Abstract

The tomato, a flowering plant of the nightshade family (Solanaceae) scientifically referred to as Solanum lycopersicum, is widely cultivated for its edible fruit. As the agricultural sector transitions from traditional to modern practices, incorporating elements of contemporary farming and greenhouse technology, the tomato remains one of the most commonly consumed vegetables. Traditional methods for identifying and detecting tomato plant diseases rely on field surveys and case studies conducted by plant protection experts; however, these approaches are time-consuming and inaccurate. This decade has seen the development of transfer learning, coupled with Convolutional Neural Network (CNN) models, which represent a noteworthy method for systematising and improving the accuracy of plant infection detection. To categorise tomato plant issues such as spider mites, septoria leaf spot, bacterial spot, and late blight, and to differentiate between sick and healthy plant leaf diseases, this study employed a CNN model in conjunction with a pre-trained VGG-16 model. The pre-trained model achieved a test accuracy of 96.21% and an accuracy of 96.82%. In any case, the proposed CNN model achieved substantially higher approval and grade scores, at 98.05% and 98.1%, respectively. These outcomes indicate that the proposed CNN model has better learned the patterns and characteristics of these diseases. Compared with the pre-trained VGG-16 model, the proposed CNN is more appropriate for diagnosing tomato plant diseases.