Map aerial photography Inpainting for missing error Using AutoEncoder

Recently, the importance of spatial data has been increasing due to the diversification of spatial data collection and utilization. Among them, aerial photography is used as a base data, not only for simulation and open platforms, but also for research purposes. However, aerial photographs can have problems in its quality due to the collection process, and there are many limitations to analyzing and re-collecting them due to large amounts of data, time and cost issues. To solve this problem, we present a system that detects and classifies missing parts in the aerial photographs using CNN models and performs inpainting using AutoEncoder. The systems presented were detected and classified error using the most accurate ResNet18 (75.69%) of the various CNN models and compared the results of the inpainting through VAE and U-Net.