The occurrence of apple leaf diseases has a significant impact on apple quality and yield. Disease monitoring is therefore an important measure to ensure the healthy development of the apple industry. Based on the ResNet structure, a lightweight disease recognition model based on multiscale feature fusion is proposed. First, the feature fusion mechanism is used to extract and fuse the high-dimensional and low-dimensional features of the network, strengthen the transmission of semantic information between convolution layers, and enhance the ability to distinguish subtle lesions. Next, multi-scale depth separable convolution is added to extract disease features of different scales by using multi-scale convolution kernel structure, which improves the richness of features and restricts the parameters of the model. Finally, a dataset containing five kinds of apple leaf diseases is used to verify the effectiveness of the proposed method. The experimental results show that the recognition accuracy of the model is 98.05%, and that the number and calculation of the model network are only 4.02 MB and 0.92 GB, respectively. Compared with other models, it also has advantages, and can provide a new scheme for the accurate identification of diseases and pests in agricultural automation.