Image processing based plant species and diseases recognition

Recently, interest in image processing-based pattern recognition and classification has increased with advances in computer technology. In this context, image processing technology; has provided innovations and conveniences in many areas such as medicine, agriculture, geographic sciences, security systems, aerospace, industrial production, and defense industry. In this thesis; The aim of study is to focus on the development of artificial intelligence and image processing based systems for the solution of agricultural problems, mainly on the recognition of plant species, the identifying of new species, the identification of harmful plants and the early detection of plant diseases. For this purpose, existing methods for the identification of plant species and diseases have been developed and new methods have been proposed. These developed and proposed methods are as follows: • Edge Step (ES) method based on the geometric shape of the plant leaf, • A shape, color and texture-based hybrid system using a divided approach to classify leaf-based plant species, • The ROM-LBP method as an improved version of the LBP method used to extract tissue properties. • A deep-textural-based feature extractor model for the recognition of plant leaf species. • Multi-Division Deep Convolutional Neural Network (MD-DCNN) model for identification and classification of plant species, • A new classification architecture based on traditional classifiers with pre-trained deep neural networks using real-time images, • A deep model including different convolution filters based on Deep Neural Networks (DNNs) for the detection of apricot diseases, • Deep Learning Based Plant Diseases and Pests Detection System (Multi-CNN PlantDiseaseNet) In order to evaluate the performance of these proposed methods, existing datasets in the literature were used. In addition, a new dataset consisting of 15 different plant disease images obtained in real-time was constructed and experimental studies were conducted using this dataset. According to the results obtained from the comprehensive experimental studies, the methods developed and proposed for the classification of plant species and diseases have achieved high success. These methods will form the basis for automatic diagnosis and detection system which will be held regarding the solution of agricultural problems in the future. In this thesis, real-time automated Plant Disease Detection Software has been developed by using plant disease images obtained from the natural environment. This system has been designed by using MATLAB GUI platform and it can be easily used by everyone as a desktop application. This software; will support experts, will allow easy and early detection of plant diseases in a short time, will prevent the formation of new infections and loss of efficiency that caused diseases. KEYWORDS: Plant recognition, plant disease detection, feature extraction, deep learning, artificial intelligence, classification.