Comparative Analysis of Segmentation Models to Detect Leaf Diseases in Tomato Plant

Abstract

Abstract In agriculture around 22% of crop yield loss is due to living and non-living organisms such as biotic and abiotic stress/disease. The early-stage diagnosis of these stresses is an important issue for farmers through naked eyes. Using computer vision technologies can detect the pattern and clustering of diseases at an early stage. However, inthe last few years, computer vision-based deep learning technology is helpful for the diagnosis of biotic stress (single biotic and multi biotic) in tomato plant leaves.In this work, the PlantVillage dataset is gathered for the segmentation of object detection. The labeled, enhanced and augmented data has been used for training the model. The proposed hybrid Deep Segmentation Convolutional Neural Network (Hybrid-DSCNN)model has been segmenting the diseased objects in the tomato plant. This Hybrid-DSCNN is assembled usingU-Net and Seg-Net pre-trained models with instance segmentation for better detection of objects. The result of the proposed Hybrid DSCNN model is compared with other modified U-Net, M-SegNet, and modified U-SegNet in terms of Accuracy, Precision, Recall, Intersection over Union (IoU), and mean Intersection over Union (mIoU). The proposed model processed 1004 images in 30ns,whichis better than other compared models. The accuracy achieved using the proposed model is 98.24%,which is far better than other modified segmentation models. In this paper, the semantic segmented data has been recognized for the single and multiple leaf diseases for identification and classification.