Using parallelized multi-scale feature extraction to minimize the network architecture, our method can run at 10fps on Raspberry Pi while achieving an accuracy of 83.64%. 

The low-level logical control system can combine with a high-level intelligent command system to enable AUV navigation control based on artificial intelligence. A hardware-in-the-loop simulation platform is beneficial to developing and testing an artificial intelligence based AUV navigation control system.

Compared to tiny-YOLO, parameter amount and calculation amount are reduced by 107 times and 114 times, detection speed is increased by 61.4 times, and the accuracy is maintained within 6%.

Data sharing mechanism of the AUV multi-agent control system improves the efficiency of the AUV control system. The hardware-in-the-loop simulation platform is beneficial to develop an AI-based AUV navigation control system.

AI fish identification can be used to protecting and reviving endangered fish to keep the complexity of marine ecology. By the way, the observation of fish ecology is also used to make movie of marine environmental protection for society propaganda. Moreover, the developed AUV with AI will be able to carry out many missions, including under-water inspection, seabed exploration, and perhaps sample collection.