A hybrid ResNet-101 and random forest framework for high-precision multiclass tomato leaf disease classification

Abstract

Tomato (Solanum lycopersicum) is among the most common horticultural crops that are grown all over the world due to its rich nutritional value as well as the wide use of tomatoes in everyday cooking. In spite of its value, the production of tomatoes suffers a lot because of the spread of leaf diseases that have the potential of lowering the crop by a huge margin and undermining the quality of the fruits. Early and proper detection of the disease is thus necessary to help in sustaining agricultural practices and reducing losses that may be incurred by farmers. The paper introduces a smart yet very precise disease prediction model which combines deep learning and conventional machine learning methods. The proposed system is a combination of ResNet-101 strong feature extraction and other customized convolutional neural network (CNN) layers to ensure the extraction of the fine-grained disease specific patterns. The features obtained are then determined as categories with the help of a Random Forest algorithm that increases the strength and trustworthiness of the final forecasts. The model was trained on a varied dataset of ten different tomato leaf diseases of which it was able to generalize over a broad scope of pathological conditions. The experimental findings indicate that the hybrid architecture attains an amazing classification accuracy of 98.72% which shows that it is effective in detecting the symptoms of a disease in its early stages. In general, the suggested solution has a high possibility of real-time implementation within agricultural settings. This system can help farmers to make timely interventions by facilitating automatic and accurate diagnosis of the disease in the field thus enhancing crop health, crop productivity, and sustainable production.