High-Performance Disease Detection in Plant Leaves Using an Enhanced Mask R-CNN

Abstract

The present research model classifies plant leaf diseases with the help of a sophisticated deep learning structure called the Enhanced Mask Region-Based Convoluted Neural Network (EnMask-R-CNN). This model is accurate in classifying and segmenting images and isolating disease-afflicted areas and defining their boundaries perfectly. A threshold-based segmentation approach is also added to enhance region detection and hence, enhance the overall accuracy and computational performance of the model. Proposed EnMask-R-CNN performance is compared with the performance of a standard Recurrent Neural Network (RNN) model in terms of 105 images as a sample size. The results of the experiment demonstrate that the suggested model has a much lower error of 2.657, which is lower than that of the RNN baseline. In addition, the significance value of 0.258 received and the reduced loss metrics proves reliability and strength of the suggested approach. The obtained results prove that the EnMask-R-CNN system is more effective and correct in detecting plant disease than the conventional neural network methods. Evaluation is done on the PlantVillage dataset, which contains over 50K images across 38 classes of plant disease. This large-scale dataset allows for diversity in disease patterns, lighting conditions and background variations, ultimately performing better on robustness and generalization of the proposed model.