Research on UAV Canopy Detection Coupled with Lightweight Network GhostNet
Although great progress has been made in the research of canopy detection based on high-resolution remote sensing images,the detection effect based on UAV aerial images is not good,and the detection speed is slow and the real-time performance is poor.In order to solve such problems,this paper aims to obtain tree canopy images from UAVs,improve the YOLOv4 network model from three aspects,and improve the detection speed and accuracy.Firstly,a variety of data augmentation methods are used to amplify experimental samples and improve the robustness of the model.Second-ly,the YOLOv4 network model is improved,and the lightweight network GhostNet is used as the backbone feature extraction network of the improved algorithm,which reduces the model parameters and makes the model lightweight.Finally,the depth separable convolution is used to replace part of the ordinary convolution,which shortens the extraction time under the condition that the extracted features are equally extracted.Experimental results show that the detection accuracy of the improved YOLOv4 model is as high as 98%,and the detection speed of a single sheet reaches 0.166 ms,which can effectively save time and labor costs.
国家科技期刊平台
NSTL
万方数据
维普
