Monitoring system for peanut leaf disease based on a lightweight deep learning model

•Introduction of lightweight DL models: the lightweight YOLOv8n.•High precision and robustness of the models.•Design of mixed lightweight convolution.•Model performance on datasets of various peanut varieties leaf disease detection.•Proposed an automatic system for disease monitoring and leaf disease localization. Plant pathogens are commonly identified in the field based on the typical disease that they can cause. For effective management measures and the selection of highly resistant breeding stock, effective early disease detection and disease identification and localisation are essential. However, traditional detection methods, which rely on field administrators’ experience, are inefficient for large-scale production. This study introduces an automated leaf disease detection system, including edge computing equipment deploying an improved lightweight detection algorithm, Huawei 5G communication equipment and specialised management software. However, existing lightweight deep learning-based methods often fall short in overall performance when considering aspects such as model size, parameters or FLOPs collectively. To enhance the performance of the system, this study introduces an improved YOLOv8n network model that incorporates advancements in the FasterNeXt, DSConv modules and the Generalized Intersection over Union loss function. This improvement aims to lightweight the model to adapt to edge computing devices and achieve real-time peanut leaf disease detection while maintaining high detection accuracy. Experimental results indicate a noteworthy reduction in the number of model parameters and FLOPS, by 31.01% and 45.40%, respectively, when compared with the original YOLOv8n, while achieving a mean average precision of 91.10% and a precision of 89.80%. Moreover, the detection speed on the central processing unit (CPU) and graphics processing unit (GPU) platforms were 19.10 and 72.30 img/s, which were 35.07% and 7.05% better than the original algorithm. Notably, our approach leverages an improved YOLOv8 algorithm for leaf disease detection, supplemented with location data acquired via QR codes for the generated region-wide disease condition map. Field trials have shown that the efficiency of the system in monitoring diseases has increased by 74.19% compared with that in humans, contributing to the early detection of diseases and breeding of peanut leaf disease-resistant varieties, improvement of monitoring efficiency and reduction of labour costs.