Research on underwater object detection based on frequency domain attention mechanism

The detection of underwater objects is of great importance in fields such as oceanography and ecological monitoring. However, traditional detection methods are hindered by significant challenges, including the prevalence of noise interference, low contrast, and complex illumination variations in underwater environments. To address these challenges, a novel frequency domain attention mechanism is proposed. This mechanism combines frequency domain processing with an attention mechanism, weighting the input feature map and subsequently fusing the processed feature outputs. In terms of frequency domain processing, the module employs a range of techniques, including frequency weighting, multi-scale enhancement, phase-preserving filtering and noise suppression. This is done with the aim of optimising the detection of complex underwater targets, which significantly improves the robustness and accuracy of target detection in complex underwater environments. Concurrently, the attention mechanism generates corresponding feature patterns and specific information by dynamically weighting global and local features, which enables the effective identification of the most important features in the current context, thus providing higher accuracy in target edge and texture recognition. The experimental results demonstrate that the target detection model integrated with the frequency-domain attention mechanism exhibits improvements in various metrics, ranging from 0.2% to 1.4%, across multiple datasets. Furthermore, it outperforms the traditional attention mechanism in the majority of cases. These findings not only validate the efficacy of the frequency domain attention mechanism in underwater target detection but also offer novel insights and avenues for future research in this field.