Underwater Fish Object Detection based on Attention Mechanism improved Ghost-YOLOv5

Abstract

Object detection is a popular research field in deep learning. People usually design large-scale deep convolutional neural networks to continuously improve the accuracy of object detection. However, in the special application scenario of using a robot for underwater fish detection, due to the computational ability and storage space are limited, which leads to the problem of low recognition accuracy of underwater fish. In this paper, an improved Ghost-YOLOv5 network based on attention mechanism is proposed, and use Ghostconvolution in GhostNet to replace the convolution in YOLOv5. Which reduces the number of parameters of the model and makes the network more lightweight. At the same time, we propose a new attention mechanism added to the feature extraction network to enhance the feature expression of fish objects and the robustness of the model. The experimental results show that compared with the original algorithm, the improved YOLOv5 network reduces the calculation amount of the model, and also has better detection performance, the mAP value increased by about 5%.