EARLY DETECTION OF FUNGAL DISEASES IN MANGO LEAVES: FUSION OF GLCM FEATURES AND VISION TRANSFORMERS WITH DEEP NEURAL NETWORK CLASSIFICATION

Abstract

Early detection of fungal diseases in mango leaves minimizes agricultural losses. Traditional methods like visual inspection and invasive techniques tend to be subjective, destructive, or costly. In contrast, technologies like thermography require controlled environments. As an alternative, automated image analysis provides a non-destructive option, although interpreting ambiguous features poses challenges. This study proposes a novel methodology that integrates Vision Transformers (ViT) and GLCM texture features within a deep neural network (DNN) framework to classify four specific fungal diseases: Anthracnose, Powdery Mildew, Sooty Mold, and Tip Dieback as well as healthy leaves, all from a single RGB image. The approach consists of four key stages: First, leaf segmentation is performed using a convolutional neural network (CNN) to isolate regions of interest (ROIs); second, GLCM metrics are calculated to identify textural anomalies; third, global image analysis is conducted with ViT to capture contextual patterns; and finally, local and global features are fused through a DNN. The experimental results are promising, demonstrating that our approach performs effectively in practical scenarios. Furthermore, the proposed methodology achieves an 8.57% increase in precision, a 36.33% improvement in recall, and an 8.57% enhancement in F1-score compared to conventional classification methods.