Digital Fish Image Segmentation Using U-Net for Shape Feature Extraction

Segmentation of digital images of fish is an important challenge in image processing in the field of marine biology and aquaculture. Extraction of fish shape features through image segmentation can improve accuracy in species identification and fish population monitoring. The U-Net method, which is based on deep learning, has been proven effective in medical image segmentation and is beginning to be applied in fish image segmentation. This study aims to develop a fish digital image segmentation method using U-Net architecture for accurate and efficient fish shape feature extraction. The dataset used consists of 500 fish images of various shapes and sizes collected from various sources. The fish images were processed using a U-Net artificial neural network, which was trained and tested to obtain the best segmentation results, with evaluation using Intersection over Union (IoU). The segmentation results show that the U-Net method can produce precise segmentation, with a high degree of accuracy in extracting fish shape features. Evaluation of the segmentation metrics resulted in an IoU value of 0.88, indicating excellent performance in distinguishing the fish object from the background and accurately mapping the fish shape. The fish digital image segmentation method using U-Net is effective for fish shape feature extraction and can be applied in fish species identification and aquatic ecosystem monitoring.