Diagnosis of plant leaf diseases using digital image processing and machine learning techniques

Reasons such as rapid population growth in the world, climate change and war have created the need to renew agricultural production processes in a way that will get maximum efficiency from the unit of agricultural area without causing environmental pollution and with the correct use of resources. Smart agriculture is the application of current technologies to agricultural production processes to achieve high efficiency and diversity in production, protect the environment and use water resources correctly. This study is about using digital image processing and machine learning technologies to diagnose plant diseases at an early stage and accurately in order to contribute to the transition to smart agricultural practices in Turkey. The study aims to save cost and time, especially in large agricultural lands. In this thesis study, the plant leaf images used were obtained from the anonymous PlantVillage dataset available on the Kaggle platform. Images of healthy and diseased leaves of potato and tomato plants were used. The quality of the leaf images was enhanced using digital image processing techniques, and a total of 37 features were extracted, including 4 shapes, 12 color, and 21 texture features. The extracted features were then grouped in different ways to create five distinct feature sets. For classification, traditional machine learning techniques Random Forest (RF) and Support Vector Machines (SVM) were selected which are commonly used in classification problems. Both algorithms were trained with the five different feature sets to develop classification models, and the algorithm and feature set combinations yielding the highest performance were identified. The performance of the developed models was evaluated using the confusion matrix, accuracy, and F1-score metrics, and the results were presented in tabular form. The Random Forest algorithm and Feature Set 1 (Color moments and GLCM) that gave the best combination. As a result of the study the model is developed for tomato by using Random Forest algorithm and feature set 1 combination and 91.44% accuracy and 91.47% F1-score success were achieved.