A new method for Tomicus classification of forest pests based on improved ResNet50 algorithm.

Area of Science:

• Forestry Science
• Entomology
• Computer Science

Background:

• Tomicus beetles are significant forestry pests, causing severe damage in regions like Yunnan Province, China.
• Accurate identification of Tomicus species is difficult due to minimal morphological differences and limitations of traditional methods.
• There is a need for rapid, efficient, and accurate Tomicus classification models, especially for non-experts.

Purpose of the Study:

• To develop a novel and efficient classification model for identifying major Tomicus species in Yunnan Province.
• To address the challenges posed by difficult morphological identification and time-consuming traditional methods.

Main Methods:

• Collected 6,371 high-resolution images of four major Tomicus species (T. yunnanensis, T. minor, T. brevipilosus, T. armandii) using a handheld microscope.
• Developed a new Tomicus classification model, DEMNet, based on an improved ResNet50 architecture.
• Evaluated DEMNet's performance against ResNet50 using key metrics like accuracy, parameter count, and inference speed.

Main Results:

• DEMNet achieved a classification accuracy of 92.8%, outperforming ResNet50.
• DEMNet significantly reduced the parameter count by 90% (to 1.6 M) while improving accuracy by 9.5%.
• The model boasts a fast inference speed of 0.1193 seconds per image.

Conclusions:

• DEMNet is a lightweight, high-precision model suitable for deployment on embedded devices for real-time pest identification.
• The developed model offers significant potential for improving Tomicus pest management strategies.
• DEMNet provides a viable solution for accurate and efficient identification of forest pests.