Design of a Fish Fry Counting Device Based on Deep Learning
Abstract：To solve the problems of low efficiency, poor accuracy, and fish fry damage in manual fish fry counting in industrial aquaculture, a non-contact high-throughput fish fry counting device based on deep learning was designed. First, the YOLOv8-nano detection model was adopted, and the small target detection performance was optimized by combining the Mixed Local Channel Attention mechanism (MLCA) with the Progressive Feature Pyramid Network (AFPN). Then, the Bytetrack multi-object tracking algorithm was combined with the counting line method to address issues of adhesion and occlusion under high-speed motion. The final model achieved an accuracy of 98.5% and a recall rate of 98.4%. In addition, based on the fish fry counting method, the mechanical structure of the fish fry counting device was designed. The mechanical structure integrates water drainage and separation modules, and is combined with an industrial camera, human-machine interface, and lightweight touchscreen host to achieve quick and accurate counting.
Keywords：AquacultureMachine VisionDeep LearningYOLOv8
Publication Date：2025-10-25
Online Publishing Date：2025-11-14(First online date of this platform, not the publication date of the document)
Pages：5( 1-5 )