外文摘要：Wheat plays a vital role in global food production, trade, and food security concerns. However, it is susceptible to Fusarium head blight (FHB), resulting in decreased yield and quality. With the aim of swiftly detecting FHB in fields to minimize economic losses, we utilize an unmanned aerial vehicle (UAV) outfitted with an RGB sensor to gather images of wheat in open field settings. The high spatial resolution, exceptional temporal effectiveness, and cost-effectiveness of the UAV effectively overcome the constraints of ground-based monitoring, although there are still several challenges associated with the use of UAV-based detection for FHB, such as overexposure caused by strong natural lighting, the diversity in the orientation of wheat spikes, the imbalance in the distribution of the scales of FHB lesions, and the difficulty of detecting tiny FHB lesions against complex backgrounds. To detect tiny FHB lesions, a parallel channel spatial attention (PCSA) module is proposed, which enhances the representation capacity of the output feature map and eliminates the interference of weight coefficients under a serial structure. It can detect tiny FHB targets against complex backgrounds. As an FHB detection algorithm, PCSA-YOLO is proposed, which is based on the YOLO detection framework. At the image preprocessing stage, a cubic power stretching illumination processing algorithm is used to tackle the issue of lost details due to image overexposure. In addition, random angle rotation and random scaling are applied to deal with the challenges arising from variations in the orientation and scale of wheat heads. Our experimental results show that our method achieves a relatively high precision of 80.6 %, a recall of 74.5 %, and a mAP@0.5 of 83.2 %; these results are higher than those of YOLOv5s, which yields values for the precision, recall, and mAP@0.5 of 11.8 %, 3.0 %, and 9.2 %, respectively. The method proposed in this paper can detect FHB in wheat using UAV remote sensing images in agricultural environments.
外文关键词：YOLOv5;object detection;Fusarium head blight;UAV remote sensing;attention mechanism
作者：Bao, Wenxia;Hu, Gensheng;Yang, Xianjun;Huang, Chengpei;Su, Biaobiao
作者单位：Chinese Acad Sci;Anhui Univ
期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE
期刊影响因子：0.0
出版年份：2024
出版刊次：217
原文传递申请：江苏省科技资源（工程技术文献）统筹服务平台