Tomato Plant Leaf Disease Detection Using Image Recognition: A Case Study of Mlali in Morogoro Region, Tanzania

Abstract

Tomato plant diseases pose a big problem as they drastically reduce the quantity of a farm’s yield and also result in poor tomato quality, which may affect users. Detecting and identifying leaf diseases in tomato plants is a big challenge for farmers and agricultural officers due to the lack of necessary knowledge and diagnosis tools. This study developed a diagnostic tool accessible through a mobile phone application that can easily be used in the field. The tool uses image recognition technology to classify tomato disease from affected plants. The methodology used to develop the image recognition model was a deep learning technique using Convolutional Neural Networks (CNN) architecture, trained and evaluated using four different models for detecting bacterial spots, late blight, early blight, and healthy tomato leaf. Those models were ResNet18, ResNet50, InceptionV3, and EfficientNet. Since the existing dataset was limited, the learning approach was used to transfer knowledge (weight and bias) of selected models and use it to train on the existing data of tomato. The dataset contains 1000 images for each class, but for unknown images only contains 100 images used in training, 50 images for each class used in validation (val), and 50 for each class used in the test. The four classes of common tomato leaf diseases, early blight, late blight, bacterial spots, healthy tomato leaf, and unknown images, were used for training, validation, and testing. The EfficientNet model achieved an F-score accuracy of 0.91%, Resnet50 achieved an F-score accuracy of 0.99%, Resnet18 achieved an F-score accuracy of 0.99%, and InceptionV3 achieved an F-score accuracy of 0.84%. The model evaluation results for all classes were efficient since the confusion matrix gave correct precision, recall, and F-score values for both test and validation datasets. The research picked the resnet18 model for integration with mobile applications because it only uses less memory, and it has given high prediction in the classification of tomato diseases compared to other models. The developed system can detect tomato plant leaf diseases and give farmers procedures on how to control and prevent the disease; also, the system has the benefit of supporting smallholder. Farmers and extension officers detect tomato plant leaf diseases, thus helping to detect diseases at an early stage and helping to increase the quality of tomatoes.

Plant phenotyping relevance

トマト葉の画像から病害状態を推定するCNNモデルとモバイル診断ツールを開発・評価しており、植物病害表現型の取得・分類手法が中心である。

Code and data availability

The paper's phenotyping analysis (tomato leaf disease classification with ResNet18/50, InceptionV3, EfficientNet) is built directly on a public Kaggle tomato leaf image dataset, explicitly cited with URL. No author code, models, or other paper-specific assets are reported.