A Detection Method of UAV Remote Sensing Image Combining RFCABAM and YOLOv7
In view of the low detection accuracy and high missed detection rate caused by factors such as small size and complex background in UAV remote sensing images,a detection method based on an improved YOLOv7 model is proposed.In the backbone network,the RFCABAM module is introduced to replace the existing convolu-tional kernel group,and by calculating the semantic features of the target at the channel and spatial levels,the model can focus more on learning the target sample features.In the feature fusion network,cross-layer and cross-scale con-nection paths are introduced,and a weighted fusion mechanism is adopted to highlight the importance of target detail features in the image.At the detection end,a larger-scale detection head is introduced to detect small-sized targets,and candidate boxes are screened using the EIoU-Adaptive-NMS algorithm to improve the model's sensitivity to dense targets.Experimental results show that all three sets of improvement strategies used can improve the detection accuracy of the model.On visible light and thermal infrared influence datasets,the improved model detection accura-cy is superior to other models in the control group,and the detection speed of improved model under test environ-ment can also reach real-time level.
万方数据
维普
