Search

The effectiveness of training-based object detection heavily depends on the amount of sample data. But in the field of remote sensing, the amount of sample data is difficult to meet the needs of network training due to the non-cooperative imaging modes and complex imaging conditions. Moreover, the imbalance of the sample data between different categories may lead to the long-tail problem during the training. Given that similar sensors, data acquisition approaches, and data structures could make the targets in different categories possess certain similarities, those categories can be modeled together within a subspace rather than the entire space to leverage the amounts of sample data in different subspaces. To this end, a subspace-dividing strategy and a subspace-based multi-branch network is proposed for object detection in remotely sensed images. Specifically, a combination index is defined to depict this kind of similarity, a generalized category consisting of similar categories is proposed to represent the subspace, and a new subspace-based loss function is devised to address the relationship between targets in one subspace and across different subspaces to integrate the sample data from similar categories within a subspace and to balance the amounts of sample data between different subspaces. Furthermore, a subspace-based multi-branch network is constructed to ensure the subspace-aware regression. Experiments on the DOTA and HRSC2016 datasets demonstrated the superiority of our proposed method.

Canola and wheat are the main oilseed crop and grain crop, respectively, and they often have similar phenological stages. The understanding of the interactions between microwave signals with wheat and canola in different stages is important for their monitoring using synthetic aperture radar (SAR) imagery. This paper investigates the characteristics of canola and wheat through the use of backscattering profiles from multi-year Sentinel-1 images. Large fluctuations are observed for the temporal backscattering profiles of canola and wheat in different growth statuses induced by agrometeorological conditions in different years. The capability and stability of Sentinel-1 for wheat and canola mapping is further investigated using single- and multi-temporal SAR images hosted in Google Earth Engine (GEE) using the random forest classifier. Although different agrometeorological conditions and field managements make the temporal profiles of backscattering variations, the large difference in canopy structure allows SAR images to make the separability of canola and wheat stable on Sentinel-1 images in different phenology stages. The classification accuracies and the feature importance scores from multi-temporal classification in different years show that the backscattering features obtained at flowering to maturity stages make more contributions to the good-quality mapping of canola and wheat than those at other stages. The F1 scores of canola and wheat achieve 0.95 during the canola flowering and podding period, and the minimum F1 scores of 0.85 were also obtained at other stages. These findings show that SAR images have great potential in the good-quality mapping of canola and wheat in a wide phenology window.

GaoFen-3 was the first Chinese civilian C-band synthetic aperture radar (SAR) satellite, launched in August 2016. The need for monitoring the satellite’s image quality has been boosted by its widespread applications in various fields. The efficient and scientific assessment of the system’s radiometric and polarimetric performance has been essential in its more than five years of service. The authors collected 90 images of the Inner Mongolia calibration site, 888 images of the Amazon rainforest, and 39,929 images of the Chinese mainland from 2017 to 2021. This was achieved whilst covering the leading imaging modes, such as the spotlight mode, stripmap mode, ultra-fine mode, wave imaging mode, etc. In this study, we derive a framework that incorporates the man-made corner reflectors (CRs) in Mongolia, the traditional Amazon rainforest datasets, and even the long-strip data in the Chinese mainland (known as CRAS) for the purposes of GaoFen-3 radiometric quality analysis and polarimetric validation over its five years of operation. Polarimetric calibration without recourse to the CRs is utilized to measure the polarimetric distortions regardless of the region, and thus requires a higher calibration accuracy for the GaoFen-3 polarimetric monitoring task. Consequently, the modified Quegan method is developed by relaxing the target azimuth symmetry constraint with the Amazon forest datasets. The experiments based on the CRAS demonstrate that the main radiometric characteristics could reach the international level, with an estimated noise-equivalent sigma zero of approximately −30 dB, a radiometric resolution that is better than 2.9 dB, and a single-imagery relative radiation accuracy that is better than 0.51 dB. For polarimetric validation, the modified Quegan method was utilized to measure the crosstalk for quad-pol products to ensure that it was than −40 dB. Meanwhile, non-negligible channel imbalance errors were found in the QPSII and WAV modes, and they were effectively well-calibrated with strip estimators to satisfy the system design.

Marine ship detection by synthetic aperture radar (SAR) is an important remote sensing technology. The rapid development of big data and artificial intelligence technology has facilitated the wide use of deep learning methods in SAR imagery for ship detection. Although deep learning can achieve a much better detection performance than traditional methods, it is difficult to achieve satisfying performance for small-sized ships nearshore due to the weak scattering caused by their material and simple structure. Another difficulty is that a huge amount of data needs to be manually labeled to obtain a reliable CNN model. Manual labeling each datum not only takes too much time but also requires a high degree of professional knowledge. In addition, the land and island with high backscattering often cause high false alarms for ship detection in the nearshore area. In this study, a novel method based on candidate target detection, boundary box optimization, and convolutional neural network (CNN) embedded with active learning strategy is proposed to improve the accuracy and efficiency of ship detection in nearshore areas. The candidate target detection results are obtained by global threshold segmentation. Then, the strategy of boundary box optimization is defined and applied to reduce the noise and false alarms caused by island and land targets as well as by sidelobe interference. Finally, a lightweight CNN embedded with active learning scheme is used to classify the ships using only a small labeled training set. Experimental results show that the performance of the proposed method for small-sized ship detection can achieve 97.78% accuracy and 0.96 F1-score with Sentinel-1 images in complex nearshore areas.

Synthetic aperture radar (SAR) is a significant application in maritime monitoring, which can provide SAR data throughout the day and in all weather conditions. With the development of artificial intelligence and big data technologies, the data-driven convolutional neural network (CNN) has become widely used in ship detection. However, the accuracy, feature visualization, and analysis of ship detection need to be improved further, when the CNN method is used. In this letter, we propose a two-stage ship detection for land-contained sea area without a traditional sea-land segmentation process. First, to decrease the possibly existing false alarms from the island, an island filter is used as the first step, and then threshold segmentation is used to quickly perform candidate detection. Second, a two-layer lightweight CNN model-based classifier is built to separate false alarms from the ship object. Finally, we discuss the CNN interpretation and visualize in detail when the ship is predicted in vertical–horizontal (VH) and vertical–vertical (VV) polarization. Experiments demonstrate that the proposed method can reach an accuracy of 99.4% and an F1 score of 0.99 based on the Sentinel-1 images for a ship with a size of less than 32 × 32.

Time series analysis (TSA) based on multi-temporal polarimetric synthetic aperture radar (PolSAR) images can deeply mine the scattering characteristics of objects in different stages and improve the interpretation effect, or help to extract the range of surface changes. However, as far as classification is concerned, it is difficult to directly generate the classification map for a new temporal image, by the use of conventional TSA or change detection methods. Once some labeled samples exist in historical temporal images, semi-supervised domain adaptation (DA) is able to use historical label information to infer the categories of pixels in the new image, which is a potential solution to the above problem. In this paper, a novel semi-supervised DA algorithm is proposed, which inherits the merits of maximum margin criterion and principal component analysis in the DA learning scenario. Using a kernel mapping function established on the statistical distribution of PolSAR data, the proposed algorithm aims to find an optimal subspace for eliminating domain influence and keeping the key information of bi-temporal images. Experiments on both UAVSAR and Radarsat-2 multi-temporal datasets show that, superior classification results with the average accuracy of about 80% can be obtained by a simple classifier trained with historical labeled samples in the learned low- dimensional subspaces.

The combination of transfer learning and remote sensing image processing technology can effectively improve the automation level of image information extraction from a remote sensing time series. However, in the processing of polarimetric synthetic aperture radar (PolSAR) time-series images, the existing transfer learning methods often cannot make full use of the time-series information of the images, relying too much on the labeled samples in the target domain. Furthermore, the speckle noise inherent in synthetic aperture radar (SAR) imagery aggravates the difficulty of the manual selection of labeled samples, so these methods have difficulty in meeting the processing requirements of large data volumes and high efficiency. In lieu of these problems and the spatio-temporal relational knowledge of objects in time-series images, this paper introduces the theory of time-series clustering and proposes a new three-phase time-series clustering algorithm. Due to the full use of the inherent characteristics of the PolSAR images, this algorithm can accurately transfer the labels of the source domain samples to those samples that have not changed in the whole time series without relying on the target domain labeled samples, so as to realize transductive sample label transfer for PolSAR time-series images. Experiments were carried out using three different sets of PolSAR time-series images and the proposed method was compared with two of the existing methods. The experimental results showed that the transfer precision of the proposed method reaches a high level with different data and different objects and it performs significantly better than the existing methods. With strong reliability and practicability, the proposed method can provide a new solution for the rapid information extraction of remote sensing image time series.

Fusion of the high-spatial-resolution hyperspectral (HHS) image using low-spatial- resolution hyperspectral (LHS) and high-spatial-resolution multispectral (HMS) image is usually formulated as a spatial super-resolution problem of LHS image with the help of an HMS image, and that may result in the loss of detailed structural information. Facing the above problem, the fusion of HMS with LHS image is formulated as a nonlinear spectral mapping from an HMS to HHS image with the help of an LHS image, and a novel cluster-based fusion method using multi-branch BP neural networks (named CF-BPNNs) is proposed, to ensure a more reasonable spectral mapping for each cluster. In the training stage, considering the intrinsic characteristics that the spectra are more similar within each cluster than that between clusters and so do the corresponding spectral mapping, an unsupervised clustering is used to divide the spectra of the down-sampled HMS image (marked as LMS) into several clusters according to spectral correlation. Then, the spectrum-pairs from the clustered LMS image and the corresponding LHS image are used to train multi-branch BP neural networks (BPNNs), to establish the nonlinear spectral mapping for each cluster. In the fusion stage, a supervised clustering is used to group the spectra of HMS image into the clusters determined during the training stage, and the final HHS image is reconstructed from the clustered HMS image using the trained multi-branch BPNNs accordingly. Comparison results with the related state-of-the-art methods demonstrate that our proposed method achieves a better fusion quality both in spatial and spectral domains.