Pub Date : 2022-12-05DOI: 10.1109/IPAS55744.2022.10053050
Jingan Liu, N. P. Chandrasiri
With the development of deep learning, super-resolution image synthesis techniques for enhancing low-resolution images have advanced remarkably. However, mainstream algorithms focus on improving the quality of the entire image on average and this may result in blurring. In this paper, we propose three key components for synthesizing super-resolution images that can reflect the fine details of an image. We synthesize super-resolution images by image classification. First, the neural network weights learned using the images in the same image category were utilized in synthesizing super-resolution images. For this purpose, image classification was performed using a transfer-trained ResNet. Second, SENet was applied to the generators in our proposed method to obtain detailed information about the images. Finally, the feature extraction network was changed from VGG to ResNet in order to get more important features. As a result, we achieved better image evaluation values (PSNR, NIQE) for the super-resolution images of dogs and cats compared to the previous studies. Furthermore, the images were generated more naturally on the benchmark dataset.
{"title":"C-ESRGAN: Synthesis of super-resolution images by image classification","authors":"Jingan Liu, N. P. Chandrasiri","doi":"10.1109/IPAS55744.2022.10053050","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10053050","url":null,"abstract":"With the development of deep learning, super-resolution image synthesis techniques for enhancing low-resolution images have advanced remarkably. However, mainstream algorithms focus on improving the quality of the entire image on average and this may result in blurring. In this paper, we propose three key components for synthesizing super-resolution images that can reflect the fine details of an image. We synthesize super-resolution images by image classification. First, the neural network weights learned using the images in the same image category were utilized in synthesizing super-resolution images. For this purpose, image classification was performed using a transfer-trained ResNet. Second, SENet was applied to the generators in our proposed method to obtain detailed information about the images. Finally, the feature extraction network was changed from VGG to ResNet in order to get more important features. As a result, we achieved better image evaluation values (PSNR, NIQE) for the super-resolution images of dogs and cats compared to the previous studies. Furthermore, the images were generated more naturally on the benchmark dataset.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"28 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":"132939856","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/IPAS55744.2022.10053044
Hamail Ayaz, D. Tormey, Ian McLoughlin, Muhammad Ahmad, S. Unnikrishnan
Glioblastoma (GB) is a malignant brain tumor and requires surgical resection. Although complete resection of GB improves prognosis, supratotal resection may cause neurological abnormalities. Therefore, intraoperative tissue classification techniques are needed to delineate infected tumor regions to remove reoccurrences. To delineate the affected regions, surgeons mostly rely on traditional magnetic resonance imaging (MRI) which often lacks accuracy and precision due to the brain-shift phenomenon. Hyperspectral Imaging (HSI) is a noninvasive advanced optical technique and has the potential to classify tissue cells accurately. However, HSI tumor classification is challenging due to overlapping regions, high interclass similarity, and homogeneous information. Additionally, HSI models using 2D Convolutional Neural Network (CNN) models works with spectral information eliminating spatial features and 3D followed by 2D hybrid model lacks abstract level spatial information. Therefore, in this study, we have used a minimal layer 3D CNN model to classify the GB tumor region from normal tissues using an intraoperative VivoHSI dataset. The HSI data have normal tissue (NT), tumor tissue (TT), hypervascularized tissue or blood vessels (BV), and background (BG) tissue cells. The proposed 3D CNN model consists of only two 3D layers using limited training samples (20%), which are further divided into 50% for training and 50% for validation and blind tested (80%) on the rest of the data. This study outperformed then state-of-the-art hybrid architecture by achieving an overall accuracy of 99.99%.
{"title":"Hyperspectral Brain Tissue Classification using a Fast and Compact 3D CNN Approach","authors":"Hamail Ayaz, D. Tormey, Ian McLoughlin, Muhammad Ahmad, S. Unnikrishnan","doi":"10.1109/IPAS55744.2022.10053044","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10053044","url":null,"abstract":"Glioblastoma (GB) is a malignant brain tumor and requires surgical resection. Although complete resection of GB improves prognosis, supratotal resection may cause neurological abnormalities. Therefore, intraoperative tissue classification techniques are needed to delineate infected tumor regions to remove reoccurrences. To delineate the affected regions, surgeons mostly rely on traditional magnetic resonance imaging (MRI) which often lacks accuracy and precision due to the brain-shift phenomenon. Hyperspectral Imaging (HSI) is a noninvasive advanced optical technique and has the potential to classify tissue cells accurately. However, HSI tumor classification is challenging due to overlapping regions, high interclass similarity, and homogeneous information. Additionally, HSI models using 2D Convolutional Neural Network (CNN) models works with spectral information eliminating spatial features and 3D followed by 2D hybrid model lacks abstract level spatial information. Therefore, in this study, we have used a minimal layer 3D CNN model to classify the GB tumor region from normal tissues using an intraoperative VivoHSI dataset. The HSI data have normal tissue (NT), tumor tissue (TT), hypervascularized tissue or blood vessels (BV), and background (BG) tissue cells. The proposed 3D CNN model consists of only two 3D layers using limited training samples (20%), which are further divided into 50% for training and 50% for validation and blind tested (80%) on the rest of the data. This study outperformed then state-of-the-art hybrid architecture by achieving an overall accuracy of 99.99%.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"51 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":"115304981","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/IPAS55744.2022.10053001
Labib Ahmed Siddique, Rabita Junhai, Tanzim Reza, Salman Khan, Tanvir Rahman
Real-time video surveillance, through CCTV camera systems has become essential for ensuring public safety which is a priority today. Although CCTV cameras help a lot in increasing security, these systems require constant human interaction and monitoring. To eradicate this issue, intelligent surveillance systems can be built using deep learning video classification techniques that can help us automate surveillance systems to detect violence as it happens. In this research, we explore deep learning video classification techniques to detect violence as they are happening. Traditional image classification techniques fall short when it comes to classifying videos as they attempt to classify each frame separately for which the predictions start to flicker. Therefore, many researchers are coming up with video classification techniques that consider spatiotemporal features while classifying. However, deploying these deep learning models with methods such as skeleton points obtained through pose estimation and optical flow obtained through depth sensors, are not always practical in an IoT environment. Although these techniques ensure a higher accuracy score, they are computationally heavier. Keeping these constraints in mind, we experimented with various video classification and action recognition techniques such as ConvLSTM, LRCN (with both custom CNN layers and VGG-16 as feature extractor) CNNTransformer and C3D. We achieved a test accuracy of 80% on ConvLSTM, 83.33% on CNN-BiLSTM, 70% on VGG16-BiLstm, 76.76% on CNN-Transformer and 80% on C3D.
{"title":"Analysis of Real-Time Hostile Activitiy Detection from Spatiotemporal Features Using Time Distributed Deep CNNs, RNNs and Attention-Based Mechanisms","authors":"Labib Ahmed Siddique, Rabita Junhai, Tanzim Reza, Salman Khan, Tanvir Rahman","doi":"10.1109/IPAS55744.2022.10053001","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10053001","url":null,"abstract":"Real-time video surveillance, through CCTV camera systems has become essential for ensuring public safety which is a priority today. Although CCTV cameras help a lot in increasing security, these systems require constant human interaction and monitoring. To eradicate this issue, intelligent surveillance systems can be built using deep learning video classification techniques that can help us automate surveillance systems to detect violence as it happens. In this research, we explore deep learning video classification techniques to detect violence as they are happening. Traditional image classification techniques fall short when it comes to classifying videos as they attempt to classify each frame separately for which the predictions start to flicker. Therefore, many researchers are coming up with video classification techniques that consider spatiotemporal features while classifying. However, deploying these deep learning models with methods such as skeleton points obtained through pose estimation and optical flow obtained through depth sensors, are not always practical in an IoT environment. Although these techniques ensure a higher accuracy score, they are computationally heavier. Keeping these constraints in mind, we experimented with various video classification and action recognition techniques such as ConvLSTM, LRCN (with both custom CNN layers and VGG-16 as feature extractor) CNNTransformer and C3D. We achieved a test accuracy of 80% on ConvLSTM, 83.33% on CNN-BiLSTM, 70% on VGG16-BiLstm, 76.76% on CNN-Transformer and 80% on C3D.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"11303 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":"114895935","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/IPAS55744.2022.10053064
Niklas Stralau, Chengxuan Fu
For driving assistance and autonomous driving systems, it is important to differentiate between dynamic objects such as moving vehicles and static objects such as guard rails. Among all the sensor modalities, RADAR and FMCW LiDAR can provide information regarding the motion state of the raw measurement data. On the other hand, perception pipelines using measurement data from ToF LiDAR typically can only differentiate between dynamic and static states on the object level. In this work, a new algorithm called DONEX was developed to classify the motion state of 3D LiDAR point cloud echoes using an occupancy grid approach. Through algorithmic improvements, e.g. 2D grid approach, it was possible to reduce the runtime. Scenarios, in which the measuring sensor is located in a moving vehicle, were also considered.
{"title":"DONEX: Real-time occupancy grid based dynamic echo classification for 3D point cloud","authors":"Niklas Stralau, Chengxuan Fu","doi":"10.1109/IPAS55744.2022.10053064","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10053064","url":null,"abstract":"For driving assistance and autonomous driving systems, it is important to differentiate between dynamic objects such as moving vehicles and static objects such as guard rails. Among all the sensor modalities, RADAR and FMCW LiDAR can provide information regarding the motion state of the raw measurement data. On the other hand, perception pipelines using measurement data from ToF LiDAR typically can only differentiate between dynamic and static states on the object level. In this work, a new algorithm called DONEX was developed to classify the motion state of 3D LiDAR point cloud echoes using an occupancy grid approach. Through algorithmic improvements, e.g. 2D grid approach, it was possible to reduce the runtime. Scenarios, in which the measuring sensor is located in a moving vehicle, were also considered.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"12 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":"114994499","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/IPAS55744.2022.10052983
Nishant Yella, Manisai Eppakayala, Tauqir Pasha
The Southern Resident Killer Whale (Orcinus Orca) is an apex predator in the oceans. Currently, these are listed as endangered species and have slowly declined in number over the past two decades. There is a lack of availability of data on audio vocalizations of killer whales, which in itself creates a demanding task to acquire labelled audio sets. The vocalizations of orcas are usually categorized into two groups namely, whistles and pulsed calls. There is a significant amount of scarcity on audio sets of these two types of vocalizations. Hence this creates a challenge to address the lack of availability of data on these vocalizations. Methods of data augmentations have proven over the years to be very effective in generating synthetically created data for the use of labelled training of a given feed-forward neural network. The Transformer based Generative Adversarial neural network (Trans-GAN) has performed phenomenally well on tasks pertaining to visual perception. In this paper, we would like to demonstrate the use of trans-GAN on audio datasets, which would be used to perform bioacoustics augmentation of the killer whale audio vocalizations obtained from existing open-source libraries to generate a synthetically substantial amount of audio data on the killer whale vocalizations for tasks pertaining to audio perception. To validate the Trans-GAN generated audio to the original killer Whale vocalization sample, we have implemented a time-sequence-based algorithm called Dynamic Time Wrapping (DTW), which compares the similarity index between these two audio samples.
{"title":"Bioacoustic augmentation of Orcas using TransGAN","authors":"Nishant Yella, Manisai Eppakayala, Tauqir Pasha","doi":"10.1109/IPAS55744.2022.10052983","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052983","url":null,"abstract":"The Southern Resident Killer Whale (Orcinus Orca) is an apex predator in the oceans. Currently, these are listed as endangered species and have slowly declined in number over the past two decades. There is a lack of availability of data on audio vocalizations of killer whales, which in itself creates a demanding task to acquire labelled audio sets. The vocalizations of orcas are usually categorized into two groups namely, whistles and pulsed calls. There is a significant amount of scarcity on audio sets of these two types of vocalizations. Hence this creates a challenge to address the lack of availability of data on these vocalizations. Methods of data augmentations have proven over the years to be very effective in generating synthetically created data for the use of labelled training of a given feed-forward neural network. The Transformer based Generative Adversarial neural network (Trans-GAN) has performed phenomenally well on tasks pertaining to visual perception. In this paper, we would like to demonstrate the use of trans-GAN on audio datasets, which would be used to perform bioacoustics augmentation of the killer whale audio vocalizations obtained from existing open-source libraries to generate a synthetically substantial amount of audio data on the killer whale vocalizations for tasks pertaining to audio perception. To validate the Trans-GAN generated audio to the original killer Whale vocalization sample, we have implemented a time-sequence-based algorithm called Dynamic Time Wrapping (DTW), which compares the similarity index between these two audio samples.","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":"129678685","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}
A glacier's mass balance and dynamics are regulated by changes in ice velocity. As a result, estimating glacier flow velocity is a crucial part of temporal glacier monitoring of its health, response to climate change parameters, and its effect on increasing the sea level rise. In this study, we estimated the Gangotri glacier surface velocities from 2014 to 2021. We used remote sensing-based techniques to estimate the Gangotri surface velocity since it provides such measurements regularly for a vast geographical area. Sub-pixel correlation of Landsat 8 imagery was used by using the COSI- Corr (co-registration of optically sensed images and correlation) tool to determine surface velocities over the Gangotri glacier. Our derived velocities values match the ground truth velocities values comparatively well. Gangotri surface velocities vary over the various regions of the glacier, and from year to year. Our study indicated that the middle region of the ablation zone and the accumulation zone had higher velocities across all the years, while the boundary regions of the glacier show lower speeds. The average velocities range varies from ∼13 m/year in the accumulation zone to ∼22m/year. For the ablation zone, the average velocities range from ∼ 11m/year in the ablation zone to ∼18m/year. The average surface velocity showed a high decrement of 26 % from 2014 to 2021. The general surface velocities in Gangotri vary from ∼8 m/y to∼ 61 m/y± 1.9 from 2014 to 2021.
{"title":"Glacier-surface velocities in Gangotri from Landsat8 satellite imagery","authors":"Reem Klaib, Hajer Alabdouli, Mritunjay Kumar Singh","doi":"10.1109/IPAS55744.2022.10052898","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052898","url":null,"abstract":"A glacier's mass balance and dynamics are regulated by changes in ice velocity. As a result, estimating glacier flow velocity is a crucial part of temporal glacier monitoring of its health, response to climate change parameters, and its effect on increasing the sea level rise. In this study, we estimated the Gangotri glacier surface velocities from 2014 to 2021. We used remote sensing-based techniques to estimate the Gangotri surface velocity since it provides such measurements regularly for a vast geographical area. Sub-pixel correlation of Landsat 8 imagery was used by using the COSI- Corr (co-registration of optically sensed images and correlation) tool to determine surface velocities over the Gangotri glacier. Our derived velocities values match the ground truth velocities values comparatively well. Gangotri surface velocities vary over the various regions of the glacier, and from year to year. Our study indicated that the middle region of the ablation zone and the accumulation zone had higher velocities across all the years, while the boundary regions of the glacier show lower speeds. The average velocities range varies from ∼13 m/year in the accumulation zone to ∼22m/year. For the ablation zone, the average velocities range from ∼ 11m/year in the ablation zone to ∼18m/year. The average surface velocity showed a high decrement of 26 % from 2014 to 2021. The general surface velocities in Gangotri vary from ∼8 m/y to∼ 61 m/y± 1.9 from 2014 to 2021.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"30 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":"123985485","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/IPAS55744.2022.10053028
Ege Ince, Sevdenur Kutuk, Rayan Abri, Sara Abri, S. Cetin
Real time processing in the context of image processing for topics like motion detection and suspicious object detection requires processing the background more times. In this field, background subtraction solutions can overcome the limitations caused by real time issues. Different methods of background subtraction have been investigated for this goal. Although more background subtraction methods provide the required efficiency, they do not make produce a real-time solution in a camera surveillance environment. In this paper, we propose a model for background subtraction using four different traditional algorithms; ViBe, Mixture of Gaussian V2 (MOG2), Two Points, and Pixel Based Adaptive Segmenter (PBAS). The presented model is a lightweight real time architecture for surveillance cameras. In this model, the dynamic programming logic is used during preprocessing of the frames. The CDnet 2014 data set is used to assess the model's accuracy, and the findings show that it is more accurate than the traditional methods whose combinations are suggested in the paper in terms of Frames per second (fps), F1 score, and Intersection over union (IoU) values by 61.31, 0.552, and 0.430 correspondingly.
{"title":"A Light Weight Approach for Real-time Background Subtraction in Camera Surveillance Systems","authors":"Ege Ince, Sevdenur Kutuk, Rayan Abri, Sara Abri, S. Cetin","doi":"10.1109/IPAS55744.2022.10053028","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10053028","url":null,"abstract":"Real time processing in the context of image processing for topics like motion detection and suspicious object detection requires processing the background more times. In this field, background subtraction solutions can overcome the limitations caused by real time issues. Different methods of background subtraction have been investigated for this goal. Although more background subtraction methods provide the required efficiency, they do not make produce a real-time solution in a camera surveillance environment. In this paper, we propose a model for background subtraction using four different traditional algorithms; ViBe, Mixture of Gaussian V2 (MOG2), Two Points, and Pixel Based Adaptive Segmenter (PBAS). The presented model is a lightweight real time architecture for surveillance cameras. In this model, the dynamic programming logic is used during preprocessing of the frames. The CDnet 2014 data set is used to assess the model's accuracy, and the findings show that it is more accurate than the traditional methods whose combinations are suggested in the paper in terms of Frames per second (fps), F1 score, and Intersection over union (IoU) values by 61.31, 0.552, and 0.430 correspondingly.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"92 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":"134519578","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/IPAS55744.2022.10052895
Amin Kargar, Mariusz P. Wilk, Dimitrios Zorbas, Michael T. Gaffney, Brendan Q'Flynn
Effective insect pest monitoring is a vital component of Integrated Pest Management (IPM) strategies. It helps to support crop productivity while minimising the need for plant protection products. In recent years, many researchers have considered the integration of intelligence into such systems in the context of the Smart Agriculture research agenda. This paper describes the development of a smart pest monitoring system, developed in accordance with specific requirements associated with the agricultural sector. The proposed system is a low-cost smart insect trap, for use in orchards, that detects specific insect species that are detrimental to fruit quality. The system helps to identify the invasive insect, Brown Marmorated Stink Bug (BMSB) or Halyomorpha halys (HH) using a Microcontroller Unit-based edge device comprising of an Internet of Things enabled, resource-constrained image acquisition and processing system. It is used to execute our proposed lightweight image analysis algorithm and Convolutional Neural Network (CNN) model for insect detection and classification, respectively. The prototype device is currently deployed in an orchard in Italy. The preliminary experimental results show over 70 percent of accuracy in BMSB classification on our custom-built dataset, demonstrating the proposed system feasibility and effectiveness in monitoring this invasive insect species.
{"title":"A Novel Resource-Constrained Insect Monitoring System based on Machine Vision with Edge AI","authors":"Amin Kargar, Mariusz P. Wilk, Dimitrios Zorbas, Michael T. Gaffney, Brendan Q'Flynn","doi":"10.1109/IPAS55744.2022.10052895","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052895","url":null,"abstract":"Effective insect pest monitoring is a vital component of Integrated Pest Management (IPM) strategies. It helps to support crop productivity while minimising the need for plant protection products. In recent years, many researchers have considered the integration of intelligence into such systems in the context of the Smart Agriculture research agenda. This paper describes the development of a smart pest monitoring system, developed in accordance with specific requirements associated with the agricultural sector. The proposed system is a low-cost smart insect trap, for use in orchards, that detects specific insect species that are detrimental to fruit quality. The system helps to identify the invasive insect, Brown Marmorated Stink Bug (BMSB) or Halyomorpha halys (HH) using a Microcontroller Unit-based edge device comprising of an Internet of Things enabled, resource-constrained image acquisition and processing system. It is used to execute our proposed lightweight image analysis algorithm and Convolutional Neural Network (CNN) model for insect detection and classification, respectively. The prototype device is currently deployed in an orchard in Italy. The preliminary experimental results show over 70 percent of accuracy in BMSB classification on our custom-built dataset, demonstrating the proposed system feasibility and effectiveness in monitoring this invasive insect species.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"136 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":"116564093","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/IPAS55744.2022.10052939
N. Tejasri, G. U. Sai, P. Rajalakshmi, B. BalajiNaik, U. B. Desai
Water is essential for any crop production. Lack of sufficient supply of water supply causes abiotic stress in crops. Accurate identification of the crops affected by drought is required for achieving sustainable agricultural yield. The image data plays a crucial role in studying the crop's response. Recent developments in aerial-based imaging methods allow us to capture RGB maize data by integrating an RGB camera with the drone. In this work, we propose a pipeline to collect data rapidly, pre-process the data and apply deep learning based models to segment drought affected/stressed and unaffected/healthy RGB maize crop grown in controlled water conditions. We develop an ensemble-based framework based on U-Net and U-Net++ architectures for the drought stress segmentation task. The ensemble framework is based on the stacking approach by averaging the predictions of fine-tuned U-Net and U-Net++ models to generate the output mask. The experimental results showed that the ensemble framework performed better than individual U-Net and U-Net++ models on the test set with a mean IoU of 0.71 and a dice coefficient of 0.74.
{"title":"Drought Stress Segmentation on Drone captured Maize using Ensemble U-Net framework","authors":"N. Tejasri, G. U. Sai, P. Rajalakshmi, B. BalajiNaik, U. B. Desai","doi":"10.1109/IPAS55744.2022.10052939","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052939","url":null,"abstract":"Water is essential for any crop production. Lack of sufficient supply of water supply causes abiotic stress in crops. Accurate identification of the crops affected by drought is required for achieving sustainable agricultural yield. The image data plays a crucial role in studying the crop's response. Recent developments in aerial-based imaging methods allow us to capture RGB maize data by integrating an RGB camera with the drone. In this work, we propose a pipeline to collect data rapidly, pre-process the data and apply deep learning based models to segment drought affected/stressed and unaffected/healthy RGB maize crop grown in controlled water conditions. We develop an ensemble-based framework based on U-Net and U-Net++ architectures for the drought stress segmentation task. The ensemble framework is based on the stacking approach by averaging the predictions of fine-tuned U-Net and U-Net++ models to generate the output mask. The experimental results showed that the ensemble framework performed better than individual U-Net and U-Net++ models on the test set with a mean IoU of 0.71 and a dice coefficient of 0.74.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"31 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":"125341284","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/IPAS55744.2022.10052995
Chinatsu Watanabe, Mayu Kaneko, N. P. Chandrasiri
In recent years, video streaming services have become increasingly popular. In general, the search function in a video sharing service site evaluates the relevance of a search query to the title, tags, description, and so on given by the creator of the video. Then, the search results with the highest relevance are displayed. Therefore, if a title is given to a video that does not match its content, there is a possibility that a video with low relevance will be found. In this research, (1) we built a new system that retrieves animal videos that are relevant to its content using image recognition. (2) By describing the relationships between the concepts of animal families and species and incorporating them into the retrieval system, it is possible to retrieve animal videos by their family names. Adding retrieval by animal family name enabled us to find species that have not been learned. In this research, (3) we confirmed the usefulness of our video retrieval system using trained neural networks, GoogLeNet and ResNet50, as animal species classifiers.
{"title":"Animal Video Retrieval System using Image Recognition and Relationships Between Concepts of Animal Families and Species","authors":"Chinatsu Watanabe, Mayu Kaneko, N. P. Chandrasiri","doi":"10.1109/IPAS55744.2022.10052995","DOIUrl":"https://doi.org/10.1109/IPAS55744.2022.10052995","url":null,"abstract":"In recent years, video streaming services have become increasingly popular. In general, the search function in a video sharing service site evaluates the relevance of a search query to the title, tags, description, and so on given by the creator of the video. Then, the search results with the highest relevance are displayed. Therefore, if a title is given to a video that does not match its content, there is a possibility that a video with low relevance will be found. In this research, (1) we built a new system that retrieves animal videos that are relevant to its content using image recognition. (2) By describing the relationships between the concepts of animal families and species and incorporating them into the retrieval system, it is possible to retrieve animal videos by their family names. Adding retrieval by animal family name enabled us to find species that have not been learned. In this research, (3) we confirmed the usefulness of our video retrieval system using trained neural networks, GoogLeNet and ResNet50, as animal species classifiers.","PeriodicalId":322228,"journal":{"name":"2022 IEEE 5th International Conference on Image Processing Applications and Systems (IPAS)","volume":"8 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":"129753809","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：