Saliency Detection of Turbid Underwater Images Based on Depth Attention Adversarial Network

The turbidity of the water leads to problems such as low contrast between the underwater optical images target and the background, and the target information is weakened. These problems severely limit the performance of existing underwater saliency detection. This paper proposes a depth attention adversarial model for turbid underwater images. Firstly, we generate the turbid underwater images data set by the turbid underwater generation model. Secondly, depth information are introduced as guide items to solve the problem of weakening target information. The depth information also assist the model in detecting salient target information and filter out background noise. Finally, high-level and low-level feature information are fused through the encoding-decoding structure to strengthen the feature extraction ability of the model. Compared with the state-of-the-art saliency detection models, the proposed model can more accurately extract low-contrast targets in turbid underwater images and is better than existing methods under different evaluation indicators.