A MobileNetV2-CBAM CNN Model for Multi-Class Classification of Fruit Crop Diseases and Pests

This study proposes a lightweight deep learning architecture that integrates the Convolutional Block Attention Module (CBAM) into MobileNetV2 for multi-class classification of citrus diseases and pests. Although MobileNetV2 provides high computational efficiency suitable for mobile and embedded plat- forms, its feature representation capability may be limited in complex field environments. To address this limitation, CBAM, which sequentially applies channel and spatial attention mechanisms, was em- bedded within the inverted residual blocks to enhance discriminative feature learning of diseased regions while preserving structural efficiency. Experiments were conducted using a citrus pest and dis- ease dataset provided by AI-Hub. The proposed model achieved an accuracy of 98.1%, maintaining performance comparable to deeper networks such as ResNet-50 and EfficientNet-B1 while significantly reducing the number of parameters and floating-point operations. In addition, the model reduced training time and memory consumption compared to the baseline MobileNetV2. These results indicate that integrating attention mechanisms effectively improves feature representation without substantial computational overhead. Therefore, the proposed architecture achieves a balanced trade-off between predictive performance and computational efficiency, demonstrating practical applicability for real-time pest diagnosis systems in resource-constrained smart farming environments.