Chili Leaf Health Classification using Xception Pretrained Model

Abstract

As one of the high-demand horticultural crops, chili peppers have a significant impact on the economy of Indonesia. However, despite the growing demand and interest in chili peppers, their production often faces disruptions due to crop failures. One of the leading causes of such failures is pests and diseases. Among all parts of the chili plant, chili leaves are the most susceptible to damage. Distinguishing between healthy and unhealthy chili leaves can serve as an early detection step for chili diseases and preventive measures to contain their spread. Convolutional Neural Network (CNN) are effective algorithms for image classification. The development of CNN has led to the use of models previously trained on large datasets to accurately classify relatively small datasets. One such pretrained model known for its exceptional classification capabilities is Xception. By utilizing the pretrained Xception model trained on the ImageNet dataset for the classification of healthy and unhealthy chili leaf images, our model achieved an accuracy of 91% on a dataset containing 2136 images. Furthermore, the model achieved a 100% success rate by correctly predicting all 10 out of 10 given images.