As the development of marine resources continues to deepen, marine ranching and land-based aquaculture are accelerating their transition from manual to digital monitoring. Underwater environments commonly contain suspended particulates, light attenuation, and wavelength-dependent absorption of visible light. These factors collectively cause acquired images to exhibit pronounced color distortion and haze. This severely constrains the effectiveness of subsequent visual analysis. Meanwhile, underwater operations typically require substantial human and material inputs. This leads to high operational costs, limited monitoring efficiency, and certain safety risks.

To address these challenges, this study integrates a self-developed underwater perception device into an unmanned underwater system. The system is remotely navigated by an operator to designated monitoring locations. This enables high-quality image acquisition of target regions in underwater net cages. On this basis, deep learning algorithms are employed to automatically identify and localize areas of net damage and biofouling attachment.

Furthermore, the perception device is equipped with multi-source sensors. During image acquisition, these sensors capture key water quality parameters in real time. These parameters are closely related to the aquaculture environment. Subsequently, through parametric modeling, the system achieves a multi-dimensional and precise monitoring of the overall aquaculture scenario.

The proposed hardware-software collaborative solution is intended to provide critical technical support for the intelligent inspection of underwater aquaculture facilities and ecological environment monitoring. It is expected to drive the steady digital transformation of marine aquaculture.