Segmentation of Plant Leaf Diseases Using a ResNet-Based Algorithm

In computer vision, image segmentation is an important process that divides an image into important and semantically consistent parts, facilitating its structural analysis. This is particularly crucial in identifying and segmenting plant leaves with abrasions, which, in severe infections, can merge to form extensive damage, leading to crop loss and defects such as “Pecky Rice”. Segmentation reveals detailed insights into plant leaf diseases by isolating affected areas, which is essential for effective disease management. Convolutional neural networks (CNNs) represent models that incorporate deep learning specially intended for tasks like image recognition and pixel-level processing. The architecture of CNNs mirrors the hierarchical structure of the human brain, with early layers capturing basic elements like textures and edges, and deeper layers learning complex features and patterns. CNNs employ convolutional layers with local receptive fields, allowing the network to focus on localized patterns through connections to specific input regions. This, combined with weight sharing across convolutional kernels, enables the network to detect features throughout the input space regardless of their position. In our study, we introduce an image segmentation algorithm inspired by the ResNet architecture. While ResNet is traditionally used for image classification, we demonstrate its adaptability to image segmentation when integrated with other segmentation techniques. Our proposed approach leverages the strength of ResNet in feature extraction and object recognition, translating these capabilities to the segmentation of diseased plant leaves, thereby offering a promising solution for agricultural applications where accurate disease detection is paramount.