Research on intelligent damage detection of far-sea cage based on machine vision and deep learning

Abstract

Far-sea cage is an essential way for aquaculture. In the process of far-sea cage aquaculture, the damage of net structure can cause severe economic property losses to farmers, so it is necessary to check the cage's integrity. At present, the typical way to inspect cages is to hire professional divers for manual inspection. This type of inspection is time-consuming and has security concerns. This paper proposes a method for detecting the damage of a far-sea cage based on machine vision and deep learning, which can detect the structure of a far-sea cage in real time and accurately detect the damaged area of the cage. Firstly, the cage image data were collected by autonomous cruising ROV. According to the characteristics of the captured images, an improved multi-scale fusion algorithm was proposed to better the performance of denoising and smoothing effect of the original method. Secondly, we use the MobileNet-SSD and key-frame extraction detection method to detect the damage of underwater cage video. The MobileNet-SSD model has been optimized in model size and detection speed compared with the SSD model. In the experiment, the simulated damaged images of the far-sea cage were used for testing. The experimental results have shown that the scheme can improve the efficiency of far-sea cage inspection and accurately detect the damaged areas in the cage in real-time.