The risk and damage of wildfires have been increasing due to various reasons including climate change, and the Republic of Korea is no exception to this situation. Burned area mapping is crucial not only to prevent further damage but also to manage burned areas. Burned area mapping using satellite data, however, has been limited by the spatial and temporal resolution of satellite data and classification accuracy. This article presents a new burned area mapping method, by which damaged areas can be mapped using semantic segmentation. For this research, PlanetScope imagery that has high-resolution images with very short revisit time was used, and the proposed method is based on U-Net which requires a unitemporal PlanetScope image. The network was trained using 17 satellite images for 12 forest fires and corresponding label images that were obtained semiautomatically by setting threshold values. Band combination tests were conducted to produce an optimal burned area mapping model. The results demonstrated that the optimal and most stable band combination is red, green, blue, and near infrared of PlanetScope. To improve classification accuracy, Normalized Difference Vegetation Index, dissimilarity extracted from Gray-Level Co-Occurrence Matrix, and Land Cover Maps were used as additional datasets. In addition, topographic normalization was conducted to improve model performance and classification accuracy by reducing shadow effects. The F1 scores and overall accuracies of the final image segmentation models are ranged from 0.883 to 0.939, and from 0.990 to 0.997, respectively. These results highlight the potential of detecting burned areas using the deep learning based approach.

Development of convolutional neural network (CNN) optimized for object detection, led to significant developments in ship detection. Although training data critically affect the performance of the CNN-based training model, previous studies focused mostly on enhancing the architecture of the training model. This study developed a sophisticated and automatic methodology to generate verified and robust training data by employing synthetic aperture radar (SAR) images and automatic identification system (AIS) data. The extraction of training data initiated from interpolating the discretely received AIS positions to the exact position of the ship at the time of image acquisition. The interpolation was conducted by applying a Kalman filter, followed by compensating the Doppler frequency shift. The bounding box for the ship was constructed tightly considering the installation of the AIS equipment and the exact size of the ship. From 18 Sentinel-1 SAR images using a completely automated procedure, 7489 training data were obtained, compared with a different set of training data from visual interpretation. The ship detection model trained using the automatic training data obtained 0.7713 of overall detection performance from 3 Sentinel-1 SAR images, which exceeded that of manual training data, evading the artificial structures of harbors and azimuth ambiguity ghost signals from detection.

Deep learning is a promising method for image classification, including satellite images acquired by various sensors. However, the synergistic use of geospatial data for water body extraction from Sentinel-1 data using deep learning and the applicability of existing deep learning models have not been thoroughly tested for operational flood monitoring. Here, we present a novel water body extraction model based on a deep neural network that exploits Sentinel-1 data and flood-related geospatial datasets. For the model, the U-Net was customised and optimised to utilise Sentinel-1 data and other flood-related geospatial data, including digital elevation model (DEM), Slope, Aspect, Profile Curvature (PC), Topographic Wetness Index (TWI), Terrain Ruggedness Index (TRI), and Buffer for the Southeast Asia region. Testing and validation of the water body extraction model was applied to three Sentinel-1 images for Vietnam, Myanmar, and Bangladesh. By segmenting 384 Sentinel-1 images, model performance and segmentation accuracy for all of the 128 cases that the combination of stacked layers had determined were evaluated following the types of combined input layers. Of the 128 cases, 31 cases showed improvement in Overall Accuracy (OA), and 19 cases showed improvement in both averaged intersection over union (IOU) and F1 score for the three Sentinel-1 images segmented for water body extraction. The averaged OA, IOU, and F1 scores of the ‘Sentinel-1 VV’ band are 95.77, 80.35, and 88.85, respectively, whereas those of ‘band combination VV, Slope, PC, and TRI’ are 96.73, 85.42, and 92.08, showing improvement by exploiting geospatial data. Such improvement was further verified with water body extraction results for the Chindwin river basin, and quantitative analysis of ‘band combination VV, Slope, PC, and TRI’ showed an improvement of the F1 score by 7.68 percent compared to the segmentation output of the ‘Sentinel-1 VV’ band. Through this research, it was demonstrated that the accuracy of deep learning-based water body extraction from Sentinel-1 images can be improved up to 7.68 percent by employing geospatial data. To the best of our knowledge, this is the first work of research that demonstrates the synergistic use of geospatial data in deep learning-based water body extraction over wide areas. It is anticipated that the results of this research could be a valuable reference when deep neural networks are applied for satellite image segmentation for operational flood monitoring and when geospatial layers are employed to improve the accuracy of deep learning-based image segmentation.

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.

Satellite-based flood monitoring for providing visual information on the targeted areas is crucial in responding to and recovering from river floods. However, such monitoring for practical purposes has been constrained mainly by obtaining and analyzing satellite data, and linking and optimizing the required processes. For these purposes, we present a deep learning-based flood area extraction model for a fully automated flood monitoring system, which is designed to continuously operate on a cloud-based computing platform for regularly extracting flooded area from Sentinel-1 data, and providing visual information on flood situations with better image segmentation accuracy. To develop the new flood area extraction model using deep learning, initial model tests were performed more than 500 times to determine optimal hyperparameters, water ratio, and best band combination. The results of this research showed that at ‘waterbody ratio 30%’, which yielded higher segmentation accuracies and lower loss, precision, overall accuracy, IOU, recall, and F1 score of ‘VV, aspect, topographic wetness index, and buffer input bands’ were 0.976, 0.956, 0.894, 0.964, and 0.970, respectively, and averaged inference time was 744.3941 s, which demonstrate improved image segmentation accuracy and reduced processing time. The operation and robustness of the fully automated flood monitoring system were demonstrated by automatically segmenting 12 Sentinel-1 images for the two major flood events in Republic of Korea during 2020 and 2022 in accordance with the hyperparameters, waterbody ratio, and band combinations determined through the intensive tests. Visual inspection of the outputs showed that misclassification of constructed facilities and mountain shadows were extremely reduced. It is anticipated that the fully automated flood monitoring system and the deep leaning-based waterbody extraction model presented in this research could be a valuable reference and benchmark for other countries trying to build a cloud-based flood monitoring system for rapid flood monitoring using deep learning.