A Hybrid IHGT–Mask R-CNN Framework for High-Accuracy Plant Disease Identification

Abstract

The model used in the study is centered on early identification of infection inflicting on leaves and the accurate division of the areas that are affected. To do it, a combined approach of Mask R-CNN and the Improved Histogram Generalization Technique (IHGT) is suggested. The preprocessing pipeline uses several convolutional networks that include five major stages, an input layer, three intermediate enhancement layers and a final output layer. The enhancement layers are aimed at improving the visual appearance of images of plant leaves by reducing noise, fixing non-clarity pixels and making the featured images look more appropriate. IHGT enhances the contrast and light consistency of the leaf images so that the network is able to detect minute patterns of diseases at an early stage of infection. The Mask R-CNN, after being improved, uses the instance segmentation to ensure that the diseased areas of every leaf are isolated perfectly. The Kaggle Plant Disease data is used to test the proposed approach and the hybrid approach proves to be effective in predicting the labeled disease classes at a better level of precision. The model can provide a powerful platform to detect early and precise plant diseases with the combination of the benefits of both modern image enhancement and deep segmentation. The proposed framework is tested on the public PlantVillage dataset containing around 54,000 images of 38 plant disease classes. With 97% overall accuracy, notable improvements in Precision, Recall, F1-score and mAP demonstrate the model's performance as both a classifier and segmentation unit. This marks the first time the Improved Histogram Generalization Technique (IHGT) has been implemented in Mask R-CNN specifically designed to work under low-contrast, noisy and real-world agricultural imaging conditions.