Citrus leaf disease detection algorithm based on lightweight YOLOv8n

The modernization of agricultural equipment is becoming an major means to promote the development of smart agriculture. Aiming at the problems such as large space occupation and insufficient detection accuracy of citrus leaf disease detection model, this paper makes a lightweight improvement on YOLOv8n model, in order that the model can be easier deployed on the mobile terminal. First, the BiFPN structure is introduced. BiFPN values more informative features, removes nodes that do not contribute much to the network, and reduces the cost of model calculation. Secondly, the weighted fusion strategy which is more suitable for the feature layer with high similarity is used to replace the concatenation operation to effectively improve the model detection accuracy. Finally, the Inner-MPDIoU loss function serves as a substitute for the CIoU loss function. Inner-MPDIoU incorporates auxiliary bounding boxes of varying sizes, aiming to minimize the distance between the top-left and bottom-right corners. that accelerates model convergence and improves detection accuracy. After the improvement, FLOPs and Params decreased by 7.4% and 33.3% respectively, while the average accuracy increased from 89.6% to 91.2%. The experimental results showed that this study realized the lightweight model, and effectively improved the accuracy of citrus leaf disease detection.