YOLO-CG-HS: A lightweight spore detection method for wheat airborne fungal pathogens

Abstract

The rapid, accurate and real-time online detection of spore concentration of various airborne pathogens in field crops is of great significance in guiding agricultural producers scientifically, enabling them to forecast disease development and implement timely preventive and control measures. This study presents a quantitative spore detection method for two prevalent wheat airborne fungal diseases using the YOLO-CG-HS lightweight model. Initially, the lightweight Context Guided module (CG) is integrated into the original Backbone of YOLOv5s to enhance the capture of global and edge information in spore images. Subsequently, the High-level Screening-feature Pyramid Networks (HS-FPN) module is incorporated into the Head to better integrate multi-scale feature information of spores, thereby improving the model’s detection performance and ability to capture spore micro-targets. The model’s robustness is then tested across various scenarios, including different shapes, densities, and complex backgrounds. Results indicate that the inclusion of both the CG module and the HS-FPN module into the original baseline model significantly reduces the number of model parameters to only 1.21 M. The model’s average precision (mAP) stands at 95.9 %, with an FPS of 152.5, maintaining performance levels similar to the original model. Moreover, the designed model effectively addresses the challenge of identifying difficult and missed cases resulting from spore adhesion and overlap in various airborne wheat diseases. The YOLO-CG-HS lightweight model developed in this study accurately detects various types of pathogen spores while balancing parameters, efficiency, and accuracy. This offers crucial technical support for the model migration and application of low-cost and high-precision embedded field spore capture instruments.