KLASIFIKASI CITRA HAMA PADA CABAI MERAH MENGGUNAKAN METODE CNN DENGAN ARSITEKTUR INCEPTION-V3 DAN RESNET-50

Red chili peppers (Capsicum annuum L.) are one of Indonesia's high-value vegetable commodities, but their existence is often threatened by pests, including caterpillars, green peach aphids, silverleaf whiteflies, and thrips. Traditional pest identification processes are generally carried out through direct observation, which requires special skills and is time-consuming. This condition has prompted the need for technology-based methods that are capable of automatic, fast, and accurate identification. This study focuses on the application of Convolutional Neural Network (CNN) using two popular architectures, namely Inception-V3 and ResNet50, for the classification of red chili pest images. The dataset used consists of 4,994 images, categorized into four classes and divided into training, validation, and test data. The research was conducted in two scenarios, namely without augmentation and with data augmentation. In the scenario without augmentation, Inception-V3 recorded an accuracy of 97%, while ResNet-50 reached 96%. After applying
augmentation, the results of both improved, with Inception-V3 achieving an accuracy of 99% and ResNet-50 reaching 97%. Analysis using precision, recall, and F1-score shows that augmentation can improve stability and expand the generalization capabilities of the model. Overall, this study confirms that augmentation strategies significantly influence the improvement of CNN performance in red chili pest image classification. From the test results, InceptionV3 showed superior performance compared to ResNet-50, making it a potential primary choice in the development of an automated image-based pest detection system.