Automatic ship detection in SAR images plays an essential role in both military and civilian fields. However, most of the existing deep learning detection methods introduce complex models and huge calculations while improving the detection accuracy, which is not conducive to the application of real-time ship detection. To solve this problem, an efficient lightweight network YOLO-Lite is proposed for SAR ship detection in this paper. First, a lightweight feature enhancement backbone (LFEBNet) is designed to reduce the amount of calculation. Additionally, a channel and position enhancement attention (CPEA) module is constructed and embedded into the backbone network to more accurately locate the target location by capturing the positional information. Second, an enhanced spatial pyramid pooling (EnSPP) module is customized to enhance the expression ability of features and address the position information loss of small SAR ships in high-level features. Third, we construct an effective multi-scale feature fusion network (MFFNet) with two feature fusion channels to obtain feature maps with more position and semantic information. Furthermore, a novel confidence loss function is proposed to effectively improve the SAR ship target detection accuracy. Extensive experiments on SSDD and SAR ship datasets verify the effectiveness of our YOLO-Lite, which can not only accurately detect SAR ships in different backgrounds but can also realize a lightweight architecture with low computation cost.

Persistent scatterer interferometric synthetic aperture radar (PSInSAR) technology provides a powerful tool for detecting ground surface displacements. However, one of its major limitations is an insufficient number of coherence points due to decorrelation. In this article, we propose an effective ground surface displacement monitoring approach based on the Stanford method for persistent scatterer (StaMPS). This approach dramatically increases the density of coherence points by jointly exploiting persistent scatterer (PS) and distributed scatterer (DS) points. In particular, we develop a new DS phase estimation method based on nonlinear optimization estimation (NLE) and exploit it in our approach. The NLE can obtain better DS phase noise reduction, can detect more coherence points, and requires slightly less computational time than conventional maximum-likelihood estimation (MLE). To demonstrate the effectiveness of our new approach, we apply it over 24 Envisat SAR images acquired over Tucson, Arizona in the USA. The coherence pixel density is increased substantially (> 6 times) compared with the conventional StaMPS-based PSInSAR approach. As for the improvement of our NLE over MLE, it can detect 11.7% more coherence pixels, can achieve better phase noise reduction, and needs approximately one third less computational time. These results validate the effectiveness of our new approach and suggest its great potential for land displacement measurements.

The MOD10A1 daily snow cover product offers a fine temporal resolution, which is particularly important for time-critical applications. Although the reliability of this product was mostly verified by ground measurements before use, the inadequate representativeness of point-scale observations may lead to errors in the assessment of MOD10A1. In this study, based on the Google Earth Engine, MOD10A1 was thoroughly evaluated by a total of 20,131 binary snow maps generated from 30 m Landsat imagery for the 2000-2016 snow seasons in three typical snow regions across China. The three typical snow regions included Northern Xinjiang, Northeast China and the Tibetan Plateau. In general, the accuracy of MOD10A1 under clear sky conditions is 86.5% based on ground observations from meteorological stations and 90.3% based on Landsat imagery. Compared with ground measurements, snow cover derived from Landsat images is in better agreement with MOD10A1 due to a better spatial match. The error analyses also indicate that 1) compared with land cover, topography has larger effects on the accuracy of MOD10A1 over the three snow regions with statistically significant negative relationships, particularly in Northern Xinjiang (R = -0.604, P < 0.05); and 2) MOD10A1 shows a relatively lower accuracy of 81.8% from 2006 to 2008 due to the decrease in snow depth during this period (R = 0.58, P < 0.05) based on the Mann-Kendall test.

Snowmelt is a natural water resource, and its distribution is essential for understanding regional climate change and hydrological cycle. In this study, using dual-polarized C-band Sentinel-1 SAR data, we propose a new wet snow detection algorithm that considers the effects of radar incidence angle, land cover types and topography for snow regions where prairie snow is the main snow cover type. We quantitatively evaluate the performance of the proposed method by optical-derived snow maps. The results show that (1) the overall accuracy of the proposed method is 80.8% (dry snow reference) and 79.0% (snow-free reference); while using a conventional threshold of −2 dB, the accuracy is 32.1% and 43.5%, respectively; (2) −2 dB can only be applied on the ratio image calculated by the snow-free reference to extract wet snow in cropland; (3) adding topographic information greatly improves the identification accuracy, especially for grassland and barren with higher topographic complexity.

The first evaluation of GaoFen-3 SAR radiation characteristics and its potential for quantitative parameter estimation in land surface is presented in this letter. Based on two triangle corner reflectors, five impulse response property parameters are calculated to assess image quality of GaoFen-3 SAR, and results show that the peak sidelobe ratio is superior to system specification and the integrated sidelobe ratio cannot be assessed by misconduct in the field campaign mainly for strong background scattering (not low enough). Five distributed target parameters are extracted from three categories and compared with Radarsat-2 and Sentinel-1A quasi-synchronous images in two days. The results indicate good radiometric resolution (RR) of 3 dB and good equivalent number of looks closed to 1 for GaoFen-3 single-look complex product, which proves GaoFen-3 SAR image is at the same quality level with the other two C-band SAR images and its RR meets system design of 3.5 dB. Then, the backscatter coefficient of these three SAR images is compared at bare soil area after GaoFen-3's incidence angle being normalized to 43° by Oh2004 empirical model, and the results reveal the problem of several decibels lower for GaoFen-3 in absolute radiometric calibration. Finally, the capability of surface parameter estimation is justified by statistical relation of GaoFen-3 co-polarization backscatter coefficient and bare soil moisture content, indicating its good potential for quantitative applications in land after solving the problem in calibration.