Efficient Insect Pests Classification by Integrating Triplet Attention with Pre-trained Deep Neural Networks

Accurate and efficient classification of insect pests is critical for sustainable agricultural practices, yet existing methods often fail to capture fine-grained morphological features due to limited cross-dimensional feature interactions. To address this challenge, this paper proposes a novel deep learning framework that integrates Triplet Attention (TA) with EfficientNetV2-S to enhance discriminative feature learning for pest classification. Experimental results on real-world field images demonstrate the framework’s superiority, achieving 98.85% training accuracy and 94.35% testing accuracy, outperforming ResNet-50, LeNet-5, and other EfficientNet models. The model exhibits exceptional robustness, with stable convergence and lower loss values (0.0465) compared to baseline methods.