MGC-YOLO: Underwater Biomimetic Shrimp With Foreign Object Recognition

摘要

We propose a low-cost, small-volume, and easy-to-use underwater bionic robot shrimp to solve the problem of complex underwater scenes and the limited ability of target feature extraction. Specifically, an improved YOLOv8s underwater bionic robot shrimp model with foreign body recognition is proposed by introducing the improved Multi-scale Ghost (MGHost) convolution into the CSPDarknet53 to 2-StageFPN (C2F) module and finally adding an improved attention mechanism Convolutional priority multi space attention module (CPMS) after the Spatial Pyramid Pooling-Fast (SPPF), the MGC-YOLO algorithm model was proposed. At the same time, the bionic robot shrimp has a detachable manipulator design and can then achieve underwater movement, lifting, and hovering functions so that the underwater robot shrimp can carry out underwater maintenance, sampling, and other operations. Secondly, the image transmission module can cooperate with the APP side to control the machine shrimp, and the APP side has a deep waveform display function. The experimental results show that compared with the original model, the improved model reduces the number of parameters and the amount of calculation by 0.94 and 3.6, respectively, and improves its accuracy and Mean Average Precision (mAP) by 5.7% and 8.9%, respectively. After deploying the equipment, the lifting speed reaches 50mm/s, the time for the manipulator to identify the foreign body is less than 0.2s, and the delay for the return of underwater information is less than 0.2s, which verifies the feasibility and effectiveness of the model in underwater foreign body recognition.