Rice Disease Detection Based on Improved YOLOv8n

Rice is one of the most important food crops in China, with an annual sowing area of about <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{3 0}$</tex> million hectares. However, pests and diseases can seriously affect the yield of rice and become an important factor restricting agricultural development. To solve the problems of difficulty in extracting small target features and low detection accuracy in complex environments in rice disease detection, this study proposes a rice disease recognition method YOLOv8-CL based on improved YOLOv8n. This method improves the detection performance of small targets by introducing a lightweight upsampling CARAFE operator (Content Aware ReAssembly of Features), and enhances multi-scale feature fusion by improving the Spatial Pyramid Pooling Fast (SPPF) module to the Large Separable Kernel Attention (LSKA) module. The experimental results show that the accuracy, recall, and average precision of the improved YOLOv8-CL model are <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{92.6} \%, \mathbf{84.7} \%$</tex>, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 2. 0 \%}$</tex>, respectively; Compared with the original base network YOLOv8n, it has improved by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1.9 \%, 2.9 \%$</tex>, and 2.7 % respectively. This study significantly improved the accuracy and deployment efficiency of rice disease detection by improving the model, providing a technical foundation for real-time disease monitoring in intelligent agriculture.