A Low Light Target Detection Algorithm Based on Deep Learning
The image acquired in complex low illumination environment is prone to low contrast and loss of detail information.Therefore,considering the challenges of low-light detection,the object detection algorithm you look only once v8(YOLOv8)was improved and YOLO-resblock selective kernel global attention mechanism(YOLO-RSG)has been proposed to improve the reliability of object detection in low-light environments.Firstly,the backbone feature extraction part of the YOLO-RSG algorithm adopts C3_ResBlock module to improve the multi-scale and weak feature extraction capability of the model.Then,squeeze-and-excitation-atrous spatial pyramid pooling structure is introduced to extract complex scene information with different expansion rates,which improves the training effect while maintaining the computational load.Finally,the selective kernel mechanism and global attention mechanism are integrated adaptively,which improves the multi-scale feature extraction,fusion and representation ability of the network model.Simulation results show that compared with YOLOv8 algorithm,YOLO-RSG algorithm improves the mean average accuracy of ExDark dataset by 3.60％ ,which can effectively improve the target detection performance in low-illumination scenes,and has good stability and applicability.
NETL
NSTL
万方数据
