The application of deep learning technology in smart agriculture: Lightweight apple leaf disease detection model

Current models for disease detection in fruit tree leaves suffer from limitations such as low recognition precision, high frequencies of missed and false detections. To address these challenges, an advanced model AppleLite-YoloV8 is proposed in this study. Built on the YOLOv8 architecture, the model incorporates a refined backbone with the EdgeNeXt network, enhancing feature extraction for improved identification precision. Also a novel C2f-SC module integrates SCCONV convolution into the C2f module, creating a lightweight architecture while reducing computational complexity. Additionally, the DySample module adaptively modifies the up-sampling process, boosting resistance to interference and improving detection of small-scale diseases. The MPDIOU module refines bounding box regression loss, enhancing accuracy and robustness for objects of varying dimensions. Experimental results demonstrate the model's effectiveness in detecting common apple leaf diseases such as Alternaria Blotch, Brown Spot, and Grey Spot, achieving precision,and recall values of 97.56%, 94.38%, respectively, with a detection speed of 124.33 fps. With just 29.3 million parameters and 57.6 GFLOPs, AppleLite-YoloV8 is computationally lightweight and suitable for resource-constrained devices. These advancements make it robust, efficient, and practical for real-time disease detection in intelligent agricultural environments.