Pub Date : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954105
C. W. D. Lumoindong, Rila Mandala
As a part of the video classification task, action recognition is also known as a task with heavy computational load, with models mostly trained on devices with multiple GPUs. The pre-trained models suffer from large size and take a long time to infer the test data, especially on devices with low specifications and mobile devices. The recent development of neural networks introduces the Binarized Neural Network (BNN), which offers a solution to these problems. BNNs are trained with binary activations and weights, which reduces the computation from 32-bits to 1-bit. Theoretically, this feature can perform using 32x less memory and hardware resource compared to the conventional, full-precision neural networks. Theoretically, the conversion from full-precision CNN to BNN should result in a smaller model size and faster inference time. In this research, a binarized 3D CNN model is built using the principles of BNN and tested against the full-precision CNN. The BNN model is able to reach 78.7% train accuracy, 76.3% validation accuracy, and 79.6% inference accuracy, which means that the model is working according to the standards defined in this research.
{"title":"Binarized and Full-Precision 3D-CNN in Action Recognition","authors":"C. W. D. Lumoindong, Rila Mandala","doi":"10.1109/ICITEE56407.2022.9954105","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954105","url":null,"abstract":"As a part of the video classification task, action recognition is also known as a task with heavy computational load, with models mostly trained on devices with multiple GPUs. The pre-trained models suffer from large size and take a long time to infer the test data, especially on devices with low specifications and mobile devices. The recent development of neural networks introduces the Binarized Neural Network (BNN), which offers a solution to these problems. BNNs are trained with binary activations and weights, which reduces the computation from 32-bits to 1-bit. Theoretically, this feature can perform using 32x less memory and hardware resource compared to the conventional, full-precision neural networks. Theoretically, the conversion from full-precision CNN to BNN should result in a smaller model size and faster inference time. In this research, a binarized 3D CNN model is built using the principles of BNN and tested against the full-precision CNN. The BNN model is able to reach 78.7% train accuracy, 76.3% validation accuracy, and 79.6% inference accuracy, which means that the model is working according to the standards defined in this research.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"84 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133487390","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 : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954078
Widyawan, Muhammad Syarif, A. R. Pratama
Mobile Positioning Data (MPD) contains information on the location of the mobile phone by approximating mobile phones’ location relative to fixed infrastructures (e.g., telecommunication towers that transmit signals). While the data query is technically straightforward, obtaining this dataset requires particular permission to protect customers’ privacy. Additionally, the dataset has large volumes of data (i.e, up to 300GB per day), resulting in not many researchers holding this data source to analyze the mobility of people. In this work, we collaborate with one of the biggest telecommunication service providers in Indonesia to collect MPD and prepare the big data infrastructure. We thus analyze mobility patterns during the early phase of COVID-19 in 2020 using actual Mobile Positioning Data in five provinces in Java. We use three metrics, namely, the number of visits, averaged travel distance, and Origin-Destination matrix. The findings indicate that the social restriction in the corresponding provinces has reduced the average traveled distance of the people, but not their number of visits. That is, while the traveled distance has declined more than eight times compared to the baseline, the number of visits may rocket up, up to nine times. It indicates that people are still having shorter trips even though their regular activities (working, schooling, etc.) have been restricted. The data also show that during Ramadhan month, the government has a successful intervention in restricting people for mudik Lebaran, The number of visits dropped to below 30 visits during Ramadhan and only small spikes exist during ‘libur lebaran’.
{"title":"Mobility of Indonesian during Early Pandemic: Insights from Mobile Positioning Data","authors":"Widyawan, Muhammad Syarif, A. R. Pratama","doi":"10.1109/ICITEE56407.2022.9954078","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954078","url":null,"abstract":"Mobile Positioning Data (MPD) contains information on the location of the mobile phone by approximating mobile phones’ location relative to fixed infrastructures (e.g., telecommunication towers that transmit signals). While the data query is technically straightforward, obtaining this dataset requires particular permission to protect customers’ privacy. Additionally, the dataset has large volumes of data (i.e, up to 300GB per day), resulting in not many researchers holding this data source to analyze the mobility of people. In this work, we collaborate with one of the biggest telecommunication service providers in Indonesia to collect MPD and prepare the big data infrastructure. We thus analyze mobility patterns during the early phase of COVID-19 in 2020 using actual Mobile Positioning Data in five provinces in Java. We use three metrics, namely, the number of visits, averaged travel distance, and Origin-Destination matrix. The findings indicate that the social restriction in the corresponding provinces has reduced the average traveled distance of the people, but not their number of visits. That is, while the traveled distance has declined more than eight times compared to the baseline, the number of visits may rocket up, up to nine times. It indicates that people are still having shorter trips even though their regular activities (working, schooling, etc.) have been restricted. The data also show that during Ramadhan month, the government has a successful intervention in restricting people for mudik Lebaran, The number of visits dropped to below 30 visits during Ramadhan and only small spikes exist during ‘libur lebaran’.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123912558","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}
Police are one of the jobs that have a heavy workload. Police are more susceptible to stress as a result. Currently, the Indonesian National Police evaluates the mental health of police officers using a questionnaire. However, this questionnaire is very prone to subjectivity bias. Electroencephalography (EEG) was studied as another method for detecting stress in humans. Participants were selected through questionnaire results, labeled, and categorized into stressed and normal. Eighteen participants were involved in this experiment. They are nine normal subjects and nine stressed subjects. The EEG data was recorded on two channels, F3 and F4. Those channels are located in the prefrontal cortex and have been recognized as channels for exploring the stress mental state. Python was used to perform EEG preprocessing, including bandstop filtering, artifact and noise removal, and ICA filtering. The cleaned EEG signal is then decomposed into Alpha, Beta, and Gamma sub-bands. Power Spectral Density (PSD) is then calculated as the feature for classifying between the two classes, the normal and stress mental state. K-Nearest Neighbor (K-NN) and Support Vector Machine (SVM) were applied to obtain accuracy. K-NN and SVM produce an accuracy of 90.8% and 74.5% consecutively.
{"title":"Classifying Stress Mental State by using Power Spectral Density of Electroencephalography (EEG)","authors":"A. Wibawa, Ulfi Widya Astuti, Nophaz Hanggara Saputra, Arbintoro Mas, Yuri Pamungkas","doi":"10.1109/ICITEE56407.2022.9954069","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954069","url":null,"abstract":"Police are one of the jobs that have a heavy workload. Police are more susceptible to stress as a result. Currently, the Indonesian National Police evaluates the mental health of police officers using a questionnaire. However, this questionnaire is very prone to subjectivity bias. Electroencephalography (EEG) was studied as another method for detecting stress in humans. Participants were selected through questionnaire results, labeled, and categorized into stressed and normal. Eighteen participants were involved in this experiment. They are nine normal subjects and nine stressed subjects. The EEG data was recorded on two channels, F3 and F4. Those channels are located in the prefrontal cortex and have been recognized as channels for exploring the stress mental state. Python was used to perform EEG preprocessing, including bandstop filtering, artifact and noise removal, and ICA filtering. The cleaned EEG signal is then decomposed into Alpha, Beta, and Gamma sub-bands. Power Spectral Density (PSD) is then calculated as the feature for classifying between the two classes, the normal and stress mental state. K-Nearest Neighbor (K-NN) and Support Vector Machine (SVM) were applied to obtain accuracy. K-NN and SVM produce an accuracy of 90.8% and 74.5% consecutively.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115150760","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 : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954097
M. Bansal, Sumit Mohanty, Anju Das, Prateek Jain
Corona Virus Disease 2019 (COVID-19) has emerged as a supreme challenge for the whole world as well as India. As of now approximately 6.5 million people died in the world. However, the major setback to the world was in 2021 as a result of the second and third waves of COVID-19, which were caused by a different variation of COVID-19 than the first variant. The governments and health sectors were not aware of the subsequent possible waves due to the lack of data analysis competency and improper forecasting models. Hence finding an inflection point of this epidemic curve for COVID-19 infection and death is very imperative to understand different waves and variants instigating these waves. Similarly predicting the epidemic curve for the future is vital to make the government and the systems aware of the impending situation and make them prepare accordingly. Hence this work attempts to demonstrate conditions for finding inflection points and intervals which helps in finding the number of waves and the variants of COVID-19. Simultaneously the forecasting of the number of infections in forthcoming wave is also done using the auto-regressive integrated moving average model to identify the number of waves in India. The prediction of the two months data was compared with actual data for proper analysis.
{"title":"Analysis and Forecasting of the COVID-19 Epidemic Curve","authors":"M. Bansal, Sumit Mohanty, Anju Das, Prateek Jain","doi":"10.1109/ICITEE56407.2022.9954097","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954097","url":null,"abstract":"Corona Virus Disease 2019 (COVID-19) has emerged as a supreme challenge for the whole world as well as India. As of now approximately 6.5 million people died in the world. However, the major setback to the world was in 2021 as a result of the second and third waves of COVID-19, which were caused by a different variation of COVID-19 than the first variant. The governments and health sectors were not aware of the subsequent possible waves due to the lack of data analysis competency and improper forecasting models. Hence finding an inflection point of this epidemic curve for COVID-19 infection and death is very imperative to understand different waves and variants instigating these waves. Similarly predicting the epidemic curve for the future is vital to make the government and the systems aware of the impending situation and make them prepare accordingly. Hence this work attempts to demonstrate conditions for finding inflection points and intervals which helps in finding the number of waves and the variants of COVID-19. Simultaneously the forecasting of the number of infections in forthcoming wave is also done using the auto-regressive integrated moving average model to identify the number of waves in India. The prediction of the two months data was compared with actual data for proper analysis.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133598439","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 : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954068
Khamzul Rifki, Irfan Dwiki Bhaswara, Andi Sulasikin, B. I. Nasution, Y. Nugraha, J. Kanggrawan, M. E. Aminanto
Computer vision research has been used in daily applications, such as art, social media app filter, and face recognition. This emergence is because of the usage of the deep learning method in the computer vision domain. Deep learning research has improved many qualities of services for various applications. Starting from recommended until detection systems are now relying on deep learning models. However, currently many models require high computational processing and storage space. Implementing such an extensive network with limited resources on an embedded device or smartphone becomes more challenging. In this study, we focus on developing a model with small computational resources with high accuracy using the knowledge distillation method. We evaluate our model on the public and private datasets of receipt and non-receipt images that we gathered from Badan Pendapatan Daerah, CORD, and Kaggle dataset. After that, we compare it with the regular convolutional neural network (CNN) and pre-trained model. We discovered that knowledge distillation only uses 12% and 5% of the total weight of the CNN and the pre-trained model, respectively. As a result, we see a possibility that knowledge distillation illustrates potential outcomes as a method that could implement for automatic receipt identification in the Jakarta Super App.
{"title":"Knowledge Distillation for Automatic Receipt Identification in Jakarta Super App Platform","authors":"Khamzul Rifki, Irfan Dwiki Bhaswara, Andi Sulasikin, B. I. Nasution, Y. Nugraha, J. Kanggrawan, M. E. Aminanto","doi":"10.1109/ICITEE56407.2022.9954068","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954068","url":null,"abstract":"Computer vision research has been used in daily applications, such as art, social media app filter, and face recognition. This emergence is because of the usage of the deep learning method in the computer vision domain. Deep learning research has improved many qualities of services for various applications. Starting from recommended until detection systems are now relying on deep learning models. However, currently many models require high computational processing and storage space. Implementing such an extensive network with limited resources on an embedded device or smartphone becomes more challenging. In this study, we focus on developing a model with small computational resources with high accuracy using the knowledge distillation method. We evaluate our model on the public and private datasets of receipt and non-receipt images that we gathered from Badan Pendapatan Daerah, CORD, and Kaggle dataset. After that, we compare it with the regular convolutional neural network (CNN) and pre-trained model. We discovered that knowledge distillation only uses 12% and 5% of the total weight of the CNN and the pre-trained model, respectively. As a result, we see a possibility that knowledge distillation illustrates potential outcomes as a method that could implement for automatic receipt identification in the Jakarta Super App.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130150585","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 : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954043
Desy Noor Permata Sari, Andy Ernesto, A. F. Sahararini, Gilang Evandyano, B. I. Nasution, Y. Nugraha, J. Kanggrawan, M. E. Aminanto
Various variants of COVID-19 have entered Indonesia, such as the delta and the omicron variants. The delta variant has a higher severity than the omicron variant, but the transmission rate for the omicron variant is much faster. The government encourages citizens to get booster vaccines to reduce the effect of the delta and omicron variants. The booster vaccine produced a better effect on citizens than on people who received only the two doses. Therefore, in this study, we observe the transmission of COVID-19 and the vaccine locations on the sub-districts level using the clustering approach. The data we use are COVID-19 positive cases, died, treated, and self-isolated cases. Meanwhile, the vaccination data are $1^{text{s}text{t}}$ dose, $2^{text{n}text{d}}$ doses, stage 3 of $1^{text{s}text{t}}$ dose, and stage 3 of $2^{text{n}text{d}}$ doses. The Dunn Index and Hubert Index methods determined the best number of clusters before the clustering process. Silhouette and Davies Bouldin are used to find better clustering between Fuzzy C-Means, K-Means, and Partition Around Medoids (PAM). The results obtained from this study showed that the K-Means method was the best with the best number of clusters, namely 3. Jagakarsa and Kebon Jeruk entered high levels at the time of the delta variant due to the COVID-19 case and vaccination spread. However, Jagakarsa and Kebon Jeruk dropped to the intermediate level during the omicron variant. The benefit of this study is to help the government pay more attention to high COVID-19 cases and low vaccine distribution.
{"title":"Lessons Learned from Delta and Omicron Variants Transmissions Leveraging Clustering Approach by Sub-Districts in Jakarta","authors":"Desy Noor Permata Sari, Andy Ernesto, A. F. Sahararini, Gilang Evandyano, B. I. Nasution, Y. Nugraha, J. Kanggrawan, M. E. Aminanto","doi":"10.1109/ICITEE56407.2022.9954043","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954043","url":null,"abstract":"Various variants of COVID-19 have entered Indonesia, such as the delta and the omicron variants. The delta variant has a higher severity than the omicron variant, but the transmission rate for the omicron variant is much faster. The government encourages citizens to get booster vaccines to reduce the effect of the delta and omicron variants. The booster vaccine produced a better effect on citizens than on people who received only the two doses. Therefore, in this study, we observe the transmission of COVID-19 and the vaccine locations on the sub-districts level using the clustering approach. The data we use are COVID-19 positive cases, died, treated, and self-isolated cases. Meanwhile, the vaccination data are $1^{text{s}text{t}}$ dose, $2^{text{n}text{d}}$ doses, stage 3 of $1^{text{s}text{t}}$ dose, and stage 3 of $2^{text{n}text{d}}$ doses. The Dunn Index and Hubert Index methods determined the best number of clusters before the clustering process. Silhouette and Davies Bouldin are used to find better clustering between Fuzzy C-Means, K-Means, and Partition Around Medoids (PAM). The results obtained from this study showed that the K-Means method was the best with the best number of clusters, namely 3. Jagakarsa and Kebon Jeruk entered high levels at the time of the delta variant due to the COVID-19 case and vaccination spread. However, Jagakarsa and Kebon Jeruk dropped to the intermediate level during the omicron variant. The benefit of this study is to help the government pay more attention to high COVID-19 cases and low vaccine distribution.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"79 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121672878","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 : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954079
R. Puriyanto, O. Wahyunggoro, A. Cahyadi
An autonomous mobile robot (AMR) with wheel drive is one of the most widely developed applications today. Navigation ability, especially path planning, is one of the problems faced in AMR development. One of the path planning algorithms that is considered reliable and can be implemented in real-time is the artificial potential field (APF). However, the weakness of APF is that the robot can be trapped in the minimum locale. The local minimums commonly encountered are goal non-reachable due to nearby obstacles (GNRON) and symmetrically aligned robot obstacle goals (SAROG). This study aims to develop an APF-based path planning algorithm to solve the local minimum problem. The Gompertz function and the cone-shaped potential field are used in the Improved APF (IAPF) algorithm. The IAPF algorithm is also implemented in the kinematic equation of a wheeled mobile robot with a differential drive type. The results show that the IAPF algorithm can be implemented in a differential drive type robot. The robot can avoid obstacles in the form of SAROG and GNRON and go to the goal with an error to the goal $(d_{rg})$ less than the tolerance value of 5%.
{"title":"Implementation of Improved Artificial Potential Field Path Planning Algorithm in Differential Drive Mobile Robot","authors":"R. Puriyanto, O. Wahyunggoro, A. Cahyadi","doi":"10.1109/ICITEE56407.2022.9954079","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954079","url":null,"abstract":"An autonomous mobile robot (AMR) with wheel drive is one of the most widely developed applications today. Navigation ability, especially path planning, is one of the problems faced in AMR development. One of the path planning algorithms that is considered reliable and can be implemented in real-time is the artificial potential field (APF). However, the weakness of APF is that the robot can be trapped in the minimum locale. The local minimums commonly encountered are goal non-reachable due to nearby obstacles (GNRON) and symmetrically aligned robot obstacle goals (SAROG). This study aims to develop an APF-based path planning algorithm to solve the local minimum problem. The Gompertz function and the cone-shaped potential field are used in the Improved APF (IAPF) algorithm. The IAPF algorithm is also implemented in the kinematic equation of a wheeled mobile robot with a differential drive type. The results show that the IAPF algorithm can be implemented in a differential drive type robot. The robot can avoid obstacles in the form of SAROG and GNRON and go to the goal with an error to the goal $(d_{rg})$ less than the tolerance value of 5%.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"272 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115962900","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 : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954106
Bayu Kusuma Atmaja, I. Mustika, Risanuri Hidayat, Hajar Nimpuno Adi, Ahmad Taufiq Musaddid
The real-time detection of earthquake occurrences in a seismic wave, drawn by seismograph, is crucial for disaster mitigation. The earlier the earthquake warning, the more lives can be saved. One approach that can monitor and detect earthquake occurrence is binary classification between earthquake and noise signals. The use of deep learning models such as CNN (Convolutional Neural Network), is considered quite accurate to perform an image classification of seismograph signals. Nevertheless, the tendency to use the large CNN model is rated to have better accuracy than smaller models. In fact, the disadvantage of utilizing large model is the inference time and the deployment of a large model to obtain real-time inference is more costly than the smaller model. This paper aims to reduce the size of a CNN model (Resnet50) by pruning the unnecessary filters and neuron on the model architecture without sacrificing the accuracy. The task of the model was to classify two classes (earthquake and noise) of spectrogram images, the dataset is STEAD (Stanford Earthquake Dataset). To prioritize which filter or neuron to be eliminated, L2-norm was calculated on each filter or neuron weights. We assumed that a filter or neuron with the lowest L2-norm had the least significant role in the model. By pruning 90% of the filter and neuron of the model and retraining the pruned model, the inference time was improved from 22. 45ms to 3. 6ms (on NVIDIA GTX 1050) per image with the accuracy of 99.405%.
{"title":"Development of CNN Pruning Method for Earthquake Signal Imagery Classification","authors":"Bayu Kusuma Atmaja, I. Mustika, Risanuri Hidayat, Hajar Nimpuno Adi, Ahmad Taufiq Musaddid","doi":"10.1109/ICITEE56407.2022.9954106","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954106","url":null,"abstract":"The real-time detection of earthquake occurrences in a seismic wave, drawn by seismograph, is crucial for disaster mitigation. The earlier the earthquake warning, the more lives can be saved. One approach that can monitor and detect earthquake occurrence is binary classification between earthquake and noise signals. The use of deep learning models such as CNN (Convolutional Neural Network), is considered quite accurate to perform an image classification of seismograph signals. Nevertheless, the tendency to use the large CNN model is rated to have better accuracy than smaller models. In fact, the disadvantage of utilizing large model is the inference time and the deployment of a large model to obtain real-time inference is more costly than the smaller model. This paper aims to reduce the size of a CNN model (Resnet50) by pruning the unnecessary filters and neuron on the model architecture without sacrificing the accuracy. The task of the model was to classify two classes (earthquake and noise) of spectrogram images, the dataset is STEAD (Stanford Earthquake Dataset). To prioritize which filter or neuron to be eliminated, L2-norm was calculated on each filter or neuron weights. We assumed that a filter or neuron with the lowest L2-norm had the least significant role in the model. By pruning 90% of the filter and neuron of the model and retraining the pruned model, the inference time was improved from 22. 45ms to 3. 6ms (on NVIDIA GTX 1050) per image with the accuracy of 99.405%.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115275307","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 : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954076
Bin Wang, Ronaldo Juanatas, Jasmin D. Niguidula
In order to avoid the problems of low accuracy of wheel hub recognition and classification and excessive dependence on template image quality in the process of automatic production of the automobile wheel hub, a vehicle wheel hub recognition method based on Histogram of Oriented Gradient (HOG) and Support Vector Machine (SVM) is proposed. Firstly, the wheel hub images under three different lighting conditions are collected, and the wheel hub images are processed in grayscale; Secondly, the positive and negative samples are made, and hog features are extracted, respectively; Finally, the extracted hog features are trained by SVM classifier, and the trained target classifier is used to recognize the wheel hub photos under three different lighting conditions. The experimental results show that this method has higher recognition accuracy than the traditional template matching method under different lighting conditions.
{"title":"Vehicle Wheel Hub Recognition Method Based on HOG Feature Extraction and SVM Classifier","authors":"Bin Wang, Ronaldo Juanatas, Jasmin D. Niguidula","doi":"10.1109/ICITEE56407.2022.9954076","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954076","url":null,"abstract":"In order to avoid the problems of low accuracy of wheel hub recognition and classification and excessive dependence on template image quality in the process of automatic production of the automobile wheel hub, a vehicle wheel hub recognition method based on Histogram of Oriented Gradient (HOG) and Support Vector Machine (SVM) is proposed. Firstly, the wheel hub images under three different lighting conditions are collected, and the wheel hub images are processed in grayscale; Secondly, the positive and negative samples are made, and hog features are extracted, respectively; Finally, the extracted hog features are trained by SVM classifier, and the trained target classifier is used to recognize the wheel hub photos under three different lighting conditions. The experimental results show that this method has higher recognition accuracy than the traditional template matching method under different lighting conditions.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128343418","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 : 2022-10-18DOI: 10.1109/ICITEE56407.2022.9954080
Alvian Tedy Aditya, R. Sigit, B. S. B. Dewantara
This paper proposes the development of a login system to a Healthcare Kiosk using facial images. The use of the face as an example of a unique biometric system other than fingerprint and iris is considered better than conventional systems using RFID cards that are prone to being lost or left behind, or passwords that are often forgotten. In this paper, we propose the use of faces as input for the login system to a healthcare kiosk by utilizing deep learning technology. We tested four types of Convolutional Neural Network (CNN) architectures such as VGG16, ResNet50, Xception and MobileNet. In the accuracy testing process, VGG16 got a total accuracy of 100% but still showed the wrong class during realtime detection testing, ResNet50 got a total accuracy of 99.531% and was able to show the correct class during realtime detection testing, Xception got a total accuracy of 80.018% but still shows the wrong class when testing realtime detection, and MobileNet gets a total accuracy of 92.934%.
{"title":"Face Recognition Using Deep Learning as User Login on Healthcare Kiosk","authors":"Alvian Tedy Aditya, R. Sigit, B. S. B. Dewantara","doi":"10.1109/ICITEE56407.2022.9954080","DOIUrl":"https://doi.org/10.1109/ICITEE56407.2022.9954080","url":null,"abstract":"This paper proposes the development of a login system to a Healthcare Kiosk using facial images. The use of the face as an example of a unique biometric system other than fingerprint and iris is considered better than conventional systems using RFID cards that are prone to being lost or left behind, or passwords that are often forgotten. In this paper, we propose the use of faces as input for the login system to a healthcare kiosk by utilizing deep learning technology. We tested four types of Convolutional Neural Network (CNN) architectures such as VGG16, ResNet50, Xception and MobileNet. In the accuracy testing process, VGG16 got a total accuracy of 100% but still showed the wrong class during realtime detection testing, ResNet50 got a total accuracy of 99.531% and was able to show the correct class during realtime detection testing, Xception got a total accuracy of 80.018% but still shows the wrong class when testing realtime detection, and MobileNet gets a total accuracy of 92.934%.","PeriodicalId":246279,"journal":{"name":"2022 14th International Conference on Information Technology and Electrical Engineering (ICITEE)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122881624","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：