دورية أكاديمية
A Fully Convolutional Network with Waterfall Atrous Spatial Pooling and Localized Active Contour Loss for Fish Segmentation
| العنوان: | A Fully Convolutional Network with Waterfall Atrous Spatial Pooling and Localized Active Contour Loss for Fish Segmentation |
|---|---|
| المؤلفون: | Thanh Viet Le, Van Yem Vu, Van Truong Pham, Thi-Thao Tran |
| المصدر: | EAI Endorsed Transactions on Industrial Networks and Intelligent Systems, Vol 10, Iss 1 (2023) |
| بيانات النشر: | European Alliance for Innovation (EAI), 2023. |
| سنة النشر: | 2023 |
| المجموعة: | LCC:Computer engineering. Computer hardware LCC:Systems engineering |
| مصطلحات موضوعية: | Fully convolutional network, DeepFish, SIUM fish data, Waterfall Atrous Spatial Pooling, Computer engineering. Computer hardware, TK7885-7895, Systems engineering, TA168 |
| الوصف: | Accurate measurements and statistics of fish data are important for sustainable development of aqua-enviroment and marine fisheries. For data measurements and statistics, automatic segmentation of fish is one of key tasks. The fish segmentation however is a challenging task due to arterfacts in underwater images. In this study, we introduce a deep-learning approach, namely FCN-WRN-WASP for automatic fish segmentation from the underwater images. In particular, we introduce a computational-efficient variation called Waterfall Atrous Spatial Pooling (WASP) module into a Fully convolutional network with Wide ResNet baseline. We also proposed a loss function inspired from active contour approach that can exploit the local intensity information from the input image. The approach has been validated on the DeepFish data and the SIUM data set. The results are promissing for fish segmentation, with higher Intersection over Union (IoU) scores compared to state of the arts. The evaluation results showed that the incorporation of the image based active contour loss helps increase the segmentation performance. In addition, the use of the WASP in the architecture is effective especially for forground fish segmentation. |
| نوع الوثيقة: | article |
| وصف الملف: | electronic resource |
| اللغة: | English |
| تدمد: | 2410-0218 |
| Relation: | https://publications.eai.eu/index.php/inis/article/view/2942; https://doaj.org/toc/2410-0218 |
| DOI: | 10.4108/eetinis.v10i1.2942 |
| URL الوصول: | https://doaj.org/article/56eabe649d3649879e2109145d4d4a3d |
| رقم الأكسشن: | edsdoj.56eabe649d3649879e2109145d4d4a3d |
| قاعدة البيانات: | Directory of Open Access Journals |
Full Text Finder | تدمد: | 24100218 |
|---|---|
| DOI: | 10.4108/eetinis.v10i1.2942 |