Transfer learning-based approach for improving cinnamon diseases and pests management

Abstract

Cinnamon is a widely cultivated species in Sri Lanka and highly valued across various industries. The cultivation of cinnamon is hindered by diseases and pests, which affect the yield and quality of cinnamon production. This study suggested a computerized solution to tackle the problems of conventional human disease diagnostic processes by classifying diseases and pests affected by cinnamon leaf images. Due to the limited availability of data samples, the investigation focused on Transfer Learning. The experiments for the proposed Transfer Learning-based approach were conducted with pre-trained deep learning models as fixed feature extractors and classifiers as Softmax with cross-entropy loss and machine learning algorithms for comparative analysis. All the experimental results of this work demonstrated better performances compared to existing studies. InceptionV3 with a Softmax classifier obtained an accuracy of 98%, precision of 98%, recall of 96%, and F1-score of 97% outperforming other models. The findings of this work highlight the efficiency of pre-trained models for leaf diseases and pests’ classification in cinnamon leaves by providing effective disease and pest management and enhancing the quality of the production.