Multilayer perception-based intelligent feeding algorithm and its deployment in mobile autonomous aquaculture systems

To address the challenges of high labor intensity, rigid feeding patterns, and imprecise feed quantities in aquaculture feeding operations, this paper presents the design and implementation of a mobile autonomous feeding device tailored for real-time environmental monitoring and adaptive feeding in aquatic farming. The system comprises three main components: a mechanical mobility module, a feed dispensing module, and a fish hunger state monitoring module. By integrating a dynamic fish behavior monitoring system, the device collects data on fish activity levels and water turbidity. These inputs are processed by a feeding decision model to assess the hunger state of the fish and determine optimal feeding times and quantities. The mechanical module enables the device to traverse the water body and perform multipoint distributed feeding, thereby overcoming the limitations of fixed-location feeding methods. Experimental results from feed dispensing trials and hunger state recognition tests demonstrate that the device achieves a 92.3% success rate in underwater dynamic fish recognition, with feeding error maintained below 10.67%. The system exhibits low feed loss and high operational reliability. This work provides a valuable reference for the development of automated and intelligent feeding systems in aquaculture.