Research on Citrus Pest and Disease Detection Based on YOLOv8-CAA

Citrus is an important cash crop in the world. Its yield and trade volume contribute greatly to the agricultural economy, but pests and diseases have a significant impact on its yield. The annual loss of citrus yield caused by pests and diseases is as high as $20 \%-30 \%$. Therefore, the detection and control of citrus pests and diseases has become an important topic in the current society. However, traditional pest detection relies on human experience, which is inefficient and costly. In this paper, a high-quality disease and pest detection model, yolov8 CAA (you only look once version 8-caa) model, was proposed to achieve efficient and accurate automatic detection of diseases and pests, promote the development of intelligent agriculture and reduce economic losses. Based on the yolov8 algorithm, the adaptive attention mechanism coordinated attention is integrated to enhance the features by emphasizing the information representation, which makes the algorithm lighter and improves the detection accuracy at the same time; Add adown module to optimize convolution, and reduce the spatial dimension of feature map by adjusting the step of convolution layer; At the same time, the ASFF adaptive spatial feature fusion head is introduced, and the yolov8 CAA model is proposed to adaptively fuse different levels of features together to improve the ability of multi scene recognition; The average recognition accuracy and detection accuracy of yolov8 CAA model on the experimental data set were $83.9 \%$ and $92 \%$, respectively, which were improved by $3.4 \%$ and 2.4% compared with the original model. Therefore, the method proposed in this study can improve the detection robustness of crop diseases and pests, and enhance the ability of the model to extract the features of small target images, so as to detect and recognize pests and diseases efficiently and accurately.