YOLOX remote sensing image object detection algorithm based on FEB
In view of the problems of reduced detection accuracy caused by complex background interference,large target scale variation,and difficulty in detecting small targets in remote sensing images,a detection algorithm is proposed to en-hance the feature extraction ability of YOLOX backbone network.By adding continuous void residual convolution and atten-tion mechanism,we design a new feature enhance block(FEB)to extract the output features of the backbone network,which allows continuous expansion residual convolution to be concatenated with 4 different expansion rates of expansion re-sidual convolution,expanding the receptive field of the algorithm,enriching contextual information,while reducing the in-fluence of background on detection.It effectively enhances the algorithm's ability to detect large target scale variations and small targets.The SA attention mechanism is used to suppress the interference of background on algorithm detection and im-prove the detection accuracy of the algorithm.Experiments on the RSOD dataset show that FEB has better feature extraction ability compared to other similar modules.
国家科技期刊平台
NSTL
万方数据
