Hybrid convolutional neural network–random forest framework for automated detection of bell pepper leaf diseases using image analysis

Abstract

Diseases that affect plants considerably lower the production of crops and pose a significant threat to global food security, particularly in high-value crops like bell pepper (Capsicum annuum L.). The timely and sensitive identification of the disease is essential in reducing losses related to yield and paving the way to sustainable management of crops. Although automated plant disease diagnosis has been improved by artificial intelligence (AI)-driven image analysis, single-model methods are not always robust or generalisable. The study proposes a hybrid convolutional neural network-random forest (CNN-RF) framework for the early detection of bell pepper leaf diseases. Deep visual features of leaf images were extracted and subsequently used with a pretrained ResNet50 model and then used to classify the images with the help of a RF ensemble algorithm. Experiments were conducted on a publicly available PlantVillage bell pepper dataset comprising 495 leaf images (199 bacterial spot and 296 healthy samples). The classification accuracy of the proposed hybrid model was 97.98 % with high class-wise precision, recall and F1-scores. Receiver operating characteristic (ROC) analysis produced an area under the curve (AUC) of 0.98, indicating excellent discriminative performance. The dataset was evaluated using a stratified 80 : 20 train-test split. The results show that deep feature extraction combined with an ensemble learning method can improve classification accuracy and offer a scalable solution to assist in precisionagriculture and timely disease control. However, the framework was evaluated only on the PlantVillage dataset and has not yet been validated under real-field agricultural conditions. The framework demonstrates potential for extension to multi-disease classification tasks, although further validation on field-acquired datasets is required.