An Efficient Pest Detection Framework with a Medium-Scale Benchmark to Increase the Agricultural Productivity

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Insect pests cause significant crop losses, impacting agricultural production globally.
• Traditional pest detection methods are often inefficient and time-consuming.
• Existing deep learning models can be computationally expensive and data-hungry.

Purpose of the Study:

• To propose an efficient and accurate pest detection method for agriculture.
• To improve upon existing deep learning models for pest identification.
• To enhance crop protection strategies through advanced computational techniques.

Main Methods:

• Modification of the YOLOv5s model, including extending the CSP module and improving the SK attention mechanism.
• Development of a medium-scale dataset featuring five common agricultural pests (ants, grasshopper, palm weevils, shield bugs, wasps).
• Comparative analysis of the proposed model against various YOLOv5 configurations.

Main Results:

• The proposed modified YOLOv5s model achieved superior performance in pest detection and classification.
• The enhancements led to accurate localization and classification of both small and large pests.
• Experimental results demonstrated the effectiveness of the modified model over standard YOLOv5 variations.

Conclusions:

• The developed pest detection model shows significant potential for real-world agricultural applications.
• This research contributes to advancing soft computing techniques for crop protection.
• Further research in pest detection can help increase overall agricultural productivity.