Objective To promote the precise and efficient application of pesticides, this study aims to develop a simple and reliable method for real-time detection of pesticide droplet deposition and spatial distribution on crop leaves.
Method This study proposes a method for detecting foliar fluorescent droplet deposition based on an improved YOLOv8. By screening fluorescent tracers and optimizing their concentrations, suitable experimental conditions for leaf droplet image acquisition were determined, and a corresponding dataset was constructed. Based on YOLOv8-seg, the AdamW optimizer was introduced, an Efficient Multi-Scale Attention (EMA) mechanism was embedded into the backbone network, and a Semantic and Detail Infusion (SDI) feature fusion module was incorporated into the neck structure to enhance detection performance.
Result The fluorescence tracer screening results showed that a 1.0 g/L Fluorescein (a yellow-green fluorescent tracer) solution clearly revealed the spatial distribution of droplets on leaf surfaces. Under this concentration condition, a handheld image acquisition device was developed to collect foliar fluorescence droplet images and construct a dataset for model training and validation. Field experiment results demonstrated that the improved model achieved mAP@0.50 and mAP@0.50–0.95 values of 95.4% and 73.1%, respectively, in object detection tasks. For segmentation mask evaluation, the mAP@0.50 and mAP@0.50–0.95 reached 92.5% and 61.3%, respectively, outperforming the baseline model in overall performance.
Conclusion The proposed foliar fluorescent droplet deposition detection method based on the improved YOLOv8 enables effective acquisition and accurate identification of droplet deposition images on leaf surfaces. The method is simple, reliable, and provides a practical solution for spray deposition evaluation and precision plant protection applications.
DownLoad: