A NeRF 3D Based Fish Weighing Model Under Fluctuating Measuring Condition

Robust fish weight estimation is vital for sustainable fisheries, preventing overfishing and conserving resources, yet traditional methods are prone to environmental factors causing significant errors. To address the challenge, a vision-based fish weight estimation model is proposed, which generates a half-fish three dimensional point cloud from multi-angle images using neural radiance fields (NeRF), then applies bayesian optimization to identify the optimal hyperplane for full-fish reconstruction. This process estimates the fish volume, which is subsequently used to predict its weight through a generalized linear model. During the validation of the established dataset, the proposed method demonstrates remarkable performance with a mean square error of 0.007, root mean square error of 0.084 and a mean absolute error of 0.070, highlight its considerable application potential.