Framework for basil disease detection using artificial intelligence

Abstract

Agriculture plays a vital role in ensuring food security and economic stability across nations. Basil (Ocimum basilicum), a widely cultivated medicinal and culinary herb, is vulnerable to a range of diseases caused by fungi, bacteria, viruses, nematodes and pests. These diseases significantly reduce crop yield and quality, especially in regions where timely and expert diagnosis is unavailable. This research addresses the challenge of early and accurate disease detection in basil by leveraging artificial intelligence, specifically machine learning and deep learning techniques. The thesis begins by examining the importance of basil in agriculture and identifying major diseases that impact its cultivation. A comprehensive literature review highlights existing ML and DL techniques applied to plant disease detection, with particular attention to basil. The research identifies gaps such as limited standardized datasets, poor scalability of existing models and challenges in real-time implementation. newlineTo address these gaps, a methodological framework is proposed that utilizes transfer learning and feature extraction through seven pre-trained convolutional neural network models: EfficientNetB3, EfficientNetV2S, InceptionResNetV2, MobileNetV2, ResNet152V2, VGG16 and VGG19. Two datasets were used: a smaller primary dataset from IEEE Dataport having 1215 images and a larger one from Kaggle having 12,000 images. The performance of each model was evaluated using key metrics including accuracy, precision, recall and loss. Experimental results reveal that the proposed models achieved high classification accuracy, with EfficientNetB3 performing best on the smaller dataset and nearly all models achieving over 94% accuracy on the larger dataset. Notably, the model tested on the Kaggle dataset achieved a peak accuracy of 99.92%, demonstrating robustness and scalability. Comparative analyses show that the proposed models outperform traditional ML and DL approaches. newline