Visual Explanation of Maize Leaf Diaseases Classification using Squeezenet and Gradient-Weighted Class Activation Map

Maize is the second most important agricultural commodity after rice. In Indonesia, maize is an alternative complimentary food, even in some areas it is used as the main food. The future prospect, maize production was increased for national sufficient. However, there are obstacles for achieving it. One of them is the attack of pests and diseases. In this article, image classification on maize leaf diseases is presented. Image classification is a common task when performing image mining. However, classification without a visual explanation certainly makes it difficult for the user to understand the results. This article aims to classify as well as visually explanation the abnormality or emergence of maize leaf diseases. The research is divided into 2 steps: classification and visual explanation. Classification uses Convolutional Neural Network (CNN) Squeezenet while visual explanation uses Gradient-Weighted Class Activation Map (Grad-CAM). The data experiment used from PlantVillage dataset with 4 classes: healthy, blight, spots, and rust. The percentage of training, validation, and testing data was 60:20:20. Validation using 10 fold cross-validation. The novelty was apply the visual explanation using GradCAM on maize leaf diseases. Performance Measure for classification are 95.2%, 94.03%, and 94.28% for accuracy, precision and recall, respectively.