Enhanced YOLO-Based Multi-Task Network for Accurate Fish Body Length Measurement

Accurate measurement of fish body length is critical for assessing growth and biomass in aquaculture. Traditional methods are invasive and labor-intensive, while image-based non-contact methods face challenges in measuring curved fish bodies. We propose an enhanced YOLO-based method that combines keypoint detection and image segmentation to form a multi-task learning network and improves accuracy and reliability by improving the loss function so that the two task predictions are aligned. We also designed a new fish body length measurement method based on the multitask model, which can effectively reduce the error caused by fish swimming freely underwater. The experimental results show that our method has good performance in body length measurement of underwater free-swimming fish.