Pub Date : 2019-12-01DOI: 10.1109/IST48021.2019.9010370
R. Hamoudi, S. Hammoudeh, Arab M. Hammoudeh, S. Rawat
The discovery of tissue heterogeneity revolutionized the existing knowledge regarding the cellular, molecular, and pathophysiological mechanisms in biomedicine. Therefore, basic science investigations were redirected to encompass observation at the classical and quantum biology levels. Various approaches have been developed to investigate and capture tissue heterogeneity; however, these approaches are costly and incompatible with all types of samples. In this paper, we propose an approach to quantify heterogeneous cellular populations through combining histology and images processing techniques. In this approach, images of immunohistochemically stained sections are processed through color binning of DAB-stained cells (in brown) and non-stained cells (in blue) to select cellular clusters expressing biomarkers of interest. Subsequently, the images were converted to a binary format through threshold modification (threshold ~ 60%) in the grey scale. The cell count was extrapolated from the binary images using the particle analysis tool in ImageJ. This approach was applied to quantify the level of progesterone receptor expression levels in a breast cancer cell line sample. The results of the proposed approach were found to closely reflect those of manual counting. Through this approach, quantitative measures can be added to qualitative observation of subcellular targets expression.
{"title":"Semi-Automated Image Analysis Methodology to Investigate Intracellular Heterogeneity in Immunohistochemical Stained Sections","authors":"R. Hamoudi, S. Hammoudeh, Arab M. Hammoudeh, S. Rawat","doi":"10.1109/IST48021.2019.9010370","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010370","url":null,"abstract":"The discovery of tissue heterogeneity revolutionized the existing knowledge regarding the cellular, molecular, and pathophysiological mechanisms in biomedicine. Therefore, basic science investigations were redirected to encompass observation at the classical and quantum biology levels. Various approaches have been developed to investigate and capture tissue heterogeneity; however, these approaches are costly and incompatible with all types of samples. In this paper, we propose an approach to quantify heterogeneous cellular populations through combining histology and images processing techniques. In this approach, images of immunohistochemically stained sections are processed through color binning of DAB-stained cells (in brown) and non-stained cells (in blue) to select cellular clusters expressing biomarkers of interest. Subsequently, the images were converted to a binary format through threshold modification (threshold ~ 60%) in the grey scale. The cell count was extrapolated from the binary images using the particle analysis tool in ImageJ. This approach was applied to quantify the level of progesterone receptor expression levels in a breast cancer cell line sample. The results of the proposed approach were found to closely reflect those of manual counting. Through this approach, quantitative measures can be added to qualitative observation of subcellular targets expression.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"203 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132963330","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010466
A. Shaffie, A. Soliman, H. A. Khalifeh, F. Taher, M. Ghazal, N. Dunlap, Adel Said Elmaghraby, R. Keynton, A. El-Baz
Lung cancer leads deaths caused by cancer for both men and women worldwide, that is why creating systems for early diagnosis with machine learning algorithms and nominal user intervention is of huge importance. In this manuscript, a new system for lung nodule diagnosis, using features extracted from one computed tomography (CT) scan, is presented. This system integrates global and local features to give an implication of the nodule prior growth rate, which is the main point for diagnosis of pulmonary nodules. 3D adjustable local binary pattern and some basic geometric features are used to extract the nodule global features, and the local features are extracted using 3D convolutional neural networks (3D-CNN) because of its ability to exploit the spatial correlation of input data in an efficient way. Finally all these features are integrated using autoencoder to give a final diagnosis for the lung nodule whether benign or malignant. The system was evaluated using 727 nodules extracted from the Lung Image Database Consortium (LIDC) dataset. The proposed system diagnosis accuracy, sensitivity, and specificity were 92.20%,93.55%, and 91.20% respectively. The proposed framework demonstrated its promise as a valuable tool for lung cancer detection evidenced by its higher accuracy.
{"title":"A New System for Lung Cancer Diagnosis based on the Integration of Global and Local CT Features","authors":"A. Shaffie, A. Soliman, H. A. Khalifeh, F. Taher, M. Ghazal, N. Dunlap, Adel Said Elmaghraby, R. Keynton, A. El-Baz","doi":"10.1109/IST48021.2019.9010466","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010466","url":null,"abstract":"Lung cancer leads deaths caused by cancer for both men and women worldwide, that is why creating systems for early diagnosis with machine learning algorithms and nominal user intervention is of huge importance. In this manuscript, a new system for lung nodule diagnosis, using features extracted from one computed tomography (CT) scan, is presented. This system integrates global and local features to give an implication of the nodule prior growth rate, which is the main point for diagnosis of pulmonary nodules. 3D adjustable local binary pattern and some basic geometric features are used to extract the nodule global features, and the local features are extracted using 3D convolutional neural networks (3D-CNN) because of its ability to exploit the spatial correlation of input data in an efficient way. Finally all these features are integrated using autoencoder to give a final diagnosis for the lung nodule whether benign or malignant. The system was evaluated using 727 nodules extracted from the Lung Image Database Consortium (LIDC) dataset. The proposed system diagnosis accuracy, sensitivity, and specificity were 92.20%,93.55%, and 91.20% respectively. The proposed framework demonstrated its promise as a valuable tool for lung cancer detection evidenced by its higher accuracy.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126772033","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010565
J. Scharcanski
On behalf of the Technical and Local Committee of the 2019 IEEE International Conference on Imaging Systems and Techniques (IST 2019) and IEEE International School on Imaging, I welcome you to Abu Dhabi, UAE.
{"title":"Welcome Message from the Chairman","authors":"J. Scharcanski","doi":"10.1109/IST48021.2019.9010565","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010565","url":null,"abstract":"On behalf of the Technical and Local Committee of the 2019 IEEE International Conference on Imaging Systems and Techniques (IST 2019) and IEEE International School on Imaging, I welcome you to Abu Dhabi, UAE.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121143841","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010115
A. Ayman, Omneya Attallah, H. Shaban
Lately, Human Activity Recognition (HAR) using wearable sensors has received extensive research attention for its great use in the human health performance evaluation across several domain. HAR methods can be embedded in a smart home healthcare model to assist patients and enhance their rehabilitation process. Several types of sensors are currently used for HAR amongst them are wearable wrist sensors, which have a great ability to deliver Valuable information about the patient's grade of ability. Some recent studies have proposed HAR using Machine Learning (ML) techniques. These studies have included non-invasive wearable wrist sensors, such as Accelerometer, Magnetometer and Gyroscope. In this paper, a novel framework for HAR using ML based on sensor-fusion is proposed. Moreover, a feature selection approach to select useful features based on Random Forest (RF), Bagged Decision Tree (DT) and Support Vector Machine (SVM) classifiers is employed. The proposed framework is investigated on two publicly available datasets. Numerical results show that our framework based on sensor-fusion outperforms other methods proposed in the literature.
{"title":"An Efficient Human Activity Recognition Framework Based on Wearable IMU Wrist Sensors","authors":"A. Ayman, Omneya Attallah, H. Shaban","doi":"10.1109/IST48021.2019.9010115","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010115","url":null,"abstract":"Lately, Human Activity Recognition (HAR) using wearable sensors has received extensive research attention for its great use in the human health performance evaluation across several domain. HAR methods can be embedded in a smart home healthcare model to assist patients and enhance their rehabilitation process. Several types of sensors are currently used for HAR amongst them are wearable wrist sensors, which have a great ability to deliver Valuable information about the patient's grade of ability. Some recent studies have proposed HAR using Machine Learning (ML) techniques. These studies have included non-invasive wearable wrist sensors, such as Accelerometer, Magnetometer and Gyroscope. In this paper, a novel framework for HAR using ML based on sensor-fusion is proposed. Moreover, a feature selection approach to select useful features based on Random Forest (RF), Bagged Decision Tree (DT) and Support Vector Machine (SVM) classifiers is employed. The proposed framework is investigated on two publicly available datasets. Numerical results show that our framework based on sensor-fusion outperforms other methods proposed in the literature.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114116049","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010068
Priscilla Dinkar Moyya, Mythili Asaithambi, A. K. Ramaniharan
Breast cancer disorders are leading cause of morbidity and mortality worldwide. Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI) is the most common method of assessing the response to chemotherapy in breast cancer treatment monitoring. Radiomic features obtained from MR images have potential in reflecting the tumor biology. In this work, an attempt has been made to investigate the clinical potential of breast DCE-MRI derived radiomic features and its response to Neoadjuvant Chemotherapy (NAC). The data used in this study (10 Patients with 20 studies (Visit-1 & Visit-2) were obtained from public domain Quantitative Imaging Network (QIN) Breast DCE-MRI database. Using Mazda software, the radiomic features were extracted from whole breast region to quantify the pathological variations during visit-1 and visit-2. Totally, 176 texture and shape features were extracted and analyzed statistically using student's t test. Result shows that, the radiomic features were able to differentiate the variations in the tumor biology during visit-1 and visit-2 due to NAC. The features such as GeoW2, GeoW3, GeoW4, GeoRs, GeoRc, GeoRm, 50 percentile of histogram intensity and Theta1 were found to be statistically significant with p values ranging from 0.03 to 0.08. Hence it appears that, the radiomic features could be used as adjunct measure in reflecting the pathological response during NAC and thus this study seems to be clinically significant.
{"title":"Extraction of Radiomic Features from Breast DCE-MRI Responds to Pathological Changes in Patients During Neoadjuvant Chemotherapy Treatment","authors":"Priscilla Dinkar Moyya, Mythili Asaithambi, A. K. Ramaniharan","doi":"10.1109/IST48021.2019.9010068","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010068","url":null,"abstract":"Breast cancer disorders are leading cause of morbidity and mortality worldwide. Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI) is the most common method of assessing the response to chemotherapy in breast cancer treatment monitoring. Radiomic features obtained from MR images have potential in reflecting the tumor biology. In this work, an attempt has been made to investigate the clinical potential of breast DCE-MRI derived radiomic features and its response to Neoadjuvant Chemotherapy (NAC). The data used in this study (10 Patients with 20 studies (Visit-1 & Visit-2) were obtained from public domain Quantitative Imaging Network (QIN) Breast DCE-MRI database. Using Mazda software, the radiomic features were extracted from whole breast region to quantify the pathological variations during visit-1 and visit-2. Totally, 176 texture and shape features were extracted and analyzed statistically using student's t test. Result shows that, the radiomic features were able to differentiate the variations in the tumor biology during visit-1 and visit-2 due to NAC. The features such as GeoW2, GeoW3, GeoW4, GeoRs, GeoRc, GeoRm, 50 percentile of histogram intensity and Theta1 were found to be statistically significant with p values ranging from 0.03 to 0.08. Hence it appears that, the radiomic features could be used as adjunct measure in reflecting the pathological response during NAC and thus this study seems to be clinically significant.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116145539","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010234
Chunfang Xue, Peng Liu, Weiping Liu
This paper proposes a new video surveillance system designed for Deep Learning. The proposed system uses three steps to transfer RTSP streams to pictures for Deep Learning. First it decapsulates the streams, then decodes and converts color space & extracts frames. The proposed system has two ways to decode RTSP streams, hardware decoding and software decoding. By checking the processor's version of CPU firstly, system chooses a better way to decode. The proposed system has GPU and CPU. CPU is used to process RTSP streams, extract frames and do human-machine interaction. GPU is used for computing and analyzing the algorithms of Deep Learning. So the complex computing does not run on the CPU. The proposed system runs on Linux system and has Python interface, so it can easily connect with the models of Deep Learning. By running on multiple machines, the result shows that the proposed system can process up to 16 channels of stream. After 7*24 hours of testing on several machines, this system can run continuously without downtime and the delay time is less than 7 seconds.
{"title":"Studies on a Video Surveillance System Designed for Deep Learning","authors":"Chunfang Xue, Peng Liu, Weiping Liu","doi":"10.1109/IST48021.2019.9010234","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010234","url":null,"abstract":"This paper proposes a new video surveillance system designed for Deep Learning. The proposed system uses three steps to transfer RTSP streams to pictures for Deep Learning. First it decapsulates the streams, then decodes and converts color space & extracts frames. The proposed system has two ways to decode RTSP streams, hardware decoding and software decoding. By checking the processor's version of CPU firstly, system chooses a better way to decode. The proposed system has GPU and CPU. CPU is used to process RTSP streams, extract frames and do human-machine interaction. GPU is used for computing and analyzing the algorithms of Deep Learning. So the complex computing does not run on the CPU. The proposed system runs on Linux system and has Python interface, so it can easily connect with the models of Deep Learning. By running on multiple machines, the result shows that the proposed system can process up to 16 channels of stream. After 7*24 hours of testing on several machines, this system can run continuously without downtime and the delay time is less than 7 seconds.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115418164","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010483
W. Shalash
Decreasing road accidents rate and increasing road safety have been the major concerns for a long time as traffic accidents expose the divers, passengers, properties to danger. Driver fatigue and drowsiness are one of the most critical factors affecting road safety, especially on highways. EEG signal is one of the reliable physiological signals used to perceive driver fatigue state but wearing a multi-channel headset to acquire the EEG signal limits the EEG based systems among drivers. The current work suggested using a driver fatigue detection system using transfer learning, depending only on one EEG channel to increase system usability. The system firstly acquires the signal and passing it through preprocessing filtering then, converts it to a 2D spectrogram. Finally, the 2D spectrogram is classified with AlexNet using transfer learning to classify it either normal or fatigue state. The current study compares the accuracy of seven EEG channel to select one of them as the most accurate channel to depend on it for classification. The results show that the channels FP1 and T3 are the most effective channels to indicate the drive fatigue state. They achieved an accuracy of 90% and 91% respectively. Therefore, using only one of these channels with the modified AlexNet CNN model can result in an efficient driver fatigue detection system.
{"title":"Driver Fatigue Detection with Single EEG Channel Using Transfer Learning","authors":"W. Shalash","doi":"10.1109/IST48021.2019.9010483","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010483","url":null,"abstract":"Decreasing road accidents rate and increasing road safety have been the major concerns for a long time as traffic accidents expose the divers, passengers, properties to danger. Driver fatigue and drowsiness are one of the most critical factors affecting road safety, especially on highways. EEG signal is one of the reliable physiological signals used to perceive driver fatigue state but wearing a multi-channel headset to acquire the EEG signal limits the EEG based systems among drivers. The current work suggested using a driver fatigue detection system using transfer learning, depending only on one EEG channel to increase system usability. The system firstly acquires the signal and passing it through preprocessing filtering then, converts it to a 2D spectrogram. Finally, the 2D spectrogram is classified with AlexNet using transfer learning to classify it either normal or fatigue state. The current study compares the accuracy of seven EEG channel to select one of them as the most accurate channel to depend on it for classification. The results show that the channels FP1 and T3 are the most effective channels to indicate the drive fatigue state. They achieved an accuracy of 90% and 91% respectively. Therefore, using only one of these channels with the modified AlexNet CNN model can result in an efficient driver fatigue detection system.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"233 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115502339","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010423
Xinmeng Yang, Chaojie Zhao, Bing Chen, Maomao Zhang, Yi Li
An efficiency electrical capacitance image reconstruction method which combines fully connected neural network and U-Net structure, is put forward for the first time in electrical capacitance tomography (ECT) area in this paper. Since the target of ECT image reconstruction can also be considered as an image segmentation problem-which U-Net structure is designed for. In this paper, the Convolutional Neural Network (CNN) based U-Net structure is used to improve the quality of images reconstructed by ECT. Firstly, about 60,000 data samples with different patterns are generated by the cosimulation of COMSOL Multiphysic and MATLAB. Then a fully connected neural network (FC) is used to pre-process these samples to get initial reconstructions which are not accurate enough. Finally, U-Net structure is used to further process those pre-trained images and will output reconstructed images with both high speed and quality. The robustness, generalization and practicability ability of the U-Net structure is proved. As stated in Section2, it illustrates that U-Net structure matches properly with ECT image reconstruction problems due to its antoencoder strcture. The preliminary results show that the image reconstruction results obtained by the U-Net network are much better than that of the fully connected neural network algorithm, the traditional Linear back projection (LBP) algorithm and the Landweber iteration method.
{"title":"Big Data driven U-Net based Electrical Capacitance Image Reconstruction Algorithm","authors":"Xinmeng Yang, Chaojie Zhao, Bing Chen, Maomao Zhang, Yi Li","doi":"10.1109/IST48021.2019.9010423","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010423","url":null,"abstract":"An efficiency electrical capacitance image reconstruction method which combines fully connected neural network and U-Net structure, is put forward for the first time in electrical capacitance tomography (ECT) area in this paper. Since the target of ECT image reconstruction can also be considered as an image segmentation problem-which U-Net structure is designed for. In this paper, the Convolutional Neural Network (CNN) based U-Net structure is used to improve the quality of images reconstructed by ECT. Firstly, about 60,000 data samples with different patterns are generated by the cosimulation of COMSOL Multiphysic and MATLAB. Then a fully connected neural network (FC) is used to pre-process these samples to get initial reconstructions which are not accurate enough. Finally, U-Net structure is used to further process those pre-trained images and will output reconstructed images with both high speed and quality. The robustness, generalization and practicability ability of the U-Net structure is proved. As stated in Section2, it illustrates that U-Net structure matches properly with ECT image reconstruction problems due to its antoencoder strcture. The preliminary results show that the image reconstruction results obtained by the U-Net network are much better than that of the fully connected neural network algorithm, the traditional Linear back projection (LBP) algorithm and the Landweber iteration method.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123104885","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010139
Kyota Higa, Hideaki Sato, Soma Shiraishi, Katsumi Kikuchi, K. Iwamoto
This paper proposes a method to accurately detect anomaly from an image by combining features extracted by discriminative and generative models. Automatic anomaly detection is a key factor for reducing operation costs of visual inspection in a wide range of domains. The proposed method consists of three sub-networks. The first sub-network is convolutional neural networks as a discriminative model for extracting features to distinguish between anomaly and normal. The second subnetwork is a variational autoencoder as a generative model to extract features representing normal. The third sub-network is neural networks to discriminate between anomaly and normal on the basis of features from the discriminative and generative models. Experiments were conducted using pseudo anomalous images generated by superimposing anomaly which was manually extracted from real images. Results of the experiments show that the proposed method improves the area under the curve by 0.08-0.33 points compared with that of a conventional method. With high accuracy, automatic visual inspection systems can be implemented for reducing operation costs.
{"title":"Anomaly Detection Combining Discriminative and Generative Models","authors":"Kyota Higa, Hideaki Sato, Soma Shiraishi, Katsumi Kikuchi, K. Iwamoto","doi":"10.1109/IST48021.2019.9010139","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010139","url":null,"abstract":"This paper proposes a method to accurately detect anomaly from an image by combining features extracted by discriminative and generative models. Automatic anomaly detection is a key factor for reducing operation costs of visual inspection in a wide range of domains. The proposed method consists of three sub-networks. The first sub-network is convolutional neural networks as a discriminative model for extracting features to distinguish between anomaly and normal. The second subnetwork is a variational autoencoder as a generative model to extract features representing normal. The third sub-network is neural networks to discriminate between anomaly and normal on the basis of features from the discriminative and generative models. Experiments were conducted using pseudo anomalous images generated by superimposing anomaly which was manually extracted from real images. Results of the experiments show that the proposed method improves the area under the curve by 0.08-0.33 points compared with that of a conventional method. With high accuracy, automatic visual inspection systems can be implemented for reducing operation costs.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"375 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124687338","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 : 2019-12-01DOI: 10.1109/IST48021.2019.9010357
Mohan Kashyap Pargi, B. Setiawan, Y. Kazama
The Fast RCNN framework utilizes the region proposals generated from the RGB images in general for object classification and detection. This paper describes about the vehicle classification employing the Fast RCNN framework and utilizing the information provided from the combination of depth images and RGB images in the form of region proposals for object detection and classification. We use this underlying system architecture to perform evaluation on the Indian and Thailand vehicle traffic datasets. Overall, we achieve a mAP of 72.91% using RGB region proposals, and mAP of 73.77% using RGB combined with depth proposals, for the Indian dataset; and mAP of 80.61% on RGB region proposals, and mAP of 81.25% on RGB combined with depth region proposals, for the Thailand dataset. Our results show that RGB combined with depth region proposals mAP performance is slightly better than the region proposals generated using RGB images only. Furthermore, we provide insights on the performance of AP(Average Precision) for each vehicle on Thailand dataset and how effective region proposals generation is crucial for object detection using the FastRCNN framework.
Fast RCNN框架通常利用RGB图像生成的区域建议进行对象分类和检测。本文介绍了采用Fast RCNN框架，利用深度图像和RGB图像结合提供的信息，以区域建议的形式进行目标检测和分类的车辆分类。我们使用这个底层系统架构对印度和泰国的车辆交通数据集进行评估。总体而言，我们使用RGB区域建议实现了72.91%的mAP，使用RGB结合深度建议实现了73.77%的mAP。对于泰国数据集，RGB区域建议的mAP值为80.61%，RGB结合深度区域建议的mAP值为81.25%。我们的研究结果表明，RGB结合深度区域建议的mAP性能略好于仅使用RGB图像生成的区域建议。此外，我们提供了关于泰国数据集上每辆车的AP(平均精度)性能的见解，以及使用FastRCNN框架生成有效的区域建议对目标检测的重要性。
{"title":"Classification of different vehicles in traffic using RGB and Depth images: A Fast RCNN Approach","authors":"Mohan Kashyap Pargi, B. Setiawan, Y. Kazama","doi":"10.1109/IST48021.2019.9010357","DOIUrl":"https://doi.org/10.1109/IST48021.2019.9010357","url":null,"abstract":"The Fast RCNN framework utilizes the region proposals generated from the RGB images in general for object classification and detection. This paper describes about the vehicle classification employing the Fast RCNN framework and utilizing the information provided from the combination of depth images and RGB images in the form of region proposals for object detection and classification. We use this underlying system architecture to perform evaluation on the Indian and Thailand vehicle traffic datasets. Overall, we achieve a mAP of 72.91% using RGB region proposals, and mAP of 73.77% using RGB combined with depth proposals, for the Indian dataset; and mAP of 80.61% on RGB region proposals, and mAP of 81.25% on RGB combined with depth region proposals, for the Thailand dataset. Our results show that RGB combined with depth region proposals mAP performance is slightly better than the region proposals generated using RGB images only. Furthermore, we provide insights on the performance of AP(Average Precision) for each vehicle on Thailand dataset and how effective region proposals generation is crucial for object detection using the FastRCNN framework.","PeriodicalId":117219,"journal":{"name":"2019 IEEE International Conference on Imaging Systems and Techniques (IST)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124243746","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：