Teknik Pembentukan Data Sintesis untuk Peningkatan Dataset Larva Nyamuk

Abstract

There are serious challenges regarding the spread of mosquito-borne diseases, such as dengue fever and malaria. Image classification of mosquito larvae can assist the public in better identifying mosquito larvae species, with the aim of minimizing exposure to mosquito bites that can transmit diseases. However, the current limitation of mosquito larvae image data poses an obstacle to the development of classification systems. This research proposes the use of Variational Autoencoder (VAE) to generate new synthetic mosquito larvae image data that can enrich the dataset. Variational Autoencoder (VAE) are a type of generative model that works by learning latent representations, thereby enabling the generation of synthetic image data. The image data is then used to improve the performance of classification and detection models, with the aim of enhancing model accuracy and generalization in recognizing various types of mosquito larvae, thereby supporting efforts to prevent and control diseases transmitted through viruses in mosquito larvae. The evaluation results are shown through loss values ranging from 29206 to 33806 and the best achieved Fréchet Inception Distance (FID) score of 0.4668. This indicates that the model successfully generated synthetic mosquito larva image data similar to the input images.