Convolutional Attention-enabled Underwater Object Detection with Electro-optical Image

Aiming at the problems of low detection accuracy and insufficient feature fusion in the underwater environment, an efficient object detection aprroach is proposed based on YOLOv5 with the convolutional attention module. Firstly, the YOLOv5s network model is optimized and improved by integrating convolutional attention, and feature extraction is performed for the input image. Secondly, the weighted bidirectional feature pyramid network is used to enhance the original structure to make multi-scale feature fusion more convenient. Finally, the post-processing algorithm of non-maximum suppression is improved. The experimental results show that the mAP of this method is 85.8%, which is 3.8% higher than that of YOLOv5s, and the accuracy is 4.7% higher than that of YOLOv5s. The proposed model can meet the real-time and accuracy requirements of seabed biological detection in the underwater environment.