Underwater Target Detection Based on Dual Inversion Dehazing and Multi-Scale Feature Fusion Network

In response to the insufficient small-scale feature extraction capabilities of traditional SSD object detection algorithms, which lead to missed detections and false alarms in the detection of small underwater targets, this paper proposes a new underwater object detection algorithm. The algorithm first uses a Dual Reverse Dehazing (DRDM) module for image enhancement. Then, on the basis of the backbone network of the SSD algorithm, a channel attention mechanism and feature extraction convolution are introduced to construct a SE-Feature-Enhancement Module (SFEM) to integrate high-resolution shallow feature maps and deep feature maps with high semantic information. This fusion method preserves the context information of the feature maps, learns the necessary features, and suppresses redundant information, thereby improving the network’s learning ability. The experimental results show that the proposed algorithm achieves a detection accuracy of 94.62% on the Wilder_fish_new dataset. Compared with the current commonly used SSD (VGG) and SSD (Mobilenetv2) algorithms, the proposed algorithm improves the accuracy by 2.7 and 3.13 percentage points, respectively.