Research on the Correlation between Image Enhancement and Underwater Object Detection

Abstract

Underwater object detection has a wide range of application scenarios. Due to the characteristics of underwater imaging, the research focus of object detection based on underwater images differs from that of general object detection algorithms. The detailed quality of underwater images can be improved effectively by using the certain image enhancement algorithm, which is helpful for human recognition of underwater objects and object detection using traditional algorithms. In recent years, research on underwater object detection algorithms based on deep learning has gradually become popular, but the influence of image enhancement algorithms on object detection based on deep learning has not yet been systematically studied. In this paper, the correlation between image enhancement and underwater object detection based on deep learning is studied. First, three methods based on image dehazing, UWCNN, and FUNIE-GAN are selected to enhance the images of the URPC dataset, Then, SSD algorithm is adopted to train and evaluate the URPC dataset and the three enhanced dataset respectively, finally, a statistical analysis of the correlation between the changes of image quality parameter after image enhancement and the object detection accuracy is carried out. Experiments show that the average object detection accuracy on the image-enhanced dataset has been improved to a certain extent, but different image quality parameter changes have no obvious statistical correlations to the final detection accuracy. The small increase in the final object detection accuracy may be the result of a combination of several factors.