Precise feeding technology for outdoor pond aquaculture based on detection and counting method

In aquaculture, both overfeeding and underfeeding pose significant challenges. Underfeeding impedes fish growth, while overfeeding leads to feed waste, increased costs, and water pollution, ultimately compromising fish health. Precise feeding strategies are thus critical for sustainable aquaculture. Existing feeding technologies are primarily limited to indoor environments and specific fish species. Our study aims to extend the research scope to outdoor natural environments, applicable across diverse fish species. We start from the basic biological characteristics of fish, that is, fish will feed when they are hungry, to evaluate fish feeding behavior. We use object detection and counting technology to measure fish appetite by dynamic changes in feeding fish populations. The feeding strategy proposed is more refined and can continuously reflect the changes in the feeding intensity of the fish school. Considering the real-time needs of the application scenario, we select the YOLOv8 object detection algorithm as the basic algorithm. In view of the complexity of natural scenes—where targets are often small and feature-poor—we enhance the YOLOv8 algorithm with a Star operation, CAA module, and DyHead mechanism. The resulting YOLOv8-FishDetect model achieves significant performance gains, improving precision by 3.30 %, [email protected] by 2.51 %, and [email protected]–0.95 by 4.28 % over baseline YOLOv8. This work can provide a scalable solution for outdoor precision feeding, advancing sustainable aquaculture practices.