Termite Pest Identification Method Based on Deep Convolution Neural Networks.

Area of Science:

• Entomology
• Computer Science
• Artificial Intelligence
• Pest Management

Background:

• Termites cause significant global economic damage annually.
• Accurate termite species identification is vital for effective pest management due to the lack of universal control methods.
• Existing automated pest recognition methods, particularly deep neural networks, have not yet been fully developed for termite identification.

Purpose of the Study:

• To develop an automated deep learning classifier for termite image recognition.
• To create a mobile-application-compatible system for identifying termite pest species.
• To evaluate the performance of the deep learning model against human expert accuracy.

Main Methods:

• Acquired 18,000 original smartphone images of four key termite pest species.
• Employed five image segmentation techniques to isolate individual termites.
• Utilized the MobileNetV2 deep learning model for termite classification, incorporating image augmentation techniques.

Main Results:

• Achieved high classification accuracies: 0.947 for soldiers, 0.946 for workers, and 0.929 for both castes.
• Model performance was comparable to that of human experts.
• Image augmentation enabled high accuracy with significantly fewer original images (200 images yielding 1,000 augmented images).

Conclusions:

• The developed deep learning system provides accurate, automated termite identification.
• The mobile-compatible system facilitates termite management for professionals and homeowners.
• Image augmentation significantly reduces the data requirements for model development.