Combination of backscatter calculation and image segmentation for denoising gated light ranging and imaging in fishing net detection

Abstract

Underwater fishing net detection and location plays a crucial role in applications such as safe navigation of unmanned underwater vehicles, protection of marine ecology and marine ranching. However, due to difficulties in detecting fishing nets and noise interference in underwater environments, underwater fishing net detection and location at long distance remains unsolved. In this paper, we use gated light ranging and imaging (LiRAI) as the detection hardware, and propose an underwater fishing net location algorithm based on the physical prior of backscatter noise. The proposed method utilizes the prior knowledge about the distributions of backscatter noise in the target and background regions. An image segmentation network based on deep learning and the initial depth information provided by gated LiRAI are employed to estimate the backscatter noise based on the prior knowledge. Our method can effectively eliminate the backscatter noise and obtain accurate fishing net location results. Field experiments show our method achieves 0.001 absolute relative error (Abs Rel) and 0.156 root mean square error (RMSE) at 27 m in water with $0\mathrm{.26}{m}^{-1}$ attenuation coefficient. Moreover, experiments in underwater environments with different turbidity further validate the effectiveness and generalization of our method.