The study aims to successfully discriminate, identify, and count mosquito and non-mosquito objects from images of collected mosquitos in non-controlled environments. Specific objectives include:
1.) To develop a computer workflow that allows successful discrimination of mosquito from non-mosquito objects across multiple variations of input images
2.) To develop a computer workflow that can identify if a discriminated mosquito is of certain species
3.) To develop a computer workflow that counts detected organisms for each mosquito species detected successfully
4.) To develop a streamlined pipeline that allows a continuous workflow of mosquito discrimination, identification, and counting in one pass.
The primary objects of this study are limited to mosquito and non-mosquito objects collected within the UPLB campus and the Los Baños municipality Image datasets are created based on the collected data done through fieldwork as well as laboratory work. Due to the relatively large scale of this study, limitations are placed into the number of mosquito species, limiting it to three species specifically: Aedes aegypti, Aedes albopictus, and Culex pipiens
There are 4 main image datasets prepared for this study. Each are used in the training of the convolutional neural network models to help detect, discriminate, and identify mosquito objects. Furthermore, due to computer hardware limitations , the convolutional neural networks are trained and modeled prioritizing computational efficiency and accuracy over speed of training to ensure that the proposed workflow is working.