Underwater object detection model based on improved YOLOv5
In complex deep-sea environments,precise detection and identification of targets are crucial for marine exploration and conservation.This study optimizes the YOLOv5 target detection algorithm to enhance the detection effects and efficiency of un-derwater targets.To augment the feature extraction capability of the model,the critical convolutional structure of YOLOv5—the C3 module—has been replaced.Moreover,the SEAttention module,which utilizes a spatial attention mechanism,is introduced,en-abling the model to focus more on recognizing important targets.For evaluation,EIoU is adopted as the standard to accurately mea-sure the performance of target detection.Experiments are primarily conducted on the DeepTrash dataset.Compared to the original YOLOv5,the optimized model has achieved a 3.1%improvement in precision,a 1.8%increase in recall rate,and a 2.4%rise in mAP@0.5.Furthermore,the parameter count of the model has been reduced by 17.4%after optimization,thereby enhancing compu-tational efficiency.Even when compared to the latest YOLOv8,the optimized model has demonstrated superior performance.
NETL
NSTL
万方数据
