Real-Time Ocean Small Target Detection Based on Improved YOLOX Network

In recent years, target recognition and detection technology, which is an important research direction in the field of computer vision, is widely used in human life. The technology has been relatively mature for the recognition of medium and large targets such as people and objects on land. However, due to some conditions, it is relatively rare in the marine field. The main reason for the analysis is that the ocean environment is complex, and many related factors, such as waves and light, can affect it. At the same time, the small targets are usually vague with little information. As a result, the feature expression ability is weak, and very few effective features can be extracted in the process of feature extraction. This problem has aroused the great attention. Based on the existing YOLOX neural network structure, this paper proposes a new neural network structure CBAM-YOLOX. The experiment results show that the AP value of CBAM-YOLOX is 2.94% higher than that of YOLOX, which effectively improves the low positioning accuracy and missing detection of ocean small targets. That has great significance to the application in the ocean field.