Pub Date : 2022-12-05DOI: 10.1109/IPAS55744.2022.10053020
Christian Mai, Jesper Liniger, A. Jensen, H. Sørensen, Simon Pedersen
Marine growth on submerged structures causes additional mechanical loads from drag and mass increases. In order to ensure structural integrity, regular inspection and cleaning procedures are carried out on the submerged structures, most commonly using remote-operated vehicles (ROVs). Often, the measurement methodology in these inspections is spot-checks using simple mechanical gauges, which yield a rough estimate of marine growth thickness. Expanding on this method, in order to optimize these inspection and cleaning procedures, modern methods for 3D surface measurement can be applied to increase inspection quality and ensure that superfluous cleaning is not carried out. This work investigates three state-of-the-art sensor technologies: a Time-of-Flight depth camera based on modulated visible blue laser illumination, a commercial stereo-vision solution based on visible-light sensors, and high-frequency imaging sonar. The sensors' performance has been compared in a laboratory environment to assess their suitability for use as a measurement device for marine-growth measurement in terms of accuracy, resolution, and noise/artifacts. It is concluded that the measurement fidelity of all evaluated sensors shows promise for the application, pending future evaluation in a real-world test.
{"title":"Experimental Investigation of Non-contact 3D Sensors for Marine-growth Cleaning Operations","authors":"Christian Mai, Jesper Liniger, A. Jensen, H. Sørensen, Simon Pedersen","doi":"10.1109/IPAS55744.2022.10053020","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10053020","url":null,"abstract":"Marine growth on submerged structures causes additional mechanical loads from drag and mass increases. In order to ensure structural integrity, regular inspection and cleaning procedures are carried out on the submerged structures, most commonly using remote-operated vehicles (ROVs). Often, the measurement methodology in these inspections is spot-checks using simple mechanical gauges, which yield a rough estimate of marine growth thickness. Expanding on this method, in order to optimize these inspection and cleaning procedures, modern methods for 3D surface measurement can be applied to increase inspection quality and ensure that superfluous cleaning is not carried out. This work investigates three state-of-the-art sensor technologies: a Time-of-Flight depth camera based on modulated visible blue laser illumination, a commercial stereo-vision solution based on visible-light sensors, and high-frequency imaging sonar. The sensors' performance has been compared in a laboratory environment to assess their suitability for use as a measurement device for marine-growth measurement in terms of accuracy, resolution, and noise/artifacts. It is concluded that the measurement fidelity of all evaluated sensors shows promise for the application, pending future evaluation in a real-world test.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116431914","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 : 2022-12-05DOI: 10.1109/IPAS55744.2022.10052903
A. Panthakkan, M. AnzarS., S. Al-Mansoori, Hussain Al-Ahmad
Using DWT-SVD and SHA3 Hash function, this research aims to develop an ownership protection and image authentication technique that embeds the watermark information and hash authentication key in a hybrid domain. The experiment was conducted with multispectral images from the KhalifaSat. The Performance of the proposed method is evaluated using wavelet domain signal to noise ratio (WSNR), structural similarity index measurement (SSIM) and peak signal to noise ratio (PSNR). To analyse the efficacy of the recovered watermark, two metrics are used: Normalized Correlation (NC) and Image Quality Index (IQI). The method presented is robust against many intended and unintended attacks. Without sacrificing transparency, our proposed watermarking approach meets the objectives of imperceptibility and robustness. It accurately detects the manipulated locations on the satellite image and is sensitive to even small changes.
{"title":"Hybrid Watermarking Algorithm to Protect and Authenticate KhalifaSat Imagery using DWT-SVD and SHA3 Hash Key","authors":"A. Panthakkan, M. AnzarS., S. Al-Mansoori, Hussain Al-Ahmad","doi":"10.1109/IPAS55744.2022.10052903","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052903","url":null,"abstract":"Using DWT-SVD and SHA3 Hash function, this research aims to develop an ownership protection and image authentication technique that embeds the watermark information and hash authentication key in a hybrid domain. The experiment was conducted with multispectral images from the KhalifaSat. The Performance of the proposed method is evaluated using wavelet domain signal to noise ratio (WSNR), structural similarity index measurement (SSIM) and peak signal to noise ratio (PSNR). To analyse the efficacy of the recovered watermark, two metrics are used: Normalized Correlation (NC) and Image Quality Index (IQI). The method presented is robust against many intended and unintended attacks. Without sacrificing transparency, our proposed watermarking approach meets the objectives of imperceptibility and robustness. It accurately detects the manipulated locations on the satellite image and is sensitive to even small changes.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"Five 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130784821","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 : 2022-12-05DOI: 10.1109/IPAS55744.2022.10052988
L. Kuncheva, Francis Williams, Samuel L. Hennessey, Juan José Rodríguez Diez
While there are multiple sources of annotated images and videos for human and vehicle re-identification, databases for individual animal recognition are still in demand. We present a database containing five annotated video clips each containing between 9 and 27 identities. The overall number of individual animals is 20,490, and the total number of classes is 93. The database can be used for testing novel methods for animal re-identification, object detection and tracking. The main challenge of the database is that multiple animals are present in the same video frame, leading to problems with occlusion and noisy, cluttered bounding boxes. To set-up a benchmark on individual animal recognition, we trained and tested 26 classification methods for the five videos and three feature representations. We also report results with state-of-the-art deep learning methods for object detection (MMDet) and tracking (Uni-Track).
{"title":"A Benchmark Database for Animal Re-Identification and Tracking","authors":"L. Kuncheva, Francis Williams, Samuel L. Hennessey, Juan José Rodríguez Diez","doi":"10.1109/IPAS55744.2022.10052988","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052988","url":null,"abstract":"While there are multiple sources of annotated images and videos for human and vehicle re-identification, databases for individual animal recognition are still in demand. We present a database containing five annotated video clips each containing between 9 and 27 identities. The overall number of individual animals is 20,490, and the total number of classes is 93. The database can be used for testing novel methods for animal re-identification, object detection and tracking. The main challenge of the database is that multiple animals are present in the same video frame, leading to problems with occlusion and noisy, cluttered bounding boxes. To set-up a benchmark on individual animal recognition, we trained and tested 26 classification methods for the five videos and three feature representations. We also report results with state-of-the-art deep learning methods for object detection (MMDet) and tracking (Uni-Track).","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134222954","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 : 2022-12-05DOI: 10.1109/ipas55744.2022.10052811
{"title":"Panel discussion I","authors":"","doi":"10.1109/ipas55744.2022.10052811","DOIUrl":"https://doi.org/10.1109/ipas55744.2022.10052811","url":null,"abstract":"","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130974087","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 : 2022-12-05DOI: 10.1109/IPAS55744.2022.10052880
Faizan Munawar, J. Donovan, Etain Kiely, Konrad Mulrennan
Monitoring the control of persons with type 1 diabetes based on their history of blood glucose levels is essential for self-management. Persons with diabetes must keep their blood glucose levels in a very narrow glycaemic region (70–180 mg/dl) to avoid hypoglycaemia and hyperglycaemia. An extended period of time in the hypoglycaemic or hyperglycaemic region can lead to short-term and long-term complications, respectively. Many measures have been proposed for the management of diabetes, such as the Glucose Management Indicator (GMI) and the Average Daily Risk Range (ADRR). A major drawback of these measures is that they only address acute (ADRR) or chronic (GMI) complications and provide no information on the trend. This paper proposes a rolling window to calculate ADRR and GMI. Calculating ADRR and GMI using a rolling window results in new data, which provide information on the efficacy of self-management of an individual and their risk trend. Use of a rolling window for the risk analysis provides novel information about the glycaemic variability and can be used for improved personal diabetes management plans. Furthermore, ADRR and GMI are combined to propose four new risk levels, which represents the lowest to the highest probable risk of complications. The analysis was performed on 12 subjects from the OhioT1DM data set. The results presented include a detailed examination and summary of all risks to the subjects and the information about their ADRR and GMI trend.
{"title":"A Combined Acute and Chronic Risk Assessment Rolling Window for Type 1 Diabetes","authors":"Faizan Munawar, J. Donovan, Etain Kiely, Konrad Mulrennan","doi":"10.1109/IPAS55744.2022.10052880","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052880","url":null,"abstract":"Monitoring the control of persons with type 1 diabetes based on their history of blood glucose levels is essential for self-management. Persons with diabetes must keep their blood glucose levels in a very narrow glycaemic region (70–180 mg/dl) to avoid hypoglycaemia and hyperglycaemia. An extended period of time in the hypoglycaemic or hyperglycaemic region can lead to short-term and long-term complications, respectively. Many measures have been proposed for the management of diabetes, such as the Glucose Management Indicator (GMI) and the Average Daily Risk Range (ADRR). A major drawback of these measures is that they only address acute (ADRR) or chronic (GMI) complications and provide no information on the trend. This paper proposes a rolling window to calculate ADRR and GMI. Calculating ADRR and GMI using a rolling window results in new data, which provide information on the efficacy of self-management of an individual and their risk trend. Use of a rolling window for the risk analysis provides novel information about the glycaemic variability and can be used for improved personal diabetes management plans. Furthermore, ADRR and GMI are combined to propose four new risk levels, which represents the lowest to the highest probable risk of complications. The analysis was performed on 12 subjects from the OhioT1DM data set. The results presented include a detailed examination and summary of all risks to the subjects and the information about their ADRR and GMI trend.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133342824","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 : 2022-12-05DOI: 10.1109/IPAS55744.2022.10052824
Omid Ghozatlou, M. Datcu
Scene classification is one of the most important tasks in the remote sensing field. In general, remotely sensed data comprises targets of different nature with many detailed classes. Therefore, the classification of patches in a satellite scene is a challenging issue. To address the problem, the preferred alternative is to transform to polar coordinates and analyze angular distances. Prior works have so far considered angular distances between points, while ignoring that the target class is not a point, but a distribution. In this paper, we take advantage of this critical fact by using a point-to-probability distribution measure rather than an $ell_{n}$ norm. In this paper, two similarity measures (Euclidean and Mahalanobis) in two different feature space are experimentally investigated through some remote sensing datasets.
{"title":"Comparative Studies on Similarity Distances for Remote Sensing Image Classification","authors":"Omid Ghozatlou, M. Datcu","doi":"10.1109/IPAS55744.2022.10052824","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052824","url":null,"abstract":"Scene classification is one of the most important tasks in the remote sensing field. In general, remotely sensed data comprises targets of different nature with many detailed classes. Therefore, the classification of patches in a satellite scene is a challenging issue. To address the problem, the preferred alternative is to transform to polar coordinates and analyze angular distances. Prior works have so far considered angular distances between points, while ignoring that the target class is not a point, but a distribution. In this paper, we take advantage of this critical fact by using a point-to-probability distribution measure rather than an $ell_{n}$ norm. In this paper, two similarity measures (Euclidean and Mahalanobis) in two different feature space are experimentally investigated through some remote sensing datasets.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"93 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132126845","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 : 2022-12-05DOI: 10.1109/IPAS55744.2022.10052984
Katia Sousa Lillo, Andrea de Maio, S. Lacroix, Amaury Nègre, M. Rombaut, Nicolas Marchand, N. Vercier
Uncertainty estimation of visual feature is essential for vision-based systems, such as visual navigation. We show that errors inherent to visual tracking, in particular using KLT tracker, can be learned using a probabilistic loss function to estimate the covariance matrix on each tracked feature position. The proposed system is trained and evaluated on synthetic data, as well as on real data, highlighting good results in comparison to the state of the art. The benefits of the tracking uncertainty estimates are illustrated for visual motion estimation.
{"title":"DUNE: Deep UNcertainty Estimation for tracked visual features","authors":"Katia Sousa Lillo, Andrea de Maio, S. Lacroix, Amaury Nègre, M. Rombaut, Nicolas Marchand, N. Vercier","doi":"10.1109/IPAS55744.2022.10052984","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052984","url":null,"abstract":"Uncertainty estimation of visual feature is essential for vision-based systems, such as visual navigation. We show that errors inherent to visual tracking, in particular using KLT tracker, can be learned using a probabilistic loss function to estimate the covariance matrix on each tracked feature position. The proposed system is trained and evaluated on synthetic data, as well as on real data, highlighting good results in comparison to the state of the art. The benefits of the tracking uncertainty estimates are illustrated for visual motion estimation.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117036629","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 : 2022-12-05DOI: 10.1109/IPAS55744.2022.10053051
Elhassan Mohamed, K. Sirlantzis, G. Howells
People with movement disabilities may find powered wheelchair driving a challenging task due to their comorbidities. Certain visually impaired persons with mobility disabilities are not prescribed a powered wheelchair because of their sight condition. However, powered wheelchairs are essential to the majority of these disabled users for commuting and social interaction. It is vital for their independence and wellbeing. In this paper, we propose to use a semantic segmentation (SS) system based on deep learning algorithms to provide environmental cues and information to visually impaired wheelchair users to aid with the navigation process. The system classifies the objects of the indoor environment and presents the annotated output on a display customised to the user's condition. The user can select a target object, for which the system can display the estimated distance from the current position of the wheelchair. The system runs in real-time, using a depth camera installed on the wheelchair, and it displays the scene in front of the wheelchair with every pixel annotated with distinguishable colour to represent the different components of the environment along with the distance to the target object. Our system has been designed, implemented and deployed on a real powered wheelchair for practical evaluation. The proposed system helped the users to estimate more accurately the distance to the target objects with a relative error of 19.8% and 18.4% for the conditions of a) semi-neglect and b) short-sightedness, respectively, compared to errors of 47.8% and 5.6% without the SS system. In our experiments, healthy participants were put in simulated conditions representing the above visual impairments using instruments commonly used in medical research for this purpose. Finally, our system helps to visualise, on the display, hidden areas of the environment and blind spots that visually impaired users would not be able to see without it.
{"title":"Real-time Powered Wheelchair Assistive Navigation System Based on Intelligent Semantic Segmentation for Visually Impaired Users","authors":"Elhassan Mohamed, K. Sirlantzis, G. Howells","doi":"10.1109/IPAS55744.2022.10053051","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10053051","url":null,"abstract":"People with movement disabilities may find powered wheelchair driving a challenging task due to their comorbidities. Certain visually impaired persons with mobility disabilities are not prescribed a powered wheelchair because of their sight condition. However, powered wheelchairs are essential to the majority of these disabled users for commuting and social interaction. It is vital for their independence and wellbeing. In this paper, we propose to use a semantic segmentation (SS) system based on deep learning algorithms to provide environmental cues and information to visually impaired wheelchair users to aid with the navigation process. The system classifies the objects of the indoor environment and presents the annotated output on a display customised to the user's condition. The user can select a target object, for which the system can display the estimated distance from the current position of the wheelchair. The system runs in real-time, using a depth camera installed on the wheelchair, and it displays the scene in front of the wheelchair with every pixel annotated with distinguishable colour to represent the different components of the environment along with the distance to the target object. Our system has been designed, implemented and deployed on a real powered wheelchair for practical evaluation. The proposed system helped the users to estimate more accurately the distance to the target objects with a relative error of 19.8% and 18.4% for the conditions of a) semi-neglect and b) short-sightedness, respectively, compared to errors of 47.8% and 5.6% without the SS system. In our experiments, healthy participants were put in simulated conditions representing the above visual impairments using instruments commonly used in medical research for this purpose. Finally, our system helps to visualise, on the display, hidden areas of the environment and blind spots that visually impaired users would not be able to see without it.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117236849","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 : 2022-12-05DOI: 10.1109/IPAS55744.2022.10052849
S. Chatterjee, Hana Haselji'c, R. Frysch, V. Kulvait, V. Semshchikov, B. Hensen, F. Wacker, Inga Brüsch, T. Werncke, O. Speck, A. Nürnberger, G. Rose
Perfusion imaging is a valuable tool for diagnosing and treatment planning for liver tumours. The time separation technique (TST) has been successfully used for modelling C-arm cone-beam computed tomography (CBCT) perfusion data. The reconstruction can be accompanied by the segmentation of the liver - for better visualisation and for generating comprehensive perfusion maps. Recently introduced Turbolift learning has been seen to perform well while working with TST reconstructions, but has not been explored for the time-resolved volumes (TRV) estimated out of TST reconstructions. The segmentation of the TRVs can be useful for tracking the movement of the liver over time. This research explores this possibility by training the multi-scale attention UNet of Turbolift learning at its third stage on the TRVs and shows the robustness of Turbolift learning since it can even work efficiently with the TRVs, resulting in a Dice score of $0.864pm 0.004$.
{"title":"Liver Segmentation in Time-resolved C-arm CT Volumes Reconstructed from Dynamic Perfusion Scans using Time Separation Technique","authors":"S. Chatterjee, Hana Haselji'c, R. Frysch, V. Kulvait, V. Semshchikov, B. Hensen, F. Wacker, Inga Brüsch, T. Werncke, O. Speck, A. Nürnberger, G. Rose","doi":"10.1109/IPAS55744.2022.10052849","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052849","url":null,"abstract":"Perfusion imaging is a valuable tool for diagnosing and treatment planning for liver tumours. The time separation technique (TST) has been successfully used for modelling C-arm cone-beam computed tomography (CBCT) perfusion data. The reconstruction can be accompanied by the segmentation of the liver - for better visualisation and for generating comprehensive perfusion maps. Recently introduced Turbolift learning has been seen to perform well while working with TST reconstructions, but has not been explored for the time-resolved volumes (TRV) estimated out of TST reconstructions. The segmentation of the TRVs can be useful for tracking the movement of the liver over time. This research explores this possibility by training the multi-scale attention UNet of Turbolift learning at its third stage on the TRVs and shows the robustness of Turbolift learning since it can even work efficiently with the TRVs, resulting in a Dice score of $0.864pm 0.004$.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124035400","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 : 2022-12-05DOI: 10.1109/IPAS55744.2022.10053047
A. Morgado, R. Carvalho, Catarina Andrade, Telmo Barbosa, Gonçalo Santos, M.J.M. Vasconcelos
The ability to incrementally learn to categorize objects is a key feature for a personalized system in real-world applications. The major constraint for such scenario relies on the catastrophic forgetting problem, which negatively impacts the performance of the models on previously learned representations. In this work, we developed an equipment classification model to be deployed on edge devices by applying regularization and memory-based class-incremental strategies, such that it can detect new classes while preserving its ability to detect previously known classes, mitigating the forgetting phenomenon. The strategies were tested on three datasets: CIFAR100 to validate the implementation, Stanford Dogs to ensure the reliability of the results as it is a more representative dataset, and SINATRA, which is the work's industrial dataset for equipment recognition. Experimental results on these datasets show that the Experience Replay strategy performed better. For the SINATRA dataset, average accuracy values of 95.57% and of 100% were achieved for Águas e Energias do Porto and Plastaze subsets, respectively. The outcomes of this work proved that by retaining only a limited number of exemplars from old classes, it is possible to update a pre-existing system to classify new devices in a shorter period and avoid catastrophic forgetting.
增量学习对象分类的能力是实际应用程序中个性化系统的一个关键特性。这种场景的主要约束依赖于灾难性遗忘问题，这对模型对先前学习表征的性能产生了负面影响。在这项工作中，我们开发了一个设备分类模型，通过应用正则化和基于记忆的类别增量策略部署在边缘设备上，这样它就可以检测新类别，同时保留检测先前已知类别的能力，从而减轻遗忘现象。这些策略在三个数据集上进行了测试:CIFAR100用于验证实施，Stanford Dogs用于确保结果的可靠性，因为它是一个更具代表性的数据集，以及SINATRA，这是该工作用于设备识别的工业数据集。在这些数据集上的实验结果表明，体验重放策略具有更好的性能。对于SINATRA数据集，Águas e Energias do Porto和Plastaze子集的平均准确率分别达到95.57%和100%。这项工作的结果证明，通过仅保留旧类中有限数量的样本，可以在更短的时间内更新现有系统以对新设备进行分类，并避免灾难性的遗忘。
{"title":"Continual Learning in an Industrial Scenario: Equipment Classification on Edge Devices","authors":"A. Morgado, R. Carvalho, Catarina Andrade, Telmo Barbosa, Gonçalo Santos, M.J.M. Vasconcelos","doi":"10.1109/IPAS55744.2022.10053047","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10053047","url":null,"abstract":"The ability to incrementally learn to categorize objects is a key feature for a personalized system in real-world applications. The major constraint for such scenario relies on the catastrophic forgetting problem, which negatively impacts the performance of the models on previously learned representations. In this work, we developed an equipment classification model to be deployed on edge devices by applying regularization and memory-based class-incremental strategies, such that it can detect new classes while preserving its ability to detect previously known classes, mitigating the forgetting phenomenon. The strategies were tested on three datasets: CIFAR100 to validate the implementation, Stanford Dogs to ensure the reliability of the results as it is a more representative dataset, and SINATRA, which is the work's industrial dataset for equipment recognition. Experimental results on these datasets show that the Experience Replay strategy performed better. For the SINATRA dataset, average accuracy values of 95.57% and of 100% were achieved for Águas e Energias do Porto and Plastaze subsets, respectively. The outcomes of this work proved that by retaining only a limited number of exemplars from old classes, it is possible to update a pre-existing system to classify new devices in a shorter period and avoid catastrophic forgetting.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"125 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131455177","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：