To reduce the risk of space mission failures due to collisions, there have been efforts to collect and maintain knowledge of the space objects orbiting the Earth. Despite these efforts, there is still lack of information on the size, shape, material, and orientation of space objects, which limits the accuracy of orbital predictions. To address this issue, a number of past works have studied various methods to perform space object identification. Photometric light curves have been exploited as a viable method as it can be applied to small and dim objects across all orbital regimes. However, there are still challenges in developing an accurate space object identification system due to the limited number of light curve dataset and the highly imbalanced datasets. To tackle with the dataset challenges, transfer learning methods can be applied with deep learning classification models to improve the models’ performance in identifying space object. To perform deep learning model pretraining, synthetic light curve data are generated through simulation. After pretrained, the model’s weights are fine-tuned by training it on real light curve data. A past study has shown that the use of transfer learning can improve the performance of a 1D-CNN (Convolutional Neural Network) model. With the rapid emergence of research in deep learning, various models and architectures for performing sequence learning have been proposed in the recent years, which showed promising results. In this presentation, we provide preliminary results of our study, discussions and analysis of our results, and suggestions for future works.