PestDetectSim: an integrated approach for crop pest diagnosis using object detection and similarity-based image...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate pest diagnosis is crucial for crop yield and agricultural productivity.
• Current deep learning methods often misclassify pests due to reliance on single highest-probability outputs.
• A need exists for user-assisted validation to improve diagnostic reliability in real-world agriculture.

Purpose of the Study:

• To develop an integrated framework combining automated pest detection with user-assisted validation.
• To enhance the reliability and practical usability of AI-driven pest diagnostic tools.
• To provide farmers with a system that offers interpretable results and facilitates verification.

Main Methods:

• Developed PestDetectSim, a framework integrating YOLO v8 object detection with similarity-based image retrieval.
• Utilized a backbone network with a Squeeze-and-Excitation (SE-Net) module for feature vector extraction.
• Evaluated the system on a real-world dataset of 30 pest species.

Main Results:

• PestDetectSim achieved 98.82% end-to-end diagnostic accuracy on a diverse pest dataset.
• The system provides object detection results and a ranked list of similar reference images for user verification.
• End-to-end inference time is approximately 60 ms per image, suitable for real-time deployment.

Conclusions:

• PestDetectSim offers practical reliability through integrated automated detection and user-assisted verification.
• The system enhances diagnostic interpretability via case-based visual evidence.
• A prototype smartphone application demonstrates the tool's potential for real-time pest monitoring and sustainable agriculture.