Enhancing Cinnamon Plants Disease Detection Using Advanced Machine Learning Techniques

Abstract

Cinnamon is a vital crop in Sri Lanka. However, some diseases can cause serious damage to the cinnamon plant. Existing methods for detecting diseases in cinnamon plantations are constrained by limited expertise and the time-intensive nature of manual testing. Because of that, there is a critical need for an efficient, comprehensive, and accurate disease identification system. The lack of a reliable and automated solution for detecting cinnamon diseases is the primary research problem. This study aims to develop an advanced deep-learning model capable of detecting and classifying six types of cinnamon diseases (Rough Bark, Stripe Canker, Leaf Blight, Black Sooty Mold, Leaf Gall Forming Louse, Leaf Gall Forming Mites), using a dataset of 1,200 images collected through field visits and Kaggle. The research explores various deep-learning models for disease classification. Baseline models starting with DenseNet, followed by VGG16 and InceptionV3, demonstrated test accuracies of 46.78%, 73.39%, and 73%, respectively. Further improvements were achieved using transfer learning, where EfficientNetB4 achieved the highest accuracy of 92%, followed by InceptionV3 (90%), ResNet50 (89%), and VGG16 (87.5%). The ultimate goal of this research is to develop an AI-powered mobile application that will enable farmers to quickly and accurately identify cinnamon diseases, facilitate timely interventions, enable effective crop management, and minimize agricultural losses. The system is designed to be user-friendly, accessible, and deployable in real-world farming environments, ensuring practical benefits to the agricultural community.