A Prediction of the Shooting Trajectory for a Tuna Purse Seine Using the Double Deep Q-Network (DDQN) Algorithm

The purse seine is a fishing method in which a net is used to encircle a fish school, capturing isolated fish by tightening a purse line at the bottom of the net. Tuna purse seine operations are technically complex, requiring the evaluation of fish movements, vessel dynamics, and their interactions, with success largely dependent on the expertise of the crew. In particular, efficiency in terms of highly complex tasks, such as calculating the shooting trajectory during fishing operations, varies significantly based on the fisher's skill level. To address this challenge, developing techniques to support less experienced fishers is necessary, particularly for operations targeting free-swimming fish schools, which are more difficult to capture compared to those utilizing Fish Aggregating Devices (FADs). This study proposes a method for predicting shooting trajectories using the Double Deep Q-Network (DDQN) algorithm. Observation states, actions, and reward functions were designed to identify optimal scenarios for shooting, and the catchability of the predicted trajectories was evaluated through gear behavior analysis. The findings of this study are expected to aid in the development of a trajectory prediction system for inexperienced fishers and serve as foundational data for automating purse seine fishing systems.