Automatic identification of medically important mosquitoes using embedded learning approach-based image-retrieval...

Area of Science:

• Medical Entomology
• Computer Science
• Artificial Intelligence

Background:

• Mosquito-borne diseases like dengue and malaria are major global health threats, particularly in low-income nations.
• Effective mosquito population control is crucial for disease prevention, yet current surveillance methods are resource-intensive.
• Traditional entomological surveillance relies on manual identification, requiring expertise and significant time.

Purpose of the Study:

• To develop an automated screening system for mosquito identification using deep metric learning.
• To enhance the efficiency and accessibility of entomological surveillance for public health initiatives.
• To evaluate the robustness and accuracy of the proposed AI model across various image sources and conditions.

Main Methods:

• Implemented a deep metric learning approach with an image-retrieval process utilizing Euclidean distance.
• Trained a ResNet34 model for mosquito image classification and identification.
• Tested the model's performance using diverse datasets, including images from stereomicroscopes and mobile phones, under varied environmental conditions.

Main Results:

• The deep learning model achieved high precision (up to 98%) and sensitivity (greater than 95%) across different image sources.
• The model demonstrated robustness against variations in lighting, scale, background, and zoom levels.
• The area under the ROC curve exceeded 0.960, indicating practical and empirical performance.

Conclusions:

• The developed deep learning model offers an accurate and efficient automated solution for mosquito identification.
• This AI tool can significantly aid public health authorities in real-time mosquito vector surveillance and control.
• The system provides a scalable and accessible alternative to traditional, labor-intensive entomological methods.