Semantic Segmentation Model for Coastal Aquaculture Extraction based on the Superpixel Algorithm
Year: 2025
Type: article
Abstract: Coastal aquaculture segmentation is essential for efficient management of seafood production and preventing maritime accidents caused by illegally established fish farms. Until recently, this task relied heavily on manual labor by skilled professionals. This study proposes a semantic segmentation model that employs a superpixel-based algorithm to enhance boundary precision in coastal aquaculture extraction. The proposed model addresses the limitation of current deep learning models which often merge adjacent aquaculture facilities due to their dense distribution. By modifying the DeepLabV3+ architecture and integrating Fast SLIC algorithm into both the model structure and loss function, we achieve more accurate segmentation results. Experiments demonstrate that our model outperforms the original DeepLabV3+ by approximately 4.7%p in Mean IoU, confirming its effectiveness for precise marine aquaculture facility detection and management.
Authors: Su-Myung Gang, Joon-Jae Lee
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
SDGs: Life below water, Zero hunger
Open Access status: diamond
OpenAlex ID: W4411360263