Related Experiment Videos
YOLO-ACBG: an enhanced deep learning model for precision monitoring of wheat stripe rust using different vegetation
Fusheng Yu1,2, Xuanyuan Tang1,2, Qi Liu1,2
1Key Laboratory of the Pest Monitoring and Safety Control of Crops and Forests of the Xinjiang Uygur Autonomous Region, College of Agronomy, Xinjiang Agricultural University, Urumqi, China.
Abstract:
Wheat stripe rust (WSR) is a fungal disease that significantly impacts wheat yield and quality. Pathogen-induced pigment degradation and cellular structural damage cause specific reflectance variations, notably, an increase in the red band due to reduced chlorophyll absorption and a concurrent decrease in the near-infrared (NIR) band caused by mesophyll tissue collapse. These physiological changes drive nonlinear shifts in spectral vegetation indices, which can serve as effective indicators for precise disease monitoring. These physiological changes drove nonlinear shifts in spectral vegetation indices, which can serve as effective indicators for precise disease monitoring. To enable precise monitoring of wheat stripe rust in the field, this study constructed 13 datasets corresponding to specific vegetation indices and performed a comparative analysis, identifying Normalized Difference Vegetation Index (NDVI) as the optimal index for detecting wheat stripe rust. The results on different vegetation index datasets demonstrate that the mean Average Precision at IoU threshold 0.5 (mAP@50) of NDVI achieved 92.11%, which is 6.06% higher than RGB. To further improve model performance in identification by fully leveraging information from vegetation indices, this study proposed a model named YOLO-ADown+ConvFormer+BiLevelRoutingAttention+CGLU (YOLO-ACBG), improved from the YOLO-ADown+ConvFormer (YOLO-AC) model, which demonstrates better performance in WSR monitoring and can be deployed on edge device. This model introduced a Bi-Level Routing Attention mechanism and further integrates a convolutional gated linear unit into the C3K2_ConvFormer module. The results on NDVI datasets demonstrate that YOLO-ACBG achieves an mAP@50 of 94.12%, which is 2.01% higher than YOLO-AC; the recall is 88.59%, representing a 3.83% improvement over YOLO-AC. These results suggest that the YOLO-ACBG model can precisely monitor wheat stripe rust in the field, providing guidance for the precise prevention and control of plant disease.
Related Concept Videos
Light Acquisition
Key Elements for Plant Nutrition
Plant Breeding and Biotechnology