The team applied instance segmentation to detect body parts of Mosquito species.

The problem

A vector is a living organism that transmits an infectious agent from an infected animal to a human or another animal. Vectors are frequently arthropods, such as mosquitoes, ticks, flies, fleas, and lice.

• 700,000 people die each year from vector-borne diseases.
• 17% of all infectious disease cases are vector-borne diseases.
• 80% of the world’s population is at risk for one or more vector-borne diseases.

Vectech’s current methods for species classification are based on images gathered with MosID, a custom imaging device designed for consistent, high-quality mosquito imaging, which is fed to a Convolutional Neural Network (CNN) based system. We want Omdena to help us develop a mosquito body part segmentation and identification method, to help us determine what parts of the mosquito are visible and intact in the image. This enables more advanced computer vision methods, serving as highly valuable prior information for the CNN, and may be paired with entomological taxonomic information for species identification. These advanced methods are required for mosquito surveillance products because captured mosquitoes from the wild are often very beaten up, missing scales, wings, legs, etc., which sometimes affects whether that mosquito can be accurately identified to its species.

The project outcome 

The team applied various data augmentation techniques and explored different machine learning models for instance segmentation. The solution segments different body parts of the mosquito visible in the image. For example, identifying the specific portions of the image that are the legs, abdomen, wings, and other important body parts of the mosquito. 

Sample prediction. Source: Omdena