A Marine Benthic Detection Algoritm Based On Improved Lightweight Network

摘要

Recently, using underwater robots to capture marine organisms have become a development trend. However, the blurred underwater images and colour distortion caused by the complex and variable underwater environment have seriously affected the accuracy and speed of underwater target detection. A marine benthic detection algorithm based on a lightweight network model is proposed in this paper, which can detect the number and distribution of marine organisms and enable the underwater robot to obtain reliable data quickly. Firstly, UGAN is added as an image enhancement network in the front end of the detection network to improve the quality of underwater images. Then a Ghost-SKYOLOv5-based target detection model is proposed, introducing GhostNet, a lightweight network composed of Ghost modules, instead of the backbone feature extraction network, significantly reducing the number of parameters and computation of the network model. In addition, an improved SK convolution is introduced in the feature extraction structure instead of the traditional convolution to improve the detection accuracy. Furthermore, the Focal loss optimization loss function is introduced to solve the positive and negative sample imbalance problem and improve detection accuracy to ensure detection speed. Finally, the algorithm’s performance is tested, and the algorithm can obtain an average accuracy of 86.12%, which is 3.96% better than the YOLOv5 algorithm; the detection speed is 60.50% better than the UGAN -YOLOv5 network.