Automatic Detection and Spline-Based Pixel-Length Estimation of Fishes from Images

Sustainable aqua farming is gaining momentum as the need for ecological preservation, at present, is being felt more intensely. For this, it is important to understand the distribution of body length and mass of the farmed fishes. It helps to make informed decisions that reduces over fishing through effective growth monitoring of the fish schools. In this paper, we present a vision based approach for automatic detection, length estimation in the pixel domain and body mass determination of fishes in an aqua farming environment. We evaluate Mask-RCNN, YOLO algorithms for detecting fishes with inexpensive single camera set up placed above the water tank. We obtain a mean average precision (mAP50) score of 0.893 with the YOLO models for fish detection. Subsequently, we use B-splines for estimating the length of fish in the pixel domain followed by mass estimation by fitting the Length-Weight relationship using regression.