In UAV photogrammetry on shallow water bottoms, there is a problem that the water depth is underestimated due to the influence of water surface refraction. To overcome this issue, a method had been proposed that solves the multi-view and two-media triangulation problem, thereby simultaneously estimating the coordinates of submerged points, water surface elevation, and slope. In this study, we conducted the first field experiment of this method to evaluate the accuracy/precision of this method under ideal conditions. As a result, it was confirmed that the coordinates of submerged points can be estimated with an RMS error of less than 0.1m in the vertical direction. On the other hand, an estimation error of about 0.2m was detected for the water surface elevation, and was attributed mainly to the estimation error of the input camera parameters. Since the true (optimal) camera parameters are unknown in the field experiments, further studies are needed in the future about the effect of errors in camera parameters.

In this study, we investigated the effect of clouds on night light using the Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) in East Asian urban areas. Focusing on the Mie scattering phenomenon that occurs when night light passes through thin clouds, we investigated the effects of clouds using the Gray Level Co-occurrence Matrix (GLCM) features of texture analysis.

To reduce the influence of non-economic activity areas at night, we proposed GLCM features with the background area treated to 0. To verify the effectiveness of the proposed features, we compared the accuracy of cloud classification by machine learning. As a result, the accuracy of GLCM features with 0-processed background area improved by 3 to 5% in Support Vector Classification (SVC) and 0.5 to 2% in Random Forest classifier (RFC). It was found that GLCM contrast and ND (co, ho) are effective features for RFC. ND is a normalized index of contrast and dissimilarity. We also found that the optimal ROI size for GLCM is 33×33 pixels. Finally, as a result of comparing the cloud mask and the RFC results, it was found that the method of this study is effective.

The Normalized Difference Vegetation Index (NDVI) has been used for evaluating various vegetation properties and then it is also effective for improving classification accuracies. However, optical remote sensing imagery is limited by cloud contamination. In this study, NDVI images were simulated using the image-to-image translation methods including CycleGAN, pix2pix and pix2pixHD and then they were evaluated for classifying crop types. A significant improvement was confirmed by adding NDVI images generated by pix2pix or pix2pixHD on Sentinel-1 C-SAR VH/VV polarization data and resulted in overall accuracies of 68.0%.