Crop insect pest detection based on dilated multi-scale attention U-Net.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Crop pests significantly reduce agricultural yield and quality.
• Accurate and rapid detection of insect pests is crucial for effective pest control.
• Identifying pests on crop leaves is a prerequisite for implementing control strategies.

Purpose of the Study:

• To develop an effective model for detecting irregular, multi-scale insect pests in field conditions.
• To enhance the accuracy and efficiency of crop insect pest detection systems.

Main Methods:

• A novel dilated multi-scale attention U-Net (DMSAU-Net) model was designed for crop insect pest detection.
• The encoder utilizes dilated Inception to capture multi-scale features from pest images.
• An attention module in the decoder focuses on pest image edges, reducing noise and accelerating convergence.

Main Results:

• The DMSAU-Net model achieved a detection accuracy of 92.16% on the IP102 dataset.
• The Intersection over Union (IoU) reached 91.2%, outperforming MSR-RCNN by 3.3% and 1.5% respectively.
• Experimental results demonstrate superior performance compared to existing methods.

Conclusions:

• The proposed DMSAU-Net model is an effective method for insect pest detection.
• Dilated Inception enhances model accuracy, while the attention module improves noise reduction and convergence speed.
• The method shows strong potential for practical application in crop insect pest monitoring systems.