Crop diseases pose a serious threat to global food security, and timely and accurate identification is essential for effective prevention and control. Deep learning technologies have advanced rapidly in the field of disease recognition, yet systematic reviews remain relatively limited. This paper aims to comprehensively summarize the research progress in crop disease recognition using deep learning, providing reference for further studies in this area. We systematically review relevant research achievements, introduce fundamental principles and evaluation metrics, summarize commonly used datasets and preprocessing methods, and detail existing work from three perspectives—image classification, object detection, and image segmentation. We analyze current challenges and outline future research directions. In image classification, convolutional neural networks (CNNs) and Transformer architectures have been widely applied. By incorporating attention mechanisms, multi-scale feature fusion, and transfer learning, these approaches alleviate issues caused by complex backgrounds and fine-grained disease confusion. In object detection, two-stage and single-stage algorithms enable simultaneous localization of candidate regions and prediction of bounding boxes, allowing joint determination of lesion locations and categories. Detection methods based on Transformers further enhance global relationship modeling under conditions of complex backgrounds and dense occlusions. In the field of image segmentation, UNet and DeepLab series methods have established fundamental frameworks, while Transformers and hybrid architectures have improved segmentation performance for irregular lesion images. Lightweight network designs reduce computational resource consumption, enabling deployment on mobile devices. Deep learning has achieved significant progress in crop disease identification, yet challenges remain in cross-scenario generalization and real-time processing. Future research can advance practical applications by enhancing generalization capability, optimizing lightweight design, integrating multimodal data, and promoting technology transfer.