Underwater Small Object Detection Model Based on Residual Connections
Due to the long distance of underwater imaging,there are a large number of small ob-jects in the existing underwater image object detection datasets,and there is little feature and se-mantic information of underwater small objects.When the existing object detection algorithms are directly applied to underwater detection,the missed detection and false detection rate of small objects is relatively high.In the preprocessing stage of underwater object detection,image en-hancement algorithms are usually used to improve the visual quality of images and improve the accuracy of small object detection.However,the data preprocessing operation of image enhance-ment can easily lead to the loss of small object features and a decrease in the performance of small object detection.Therefore,a structural model for underwater small object detection based on re-sidual connections is proposed,and the application methods of image enhancement to improve ob-ject detection performance are discussed.The enhancement algorithm and object detection algo-rithm are jointly optimized through residual connections,avoiding the problem of feature loss caused by excessive enhancement and improving the accuracy of underwater small object detec-tion.The proposed algorithm was tested on the DUO dataset,and the experimental results showed that compared to YOLOv7,the algorithm improved the accuracy of small object detection by 7.2％.Ablation experiments were conducted on two underwater datasets,verifying that the proposed residual connection method has a promoting effect on improving the performance of small object detection.
small object detectionimage enhancementresidual connectionjoint training
万方数据
维普
