Classification of Citrus Leaf Diseases Using DenseNet169: A Deep Learning Approach
Abhishek Bhattacherjee, Tejinder Pal Singh Brar, Monica Gupta, Pratham Kaushik
Abstract
This research explores the use of the DenseNet169 deep learning model for the classification of citrus leaf diseases, with a focus on five disease classes: Black Spot, Canker, Greening, Healthy, and Melanose. The dataset, consisting of 607 images, was split into a 70 % training set and a 30 % testing set. After preprocessing, including resizing, normalization, and grayscale conversion, the model achieved an overall accuracy of 93 % on the testing set of 182 images. The model performed particularly well in classifying Black Spot and Greening, with high precision and recall. However, challenges arose in distinguishing Healthy leaves from diseased ones, leading to some misclassifications. Melanose, with fewer samples, had a perfect recall but lower precision. Evaluation metrics, including precision, recall, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{F 1}$</tex>-score, indicated the model's ability to generalize well, although class imbalance affected performance for rare classes. This study demonstrates the potential of DenseNet169 in plant disease classification and outlines opportunities for future improvements, such as addressing class imbalance and incorporating additional data sources.