Classifying Nutritional Deficiencies in Coffee Leaf Using Transfer Learning and Gradient-Weighted Class Activation Mapping (Grad-CAM) Visualization

Published in ICCCT 2024, IEEE Conference,DOI:10.1109/ICCCT63501.2025.11019090, Link

Abstract: Coffee is a very popular beverage around the world. Theproduction coffee can be affected by nutrition deficiencies andresult in massive underproduction. This makes early detection ofnutrition deficiency so important. In this research we will usetransfer learning approach to create an automated system that candetect and classify different kind of nutrition deficiency fromcoffee leaf images. The dataset used in the research has 1006 leafimages with 10 classes in total; one of which is a healthy class, Thedataset was pre-processed thoroughly by applying propertechniques. DenseNet 201, InceptionV3 and Xception, these threeCNN models have been utilized for the classification task. Thedeep learning models were fine-tuned by freezing model layers,and hyperparameters were tuned. Those transfer learning modelsboth mainly work as feature extractor with only their lastconvolutional block being trainable and a custom model of ourown was used for classification. The Xception model achieved thehighest accuracy of 83.32%. To interpret how the model is workingwe integrated Explainable AI (XAI) named Gradient-WeightedClass Activation Mapping (GRAD-CAM)
(PDF) Classifying Nutritional Deficiencies in Coffee Leaf Using Transfer Learning and Gradient-Weighted Class Activation Mapping (Grad-CAM) Visualization.