BorB: A Novel Image Segmentation Technique for Improving Plant Disease Classification with Deep Learning Models

This study presents BorB, a novel image segmentation technique designed to enhance plant disease classification using deep learning. The method utilizes the b-channel from both RGB and Lab color spaces to effectively isolate diseased leaf regions. Initially, the blue (B) channel from the RGB image is extracted, followed by Otsu’s thresholding to identify significant intensity boundaries. Parallelly, the image is transformed to Lab color space, where the b-channel is similarly processed and binarized to segment the affected regions. This segmentation improves focus on disease-specific areas, minimizing background noise and enhancing feature relevance. The segmented output is then employed to train and test a VGG16-based deep convolutional neural network. A custom classification layer is added to adapt the model for 15 specific plant disease classes. Data augmentation and normalization ensure robustness during training. Performance evaluation, including accuracy, precision, recall, and F1-score, demonstrates the effectiveness of the BorB approach in improving classification accuracy. Experimental results confirm that leveraging segmented regions guided by dual-channel thresholding significantly boosts classification performance, providing a promising direction for real-world plant disease diagnosis in precision agriculture. The proposed BorB framework serves as an efficient preprocessing step for deep learning pipelines in plant pathology. 

Index Terms— Plant Disease Classification, Image Segmentation, Deep Learning, VGG16, RGB and Lab Color Spaces-Channel Thresholding, Otsu’s Method, Precision Agriculture, Convolutional Neural Networks (CNNs)