Agriculture is essential to human survival, yet it faces serious threats from plant diseases that reduce crop quality and yield. Early detection is key to managing these diseases effectively and preventing extensive damage. However, traditional machine learning methods often struggle to detect plant diseases at early stages, limiting their effectiveness. To address this, we propose a novel method combining Harris Corner Detection and Graph Neural Networks (GNNs) for early disease detection in tomato and potato plants. The proposed model was tested on two datasets, demonstrating significantly improved performance compared to traditional approaches. Our method achieved accuracy rates of 97% for tomatoes and 99% for potatoes leafs in the Plantvillage dataset, along with high precision, recall, and F1 scores. These results highlight the superior capability of our approach in accurately detecting plant diseases early, offering a robust solution for improving agricultural productivity and disease management.