Image classification of Aedes mosquitoes using transfer learning / Zetty Ilham Abdullah

The first and most important step in controlling the deadly air-borne diseases like dengue, chikungunya, Zika, and yellow fever is to track the spread of disease-carrying mosquitos. In recent years, the importance, and uses of computer vision, notably image classification, to tackle real-world issues have grown. With roughly 3,500 distinct species of mosquitoes on the planet today, classification is a time-consuming and difficult operation. The advancement and rapid growth of machine learning field should not overlook this issue. Transfer learning concept in machine learning has been shown to improve learning of the targeted task by extending the original algorithm with knowledge gathered from the initial training to improve the performance of new model. This project's model framework utilizes the concept of transfer learning by using pretrained models to classify images of Aedes Mosquitoes according to its species. The architecture is also evaluated based on the performance produced by experiments conducted using different combination of hyperparameters. In all combinations of the hyperparameters employed in the experiment, the use of the pretrained model MobileNetV2 for transfer learning surpasses the use of the pretrained model VGG16. The experimental results demonstrate the success of the model architecture to meet initial objectives of the project and be beneficial for future health workers, entomologists and potentially non-experts.