UAV Control Based on Pattern Recognition in Aquaculture Application

Abstract

This study proposes a drone application for the net cage aquaculture industry. A visual control structure is applied to the drone to obtain water-quality information surrounding the net cages. This study integrates a hexacopter, camera, onboard computer, flight control board, servo motor, and global positioning system’s auto-cruise function to adjust the drone position and control the servo motor retractable sensor to reach the desired target at an accurate location. In object identification, a deep learning neural network is used to identify the net cages. An onboard computer calculates the horizontal distance between the drone and the net cage. A “You only look once” (YOLO) neural network is used to detect the net cage images. Considering the hardware calculation speed and ability, an onboard computer is applied to process the flight control board and control the drone. In the mission, an aerial camera detects targets (net cage) and provides visual information to the drone for the target approaching control process. After executing the water-quality measurement, the drone will end the mission and return to the base. This study modifies the architecture of YOLO, compares it with the original model, and then finds a proper architecture for this mission. This study aims to assist cage aquaculture operators by using drones to measure water quality, which can reduce aquaculture’s labor costs.