In this study, we propose a novel deep learning framework combining MobileNetV2 and DenseNet121 architectures with a Transformer attention mechanism for multi-crop and multi-disease classification using the CCMT dataset. The MobileNet model, known for its lightweight design and efficient feature extraction, is concatenated with the DenseNet model, which excels in preserving feature continuity and improving gradient flow. To enhance feature refinement and boost performance, a transformer-based attention module is integrated, enabling the model to effectively capture intricate patterns and contextual dependencies across crops and diseases. The CCMT dataset, characterized by its diverse range of crop types and complex disease symptoms, presents significant challenges that are effectively addressed by the proposed hybrid model. The model is trained and evaluated on key performance metrics such as training accuracy \(99.30\%\) , test accuracy \(92.04\%\) , precision \(92\%\) , recall \(93\%\) , and \(F_1\) -score \(93\%\) . Experimental results demonstrate that the proposed MDTACNet model outperforms traditional models, achieving superior classification accuracy and robustness. This research contributes to the advancement of intelligent agricultural systems, providing a reliable solution for early detection and effective management of crop diseases.