水下智能识别与自主抓取机器人设计与实现

设计了一款面向海珍品捕捞的水下智能识别与自主抓取机器人. 首先通过YOLOv4-tiny网络对海珍品图像 离线训练, 设计单双目自适应切换与多目标选择算法以实现海珍品在线识别与持续定位. 进一步, 采用声呐与深度 传感器融合策略获取水下机器人深度信息, 设计基于模糊比例–积分–微分控制的定深抓取控制器, 以确保目标定位 与抓取过程中深度信息的有效反馈. 所提目标识别算法, 具有实时性强、复杂度低优点; 同时, 定深与抓取控制器, 不依赖于系统复杂模型, 可适应不同海况下的精确抓取. 最后, 通过试验验证了方法的有效性.

Design and implementation of underwater intelligent recognition and autonomous grasp vehicle

This paper presents an underwater intelligent recognition and autonomous grasp vehicle for the fishing of precious seafood. Firstly, we employ the YOLOv4-tiny network to off-line train the images of precious seafood, through which monocular-binocular adaptive switch and multi-target selection algorithms are conducted to achieve on-line recognition and persistent localization. Moreover, the depth information of the underwater vehicle is acquired by the fusion of sonar and depth sensor, and fuzzy proportional integral differential based depth-fixed and grasp controllers are designed respectively to guarantee effective feedback for the depth information during the procedures of target localization and grasp. The employed seafood recognition algorithm has the advantages of high real time and low complexity. On the other hand, the proposed depth-fixed and grasp controllers do not rely on complex system model, which can suit the target accurate grasp in different sea states. Finally, the experiments are conducted to verify the effectiveness of the developed approach.