Identifying Insect Pest in Horticultural Crops Using CNN Algorithm

Abstract

This study proposes a Convolutional Neural Network (CNN)-based approach to identify insect pests in horticultural crop images. We apply systematic data preprocessing, augmentation, dataset partitioning, and structured model training to improve classification performance across 16 pest categories from chili, tomato, cabbage, and potato crops. We obtain strong results, where the models successfully learn discriminative visual features and perform reliable automatic classification. This study harvests high accuracy from the Xception model at 99.1%, followed by MobileNetV2 at 97.4%, while EfficientNetB3 shows significantly lower performance at 17.05%. These findings highlight the critical role of model architecture, where Xception demonstrates superior capability in reducing prediction errors and improving overall performance. This study concludes that CNN-based methods, particularly Xception, provide an effective solution for image-based insect pest identification. The results also indicate strong potential for real-world deployment, especially when integrated into mobile platforms for real-time agricultural monitoring.