OPTIMAL CLASSIFICATION MODEL FOR INFECTED LEAF DETECTION

Abstract

Plants are considered to be important because they are one of the major source to generate energy for mankind, although providing nutritional, medicinal, and other benefits. Plant diseases can affect the leaf at any time between crop farming, resulting in massive crop production losses and economic market value. As a result, in the farming industry, identifying leaf disease is important. It does, however, necessitate a lot of work, a lot of prep time, and a lot of plant-pathogen awareness. The research is mainly focused on detecting, measuring and classifying the plant diseases by using the machine learning techniques applied on the digital images. Totally, 10k tomato leaf images are collected under ten different infection classes such as Leaf Mould, Bacterial blemish, Early blight, late blight, blight, blight, blight, blight, Spider mites, Septoria leaf spot Target Spot, Mosaic virus, Yellow Leaf Curl Virus, and the healthy leaf are all examples of two-spotted spider mites. The infected leaf detection process is done with five sequential modules such as acquisition of image data, image pre-processing, and image feature extraction, infected leaf classification by applying deep Learning models and various Supervised models and finally, optimal model selection based on the classification accuracy and loss factors. Classification using the Deep learning process is applied on two different data set size such as 5K and 10K images. Out of which, the 10K dataset size produces the accuracy score as 94.49 than 88.75 the accuracy score for 5K data size, which helps to fix the data set size as 10K for the entire classification process. Deep learning model is done with two different approaches such as with and without image augmentation with the accuracy score as 97.50 and 94.49. Support Vector Machine (SVM), K-Nearest Neighbor (KNN) methods, and Decision Tree Classification algorithms are used from classical supervised classifiers. To advance the precision score, a voting classifier method is also applied. Out of the four supervised models used, the SVM classifier model produced the highest accuracy score 86,75, which is lesser than the Deep learning classification with augmentation process.