Intelligent Diagnosis of Potato Leaf Diseases Using NLSCTAN and GPT-4 Powered Multimodal Framework

To enhance crop production and ensure food security as well as reduce excessive use of chemical pesticides in the modern farming, it is important to detect potato leaf diseases early and precisely. However, environmental changes, the shortcomings of unimodal image based techniques, and visual symptom overlap have posed a hard task in disease recognmethition. The study offers an intelligent multimodal network of potent segmentation and classification of potato leaf diseases. It is constructed on the basis of the Normalized Long-Short Convoluted Temporal Attention Network and implemented thereof with a decision support system powered by GPT-4. This is aimed at addressing these issues. It has three primary divisions namely, the deep visual feature extraction branch (MSC-ResViT), the semantic and textual feature learning branch (MSC-TextCNN), and the color-texture representation branch (CT-CNN). These features are fused and the NLSCTAN classifies them into complex spatial-temporal connections. The outcomes of the experiments on the PlantVillage and Potato Leaf Disease datasets are promising. The proposed model has 98.2 and 95.6% accuracy rate on the PlantVillage and Potato Leaf Disease datasets respectively. The class-wise analysis shows that, the F1-score of the Healthy class is 0.95, the Early Blight class is 0.90, and the Late Blight class is 0.91. Moreover, the connection with GPT-4 enables the PotatoGPT platform to interpret illnesses and give practical treatment guidance in real-time, which can add greater practicality to the platform to be applied to precision agriculture.