Hybrid AI Model With CNNs and Vision Transformers for Precision Pest Classification in Crops.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Crop pests significantly threaten agricultural productivity and food security.
• Effective pest management relies on timely and accurate pest identification.
• Existing methods struggle with visually similar species, background clutter, and variable lighting.

Purpose of the Study:

• To develop a novel hybrid deep learning model for accurate and efficient pest classification.
• To address the limitations of standalone Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) in pest identification.
• To provide a practical solution for real-time pest classification in precision agriculture.

Main Methods:

• Proposed HyPest-Net, a hybrid architecture integrating CNNs for local features, attention mechanisms for salient feature refinement, and a Vision Transformer (ViT-B/16) for long-range dependencies.
• Employed data preprocessing and augmentation techniques to enhance model generalizability.
• Evaluated the model on two benchmark datasets: a rice pest dataset (5 classes) and a farm insects dataset (15 classes).

Main Results:

• HyPest-Net achieved 0.95 accuracy on the rice pest dataset.
• On the rice pest dataset, the model obtained 0.95 precision, 0.95 sensitivity, 0.94 specificity, and 0.94 F1 score.
• The model achieved 0.93 accuracy on the dangerous farm insects dataset.

Conclusions:

• HyPest-Net demonstrates superior performance in pest classification compared to standalone models.
• The hybrid architecture effectively handles challenges like visual similarity and background complexity.
• HyPest-Net offers a lightweight, explainable, and powerful solution for real-time pest classification in precision agriculture.