To solve the problems of missing detection, false detection and low accuracy in the detection process, aiming at the different sizes of the disease manifestations of citrus leaves, the model of improved CenterNet is proposed. The feature enhancement IASPP module was introduced into a series of residual structures of the first two residual layers of the feature extraction network RestNet50 to expand the shallow receptive field, obtain more detailed information of small target leaf diseases, and enhance the significance of shallow features. A bidirectional weighted feature fusion BiFPN module was introduced to effectively integrate the shallow and deep leaf disease information of the feature extraction network. In order to improve the overall detection effect, the multi-scale attention mechanism MS-CAM was introduced. The trained model was used to detect citrus disease leaves. The experimental results showed that, compared with the original model CenterNet, the R-value of the proposed model increased by 5.2%, mAP increased by 4.1%, and AP0.5:0.95 increased by 22.7%. It can achieve accurate detection of small target, medium target and large target diseases in citrus leaf planting.