A Deep Learning Approach for Classification and Segmentation of Leafy Vegetables and Diseases
Ashraful Islam, Syeda Rafiatus Sama Raisa, Nahiyan Habib Khan, Ariful Islam Rifat
Abstract
Bangladesh's economy is heavily dependent on agriculture, making the region an exceedingly crucial sector that requires great vigilance. In comparison to other vegetables, leafy greens have yet to garner as much attention as other crops that this study's large-scale analysis comes up with, particularly when categorizing readily available leafy vegetables and identifying the affected area and whether they are affected or not. Plant diseases can cause substantial losses in agricultural productivity, and many researchers are working to address the difficulties farmers encounter in identifying and treating these conditions while diagnosing their diseases. The purpose of this study is to develop a dependable and effective approach for classifying seven varieties of leafy vegetables and detecting seven types of plant diseases precisely in Bangladesh using machine learning and deep learning CNN models. To this end, this paper analyzed and provided solutions using a total of eight models: five classification models (VGG16, VGG19, ResNet50, YOLOv5, and YOLOv8) and three instance segmentation models (YOLOv5, YOLOv7, and YOLOv8). Two distinct datasets, “LeafyVclassify7BD” with 3306 images and “LeafyVdisease7BD” with 4493 images, were developed to classify leaves, segment the diseased area, and classify and detect damaged leaves. The research outputs will help strengthen the region's ability to control plant diseases more effectively while also cutting diagnostic costs and enhancing disease detection accuracy in the country's agricultural surroundings.