Search results for: deep convolutional features

Authors: Thiago Spilborghs Bueno Meyer, Plinio Thomaz Aquino Junior

Abstract:

Through speech, which privileges the functional and interactive nature of the text, it is possible to ascertain the spatiotemporal circumstances, the conditions of production and reception of the discourse, the explicit purposes such as informing, explaining, convincing, etc. These conditions allow bringing the interaction between humans closer to the human-robot interaction, making it natural and sensitive to information. However, it is not enough to understand what is said; it is necessary to recognize emotions for the desired interaction. The validity of the use of neural networks for feature selection and emotion recognition was verified. For this purpose, it is proposed the use of neural networks and comparison of models, such as recurrent neural networks and deep neural networks, in order to carry out the classification of emotions through speech signals to verify the quality of recognition. It is expected to enable the implementation of robots in a domestic environment, such as the HERA robot from the RoboFEI@Home team, which focuses on autonomous service robots for the domestic environment. Tests were performed using only the Mel-Frequency Cepstral Coefficients, as well as tests with several characteristics of Delta-MFCC, spectral contrast, and the Mel spectrogram. To carry out the training, validation and testing of the neural networks, the eNTERFACE’05 database was used, which has 42 speakers from 14 different nationalities speaking the English language. The data from the chosen database are videos that, for use in neural networks, were converted into audios. It was found as a result, a classification of 51,969% of correct answers when using the deep neural network, when the use of the recurrent neural network was verified, with the classification with accuracy equal to 44.09%. The results are more accurate when only the Mel-Frequency Cepstral Coefficients are used for the classification, using the classifier with the deep neural network, and in only one case, it is possible to observe a greater accuracy by the recurrent neural network, which occurs in the use of various features and setting 73 for batch size and 100 training epochs.

Keywords: emotion recognition, speech, deep learning, human-robot interaction, neural networks

Procedia PDF Downloads 170

Authors: Shi-Uk Lee, Chae Young Lim

Abstract:

Objective: The aim of this study was to evaluate the changes of lumbar deep muscle thickness and cross-sectional area using ultrasonography with magnetic stimulation. Methods: To evaluate the changes of lumbar deep muscle by using magnetic stimulation, 12 healthy volunteers (39.6±10.0 yrs) without low back pain during 3 months participated in this study. All the participants were checked with X-ray and electrophysiologic study to confirm that they had no problems with their back. Magnetic stimulation was done on the L5 and S1 root with figure-eight coil as previous study. To confirm the proper motor root stimulation, the surface electrode was put on the tibialis anterior (L5) and abductor hallucis muscles (S1) and the hot spots of magnetic stimulation were found with 50% of maximal magnetic stimulation and determined the stimulation threshold lowering the magnetic intensity by 5%. Ultrasonography was used to assess the changes of L5 and S1 lumbar multifidus (superficial and deep) cross-sectional area and thickness with maximal magnetic stimulation. Cross-sectional area (CSA) and thickness was evaluated with image acquisition program, ImageJ software (National Institute of Healthy, USA). Wilcoxon signed-rank was used to compare outcomes between before and after stimulations. Results: The mean minimal threshold was 29.6±3.8% of maximal stimulation intensity. With minimal magnetic stimulation, thickness of L5 and S1 deep multifidus (DM) were increased from 1.25±0.20, 1.42±0.23 cm to 1.40±0.27, 1.56±0.34 cm, respectively (P=0.005, P=0.003). CSA of L5 and S1 DM were also increased from 2.26±0.18, 1.40±0.26 cm2 to 2.37±0.18, 1.56±0.34 cm2, respectively (P=0.002, P=0.002). However, thickness of L5 and S1 superficial multifidus (SM) were not changed from 1.92±0.21, 2.04±0.20 cm to 1.91±0.33, 1.96±0.33 cm (P=0.211, P=0.199) and CSA of L5 and S1 were also not changed from 4.29±0.53, 5.48±0.32 cm2 to 4.42±0.42, 5.64±0.38 cm2. With maximal magnetic stimulation, thickness of L5, S1 of DM and SM were increased (L5 DM, 1.29±0.26, 1.46±0.27 cm, P=0.028; L5 SM, 2.01±0.42, 2.24±0.39 cm, P=0.005; S1 DM, 1.29±0.19, 1.67±0.29 P=0.002; S1 SM, 1.90±0.36, 2.30±0.36, P=0.002). CSA of L5, S1 of DM and SM were also increased (all P values were 0.002). Conclusions: Deep lumbar muscles could be stimulated with lumbar motor root magnetic stimulation. With minimal stimulation, thickness and CSA of lumbosacral deep multifidus were increased in this study. Further studies are needed to confirm whether the similar results in chronic low back pain patients are represented. Lumbar magnetic stimulation might have strengthening effect of deep lumbar muscles with no discomfort.

Keywords: magnetic stimulation, lumbar multifidus, strengthening, ultrasonography

Procedia PDF Downloads 371

Authors: Sam Khozama, Ali M. Mayya

Abstract:

Breast cancer is one of the most common types in women. Early prediction of breast cancer helps physicians detect cancer in its early stages. Big cancer data needs a very powerful tool to analyze and extract predictions. Machine learning and deep learning are two of the most efficient tools for predicting cancer based on textual data. In this study, we developed a fusion model of two machine learning and deep learning models. To obtain the final prediction, Long-Short Term Memory (LSTM) and ensemble learning with hyper parameters optimization are used, and score-level fusion is used. Experiments are done on the Breast Cancer Surveillance Consortium (BCSC) dataset after balancing and grouping the class categories. Five different training scenarios are used, and the tests show that the designed fusion model improved the performance by 3.3% compared to the individual models.

Keywords: machine learning, deep learning, cancer prediction, breast cancer, LSTM, fusion

Procedia PDF Downloads 163

Authors: Yan Bachtiar Muslih, Hangga Wijaya, Trio Fani, Putri Agustin

Abstract:

Demand of Energy in Indonesia always increases 5-6% a year, but production of conventional energy always decreases 3-5% a year, it means that conventional energy in 20-40 years ahead will not able to complete all energy demand in Indonesia, one of the solve way is using unconventional energy that is gas hydrate, gas hydrate is gas that form by biogenic process, gas hydrate stable in condition with extremely depth and low temperature, gas hydrate can form in two condition that is in pole condition and in deep-sea condition, wherein this research will focus in gas hydrate that association with methane form methane hydrate in deep-sea condition and usually form in depth between 150-2000 m, this research will focus in process of methane hydrate formation that is biogenic process and the important role of deep-sea sediment so can produce accumulation of methane hydrate, methane hydrate usually will be accumulated in find sediment in deep-sea environment with condition high-pressure and low-temperature this condition too usually make methane hydrate change into white nodule, methodology of this research is geology field work and laboratory analysis, from geology field work will get sample data consist of 10-15 samples from Kerek Formation outcrops as random for imagine the condition of deep-sea environment that influence the methane hydrate formation and also from geology field work will get data of measuring stratigraphy in outcrops Kerek Formation too from this data will help to imagine the process in deep-sea sediment like energy flow, supply sediment, and etc, and laboratory analysis is activity to analyze all data that get from geology field work, the result of this research can used to exploration activity of methane hydrate in another prospect deep-sea environment in Indonesia.

Keywords: methane hydrate, deep-sea sediment, kerek formation, sub-basin of kendeng, central java, Indonesia

Procedia PDF Downloads 462

Authors: Bashayer Almalki, Mayar Bajrai, Dana Mirah, Kholood Alghamdi, Hala Sanyour

Abstract:

This research aims to investigate the features of an AI chess app that are most preferred by users. A questionnaire was used as the methodology to gather responses from a varied group of participants. The questionnaire consisted of several questions related to the features of the AI chess app. The responses were analyzed using descriptive statistics and factor analysis. The findings indicate that the most preferred features of an AI chess app are the ability to play against the computer, the option to adjust the difficulty level, and the availability of tutorials and puzzles. The results of this research could be useful for developers of AI chess apps to enhance the user experience and satisfaction.

Keywords: chess, game, application, computics

Procedia PDF Downloads 69

Authors: Xiaolin Zeng, Lei Cheng, Zhirong Li, Xueping Liu

Abstract:

Aiming at the problem of low accuracy and susceptibility to cultural relic diseases in the generation of high-resolution archaeology drawings by current image generation algorithms, an archaeology drawings generation algorithm based on a conditional generative adversarial network is proposed. An attention mechanism is added into the high-resolution image generation network as the backbone network, which enhances the line feature extraction capability and improves the accuracy of line drawing generation. A dual-branch parallel architecture consisting of two backbone networks is implemented, where the semantic translation branch extracts semantic features from orthophotographs of cultural relics, and the gradient screening branch extracts effective gradient features. Finally, the fusion fine-tuning module combines these two types of features to achieve the generation of high-quality and high-resolution archaeology drawings. Experimental results on the self-constructed archaeology drawings dataset of grotto temple statues show that the proposed algorithm outperforms current mainstream image generation algorithms in terms of pixel accuracy (PA), structural similarity (SSIM), and peak signal-to-noise ratio (PSNR) and can be used to assist in drawing archaeology drawings.

Keywords: archaeology drawings, digital heritage, image generation, deep learning

Procedia PDF Downloads 59

Authors: Yuanxiang Miao

Abstract:

This paper aims to examine the perceptual features of couchsurfers on a new hospitality tourism platform, the free homestay website couchsurfing. As a local host, the author has accepted 61 couchsurfers in Kyoto, Japan, and attempted to figure out couchsurfers' characteristics on perception by hosting them. Moreover, the methodology of this research is mainly based on in-depth interviews, by talking with couchsurfers, observing their behaviors, doing questionnaires, etc. Five dominant perceptual features of couchsurfers were identified: (1) Trusting; (2) Meeting; (3) Sharing; (4) Reciprocity; (5) Worries. The value of this research lies in figuring out a deeper understanding of the perceptual features of couchsurfers, and the author indeed hosted and stayed with 61 couchsurfers from 30 countries and areas over one year. Lastly, the author offers practical suggestions for future research.

Keywords: couchsurfing, depth interview, hospitality tourism, perceptual features

Procedia PDF Downloads 145

Authors: Cahit Perkgöz

Abstract:

Artificial intelligence has brought new approaches to solving problems in almost every research field in recent years. One of these topics is shape design and optimization, which has the possibility of applications in many fields, such as nanotechnology and electronics. A properly constructed cost function can eliminate the need for labeled data required in deep learning and create desired shapes. In this work, the network parameters are optimized differentially, which differs from traditional approaches. The methods are tested for physics-related structures and successful results are obtained. This work is supported by Eskişehir Technical University scientific research project (Project No: 20ADP090)

Keywords: deep learning, shape design, optimization, artificial intelligence

Procedia PDF Downloads 153

Authors: Nubwimana Rachel, Mugabowindekwe Maurice

Abstract:

Trees play a crucial role in maintaining biodiversity and providing various ecological services. Traditional methods of counting trees are time-consuming, and there is a need for more efficient techniques. However, deep learning makes it feasible to identify the multi-scale elements hidden in aerial imagery. This research focuses on the application of deep learning techniques for tree detection and counting in both forest and non-forest areas through the exploration of the deep learning application for automated tree detection and counting using satellite imagery. The objective is to identify the most effective model for automated tree counting. We used different deep learning models such as YOLOV7, SSD, and UNET, along with Generative Adversarial Networks to generate synthetic samples for training and other augmentation techniques, including Random Resized Crop, AutoAugment, and Linear Contrast Enhancement. These models were trained and fine-tuned using satellite imagery to identify and count trees. The performance of the models was assessed through multiple trials; after training and fine-tuning the models, UNET demonstrated the best performance with a validation loss of 0.1211, validation accuracy of 0.9509, and validation precision of 0.9799. This research showcases the success of deep learning in accurate tree counting through remote sensing, particularly with the UNET model. It represents a significant contribution to the field by offering an efficient and precise alternative to conventional tree-counting methods.

Keywords: remote sensing, deep learning, tree counting, image segmentation, object detection, visualization

Procedia PDF Downloads 71

Authors: A. Degale Desta, Tamirat Kebamo

Abstract:

Background: Recent advances in computer vision and pattern recognition have significantly improved activity recognition through video analysis, particularly with the application of Deep Convolutional Neural Networks (CNNs). One-stage detectors now enable efficient video-based recognition by simultaneously predicting object categories and locations. Such advancements are highly relevant in educational settings where CCTV surveillance could automatically monitor academic activities, enhancing security and classroom management. However, current datasets and recognition systems lack the specific focus on campus environments necessary for practical application in these settings.Objective: This study aims to address this gap by developing a dataset and testing an automated activity recognition system specifically tailored for educational campuses. The EthioCAD dataset was created to capture various classroom activities and teacher-student interactions, facilitating reliable recognition of academic activities using deep learning models. Method: EthioCAD, a novel video-based dataset, was created with a design science research approach to encompass teacher-student interactions across three domains and 18 distinct classroom activities. Using the Roboflow AI framework, the data was processed, with 4.224 KB of frames and 33.485 MB of images managed for frame extraction, labeling, and organization. The Ultralytics YOLOv8 model was then implemented within Google Colab to evaluate the dataset’s effectiveness, achieving high mean Average Precision (mAP) scores. Results: The YOLOv8 model demonstrated robust activity recognition within campus-like settings, achieving an mAP50 of 90.2% and an mAP50-95 of 78.6%. These results highlight the potential of EthioCAD, combined with YOLOv8, to provide reliable detection and classification of classroom activities, supporting automated surveillance needs on educational campuses. Discussion: The high performance of YOLOv8 on the EthioCAD dataset suggests that automated activity recognition for surveillance is feasible within educational environments. This system addresses current limitations in campus-specific data and tools, offering a tailored solution for academic monitoring that could enhance the effectiveness of CCTV systems in these settings. Conclusion: The EthioCAD dataset, alongside the YOLOv8 model, provides a promising framework for automated campus activity recognition. This approach lays the groundwork for future advancements in CCTV-based educational surveillance systems, enabling more refined and reliable monitoring of classroom activities.

Keywords: deep CNN, EthioCAD, deep learning, YOLOv8, activity recognition

Procedia PDF Downloads 10

Authors: Jiyoung Bae, Gyoomi Kim

Abstract:

This study explores a range of linguistic features used in Korean college students’ argumentative writings for the purpose of developing a model that identifies variables which predict writing proficiencies. This study investigated the latent variable structure of L2 linguistic features, including syntactic complexity, the lexical complexity, and fluency. One hundred forty-six university students in Korea participated in this study. The results of the study’s confirmatory factor analysis (CFA) showed that indicators of linguistic features from this study-provided a foundation for re-categorizing indicators found in extant research on L2 Korean writers depending on each latent variable of linguistic features. The CFA models indicated one measurement model of L2 syntactic complexity and L2 learners’ writing proficiency; these two latent factors were correlated with each other. Based on the overall findings of the study, integrated linguistic features of L2 writings suggested some pedagogical implications in L2 writing instructions.

Keywords: linguistic features, syntactic complexity, lexical complexity, fluency

Procedia PDF Downloads 170

Authors: Larisa Liamina, Vasily Lavrushin, Salvatore Inguaggiato

Abstract:

This study presents and summarizes the results of researching the isotopic signature of helium in the deep fluid eastern part of the Southern slope of the Greater Caucasus and the Lesser Caucasus (Azerbaijan and Armenia) for the period from 2010 to 2016. The results of isotope ratios of 3He/4He in 59 samples of the gas phase of geothermal fluids and mud volcanoes are presented. New data have been obtained not only on the isotopic ratios of helium, but also neon and argon. The R/Ra ratio was analyzed along the Ankara-Sevan ophiolite structure. The patterns of lateral variations of the 3He/4He ratio of different geological structural elements of the studied region are revealed.

Keywords: isotopes helium, deep fluids, tectonic structures, Caucasus

Procedia PDF Downloads 45

Authors: Farshad Zeinali, Sajedeh Norouzi, Nader Mokari, Eduard Jorswieck

Abstract:

The capacity of fifth-generation (5G) vehicle-to-everything (V2X) networks poses significant challenges. To ad- dress this challenge, this paper utilizes New Radio (NR) and New Radio Unlicensed (NR-U) networks to develop a heterogeneous vehicular network (HetNet). We propose a new framework, named joint BS assignment and resource allocation (JBSRA) for mobile V2X users and also consider coexistence schemes based on flexible duty cycle (DC) mechanism for unlicensed bands. Our objective is to maximize the average throughput of vehicles while guaranteeing the WiFi users' throughput. In simulations based on deep reinforcement learning (DRL) algorithms such as deep deterministic policy gradient (DDPG) and deep Q network (DQN), our proposed framework outperforms existing solutions that rely on fixed DC or schemes without consideration of unlicensed bands.

Keywords: vehicle-to-everything (V2X), resource allocation, BS assignment, new radio (NR), new radio unlicensed (NR-U), coexistence NR-U and WiFi, deep deterministic policy gradient (DDPG), deep Q-network (DQN), joint BS assignment and resource allocation (JBSRA), duty cycle mechanism

Procedia PDF Downloads 103

Authors: Donya Ashtiani Haghighi, Amirali Baniasadi

Abstract:

Coronavirus (COVID-19) disease has spread abruptly all over the world since the end of 2019. Computed tomography (CT) scans and X-ray images are used to detect this disease. Different Deep Neural Network (DNN)-based diagnosis solutions have been developed, mainly based on Convolutional Neural Networks (CNNs), to accelerate the identification of COVID-19 cases. However, CNNs lose important information in intermediate layers and require large datasets. In this paper, Capsule Network (CapsNet) is used. Capsule Network performs better than CNNs for small datasets. Accuracy of 0.9885, f1-score of 0.9883, precision of 0.9859, recall of 0.9908, and Area Under the Curve (AUC) of 0.9948 are achieved on the Capsule-based framework with hyperparameter tuning. Moreover, different dropout rates are investigated to decrease overfitting. Accordingly, a dropout rate of 0.1 shows the best results. Finally, we remove one convolution layer and decrease the number of trainable parameters to 146,752, which is a promising result.

Keywords: capsule network, dropout, hyperparameter tuning, classification

Procedia PDF Downloads 77

Authors: Essam Al Daoud

Abstract:

Gradient boosting methods have been proven to be a very important strategy. Many successful machine learning solutions were developed using the XGBoost and its derivatives. The aim of this study is to investigate and compare the efficiency of three gradient methods. Home credit dataset is used in this work which contains 219 features and 356251 records. However, new features are generated and several techniques are used to rank and select the best features. The implementation indicates that the LightGBM is faster and more accurate than CatBoost and XGBoost using variant number of features and records.

Keywords: gradient boosting, XGBoost, LightGBM, CatBoost, home credit

Procedia PDF Downloads 171

Authors: Vahid Bayrami Rad

Abstract:

In recent years, the use of artificial intelligence techniques, tools, and methods in processing medical images and health-related applications has been highlighted and a lot of research has been done in this regard. For example, colonoscopy and diagnosis of colon lesions are some cases in which the process of diagnosis of lesions can be improved by using image processing and artificial intelligence algorithms, which help doctors a lot. Due to the lack of accurate measurements and the variety of injuries in colonoscopy images, the process of diagnosing the type of lesions is a little difficult even for expert doctors. Therefore, by using different software and image processing, doctors can be helped to increase the accuracy of their observations and ultimately improve their diagnosis. Also, by using automatic methods, the process of diagnosing the type of disease can be improved. Therefore, in this paper, a deep learning framework called KCBA is proposed to classify colonoscopy lesions which are composed of several methods such as K-means clustering, a bag of features and deep auto-encoder. Finally, according to the experimental results, the proposed method's performance in classifying colonoscopy images is depicted considering the accuracy criterion.

Keywords: colorectal cancer, colonoscopy, region of interest, narrow band imaging, texture analysis, bag of feature

Procedia PDF Downloads 57

Authors: Boris Kolev, Matt Kokan, Mohammad Deriszadeh, Farshid Bateni

Abstract:

Due to the increasing demand for real estate and the need for efficient land utilization in Greater Vancouver, developers have been increasingly considering the construction of high-rise structures with multiple below-grade parking. The temporary excavations required to allow for the construction of underground levels have recently reached up to 40 meters in depth. One of the challenges with deep excavations is the prediction of wall displacements and ground settlements due to their effect on the integrity of City utilities, infrastructure, and adjacent buildings. A large database of survey monitoring data has been collected for deep excavations in various soil conditions and shoring systems. The majority of the data collected is for tie-back anchors and shotcrete lagging systems. The data were categorized, analyzed and the results were evaluated to find a relationship between the most dominant parameters controlling the displacement, such as depth of excavation, soil properties, and the tie-back anchor loading and arrangement. For a select number of deep excavations, finite element modeling was considered for analyses. The lateral displacements from the simulation results were compared to the recorded survey monitoring data. The study concludes with a discussion and comparison of the available empirical and numerical modeling methodologies for evaluating lateral displacements in deep excavations.

Keywords: deep excavations, lateral displacements, numerical modeling, shoring walls, tieback anchors

Procedia PDF Downloads 181

Authors: Sergey Ermolin, Olga Ermolin

Abstract:

A Deep Convolution Neural Network with Triplet Loss is used to identify duplicate images in real-estate advertisements in the presence of image artifacts such as watermarking, cropping, hue/brightness adjustment, and others. The effects of batch normalization, spatial dropout, and various convergence methodologies on the resulting detection accuracy are discussed. For comparative Return-on-Investment study (per industry request), end-2-end performance is benchmarked on both Nvidia Titan GPUs and Intel’s Xeon CPUs. A new real-estate dataset from San Francisco Bay Area is used for this work. Sufficient duplicate detection accuracy is achieved to supplement other database-grounded methods of duplicate removal. The implemented method is used in a Proof-of-Concept project in the real-estate industry.

Keywords: visual recognition, convolutional neural networks, triplet loss, spatial batch normalization with dropout, duplicate removal, advertisement technologies, performance benchmarking

Procedia PDF Downloads 338

Authors: Shervin Alaei, Reza Moradi

Abstract:

Trading in financial markets is subject to risks due to their high volatilities. Here, using an LSTM neural network, and by doing some risk-based feature engineering tasks, we developed a method that can accurately predict trends of the Tehran stock exchange market index from a few days ago. Our test results have shown that the proposed method with an average prediction accuracy of more than 94% is superior to the other common machine learning algorithms. To the best of our knowledge, this is the first work incorporating deep learning and risk factors to accurately predict market trends.

Keywords: deep learning, LSTM, trend prediction, risk management, artificial neural networks

Procedia PDF Downloads 156

Authors: Saba Rahimi, Ozan Oktay, Javier Alvarez-Valle, Sujeeth Bharadwaj

Abstract:

Successful application of deep learning in medical image analysis necessitates unprecedented amounts of labeled training data. Unlike conventional 2D applications, radiological images can be three-dimensional (e.g., CT, MRI), consisting of many instances within each image. The problem is exacerbated when expert annotations are required for effective pixel-wise labeling, which incurs exorbitant labeling effort and cost. Active learning is an established research domain that aims to reduce labeling workload by prioritizing a subset of informative unlabeled examples to annotate. Our contribution is a cost-effective approach for U-Net 3D models that uses Monte Carlo sampling to analyze pixel-wise uncertainty. Experiments on the AAPM 2017 lung CT segmentation challenge dataset show that our proposed framework can achieve promising segmentation results by using only 42% of the training data.

Keywords: image segmentation, active learning, convolutional neural network, 3D U-Net

Procedia PDF Downloads 155

Authors: Yasmeen Bassas, Sandra Kuebler, Allen Riddell

Abstract:

Native language identification is one of the growing subfields in natural language processing (NLP). The task of native language identification (NLI) is mainly concerned with predicting the native language of an author’s writing in a second language. In this paper, we investigate the performance of two types of features; content-based features vs. content independent features, when they are evaluated on a different corpus (using social media data “Reddit”). In this NLI task, the predefined models are trained on one corpus (TOEFL), and then the trained models are evaluated on different data using an external corpus (Reddit). Three classifiers are used in this task; the baseline, linear SVM, and logistic regression. Results show that content-based features are more accurate and robust than content independent ones when tested within the corpus and across corpus.

Keywords: NLI, NLP, content-based features, content independent features, social media corpus, ML

Procedia PDF Downloads 137

Authors: Harold Mauricio Díaz-Vargas, Cristian Alfonso Jimenez-Castaño, David Augusto Cárdenas-Peña, Guillermo Alberto Ortiz-Gómez, Alvaro Angel Orozco-Gutierrez

Abstract:

Precise identification of the nerves is a crucial task performed by anesthesiologists for an effective Peripheral Nerve Blocking (PNB). Now, anesthesiologists use ultrasound imaging equipment to guide the PNB and detect nervous structures. However, visual identification of the nerves from ultrasound images is difficult, even for trained specialists, due to artifacts and low contrast. The recent advances in deep learning make neural networks a potential tool for accurate nerve segmentation systems, so addressing the above issues from raw data. The most widely spread U-Net network yields pixel-by-pixel segmentation by encoding the input image and decoding the attained feature vector into a semantic image. This work proposes a conditioning approach and encoder pre-training to enhance the nerve segmentation of traditional U-Nets. Conditioning is achieved by the one-hot encoding of the kind of target nerve a the network input, while the pre-training considers five well-known deep networks for image classification. The proposed approach is tested in a collection of 619 US images, where the best C-UNet architecture yields an 81% Dice coefficient, outperforming the 74% of the best traditional U-Net. Results prove that pre-trained models with the conditional approach outperform their equivalent baseline by supporting learning new features and enriching the discriminant capability of the tested networks.

Keywords: nerve segmentation, U-Net, deep learning, ultrasound imaging, peripheral nerve blocking

Procedia PDF Downloads 106

Authors: Syed Atif Hassan, Luv Mehta, Syed Asif Hassan

Abstract:

Hit song science is a field of investigation wherein machine learning techniques are applied to music tracks in order to extract such features from audio signals which can capture information that could explain the popularity of respective tracks. Record companies invest huge amounts of money into recruiting fresh talents and churning out new music each year. Gaining insight into the basis of why a song becomes popular will result in tremendous benefits for the music industry. This paper aims to extract basic musical and more advanced, acoustic features from songs while also taking into account external factors that play a role in making a particular song popular. We use a dataset derived from popular Spotify playlists divided by genre. We use ten genres (blues, classical, country, disco, hip-hop, jazz, metal, pop, reggae, rock), chosen on the basis of clear to ambiguous delineation in the typical sound of their genres. We feed these features into three different classifiers, namely, SVM with RBF kernel, a deep neural network, and a recurring neural network, to build separate predictive models and choosing the best performing model at the end. Predicting song popularity is particularly important for the music industry as it would allow record companies to produce better content for the masses resulting in a more competitive market.

Keywords: classifier, machine learning, music tracks, popularity, prediction

Procedia PDF Downloads 663

Authors: Parthiban Srinivasan

Abstract:

With recent advances in computational power and access to enough data in biosciences, artificial intelligence methods are increasingly being used in drug discovery research. These methods are essentially a series of advanced statistics based exercises that review the past to indicate the likely future. Our goal is to develop a model that accurately predicts biological activity and toxicity parameters for novel compounds. We have compiled a robust library of over 150,000 chemical compounds with different pharmacological properties from literature and public domain databases. The compounds are stored in simplified molecular-input line-entry system (SMILES), a commonly used text encoding for organic molecules. We utilize an automated process to generate an array of numerical descriptors (features) for each molecule. Redundant and irrelevant descriptors are eliminated iteratively. Our prediction engine is based on a portfolio of machine learning algorithms. We found Random Forest algorithm to be a better choice for this analysis. We captured non-linear relationship in the data and formed a prediction model with reasonable accuracy by averaging across a large number of randomized decision trees. Our next step is to apply deep neural network (DNN) algorithm to predict the biological activity and toxicity properties. We expect the DNN algorithm to give better results and improve the accuracy of the prediction. This presentation will review all these prominent machine learning and deep learning methods, our implementation protocols and discuss these techniques for their usefulness in biomedical and health informatics.

Keywords: deep learning, drug discovery, health informatics, machine learning, toxicity prediction

Procedia PDF Downloads 357

Authors: Ousmane Youme, Jean Marie Dembele, Eugene Ezin, Christophe Cambier

Abstract:

In recent years, the convolutional neural networks (CNN) architectures designed by evolution algorithms have proven to be competitive with handcrafted architectures designed by experts. However, these algorithms need a lot of computational power, which is beyond the capabilities of most researchers and engineers. To overcome this problem, we propose an evolution architecture under length constraints. It consists of two algorithms: a search length strategy to find an optimal space and a search architecture strategy based on a genetic algorithm to find the best individual in the optimal space. Our algorithms drastically reduce resource costs and also keep good performance. On the Cifar-10 dataset, our framework presents outstanding performance with an error rate of 5.12% and only 4.6 GPU a day to converge to the optimal individual -22 GPU a day less than the lowest cost automatic evolutionary algorithm in the peer competition.

Keywords: CNN architecture, genetic algorithm, evolution algorithm, length constraints

Procedia PDF Downloads 128

Authors: Tian Lan, Yixiang Wang, Wenxin Tai, Yilan Lyu, Zufeng Wu

Abstract:

The monaural speech enhancement task in the time-frequencydomain has a myriad of approaches, with the stacked con-volutional neural network (CNN) demonstrating superiorability in feature extraction and selection. However, usingstacked single convolutions method limits feature represen-tation capability and generalization ability. In order to solvethe aforementioned problem, we propose an attention-basedadaptive convolutional network that integrates the multi-scale convolutional operations into a operation-specific blockvia input dependent attention to adapt to complex auditoryscenes. In addition, we introduce a two-stage progressivelearning method to enlarge the receptive field without a dra-matic increase in computation burden. We conduct a series ofexperiments based on the TIMIT corpus, and the experimen-tal results prove that our proposed model is better than thestate-of-art models on all metrics.

Keywords: speech enhancement, adaptive convolu-tion, progressive learning, time-frequency domain

Procedia PDF Downloads 122

Authors: Degale Desta, Cheng Jian

Abstract:

The most important factor in identifying a person is their face. Even identical twins have their own distinct faces. As a result, identification and face recognition are needed to tell one person from another. A face recognition system is a verification tool used to establish a person's identity using biometrics. Nowadays, face recognition is a common technique used in a variety of applications, including home security systems, criminal identification, and phone unlock systems. This system is more secure because it only requires a facial image instead of other dependencies like a key or card. Face detection and face identification are the two phases that typically make up a human recognition system.The idea behind designing and creating a face recognition system using deep learning with Azure ML Python's OpenCV is explained in this paper. Face recognition is a task that can be accomplished using deep learning, and given the accuracy of this method, it appears to be a suitable approach. To show how accurate the suggested face recognition system is, experimental results are given in 98.46% accuracy using Fast-RCNN Performance of algorithms under different training conditions.

Keywords: deep learning, face recognition, identification, fast-RCNN

Procedia PDF Downloads 140

Authors: Donghee Noh, Seonhyeong Kim, Junhwan Choi, Heegon Kim, Sooho Jung, Keunho Park

Abstract:

In this paper, we describe an image post-processing method to enhance the performance of the parcel segmentation method using deep learning-based aerial images conducted in previous studies. The study results were evaluated using a confusion matrix, IoU, Precision, Recall, and F1-Score. In the case of the confusion matrix, it was observed that the false positive value, which is the result of misclassification, was greatly reduced as a result of image post-processing. The average IoU was 0.9688 in the image post-processing, which is higher than the deep learning result of 0.8362, and the F1-Score was also 0.9822 in the image post-processing, which was higher than the deep learning result of 0.8850. As a result of the experiment, it was found that the proposed technique positively complements the deep learning results in segmenting the parcel of interest.

Keywords: aerial image, image process, machine vision, open field smart farm, segmentation

Procedia PDF Downloads 81

Authors: Marcos Silva Tavares, Jamile Raquel Regazzo, Edson José de Souza Sardinha, Murilo Mesquita Baesso

Abstract:

Common beans are a widely cultivated and consumed legume globally, serving as a staple food for humans, especially in developing countries, due to their nutritional characteristics. Nitrogen (N) is the most limiting nutrient for productivity, and foliar analysis is crucial to ensure balanced nitrogen fertilization. Excessive N applications can cause, either isolated or cumulatively, soil and water contamination, plant toxicity, and increase their susceptibility to diseases and pests. However, the quantification of N using conventional methods is time-consuming and costly, demanding new technologies to optimize the adequate supply of N to plants. Thus, it becomes necessary to establish constant monitoring of the foliar content of this macronutrient in plants, mainly at the V4 stage, aiming at precision management of nitrogen fertilization. In this work, the objective was to evaluate the performance of a deep learning model, Resnet-50, in the classification of foliar nitrogen in common beans using RGB images. The BRS Estilo cultivar was sown in a greenhouse in a completely randomized design with four nitrogen doses (T1 = 0 kg N ha-1, T2 = 25 kg N ha-1, T3 = 75 kg N ha-1, and T4 = 100 kg N ha-1) and 12 replications. Pots with 5L capacity were used with a substrate composed of 43% soil (Neossolo Quartzarênico), 28.5% crushed sugarcane bagasse, and 28.5% cured bovine manure. The water supply of the plants was done with 5mm of water per day. The application of urea (45% N) and the acquisition of images occurred 14 and 32 days after sowing, respectively. A code developed in Matlab© R2022b was used to cut the original images into smaller blocks, originating an image bank composed of 4 folders representing the four classes and labeled as T1, T2, T3, and T4, each containing 500 images of 224x224 pixels obtained from plants cultivated under different N doses. The Matlab© R2022b software was used for the implementation and performance analysis of the model. The evaluation of the efficiency was done by a set of metrics, including accuracy (AC), F1-score (F1), specificity (SP), area under the curve (AUC), and precision (P). The ResNet-50 showed high performance in the classification of foliar N levels in common beans, with AC values of 85.6%. The F1 for classes T1, T2, T3, and T4 was 76, 72, 74, and 77%, respectively. This study revealed that the use of RGB images combined with deep learning can be a promising alternative to slow laboratory analyses, capable of optimizing the estimation of foliar N. This can allow rapid intervention by the producer to achieve higher productivity and less fertilizer waste. Future approaches are encouraged to develop mobile devices capable of handling images using deep learning for the classification of the nutritional status of plants in situ.

Keywords: convolutional neural network, residual network 50, nutritional status, artificial intelligence

Procedia PDF Downloads 19

Authors: Getaneh Berie Tarekegn, Li-Chia Tai

Abstract:

With the rapid advancement of artificial intelligence, low-power built-in sensors on Internet of Things devices, and communication technologies, location-aware services have become increasingly popular and have permeated every aspect of people’s lives. Global navigation satellite systems (GNSSs) are the default method of providing continuous positioning services for ground and aerial vehicles, as well as consumer devices (smartphones, watches, notepads, etc.). However, the environment affects satellite positioning systems, particularly indoors, in dense urban and suburban cities enclosed by skyscrapers, or when deep shadows obscure satellite signals. This is because (1) indoor environments are more complicated due to the presence of many objects surrounding them; (2) reflection within the building is highly dependent on the surrounding environment, including the positions of objects and human activity; and (3) satellite signals cannot be reached in an indoor environment, and GNSS doesn't have enough power to penetrate building walls. GPS is also highly power-hungry, which poses a severe challenge for battery-powered IoT devices. Due to these challenges, IoT applications are limited. Consequently, precise, seamless, and ubiquitous Positioning, Navigation and Timing (PNT) systems are crucial for many artificial intelligence Internet of Things (AI-IoT) applications in the era of smart cities. Their applications include traffic monitoring, emergency alarms, environmental monitoring, location-based advertising, intelligent transportation, and smart health care. This paper proposes a generative AI-based positioning scheme for large-scale wireless settings using fingerprinting techniques. In this article, we presented a semi-supervised deep convolutional generative adversarial network (S-DCGAN)-based radio map construction method for real-time device localization. We also employed a reliable signal fingerprint feature extraction method with t-distributed stochastic neighbor embedding (t-SNE), which extracts dominant features while eliminating noise from hybrid WLAN and long-term evolution (LTE) fingerprints. The proposed scheme reduced the workload of site surveying required to build the fingerprint database by up to 78.5% and significantly improved positioning accuracy. The results show that the average positioning error of GAILoc is less than 0.39 m, and more than 90% of the errors are less than 0.82 m. According to numerical results, SRCLoc improves positioning performance and reduces radio map construction costs significantly compared to traditional methods.

Keywords: location-aware services, feature extraction technique, generative adversarial network, long short-term memory, support vector machine

Procedia PDF Downloads 42