Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9558946
Melike Dokgöz, Y. Yaslan
With the increase of conventional vehicles and carbon emission from them boosted the need for electrical vehicles (EV). One of the major components of the EVs are their batteries and the commercialization of EVs are affected by their battery technology and performance. It is also obvious that the range of an EV is mainly affected by the lifetime of its battery. Estimation of the battery cycle life in the early cycles is one of the most important challenges for maximization of the EVs range. Charge-discharge cycles affect battery lifetime of the EV which also made the estimation of battery life cycle a matter of interest. In this study, different machine learning models are applied to predict the lifecycle of a battery at early stages of usage. Detailed experiments have been performed to analyze the prediction accuracy at early cycle numbers. Experimental results show that the error rate in cycle life estimation decreased from 9.2 to 2.4% using Adaptive Boosting method.
{"title":"A Comparison of Machine Learning Algorithms on Lithium-ion Battery Cycle Life Prediction","authors":"Melike Dokgöz, Y. Yaslan","doi":"10.1109/UBMK52708.2021.9558946","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9558946","url":null,"abstract":"With the increase of conventional vehicles and carbon emission from them boosted the need for electrical vehicles (EV). One of the major components of the EVs are their batteries and the commercialization of EVs are affected by their battery technology and performance. It is also obvious that the range of an EV is mainly affected by the lifetime of its battery. Estimation of the battery cycle life in the early cycles is one of the most important challenges for maximization of the EVs range. Charge-discharge cycles affect battery lifetime of the EV which also made the estimation of battery life cycle a matter of interest. In this study, different machine learning models are applied to predict the lifecycle of a battery at early stages of usage. Detailed experiments have been performed to analyze the prediction accuracy at early cycle numbers. Experimental results show that the error rate in cycle life estimation decreased from 9.2 to 2.4% using Adaptive Boosting method.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125548174","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9559004
Marwan S. Jameel, Oguz Yayla
Implementing blockchain systems with less amount of decentralized computation power is a challenging task. One of the most essential solution in scaling the blockchain is selection of a trusted committee (TC). We intend to develop and apply a new method for selecting committee members depending on their reputation via particle swarm optimization (PSO) based update of the behavior of nodes. Indeed, this is the same concept as accumulating participants in the blockchain throughout the previous rounds of the consensus procedure.
{"title":"PSO based Blockchain Committee Member Selection","authors":"Marwan S. Jameel, Oguz Yayla","doi":"10.1109/UBMK52708.2021.9559004","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9559004","url":null,"abstract":"Implementing blockchain systems with less amount of decentralized computation power is a challenging task. One of the most essential solution in scaling the blockchain is selection of a trusted committee (TC). We intend to develop and apply a new method for selecting committee members depending on their reputation via particle swarm optimization (PSO) based update of the behavior of nodes. Indeed, this is the same concept as accumulating participants in the blockchain throughout the previous rounds of the consensus procedure.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"93 3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129390336","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9559032
Cavide Balkı Gemirter, Çağatay Şenturca, S. Baydere
MQTT, AMQP and HTTP are messaging protocols that are commonly used for communicating with resource-constrained IoT devices. HTTP is the standard reference protocol for the REST transportation based on the request/response model, whereas both AMQP and MQTT are message-oriented protocols that use the publish/subscribe model. Message-oriented protocols enhance some of the shortcomings of the complex HTTP protocol by using asynchronous communication, changing the design from a document-centric to a data-centric approach and decreasing the header and message sizes. Although significant technical detail is present on these protocols, their real-time performance is insufficiently elaborated. In this paper, we present an experimental evaluation of these protocols conducted in a homogeneous IoT testbed using a real-time Smart City public data set. We provide the behavioral differences between messaging-based protocols and the REST-based HTTP protocol in terms of message latency and CPU usage for varying traffic loads and message sizes. The results showed that MQTT and AMQP are four times faster than HTTP protocol when comparing the message sent latencies. HTTP uses four times more CPU than the AMQP and MQTT protocols. In summary, message-oriented protocols give more stable and improved results as compared to the REST model-based HTTP protocol for all evaluation scenarios.
{"title":"A Comparative Evaluation of AMQP, MQTT and HTTP Protocols Using Real-Time Public Smart City Data","authors":"Cavide Balkı Gemirter, Çağatay Şenturca, S. Baydere","doi":"10.1109/UBMK52708.2021.9559032","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9559032","url":null,"abstract":"MQTT, AMQP and HTTP are messaging protocols that are commonly used for communicating with resource-constrained IoT devices. HTTP is the standard reference protocol for the REST transportation based on the request/response model, whereas both AMQP and MQTT are message-oriented protocols that use the publish/subscribe model. Message-oriented protocols enhance some of the shortcomings of the complex HTTP protocol by using asynchronous communication, changing the design from a document-centric to a data-centric approach and decreasing the header and message sizes. Although significant technical detail is present on these protocols, their real-time performance is insufficiently elaborated. In this paper, we present an experimental evaluation of these protocols conducted in a homogeneous IoT testbed using a real-time Smart City public data set. We provide the behavioral differences between messaging-based protocols and the REST-based HTTP protocol in terms of message latency and CPU usage for varying traffic loads and message sizes. The results showed that MQTT and AMQP are four times faster than HTTP protocol when comparing the message sent latencies. HTTP uses four times more CPU than the AMQP and MQTT protocols. In summary, message-oriented protocols give more stable and improved results as compared to the REST model-based HTTP protocol for all evaluation scenarios.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121945733","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9558949
A. Okutan, Can Eyüpoğlu
Cyber attacks are carried out by unauthorized access to a system connected to an internet network anywhere in the world or by taking control of this system. These attacks continue to increase exponentially with the introduction of artificial intelligence technology, which has been used effectively in the field of science and engineering. With the use of artificial intelligence in the cyber field, conventional cyber attacks have begun to emerge as smart cyber attacks. Software development with traditional algorithms is insufficient to defend against attacks. In this study, studies in the literature on artificial intelligence methods used in the field of cyber security were examined. On the subjects discussed from different perspectives, solutions in terms of defense and attack, the applications used, the contexts between different disciplines, analyzes and examples on the basis of countries will be presented.
{"title":"A Review on Artificial Intelligence and Cyber Security","authors":"A. Okutan, Can Eyüpoğlu","doi":"10.1109/UBMK52708.2021.9558949","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9558949","url":null,"abstract":"Cyber attacks are carried out by unauthorized access to a system connected to an internet network anywhere in the world or by taking control of this system. These attacks continue to increase exponentially with the introduction of artificial intelligence technology, which has been used effectively in the field of science and engineering. With the use of artificial intelligence in the cyber field, conventional cyber attacks have begun to emerge as smart cyber attacks. Software development with traditional algorithms is insufficient to defend against attacks. In this study, studies in the literature on artificial intelligence methods used in the field of cyber security were examined. On the subjects discussed from different perspectives, solutions in terms of defense and attack, the applications used, the contexts between different disciplines, analyzes and examples on the basis of countries will be presented.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"97 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127027758","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9558983
Oğuz Emre Kural, Durmuş Özkan Şahin, S. Akleylek, E. Kılıç, Murat Ömüral
In this study, the Apk2Img4AndMal framework, which provides information about the application without the need for static or dynamic attributes, is recommended. The proposed framework reads APK files in binary format and converts them to grayscale images. In the classification phase of the framework, the convolutional neural network (CNN) is used, which gives successful results in image classification. In this way, the required features are obtained through a CNN. Therefore, there is also no feature extraction phase as other dynamic or static analysis-based frameworks. This property is the most important advantage of the Apk2Img4AndMal framework. The proposed framework is tested with 24588 Android malware and 3000 benign applications. The highest performance achieved in the study is up to 94%, according to the accuracy metric.
{"title":"Apk2Img4AndMal: Android Malware Detection Framework Based on Convolutional Neural Network","authors":"Oğuz Emre Kural, Durmuş Özkan Şahin, S. Akleylek, E. Kılıç, Murat Ömüral","doi":"10.1109/UBMK52708.2021.9558983","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9558983","url":null,"abstract":"In this study, the Apk2Img4AndMal framework, which provides information about the application without the need for static or dynamic attributes, is recommended. The proposed framework reads APK files in binary format and converts them to grayscale images. In the classification phase of the framework, the convolutional neural network (CNN) is used, which gives successful results in image classification. In this way, the required features are obtained through a CNN. Therefore, there is also no feature extraction phase as other dynamic or static analysis-based frameworks. This property is the most important advantage of the Apk2Img4AndMal framework. The proposed framework is tested with 24588 Android malware and 3000 benign applications. The highest performance achieved in the study is up to 94%, according to the accuracy metric.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"50 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124099710","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9558978
Şeyda Melis Türkkahraman, Dindar Öz
The quadratic assignment problem (QAP) is a well-known optimization problem that has many applications in various engineering areas. Due to its NP-hard nature, rather than the exact methods, heuristic and metaheuristic approaches are commonly adapted. In this study, we propose an improved hybrid genetic algorithm which mainly combines a greedy heuristic, and a simulated annealing algorithm with the classical genetic algorithm. We test our algorithm on the well-known benchmark for the QAP and compare the results with four different algorithms: a greedy algorithm, simulated annealing algorithm (SA), demon algorithm (DA), and a classical genetic algorithm (GA). The results of the experiments validate that our hybridization significantly improves the performance of the algorithms comparing to their standalone executions.
{"title":"An Improved Hybrid Genetic Algorithm for the Quadratic Assignment Problem","authors":"Şeyda Melis Türkkahraman, Dindar Öz","doi":"10.1109/UBMK52708.2021.9558978","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9558978","url":null,"abstract":"The quadratic assignment problem (QAP) is a well-known optimization problem that has many applications in various engineering areas. Due to its NP-hard nature, rather than the exact methods, heuristic and metaheuristic approaches are commonly adapted. In this study, we propose an improved hybrid genetic algorithm which mainly combines a greedy heuristic, and a simulated annealing algorithm with the classical genetic algorithm. We test our algorithm on the well-known benchmark for the QAP and compare the results with four different algorithms: a greedy algorithm, simulated annealing algorithm (SA), demon algorithm (DA), and a classical genetic algorithm (GA). The results of the experiments validate that our hybridization significantly improves the performance of the algorithms comparing to their standalone executions.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"53 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127568034","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9559008
Nurşah Çevi̇k, S. Akleylek, Kadir Koc
Nowadays, many companies use biometric technologies for the security of critical systems as well as username-password methods. In the literature, biometric systems are the most commonly used systems among the two-factor authentication systems. There are two different approaches to biometric systems: physical and behavioral biometric systems. In the last decade, the accuracy of behavioral biometric systems has significantly increased with the use of machine learning methods in these systems. For this reason, the usage areas of the studies in this field have expanded. In this study, we focus on keystroke dynamics based on behavioral methods. Firstly, we make a web application to collect keystroke data from 54 employees in a company. Then, we use the benchmark database and our database to train-test machine learning algorithms, which have the highest accuracy in this field in the literature. Among them, tree-based algorithms have the highest accuracy score, with an average of 0.94.
{"title":"Keystroke Dynamics Based Authentication System","authors":"Nurşah Çevi̇k, S. Akleylek, Kadir Koc","doi":"10.1109/UBMK52708.2021.9559008","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9559008","url":null,"abstract":"Nowadays, many companies use biometric technologies for the security of critical systems as well as username-password methods. In the literature, biometric systems are the most commonly used systems among the two-factor authentication systems. There are two different approaches to biometric systems: physical and behavioral biometric systems. In the last decade, the accuracy of behavioral biometric systems has significantly increased with the use of machine learning methods in these systems. For this reason, the usage areas of the studies in this field have expanded. In this study, we focus on keystroke dynamics based on behavioral methods. Firstly, we make a web application to collect keystroke data from 54 employees in a company. Then, we use the benchmark database and our database to train-test machine learning algorithms, which have the highest accuracy in this field in the literature. Among them, tree-based algorithms have the highest accuracy score, with an average of 0.94.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127461707","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9558941
M. Nergiz
The automated weed detection is an important research field in terms of agricultural productivity and economy. This study aims to apply RepVGG which is a new deep learning architecture developed on PyTorch framework and has promising results when trained and tested on ImageNet1K dataset. 920 images of the small sized Dandelion Images dataset is used for this study. Pretrained vanilla, pretrained and dropout regularized, squeeze and excitation block added and spatial attention block added versions of RepVGG are tested on the dataset. VGG16 method is also applied to the dataset and the results of the MobileNetV2 method is taken from the Kaggle Competition to get an insight about the baseline results of the classical state of the art models. The proposed RepVGG modifications could not outperform the state of the art methods on this dataset but the effect of the modifications are deeply analyzed and the best configuration is obtained by Squeeze and Excitation block added RepVGG-A0 architecture which is trained from scratch for 5 epochs and provided results of 0,875, 0,665, 0,89 and 0,74 for Accuracy, Recall, Precision and F1 metrics respectively.
{"title":"Analysis of RepVGG on Small Sized Dandelion Images Dataset in terms of Transfer Learning, Regularization, Spatial Attention as well as Squeeze and Excitation Blocks","authors":"M. Nergiz","doi":"10.1109/UBMK52708.2021.9558941","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9558941","url":null,"abstract":"The automated weed detection is an important research field in terms of agricultural productivity and economy. This study aims to apply RepVGG which is a new deep learning architecture developed on PyTorch framework and has promising results when trained and tested on ImageNet1K dataset. 920 images of the small sized Dandelion Images dataset is used for this study. Pretrained vanilla, pretrained and dropout regularized, squeeze and excitation block added and spatial attention block added versions of RepVGG are tested on the dataset. VGG16 method is also applied to the dataset and the results of the MobileNetV2 method is taken from the Kaggle Competition to get an insight about the baseline results of the classical state of the art models. The proposed RepVGG modifications could not outperform the state of the art methods on this dataset but the effect of the modifications are deeply analyzed and the best configuration is obtained by Squeeze and Excitation block added RepVGG-A0 architecture which is trained from scratch for 5 epochs and provided results of 0,875, 0,665, 0,89 and 0,74 for Accuracy, Recall, Precision and F1 metrics respectively.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130812776","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9558896
Ensar Arif Sağbaş, Serdar Çorukoğlu, Serkan Balli
Stress is an important problem to deal with in today’s society. Thanks to the built-in sensors of smartphones, many operations can be performed unobtrusively. Accordingly, smartphones are among the indispensable data sources of research subjects. In this study, the problem of stress classification was discussed with the data obtained from smartphones. Sensor data were collected to examine users’ writing behavior. The features extracted from the obtained raw data were classified with the random subspace-based structures and their performances were compared. As a result of the study, it was observed that the stacking method showed promising results.
{"title":"A Random Subspace Based Approach for Stress Classification from Smartphone Data","authors":"Ensar Arif Sağbaş, Serdar Çorukoğlu, Serkan Balli","doi":"10.1109/UBMK52708.2021.9558896","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9558896","url":null,"abstract":"Stress is an important problem to deal with in today’s society. Thanks to the built-in sensors of smartphones, many operations can be performed unobtrusively. Accordingly, smartphones are among the indispensable data sources of research subjects. In this study, the problem of stress classification was discussed with the data obtained from smartphones. Sensor data were collected to examine users’ writing behavior. The features extracted from the obtained raw data were classified with the random subspace-based structures and their performances were compared. As a result of the study, it was observed that the stacking method showed promising results.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"94 4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131025747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-09-15DOI: 10.1109/UBMK52708.2021.9558899
Oguzhan Can, Sezai Burak Kantarci, Gozde Unal
DL architectures rely on extensive usage on powerful computer systems to operate in real-time. Therefore, cooperative and constructive optimizations should be made in both architecture and software parts of the related DL system. In this work, input system of the YOLO architecture is modified to accept several sources at the same time with two effective methods to increase the efficiency of the hardware system. First method is to design a scheduler which will allow YOLO architecture to process several input sources sequentially, allowing the architecture to use its full potential. Second method is to design a preprocessing algorithm to combine 4 or 9 input sources in a single input source as a 2x2 or 3x3 image matrix. In this way, YOLO architecture processes four or nine times more images in the same time, increasing its practical frame per second (FPS) value by four or nine folds. Experiment results on our machine show that the used YOLO architecture can process 3 input sources at the same time with only minimal loss of accuracy of 0.002 in terms of Mean Average Precision (mAP) while using the proposed scheduler. Additionally, using 4 inputs combined increases the practical FPS value from 31 to 120 and using 9 inputs increases the practical FPS value from 13 to 108, all while decreasing the mAP value by only 0.008 for 4 inputs and by only 0.034 for 9 inputs. Considering the obtained FPS values and achieved hardware efficiency, these minimal losses of mAP are easily acceptable.
{"title":"A New Approach to Use Modern Object Detection Methods More Efficiently on CCTV Systems","authors":"Oguzhan Can, Sezai Burak Kantarci, Gozde Unal","doi":"10.1109/UBMK52708.2021.9558899","DOIUrl":"https://doi.org/10.1109/UBMK52708.2021.9558899","url":null,"abstract":"DL architectures rely on extensive usage on powerful computer systems to operate in real-time. Therefore, cooperative and constructive optimizations should be made in both architecture and software parts of the related DL system. In this work, input system of the YOLO architecture is modified to accept several sources at the same time with two effective methods to increase the efficiency of the hardware system. First method is to design a scheduler which will allow YOLO architecture to process several input sources sequentially, allowing the architecture to use its full potential. Second method is to design a preprocessing algorithm to combine 4 or 9 input sources in a single input source as a 2x2 or 3x3 image matrix. In this way, YOLO architecture processes four or nine times more images in the same time, increasing its practical frame per second (FPS) value by four or nine folds. Experiment results on our machine show that the used YOLO architecture can process 3 input sources at the same time with only minimal loss of accuracy of 0.002 in terms of Mean Average Precision (mAP) while using the proposed scheduler. Additionally, using 4 inputs combined increases the practical FPS value from 31 to 120 and using 9 inputs increases the practical FPS value from 13 to 108, all while decreasing the mAP value by only 0.008 for 4 inputs and by only 0.034 for 9 inputs. Considering the obtained FPS values and achieved hardware efficiency, these minimal losses of mAP are easily acceptable.","PeriodicalId":106516,"journal":{"name":"2021 6th International Conference on Computer Science and Engineering (UBMK)","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128314310","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：