Land Cover Map Production of the Sakarya Basin from Multi-Temporal Satellite Images

The proliferation of multi-temporal remote sensing imagery, especially through Sentinel 2 satellites, has reinforced efforts towards the processing of multi-spectral and multi-temporal images. In this paper, we present the results of our study on the production of land cover/land use of the Sakarya basin, employing CORINE ground truths. The main contribution of our study is the exploitation of the temporal dimension through 3 dimensional convolutional neural networks, motivated by their capacity to process data across the temporal dimension of the input patch cube. Our experiments spanning 26 classes at a region-wide scale, show that 3D convolutional neural networks possess a strong potential in this regard.