenting our findings comprehensively, we aim to contribute substantially to the field of OCR and pave the way for advancements in Arabic dot-matrix character recognition. Our proposed approach is not limited to Arabic dot matrix digit recognition but can also be extended to text recognition tasks, such as text classification and sentiment analysis.

Keywords: computer vision, pattern recognition, optical character recognition, deep learning

Procedia PDF Downloads 62

Authors: Tianle Yang

Abstract:

As a common mental disorder, depression is a leading cause of various diseases worldwide. Early detection and treatment of depression can dramatically promote remission and prevent relapse. However, conventional ways of depression diagnosis require considerable human effort and cause economic burden, while still being prone to misdiagnosis. On the other hand, recent studies report that physical characteristics are major contributors to the diagnosis of depression, which inspires us to mine the internal relationship by neural networks instead of relying on clinical experiences. In this paper, neural networks are constructed to predict depression from physical characteristics. Two initialization methods are examined - Xaiver and Kaiming initialization. Experimental results show that a 3-layers neural network with Kaiming initialization achieves 83% accuracy.

Keywords: depression, neural network, Xavier initialization, Kaiming initialization

Procedia PDF Downloads 105

Authors: A. Shebani, C. Pislaru

Abstract:

Wear of materials is an everyday experience and has been observed and studied for long time. The prediction of wear is a fundamental problem in the industrial field, mainly correlated to the planning of maintenance interventions and economy. Pin-on-disc test is the most common test which is used to study the wear behaviour. In this paper, the pin-on-disc (AEROTECH UNIDEX 11) is used for the investigation of the effects of normal load and hardness of material on the wear under dry and sliding conditions. In the pin-on-disc rig, two specimens were used; one, a pin which is made of steel with a tip, is positioned perpendicular to the disc, where the disc is made of aluminium. The pin wear and disc wear were measured by using the following instruments: The Talysurf instrument, a digital microscope, and the alicona instrument; where the Talysurf profilometer was used to measure the pin/disc wear scar depth, and the alicona was used to measure the volume loss for pin and disc. After that, the Archard model, American Society for Testing and Materials model (ASTM), and neural network model were used for pin/disc wear modelling and the simulation results are implemented by using the Matlab program. This paper focuses on how the alicona can be considered as a powerful tool for wear measurements and how the neural network is an effective algorithm for wear estimation.

Keywords: wear modelling, Archard Model, ASTM Model, Neural Networks Model, Pin-on-disc Test, Talysurf, digital microscope, Alicona

Procedia PDF Downloads 431

Authors: Ahmad Saadiq, Neeraj Sahu

Abstract:

Sediment, which comprises of solid particles of mineral and organic material are transported by water. In river systems, the amount of sediment transported is controlled by both the transport capacity of the flow and the supply of sediment. The transport of sediment in rivers is important with respect to pollution, channel navigability, reservoir ageing, hydroelectric equipment longevity, fish habitat, river aesthetics and scientific interests. The sediment load transported in a river is a very complex hydrological phenomenon. Hence, sediment transport has attracted the attention of engineers from various aspects, and different methods have been used for its estimation. So, several experimental equations have been submitted by experts. Though the results of these methods have considerable differences with each other and with experimental observations, because the sediment measures have some limits, these equations can be used in estimating sediment load. In this present study, two black box models namely, an SRC (Sediment Rating Curve) and ANN (Artificial Neural Network) are used in the simulation of the suspended sediment load. The study is carried out for Seonath subbasin. Seonath is the biggest tributary of Mahanadi river, and it carries a vast amount of sediment. The data is collected for Jondhra hydrological observation station from India-WRIS (Water Resources Information System) and IMD (Indian Meteorological Department). These data include the discharge, sediment concentration and rainfall for 10 years. In this study, sediment load is estimated from the input parameters (discharge, rainfall, and past sediment) in various combination of simulations. A sediment rating curve used the water discharge to estimate the sediment concentration. This estimated sediment concentration is converted to sediment load. Likewise, for the application of these data in ANN, they are normalised first and then fed in various combinations to yield the sediment load. RMSE (root mean square error) and R² (coefficient of determination) between the observed load and the estimated load are used as evaluating criteria. For an ideal model, RMSE is zero and R² is 1. However, as the models used in this study are black box models, they don’t carry the exact representation of the factors which causes sedimentation. Hence, a model which gives the lowest RMSE and highest R² is the best model in this study. The lowest values of RMSE (based on normalised data) for sediment rating curve, feed forward back propagation, cascade forward back propagation and neural network fitting are 0.043425, 0.00679781, 0.0050089 and 0.0043727 respectively. The corresponding values of R² are 0.8258, 0.9941, 0.9968 and 0.9976. This implies that a neural network fitting model is superior to the other models used in this study. However, a drawback of neural network fitting is that it produces few negative estimates, which is not at all tolerable in the field of estimation of sediment load, and hence this model can’t be crowned as the best model among others, based on this study. A cascade forward back propagation produces results much closer to a neural network model and hence this model is the best model based on the present study.

Keywords: artificial neural network, Root mean squared error, sediment, sediment rating curve

Procedia PDF Downloads 309

Authors: Jia-Hong Lee, Mei-Yi Wu, Hsien-Tsung Kuo

Abstract:

Texture is an important characteristic in real and synthetic scenes. Texture analysis plays a critical role in inspecting surfaces and provides important techniques in a variety of applications. Although several descriptors have been presented to extract texture features, the development of object recognition is still a difficult task due to the complex aspects of texture. Recently, many robust and scaling-invariant image features such as SIFT, SURF and ORB have been successfully used in image retrieval and object recognition. In this paper, we have tried to compare the performance for texture classification using these feature descriptors with k-means clustering. Different classifiers including K-NN, Naive Bayes, Back Propagation Neural Network , Decision Tree and Kstar were applied in three texture image sets － UIUCTex, KTH-TIPS and Brodatz, respectively. Experimental results reveal SIFTS as the best average accuracy rate holder in UIUCTex, KTH-TIPS and SURF is advantaged in Brodatz texture set. BP neuro network works best in the test set classification among all used classifiers.

Keywords: texture classification, texture descriptor, SIFT, SURF, ORB

Procedia PDF Downloads 346

Authors: Khalid A. Al-Afandy, El-Sayyed El-Rabaie, Osama Salah, Ahmed El-Mhalaway

Abstract:

This paper presents a high secure data hiding technique using image cropping and Least Significant Bit (LSB) steganography. The predefined certain secret coordinate crops will be extracted from the cover image. The secret text message will be divided into sections. These sections quantity is equal the image crops quantity. Each section from the secret text message will embed into an image crop with a secret sequence using LSB technique. The embedding is done using the cover image color channels. Stego image is given by reassembling the image and the stego crops. The results of the technique will be compared to the other state of art techniques. Evaluation is based on visualization to detect any degradation of stego image, the difficulty of extracting the embedded data by any unauthorized viewer, Peak Signal-to-Noise Ratio of stego image (PSNR), and the embedding algorithm CPU time. Experimental results ensure that the proposed technique is more secure compared with the other traditional techniques.

Keywords: steganography, stego, LSB, crop

Procedia PDF Downloads 252

Authors: Jurijs Salijevs, Katrina Bolocko

Abstract:

The integration of artificial intelligence and neural networks has significantly changed full-body imaging and high-resolution 3D mapping systems, and this paper reviews research in these areas. With an emphasis on their use in the early identification of melanoma and other disorders, the goal is to give a wide perspective on the current status and potential future of these medical imaging technologies. Authors also examine methodologies such as machine learning and deep learning, seeking to identify efficient procedures that enhance diagnostic capabilities through the analysis of 3D body scans. This work aims to encourage further research and technological development to harness the full potential of AI in disease diagnosis.

Keywords: artificial intelligence, neural networks, 3D scan, body scan, 3D mapping system, healthcare

Procedia PDF Downloads 73

Authors: S. Curteanu, F. Leon, M. Gavrilescu, S. A. Floria

Abstract:

Optimization algorithms inspired from human behavior were applied in this approach, associated with neural networks models. The algorithms belong to human behaviors of learning and cooperation and human competitive behavior classes. For the first class, the main strategies include: random learning, individual learning, and social learning, and the selected algorithms are: simplified human learning optimization (SHLO), social learning optimization (SLO), and teaching-learning based optimization (TLBO). For the second class, the concept of learning is associated with competitiveness, and the selected algorithms are sports-inspired algorithms (with Football Game Algorithm, FGA and Volleyball Premier League, VPL) and Imperialist Competitive Algorithm (ICA). A real process, the synthesis of polyacrylamide-based multicomponent hydrogels, where some parameters are difficult to obtain experimentally, is considered as a case study. Reaction yield and swelling degree are predicted as a function of reaction conditions (acrylamide concentration, initiator concentration, crosslinking agent concentration, temperature, reaction time, and amount of inclusion polymer, which could be starch, poly(vinyl alcohol) or gelatin). The experimental results contain 175 data. Artificial neural networks are obtained in optimal form with biologically inspired algorithm; the optimization being perform at two level: structural and parametric. Feedforward neural networks with one or two hidden layers and no more than 25 neurons in intermediate layers were obtained with values of correlation coefficient in the validation phase over 0.90. The best results were obtained with TLBO algorithm, correlation coefficient being 0.94 for an MLP(6:9:20:2) – a feedforward neural network with two hidden layers and 9 and 20, respectively, intermediate neurons. Good results obtained prove the efficiency of the optimization algorithms. More than the good results, what is important in this approach is the simulation methodology, including neural networks and optimization biologically inspired algorithms, which provide satisfactory results. In addition, the methodology developed in this approach is general and has flexibility so that it can be easily adapted to other processes in association with different types of models.

Keywords: artificial neural networks, human behaviors of learning and cooperation, human competitive behavior, optimization algorithms

Procedia PDF Downloads 91

Authors: Dionisio Andujar, Juan lópez-Correa, Hugo Moreno, Angela Ri

Abstract:

The increased number of weeds in tomato crops highly lower yields. Weed identification with the aim of machine learning is important to carry out site-specific control. The last advances in computer vision are a powerful tool to face the problem. The analysis of RGB (Red, Green, Blue) images through Artificial Neural Networks had been rapidly developed in the past few years, providing new methods for weed classification. The development of the algorithms for crop and weed species classification looks for a real-time classification system using Object Detection algorithms based on Convolutional Neural Networks. The site study was located in commercial corn fields. The classification system has been tested. The procedure can detect and classify weed seedlings in tomato fields. The input to the Neural Network was a set of 10,000 RGB images with a natural infestation of Cyperus rotundus l., Echinochloa crus galli L., Setaria italica L., Portulaca oeracea L., and Solanum nigrum L. The validation process was done with a random selection of RGB images containing the aforementioned species. The mean average precision (mAP) was established as the metric for object detection. The results showed agreements higher than 95 %. The system will provide the input for an online spraying system. Thus, this work plays an important role in Site Specific Weed Management by reducing herbicide use in a single step.

Keywords: deep learning, object detection, cnn, tomato, weeds

Procedia PDF Downloads 87

Authors: Khadija Refouh

Abstract:

Culture bound-expressions have been a bottleneck for Natural Language Processing (NLP) and comprehension, especially in the case of machine translation (MT). In the last decade, the field of machine translation has greatly advanced. Neural machine translation NMT has recently achieved considerable development in the quality of translation that outperformed previous traditional translation systems in many language pairs. Neural machine translation NMT is an Artificial Intelligence AI and deep neural networks applied to language processing. Despite this development, there remain some serious challenges that face neural machine translation NMT when translating culture bounded-expressions, especially for low resources language pairs such as Arabic-English and Arabic-French, which is not the case with well-established language pairs such as English-French. Machine translation of opaque idioms from English into French are likely to be more accurate than translating them from English into Arabic. For example, Google Translate Application translated the sentence “What a bad weather! It runs cats and dogs.” to “يا له من طقس سيء! تمطر القطط والكلاب” into the target language Arabic which is an inaccurate literal translation. The translation of the same sentence into the target language French was “Quel mauvais temps! Il pleut des cordes.” where Google Translate Application used the accurate French corresponding idioms. This paper aims to perform NMT experiments towards better translation of opaque idioms using high quality clean multilingual corpus. This Corpus will be collected analytically from human generated idiom translation. AutoML translation, a Google Neural Machine Translation Platform, is used as a custom translation model to improve the translation of opaque idioms. The automatic evaluation of the custom model will be compared to the Google NMT using Bilingual Evaluation Understudy Score BLEU. BLEU is an algorithm for evaluating the quality of text which has been machine-translated from one natural language to another. Human evaluation is integrated to test the reliability of the Blue Score. The researcher will examine syntactical, lexical, and semantic features using Halliday's functional theory.

Keywords: multilingual corpora, natural language processing (NLP), neural machine translation (NMT), opaque idioms

Procedia PDF Downloads 122

Authors: Xiaoqi Jia, Qing Zhu, Hao Zhang, Su Yang

Abstract:

A gender recognition system is able to tell the gender of the given person through a few of frontal facial images. An effective gender recognition approach enables to improve the performance of many other applications, including security monitoring, human-computer interaction, image or video retrieval and so on. In this paper, we present an effective method for gender classification task in frontal facial images based on deep belief networks (DBNs), which can pre-train model and improve accuracy a little bit. Our experiments have shown that the pre-training method with DBNs for gender classification task is feasible and achieves a little improvement of accuracy on FERET and CAS-PEAL-R1 facial datasets.

Keywords: gender recognition, beep belief net-works, semi-supervised learning, greedy-layer wise RBMs

Procedia PDF Downloads 427

Authors: Anthony D. Rhodes, Manan Goel

Abstract:

We provide a high fidelity deep learning algorithm (HyperSeg) for interactive video segmentation tasks using a dense convolutional network with context-aware skip connections and compressed, 'hypercolumn' image features combined with a convolutional tessellation procedure. In order to maintain high output fidelity, our model crucially processes and renders all image features in high resolution, without utilizing downsampling or pooling procedures. We maintain this consistent, high grade fidelity efficiently in our model chiefly through two means: (1) we use a statistically-principled, tensor decomposition procedure to modulate the number of hypercolumn features and (2) we render these features in their native resolution using a convolutional tessellation technique. For improved pixel-level segmentation results, we introduce a boundary loss function; for improved temporal coherence in video data, we include temporal image information in our model. Through experiments, we demonstrate the improved accuracy of our model against baseline models for interactive segmentation tasks using high resolution video data. We also introduce a benchmark video segmentation dataset, the VFX Segmentation Dataset, which contains over 27,046 high resolution video frames, including green screen and various composited scenes with corresponding, hand-crafted, pixel-level segmentations. Our work presents a improves state of the art segmentation fidelity with high resolution data and can be used across a broad range of application domains, including VFX pipelines and medical imaging disciplines.

Keywords: computer vision, object segmentation, interactive segmentation, model compression

Procedia PDF Downloads 102

Authors: K. Suganya Devi, P. Srinivasan, M. P. Vaishnave, G. Arutperumjothi

Abstract:

Today’s internet world is highly prone to various online attacks, of which the most harmful attack is phishing. The attackers host the fake websites which are very similar and look alike. We propose an image based authentication using steganography and visual cryptography to prevent phishing. This paper presents a secure steganographic technique for true color (RGB) images and uses Discrete Cosine Transform to compress the images. The proposed method hides the secret data inside the cover image. The use of visual cryptography is to preserve the privacy of an image by decomposing the original image into two shares. Original image can be identified only when both qualified shares are simultaneously available. Individual share does not reveal the identity of the original image. Thus, the existence of the secret message is hard to be detected by the RS steganalysis.

Keywords: image security, random LSB, steganography, visual cryptography

Procedia PDF Downloads 313

Authors: Minakshi Mainaji Sonawane, Bharti W. Gawali, Sudhir Mendhekar, Ramesh R. Manza

Abstract:

People's lives and health are severely impacted by skin diseases. A new study proposes an effective method for identifying the different forms of skin diseases. Image denoising is a technique for improving image quality after it has been harmed by noise. The proposed technique is based on the usage of the wavelet transform. Wavelet transform is the best method for analyzing the image due to the ability to split the image into the sub-band, which has been used to estimate the noise ratio at the noisy image. According to experimental results, the proposed method presents the best values for MSE, PSNR, and Entropy for denoised images. we can found in Also, by using different types of wavelet transform filters is make the proposed approach can obtain the best results 23.13, 20.08, 50.7 for the image denoising process

Keywords: MSE, PSNR, entropy, Gaussian filter, DWT

Procedia PDF Downloads 197

Authors: Ananya Ananya, Karthik Rao

Abstract:

Accurate segmentation of knee cartilage in 3-D magnetic resonance (MR) images for quantitative assessment of volume is crucial for studying and diagnosing osteoarthritis (OA) of the knee, one of the major causes of disability in elderly people. Radiologists generally perform this task in slice-by-slice manner taking 15-20 minutes per 3D image, and lead to high inter and intra observer variability. Hence automatic methods for knee cartilage segmentation are desirable and are an active field of research. This paper presents design and experimental evaluation of 2D convolutional neural networks based fully automated methods for knee cartilage segmentation in 3D MRI. The architectures are validated based on 40 test images and 60 training images from SKI10 dataset. The proposed methods segment 2D slices one by one, which are then combined to give segmentation for whole 3D images. Proposed methods are modified versions of U-net and dilated convolutions, consisting of a single step that segments the given image to 5 labels: background, femoral cartilage, tibia cartilage, femoral bone and tibia bone; cartilages being the primary components of interest. U-net consists of a contracting path and an expanding path, to capture context and localization respectively. Dilated convolutions lead to an exponential expansion of receptive field with only a linear increase in a number of parameters. A combination of modified U-net and dilated convolutions has also been explored. These architectures segment one 3D image in 8 – 10 seconds giving average volumetric Dice Score Coefficients (DSC) of 0.950 - 0.962 for femoral cartilage and 0.951 - 0.966 for tibia cartilage, reference being the manual segmentation.

Keywords: convolutional neural networks, dilated convolutions, 3 dimensional, fully automated, knee cartilage, MRI, segmentation, U-net

Procedia PDF Downloads 241

Authors: Teja Koduru

Abstract:

Untreated Lyme disease can lead to neurological, cardiac, and dermatological complications. Rapid diagnosis of the erythema migrans (EM) rash, a characteristic symptom of Lyme disease is therefore crucial to early diagnosis and treatment. In this study, we aim to utilize deep learning frameworks including Tensorflow and Keras to create deep convolutional neural networks (DCNN) to detect images of acute Lyme Disease from images of erythema migrans. This study uses a custom database of erythema migrans images of varying quality to train a DCNN capable of classifying images of EM rashes vs. non-EM rashes. Images from publicly available sources were mined to create an initial database. Machine-based removal of duplicate images was then performed, followed by a thorough examination of all images by a clinician. The resulting database was combined with images of confounding rashes and regular skin, resulting in a total of 683 images. This database was then used to create a DCNN with an accuracy of 93% when classifying images of rashes as EM vs. non EM. Finally, this model was converted into a web and mobile application to allow for rapid diagnosis of EM rashes by both patients and clinicians. This tool could be used for patient prescreening prior to treatment and lead to a lower mortality rate from Lyme disease.

Keywords: Lyme, untreated Lyme, erythema migrans rash, EM rash

Procedia PDF Downloads 217

Authors: Cristian Pauna

Abstract:

Trading the financial markets is a widespread activity today. A large number of investors, companies, public of private funds are buying and selling every day in order to make profit. Algorithmic trading is the prevalent method to make the trade decisions after the electronic trading release. The orders are sent almost instantly by computers using mathematical models. This paper will present a price prediction methodology based on a multi-dimensional neural network. Using the Fisher transform, the neural network will be instructed for a low-latency auto-adaptive process in order to predict the price evolution for the next period of time. The model is designed especially for algorithmic trading and uses the real-time price series. It was found that the characteristics of the Fisher function applied at the nodes scale level can generate reliable trading signals using the neural network methodology. After real time tests it was found that this method can be applied in any timeframe to trade the financial markets. The paper will also include the steps to implement the presented methodology into an automated trading system. Real trading results will be displayed and analyzed in order to qualify the model. As conclusion, the compared results will reveal that the neural network methodology applied together with the Fisher transform at the nodes level can generate a good price prediction and can build reliable trading signals for algorithmic trading.

Keywords: algorithmic trading, automated trading systems, financial markets, high-frequency trading, neural network

Procedia PDF Downloads 140

Authors: Ola Hall, Ibrahim Wahab, Thorsteinn Rognvaldsson, Mattias Ohlsson

Abstract:

The subfield of poverty and welfare estimation that applies machine learning tools and methods on satellite imagery is a nascent but rapidly growing one. This is in part driven by the sustainable development goal, whose overarching principle is that no region is left behind. Among other things, this requires that welfare levels can be accurately and rapidly estimated at different spatial scales and resolutions. Conventional tools of household surveys and interviews do not suffice in this regard. While they are useful for gaining a longitudinal understanding of the welfare levels of populations, they do not offer adequate spatial coverage for the accuracy that is needed, nor are their implementation sufficiently swift to gain an accurate insight into people and places. It is this void that satellite imagery fills. Previously, this was near-impossible to implement due to the sheer volume of data that needed processing. Recent advances in machine learning, especially the deep learning subtype, such as deep neural networks, have made this a rapidly growing area of scholarship. Despite their unprecedented levels of performance, such models lack transparency and explainability and thus have seen limited downstream applications as humans generally are apprehensive of techniques that are not inherently interpretable and trustworthy. While several studies have demonstrated the superhuman performance of AI models, none has directly compared the performance of such models and human readers in the domain of poverty studies. In the present study, we directly compare the performance of human readers and a DL model using different resolutions of satellite imagery to estimate the welfare levels of demographic and health survey clusters in Tanzania, using the wealth quintile ratings from the same survey as the ground truth data. The cluster-level imagery covers all 608 cluster locations, of which 428 were classified as rural. The imagery for the human readers was sourced from the Google Maps Platform at an ultra-high resolution of 0.6m per pixel at zoom level 18, while that of the machine learning model was sourced from the comparatively lower resolution Sentinel-2 10m per pixel data for the same cluster locations. Rank correlation coefficients of between 0.31 and 0.32 achieved by the human readers were much lower when compared to those attained by the machine learning model – 0.69-0.79. This superhuman performance by the model is even more significant given that it was trained on the relatively lower 10-meter resolution satellite data while the human readers estimated welfare levels from the higher 0.6m spatial resolution data from which key markers of poverty and slums – roofing and road quality – are discernible. It is important to note, however, that the human readers did not receive any training before ratings, and had this been done, their performance might have improved. The stellar performance of the model also comes with the inevitable shortfall relating to limited transparency and explainability. The findings have significant implications for attaining the objective of the current frontier of deep learning models in this domain of scholarship – eXplainable Artificial Intelligence through a collaborative rather than a comparative framework.

Keywords: poverty prediction, satellite imagery, human readers, machine learning, Tanzania

Procedia PDF Downloads 78

Authors: Cheng Zeng, George Michailidis, Hitoshi Iyatomi, Leo L. Duan

Abstract:

The conditional density characterizes the distribution of a response variable y given other predictor x and plays a key role in many statistical tasks, including classification and outlier detection. Although there has been abundant work on the problem of Conditional Density Estimation (CDE) for a low-dimensional response in the presence of a high-dimensional predictor, little work has been done for a high-dimensional response such as images. The promising performance of normalizing flow (NF) neural networks in unconditional density estimation acts as a motivating starting point. In this work, the authors extend NF neural networks when external x is present. Specifically, they use the NF to parameterize a one-to-one transform between a high-dimensional y and a latent z that comprises two components [zₚ, zₙ]. The zₚ component is a low-dimensional subvector obtained from the posterior distribution of an elementary predictive model for x, such as logistic/linear regression. The zₙ component is a high-dimensional independent Gaussian vector, which explains the variations in y not or less related to x. Unlike existing CDE methods, the proposed approach coined Augmented Posterior CDE (AP-CDE) only requires a simple modification of the common normalizing flow framework while significantly improving the interpretation of the latent component since zₚ represents a supervised dimension reduction. In image analytics applications, AP-CDE shows good separation of 𝑥-related variations due to factors such as lighting condition and subject id from the other random variations. Further, the experiments show that an unconditional NF neural network based on an unsupervised model of z, such as a Gaussian mixture, fails to generate interpretable results.

Keywords: conditional density estimation, image generation, normalizing flow, supervised dimension reduction

Procedia PDF Downloads 75

Authors: Mamadou I. Wade, Henry C. Ogworonjo, Madiha Gul, Mandoye Ndoye, Mohamed Chouikha, Wayne Patterson

Abstract:

In this paper, we present a novel application of the Paillier cryptographic system to the encryption of RGB (Red Green Blue) images. In this method, an RGB image is first separated into its constituent channel images, and the Paillier encryption function is applied to each of the channels pixel intensity values. Next, the encrypted image is combined and compressed if necessary before being transmitted through an unsecured communication channel. The transmitted image is subsequently recovered by a decryption process. We performed a series of security and performance analyses to the recovered images in order to verify their robustness to security attack. The results show that the proposed image encryption scheme produces highly secured encrypted images.

Keywords: image encryption, Paillier cryptographic system, RBG image encryption, Paillier

Procedia PDF Downloads 213

Authors: Rabi Mouhcine, Amrouch Mustapha, Mahani Zouhir, Mammass Driss

Abstract:

In this paper, we present a system for offline recognition cursive Arabic handwritten text based on Hidden Markov Models (HMMs). The system is analytical without explicit segmentation used embedded training to perform and enhance the character models. Extraction features preceded by baseline estimation are statistical and geometric to integrate both the peculiarities of the text and the pixel distribution characteristics in the word image. These features are modelled using hidden Markov models and trained by embedded training. The experiments on images of the benchmark IFN/ENIT database show that the proposed system improves recognition.

Keywords: recognition, handwriting, Arabic text, HMMs, embedded training

Procedia PDF Downloads 331

Authors: Chin-Chen Chang, I-Ta Lee, Tsung-Ta Ke, Wen-Kai Tai

Abstract:

Common methods for resizing image size include scaling and cropping. However, these two approaches have some quality problems for reduced images. In this paper, we propose an image resizing algorithm by separating the main objects and the background. First, we extract two feature maps, namely, an enhanced visual saliency map and an improved gradient map from an input image. After that, we integrate these two feature maps to an importance map. Finally, we generate the target image using the importance map. The proposed approach can obtain desired results for a wide range of images.

Keywords: energy map, visual saliency, gradient map, seam carving

Procedia PDF Downloads 464

Authors: Natalia Espinosa, Arthur Amorim, Rudolf Huebner

Abstract:

Epilepsy is a chronic neural disease and around 50 million people in the world suffer from this disease, however, in many cases, the individual acquires resistance to the medication, which is known as drug-resistant epilepsy, where a detection system is necessary. This paper showed the development of an automatic system for seizure detection based on artificial neural networks (ANN), which are common techniques of machine learning. Discrete Wavelet Transform (DWT) is used for decomposing electroencephalogram (EEG) signal into main brain waves, with these frequency bands is extracted features for training a feedforward neural network with backpropagation, finally made a pattern classification, seizure or non-seizure. Obtaining 95% accuracy in epileptic EEG and 100% in normal EEG.

Keywords: Artificial Neural Network (ANN), Discrete Wavelet Transform (DWT), Epilepsy Detection , Seizure.

Procedia PDF Downloads 192

Authors: Chen Xi, Liu Xuebing, Lao Xueru, Kuan Sinman, Jiang Yike, Wang Hanwei, Yang Xiaolang, Zhou Junjie, Xie Jinpeng

Abstract:

To further promote the development of smart cities, the microscopic "nerve endings" of the City Intelligent Model (CIM) are extended to be more sensitive. In this paper, we develop a pedestrian trajectory twin terminal based on the CIM and CNN technology. It also uses 5G networks, architectural and geoinformatics technologies, convolutional neural networks, combined with deep learning networks for human behavior recognition models, to provide empirical data such as 'pedestrian flow data and human behavioral characteristics data', and ultimately form spatial performance evaluation criteria and spatial performance warning systems, to make the empirical data accurate and intelligent for prediction and decision making.

Keywords: urban planning, urban governance, CIM, artificial intelligence, sustainable development

Procedia PDF Downloads 377