Along with vessel detection, vessel recognition in high-resolution SAR images was necessary in order to monitor marine vessels effectively. However, lack of target data and phase defocusing of target from its velocity limited the recognition performance, especially when using detectors based on artificial intelligence. This study accordingly proposed effective vessel recognition in high-resolution ICEYE spotlight SAR images consecutively utilizing (i) vessel detector robust to defocused moving vessels and (ii) mitigation of moving target phase distortion. In order to apply quantitative and qualitative training data enhancement, a target velocity SAR phase refocusing function was developed. The proposed target velocity SAR phase refocusing function generated defocused SLC image with respect to different target azimuth velocity, which can be utilized for both training data augmentation and refocusing of velocity-induced phase distortion. Achievement of stable vessel recognition performance was enabled from (i) robust vessel detection on defocused moving vessels and (ii) well-focused detected vessel targets, both of which were consecutively applied using the proposed target velocity SAR phase refocusing function. Vessel detection results demonstrated robust performance regardless of vessel motion and vessel recognition results significantly improved after phase refocusing, both of which were subject to quantitative and qualitative training data enhancement. Performance of the proposed algorithm was analyzed both in terms of phase focusing and velocity estimation. Refocusing performance outperformed that of conventional state-of-the-art autofocusing algorithm, modified Phase Gradient Autofocusing, while azimuth velocity estimation derived the average offset of 0.68 m/s, which was regarded more accurate than previous azimuth velocity estimators based on single-channel SAR image.

This study quantitatively measured vessel rotational motion from satellite SAR data using a target motion phase refocusing function. In contrast to horizontal linear motion of vessel, vessel rotational motion caused by ocean wave force was not able to be monitored from AIS information and accordingly rarely studied. Nevertheless, extensive vessel rotational motion was directly related to vessel motion hazard and potential of maritime accident, which was required to be measured in order to thoroughly analyze vessel movement. A target motion SAR phase refocusing function was proposed, which effectively measured vessel velocity, acceleration, and accordingly its rotational motion, conventionally named as yawing, pitching, and rolling motion. A total of 29 vessels whose velocity exceeding 1 m/s were selected from three different satellite SAR images and analyzed using the target motion SAR phase refocusing function. It precisely measured vessel velocity and acceleration, and subsequently derived horizontal and vertical vessel angular accelerations, which reorganized vessel yawing, pitching, and rolling motion. When compared with AIS-driven motion, estimated azimuth velocity and horizontally projected range acceleration respectively derived RMSE offset of 0.49 m/s and 0.0032 m/s2. Moreover, the proposed phase refocusing function outperformed the conventional SAR phase focusing algorithms in aspect of focusing performance. As the measurement of the vessel rotational motion using the proposed refocusing function was presented, it can be practically applied to monitoring vessel motion hazard in ocean using satellite SAR data.

High-resolution SAR images have become essential data in various fields such as object detection and disaster analysis. However, obtaining high-resolution images is relatively difficult and expensive, leading to ongoing research on methods to generate them. Recently, deep learning has been explored to convert low-resolution images into high-resolution ones. In this paper, we utilize deep learning to generate high-resolution images, and we modify and combine loss functions to enhance the model’s performance. The employed loss function incorporates various functions such as SSIM, MS-SSIM, L1, and L2. The experimental results demonstrate that utilizing the combined loss function outperforms using the existing loss function alone for image generation. When comparing the indicators, the combined loss function yields an SSIM value of 0.6098 and a DISTS value of 0.0542, indicating a 5% improvement over the original loss function.

SAR images have been widely used in the detection of oil spills due to their differing backscattering coefficients allowing them to be easily distinguished from the oil-free surface of the sea. However, the insufficient training data and the presence of look-alikes have always been challenges in oil spill detection. A novel SAR-based self-evolving oil spill detection algorithm was proposed which can automatically generate new training data from whole SAR images and improve its performance through model retraining with training dataset updates. The algorithm was tested on a total of eight SAR images with satisfactory results. We expect that our algorithm will be an effective tool for oil spill detection. Significantly, the self-evolving nature of our algorithm allows it to improve over time with more SAR data.

Mitigation of geometric calibration offset in ocean without utilizing ground control points was investigated in this study. Real-time AIS information on vessels was exploited after preprocessing and accordingly tested against the detected vessels in the SAR image. Repetitive procedure of measuring the offset between the AIS sensor and the vessel detection was conducted and derived the SAR image of which the positioning offset was ameliorated. The proposed geo-location enhancement algorithm demonstrated the possibility of application in real-time vessel monitoring from remote sensing.

Flooding is one of the most serious disasters in Asian counties. Satellites can detect such floods well over large areas, and SAR satellites are particularly useful in determining the extent of flood damage at night or in cloudy weather. However, in order for the SAR satellite to be of practical help in reducing flood damage, the information must be provided to decision makers quickly, and such a real-time monitoring system for Asian countries has not yet been operated. In this study, a fully automated warning system that can quickly detect and display flood events occurring in East Asia was developed using AWS's S3, Lambda, and EC2.

This paper suggests a novel and reliable method to detect water body from Sentinel-1 SAR satellite data using deep learning technique. There have been a lot of studies to extract water body from SAR images with deep learning. Although they achieved good performance, most of them used training data without guaranteeing good quality. In this study, land cover map generated by an official government agency were used for labelling ground truth data. After identifying the acquisition date of aerial photo used for generating the land cover map, vector polygons for river or reservoir were extracted and used as label data. This new method reduced producing time and cost to generate reliable training data. After training our deep learning model, it showed 0.874 of f1score. We also tested our deep learning model to the heavy rain season in Korea (August 2020) and successfully detected river flooding.

In this paper, we propose a partial differential equation model for three-dimensional (3-D) volume reconstruction from 2-D slices. The proposed method is based on the modified Cahn--Hilliard equation for 3-D binary inpainting. In order to accurately satisfy the constraints while obtaining a smooth result, we apply a presmoothing procedure based on anisotropic diffusion to the slices. We discuss the justification for our inpainting model using a $\Gamma$-convergence analysis. After splitting a grayscale image into binary channels, we perform multichannel Cahn--Hilliard inpainting. Then we adopt smoothing and a shock filter as postprocessing to combine the binary inpainting results. We then employ our method to reconstruct a 3-D human body from parallel slices of CT images.