Accurately diagnosing crop diseases is crucial for agricultural productivity and food safety. This study addresses the challenge by developing an AI crop disease diagnosis platform, leveraging the strengths of Convolution Neural Networks (CNNs) and Vision Transformers. The proposed Convolution Self-Guided Transformer (CSGT) model integrates CNN’s local feature extraction with the Self-Guided Transformer’s(SGT) global information processing, enhancing the precision and efficiency of agricultural diagnostics.We selected two datasets with characteristics from Jiangsu and Guangdong provinces to validate our model, respectively representing controlled and uncontrolled agricultural environments. The CSGT model demonstrates exceptional performance in diagnosing crop diseases across diverse backgrounds, overcoming the limitations of current models, especially in complex settings. The CSGT model’s CNN layer, hybrid-scale, and self-guided attention mechanisms ensure accurate diagnoses, even in the presence of background clutter. CSGT has an accuracy of 96.9%(Apple),95.8%(Corn), 96.1%(Grape), and 96.5%(Tomato) when classifying crop diseases in a stable environment, and 95.8%(Rice) when classifying diseases in a complex environment. The research results are anticipated to enhance the effectiveness and stability of natural crop disease recognition applications.

Tomato leaf diseases cause substantial yield and quality losses worldwide, yet reliable detection in real fields remains challenging. Two practical bottlenecks dominate current research: (i) limited data, including small samples for rare diseases, class imbalance, and noisy field images, and (ii) multimodal heterogeneity, where RGB images, textual symptom descriptions, spectral cues, and optional molecular assays provide complementary but hard-to-align evidence. This review summarizes recent advances in tomato leaf disease detection under these constraints. We first formalize the problem settings of limited and multimodal data and analyze their impacts on model generalization. We then survey representative solutions for limited data (transfer learning, data augmentation, few-/zero-shot learning, self-supervised learning, and knowledge distillation) and multimodal fusion (feature-, decision-, and hybrid-level strategies, with attention-based alignment). Typical model–dataset pairs are compared, with emphasis on cross-domain robustness and deployment cost. Finally, we outline open challenges—including weak generalization in complex field environments, limited interpretability of multimodal models, and the absence of unified multimodal benchmarks—and discuss future opportunities toward lightweight, edge-ready, and scalable multimodal systems for precision agriculture.