Data Augmentation for Wildlife Animal Recognition Using Style Transfer
Seungjin Jung, Yuyeon Lee, Inbok Lee, Joohyun Kang, Seungwook Lim, Jongwon Choi · 2022 IEEE International Conference on Consumer Electronics-Asia (ICCE-Asia) · 2022
Many military facilities are located in the mountain and the forests, so the wildlife animals easily pass through the facilities. To prevent their invasion, military facilities developed an advanced surveillance system to detect wildlife animals, but the insufficient data for nighttime animals has been a challenging problem. To solve the issue, we design two methods to augment and utilize the training data for nighttime wildlife animals by using a style transfer. The first method is designed to transfer the daytime data to be a style of nighttime data, and the second method exchanges the style of nighttime data with that of daytime data. Through the experiments, we show the effectiveness of the two methods, analyzing the augmented training data.