<![CDATA[This paper investigates the effectiveness of six pre-trained deep learning models to classify images of agricultural plant insects. We utilized the BAUInsectv2 dataset, which includes images from nine classes. Aphids, Armyworm, Beetle, Bollworm, Grasshopper, Mites, Mosquito, Sawfly, and Stem borer. The models, namely Xception, MobileNetV2, ResNet50, EfficientNetV2B3, ResNet101, and DenseNet121, are fine-tuned by transfer learning from ImageNet. This approach significantly reduces training time while improving classification accuracy. Our experiments reveal that each model reliably distinguishes between insect species even when faced with varying lighting conditions and diverse viewpoints. To further clarify how these models make predictions, we employ Gradient-weighted Class Activation Mapping (Grad-CAM) to highlight critical regions in the images. The results demonstrate that each model focuses on unique biological features and offers clear explanations for its decisions. The research results contribute to demonstrating the potential of pre-trained deep learning architectures for agricultural monitoring and pest management, paving the way for promising future applications.]]>
Copyrights © 2025
