LuTan-1 (LT-1) is the first civil L-band synthetic aperture radar (SAR) satellite constellation and comprises two identical satellites. LT-1 is designed to fulfill two main requirements, one of which the main tasks is to provide digital surface model (DSM) products covering the areas that are not available using optical satellites. The second task is to provide deformation products to support the geohazard monitoring task. For the first task, LT-1 provides a novel noninterrupted imaging technology to facilitate phase synchronization. For the second task, the orbit maintenance is implemented using a newly proposed in-plane offset semimajor axis and out-of-plane control triggered strategy. Besides, we provide the system performance analysis for the two main tasks. Finally, the first results are produced with root-mean-square-error (RMSE) of the DSM less than 0.7 m in the flat region and 6.7 m in the mountainous region. The RMSE of the first deformation products is less than 2.7 mm in the test region. The results indicate a favorable potential for terrain mapping and deformation monitoring applications.

LuTan-1(LT-1) has finished its in-orbit test in 2023. The inorbit test results show that LT-1 performed well and fulfilled the needs of the users. In order to use the data operationally for natural resources monitoring, especially for geohazard monitoring, we have established the application system for LT-1. In this paper, we have introduced the application system including the geometric technology, the standard products generation method, the basic deformation products generation method. Demonstrations of LT-1 are also provided to support the applications of the satellite data.

The Kinematics of rock glaciers is a key indicator for studying periglacial environment hydrology, climate, and disasters. Differential Interferometric Synthetic Aperture Radar (DInSAR) has been demonstrated to be an effective technique for detecting the deformation of rock glaciers. In this study, we present an investigation of the potential of the Lutan-1 (LT-1) SAR constellation for rock glacier topography mapping and deformation monitoring over a periglacial environment with complex topography using Stacking InSAR methods. Firstly, a high-resolution digital surface model (DSM) is generated by the bistatic formation data of the LT-1 SAR constellation. Compared with the ICESat GLAS points elevation product distributed in the study area, the R2reached 0.96, and the root mean square error (RMSE) reached 1.6m, indicating the accuracy of the LT-1 derived DSM product. Then, the deformation velocities of the rock glaciers were acquired from the Stacking InSAR method over the study area. The results indicated that the LOS deformation velocities of rock glaciers in the study area ranged from −0.8 m/y to approximately 0.6 m/y, with an average velocity amplitude of 0.06 m/y. With 241 of the rock glaciers exhibit active zones. The results demonstrated that the LT-1 SAR constellation could provide a new SAR data source for periglacial landform topography mapping and deformation monitoring.

The L-band differential interferometric SAR satellite, also named as LuTan-1, is the first bistatic spaceborne L-band SAR constellation for multiple applications in China. The Datong, Yulin and Liupanshui coalfields are the main coal production base in China which produce over 100 million tons coal each year and play a significant role for the development of national economy. However, a series of issues such as ground subsidence, landslides and damage of structures are induced by extensive coal mining. Therefore, the preliminary assessment of LuTan-1 SAR data for mining deformation monitoring, including Datong, Yulin and Liupanshui coalmines in China, were implemented in this research. Owing to the high revisit ability and long wavelength of LuTan-1 satellites, obvious subsidence was detected for different coalmines using LuTan-1 SAR data. In comparison with the simultaneous leveling measurements, deformation monitoring with cm accuracy were achieved over the study area.

Lutan-l (LT-1) is the first Chinese SAR satellite constellation that provide continuous deformation images covering the whole China. Three deformation products are designed to conduct the geohazard monitoring tasks. The first is DInSAR deformation field product which can be used for general geohazards investigation. The second is stacking deformation velocity field that can be used for geohazards screening. The third is deformation time-series obtained from multi-temporal InSAR and is specially designed for the continuous deformation regions to discover the deformation rules in temporal domain. In this paper, the LT-1 key features for deformation monitoring are provided. Product examples are shown with the Setninel-1A SAR images covering Datong City, Shanxi province. The three deformation products are expected to be used for geohazard monitoring in China.

The convolutional sequence-to-sequence (ConvS2S) machine translation system is one of the typical neural machine translation (NMT) systems. Training the ConvS2S model tends to get stuck in a local optimum in our pre-studies. To overcome this inferior behavior, we propose to de-train a trained ConvS2S model in a mild way and retrain to find a better solution globally. In particular, the trained parameters of one layer of the NMT network are abandoned by re-initialization while other layers’ parameters are kept at the same time to kick off re-optimization from a new start point and safeguard the new start point not too far from the previous optimum. This procedure is executed layer by layer until all layers of the ConvS2S model are explored. Experiments show that when compared to various measures for escaping from the local optimum, including initialization with random seeds, adding perturbations to the baseline parameters, and continuing training (con-training) with the baseline models, our method consistently improves the ConvS2S translation quality across various language pairs and achieves better performance.

While neural models based on the Transformer architecture achieve the State-of-the-Art translation performance, it is well known that the learned target-to-source attentions do not correlate well with word alignment. There is an increasing interest in inducing accurate word alignment in Transformer, due to its important role in practical applications such as dictionary-guided translation and interactive translation. In this article, we extend and improve the recent work on unsupervised learning of word alignment in Transformer on two dimensions: a) parameter initialization from a pre-trained cross-lingual language model to leverage large amounts of monolingual data for learning robust contextualized word representations, and b) regularization of the training objective to directly model characteristics of word alignments which results in favorable word alignments receiving more concentrated probabilities. Experiments on benchmark data sets of three language pairs show that the proposed methods can significantly reduce alignment error rate (AER) by at least 3.7 to 7.7 points on each language pair over two recent works on improving the Transformer's word alignment. Moreover, our methods can achieve better alignment results than GIZA++ on certain test sets.