Deep Learning-Based Image Classification for Major Mosquito Species Inhabiting Korea

Area of Science:

• Entomology
• Computer Science
• Public Health

Background:

• Mosquitoes pose a significant global health threat, transmitting deadly diseases.
• Manual mosquito identification is labor-intensive, time-consuming, and prone to errors.
• Accurate species identification is crucial for effective disease prevention and control.

Purpose of the Study:

• To develop an automated image analysis method for rapid and accurate mosquito species identification.
• To leverage deep learning object detection for enhanced mosquito surveillance.
• To reduce human error and labor costs associated with traditional identification methods.

Main Methods:

• Acquisition of color and fluorescence images of live mosquitoes using a specialized capture device.
• Development and training of a deep learning object detection model.
• Evaluation of various deep learning models, including transformer and Faster R-CNN architectures.

Main Results:

• The combined swine transformer and Faster R-CNN model achieved the highest performance.
• An F1-score of 91.7% was obtained, demonstrating high accuracy in mosquito identification.
• The method proved effective in analyzing mosquito species and populations.

Conclusions:

• The proposed automated identification method offers a significant advancement over manual techniques.
• This deep learning approach enables efficient and rapid analysis of vector-borne mosquito populations.
• The technology has the potential to reduce fieldwork labor and improve disease surveillance.