Identifying Mosquito Species Using Computer Vision

Background

Mosquito species identification plays a critical role in combating vector-borne diseases, which account for 700,000 deaths annually and make up 17% of all infectious diseases worldwide. Mosquitoes, as vectors, transmit infections that threaten 80% of the global population. Identifying mosquito species accurately is essential for disease surveillance and control.

Vectech, with its custom MosID imaging device, collects high-quality mosquito images for classification using Convolutional Neural Network (CNN)-based systems. However, challenges arise when processing damaged specimens captured in the wild, as missing parts like wings, legs, or scales can impact identification accuracy. To address this, Vectech collaborated with Omdena to develop a mosquito body part segmentation and identification method that would enhance computer vision methods, providing critical prior information for species classification.

Objective

The project aimed to:

• Develop a robust mosquito body part segmentation and identification method.
• Enhance the CNN-based mosquito species identification system by identifying visible and intact mosquito body parts.
• Improve the accuracy of mosquito detection and species identification, even with damaged specimens.

Approach

To tackle the problem, the team adopted the following steps:

1. Data Augmentation: The team utilized advanced augmentation techniques to expand the dataset and simulate various conditions, including damaged body parts.
2. Instance Segmentation Models: Various machine learning models were tested and fine-tuned for identifying specific mosquito body parts.
3. Segmentation Methodology: The solution segmented body parts such as wings, legs, and abdomen, ensuring high-resolution differentiation in images.
4. Tool Integration: Integrated the segmentation data into Vectech’s MosID pipeline, providing actionable insights for species identification using CNN.

Results and Impact

The collaboration led to the development of a highly accurate segmentation model, which successfully identifies and classifies mosquito body parts. Key outcomes include:

• Improved Identification: Enhanced mosquito species identification by isolating intact and visible body parts for analysis.
• Scalability: The solution can process images of damaged mosquitoes, significantly increasing its utility in real-world applications.
• Support for Mosquito Surveillance: The improved methods ensure better detection and classification, critical for monitoring vector-borne disease risks globally.

This project provides a breakthrough in mosquito identification, paving the way for more reliable and scalable mosquito surveillance systems.

Future Implications

The findings from this project can:

• Influence vector surveillance policies by providing accurate data on mosquito species in affected regions.
• Enhance public health strategies aimed at controlling outbreaks of diseases such as malaria and dengue.
• Serve as a foundation for future research in applying AI and computer vision for biological specimen identification, potentially extending to other disease vectors.

This innovative approach ensures that even damaged mosquito specimens can be effectively analyzed, reinforcing the global fight against vector-borne diseases.