Development Of Online Multi-Machine Fish Pose Recognition System Based On Deep Learning

In recent years,as the economic value and strategic status of the ocean have gradually warmed up,research on underwater unmanned vehicles has also increased year by year.Compared to the traditional propeller-driven underwater unmanned vehicle,the bionic robotic fish has received more attention from researchers because of its flexible movement,small size,low cost,and strong camouflage characteristics.In order to optimize the movement characteristics of the bionic robotic fish and make its movement closer to real fish,research on real fish is also essential.The main methods for studying real fish are observing the corresponding laws of motion under different flow conditions,exploring its dynamic model and using computational fluid dynamics to explore the advantages of its corresponding flow field.To observe the laws of fish,we need to have corresponding programs and algorithms that can calculate the real-time position and attitude of fish.At the same time,after finding out the movement law of fish movement,it needs to be transformed into the dynamic equation of bionic robotic fish to drive the movement of fish like bionic robotic fish.At the same time,in order to build a global visual bionic robotic fish control platform,global vision algorithms for robotic fish and target recognition are also essential.In response to the above issues,the research done in this paper is mainly divided into the following sections:1)For bionic multi-joint robotic fish,a complete set of control system is designed,including the central controller,wireless communication unit,power management unit,and motion execution unit;2)Aiming at bionic multi-joint robotic fish,a nonlinear central-mode controller algorithm is proposed.The advantages and disadvantages of the proposed algorithm and the traditional robotic fish control algorithm are compared.The proposed algorithm is discretely deployed to the controller;3)In order to facilitate the design and development of bionic multi-joint fish algorithm,a set of bionic kinematics simulation platform for multi-joint robotic fish was built.The simulation effect was compared with the real machine fish motion curve driven by the same set of parameters,and the simulation platform was verified.4)According to the static water environment of the turbulent simulation water tunnel and the dynamic water environment of the global visual bionic robotic fish control platform,two visual segmentation algorithms are proposed respectively to separate fish and robotic fish from the background.For dynamic water environment The water surface noise can also be removed,and the phase of the fish is calculated based on the extracted results,that is,the fish posture.Combined with the position information obtained by the segmented fish body,the posture information of the test fish is formed.5)An image semantic segmentation network model based on confrontation generation network is proposed to solve the problem of fish segmentation in multiple water environment at the same time.At the same time,the bionic robotic fish and real fish species can be classified,and at the same time,the biomimetic single The joint fish tested split and output phase calculation point information.