Diagnosis of Tomato Plant Diseases Using Pre-trained Architectures and A Proposed Convolutional Neural Network Model
Abstract
Tomatoes are of the most important vegetables in the world. Presence of diseases and pests in the growing area significantly affect the choice of variety in tomato. The aim of this study is to diagnose tomato plant diseases faster and with higher degrees of accuracy. For this purpose, deep learning was used to diagnose some diseases in tomatoes, including bacterial spot, early blight, leaf mold, septoria leaf spot, target spot, mosaic virus, and yellow leaf curl virus were analyzed CNN models. A CNN model with a 2D convolutional three layers, one flatten layer approach and several Keras models, including DenseNet201, InceptionResNetV2, MobileNet, Visual Geometry Group 16 architectures were proposed. The experimental results showed that the accuracy scores were 99.82%, 92.12%, 92.75%, 91.50% and 84.12% training accuracy, respectively. The proposed CNN model provided the opportunity for rapid diagnosis for approximately 14.9 minutes. The results obtained in this study can be used in robotic spraying and harvesting operations.
- Introduction
Tomatoes are among the most widely produced and consumed foods in the world (Solanum lycopersicum L.). According to Turkish Statistical Institute (TSI) 2021 data, 13,095,258 tons of tomatoes were produced (TSI 2022) and they form an important agricultural product in Turkey (FAO 2019).
The cultivation of tomatoes is commonly carried out under greenhouse and in field conditions. In order to increase efficiency in production, many parameters such as climate conditions, diseases and pests should be taken into consideration. Tomato plant diseases can be divided into two groups according to their factors, physiological and pathogens. Physiological factors are typically caused by climatic conditions such as excessive irrigation and malnutrition, sunburn, cracking in fruits, and rot. Viral diseases cause damage such as black spots, brown spots, curl and deformity on the leaves. Early diagnosis of plant diseases is critical in preventing enormous economic and agricultural loss. Every disease in tomato manifests itself in different ways. Tomato early blight disease (Alternaria solani) occurs as spots on the leaves, stems and fruits that can be seen in every phase of the plant. Tomato mildew (Phytophthora infestans) disease starts as pale green spots which later turn brown and then black. Tomato leaf mold (Cladosporium fulvum = Fulvia fulva) is seen as yellow spots on the leaves, and brown mold occurs in the lower parts of these spots in the later stages (Griffiths et al. 2018). Viruses in tomato plants can cause diseases (Nitzany, 1960), one of which is the Tomato mosaic virus disease (ToMV) the symptoms of which manifest as light green and yellow irregular mosaic spots. The misshapen fruit causes damages such as the formation of smaller fruit than normal. Tomato yellow leaf curl virus disease, on the other hand, manifests itself in the form of shrinkage and swelling in the infected leaves, inward curving, stunting and deformity in the plant (Sade et al. 2020). According to Richard et al. (2017), the bacterial spot disease (Xanthomonas vesicatoria) can affect any part of the plant, including the stem, leaf, and fruit. The symptoms of this disease (Pseudomonas syringae pv. tomato) include the forming of stains on all above-ground organs of the plant (Abramovitch et al. 2006).
This symptom begins in the seedling period, and many brown-black spots appear on the leaves and stems of the seedlings. These spots cause drying of the entire seedling over time. Various methods are used to combat these diseases and include cultural measures if the disease is detected early, and chemical methods, in other words pesticides, are applied in the future. The early detection of diseases and determining and distinguishing exactly what factor caused the disease is critical in order to prevent devastation to large numbers of crops. Rapid diagnosis is crucial in preventing serious economic losses for tomato cultivators.
A wide variety of techniques are used for disease diagnosis and classification in tomatoes. Deep learning is one such technique and is based on Artificial Neural Networks, using multiple layers of neurons to extract attributes from raw data, and is designed to simulate the working mechanism of the human brain. Although deep learning, also known as deep neural networks or hierarchical learning, emerged in 2006 as a new field of machine learning, its foundations date back to 1940 (Deng & Yu 2014).
Diagnosing plant diseases via deep learning is popular diagnostic method for many researchers. A higher disease diagnosis rate can be obtained if this dataset is used by comparing it with images from the real environment. A further way to increase the success rate in disease diagnosis is to use hyperspectral/multispectral imaging techniques in deep learning studies. In particular, the diagnosis of diseases in their early stages is possible through the use of these technologies. Disease detection in the early stages means less pesticide application, greater economic gain, and less environmental pollution. In recent years, a number of algorithms have been used to determine plant diseases through the deep learning method (Dhakal & Shakya 2018). Ferentinos (2018) used a database consisting of 87,848 photographs taken in different conditions, both in the laboratory environment and in the field. In the database created, 58 diseases belonging to 25 plant species were examined. The trials highlighted that the VGG CNN model architecture, with a success rate of 99.53%, had the highest classification success. Another study in which the VGG19 model classified with the highest accuracy with a rate of 97.86% was carried out by Turkoglu & Hanbay (2019). Studies on real-time detection of plant diseases and the development of appropriate automation systems have made significant progress in recent years.
Some of the deep learning architectures currently in use have been successfully trained to accurately identify diseases. It is difficult to keep its generalisability because high accuracy must be addressed to the output, which may differ when implemented to real items (PC, tablet, mobile phone, various camera systems, etc.). The goal of this research is to develop a simpler convolutional neural network for diagnosing tomato plant diseases. Various deep learning models [Visual Geometry Group 16 (VGG16), MobileNet, DenseNet, Inception-ResNet] and proposed methods were investigated for this purpose. The proposed model’s results were also compared to existing state-of-the-art methods.
Diagnosis of Tomato Plant Diseases Using Pre-trained Architectures and A Proposed Convolutional Neural Network Model
Dilara GERDAN Caner KOÇ Mustafa VATANDAŞ
Year 2023, Volume 29, Issue 2, 618–629
