Fengkai Lang

Synthetic Aperture Radar (SAR) possesses all-weather and all-time imaging capabilities, which play a crucial role in ship monitoring. Meanwhile, convolutional neural networks (CNNs) are extensively employed in SAR ship detection because of their strong ability to extract features. However, existing CNN-based SAR ship detectors often exhibit insufficient sensitivity to small-scale ship targets, primarily due to inadequate multi-scale feature representation and limited feature expressiveness in complex environments. Moreover, some models rely on large network architectures, which restricts their deployment in real-time applications. To address these challenges, we propose a lightweight dilated feature enhanced network (LDFENet). First, we design the CSBS module, in which a space-to-depth convolution (SPDConv) is incorporated to reduce the loss of small ships information. Second, we introduce the augmented dilated efficient layer aggregation network module (AD-ELAN) to improve the model's adaptability for targets of varying scales. Third, we utilize a mixed attention module to enhance features while suppressing background noise. Finally, the Powerful-IoU loss function (PIoU) is adopted to improve target localization performance and accelerate model convergence. Experimental results on the HRSID and SSDD datasets demonstrate the effectiveness of LDFENet in SAR ship detection. With only 4.3M parameters and 10.8 GFLOPs, LDFENet achieves superior accuracy and efficiency compared to the latest methods. Furthermore, LDFENet exhibits strong generalization capability on large-scale SAR images, highlighting its potential for practical applications. The code will be available at https://github.com/Z-LM-10/LDFENet

Multi-source remote sensing images (RSIs) can capture the complementary information of ground objects for use in semantic segmentation. However, there can be inconsistency and interference noise among the multimodal data from different sensors. Therefore, it is a challenge to effectively reduce the differences and noise between the different modalities and fully utilize their complementary features. In this paper, we propose a universal cross-fusion transformer framework (CFFormer) for the semantic segmentation of multi-source RSIs, adopting a parallel dual-stream structure to extract features separately from the different modalities. We introduce a feature correction module (FCM) that corrects the features of the current modality by combining features from the other modalities in both the spatial and channel dimensions. In the feature fusion module (FFM), we employ a multi-head cross-attention mechanism to interact globally and fuse features from the different modalities, enabling the comprehensive utilization of the complementary information in multi-source RSIs. Finally, comparative experiments demonstrate that the proposed CFFormer framework not only achieves a state-of-the-art accuracy but also exhibits outstanding robustness when compared to the current advanced networks for semantic segmentation of multi-source RSIs. Specifically, CF-Former achieves a mean intersection over union (mIOU) of 58% and an overall accuracy (OA) of 85.35% on the WHU-OPT-SAR dataset, outperforming the second-ranked network by 4.71% and 1.74%, respectively. On the Vaihingen and Potsdam datasets, CFFormer also achieves the best results, with mIoU and OA values of 84.31%/91.88% and 88.62%/92.64%, respectively. The source code is available at https://github.com/masurq/CFFormer.

Synthetic aperture radar (SAR) technology has become an important means of flood monitoring because of its large coverage, repeated observation, and all-weather and all-time working capabilities. The commonly used thresholding and change detection methods in emergency monitoring can quickly and simply detect floods. However, these methods still have some problems: (1) thresholding methods are easily affected by low backscattering regions and speckle noise; (2) changes from multi-temporal information include urban renewal and seasonal variation, reducing the precision of flood monitoring. To solve these problems, this paper presents a new flood mapping framework that combines semi-automatic thresholding and change detection. First, multiple lines across land and water are drawn manually, and their local optimal thresholds are calculated automatically along these lines from two ends towards the middle. Using the average of these thresholds, the low backscattering regions are extracted to generate a preliminary inundation map. Then, the neighborhood-based change detection method combined with entropy thresholding is adopted to detect the changed areas. Finally, pixels in both the low backscattering regions and the changed regions are marked as inundated terrain. Two flood datasets, one from Sentinel-1 in the Wharfe and Ouse River basin and another from GF-3 in Chaohu are chosen to verify the effectiveness and practicality of the proposed method.

Sea ice is regarded as an indicator of temperature change. In recent years, the spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) technology has made remarkable progress in sea ice detection. Delay-Doppler maps (DDMs) as one of significant observations can reflect different characteristics for sea ice and open water, and a single DDM is usually viewed as the unit of feature extraction; however, it is easily influenced by wave height, wind and other factors. Therefore, this paper proposes building scene semantic objects to enhance the reliability of observation and reflect the object characteristics. The synergism between DDMs and the spatial correlation of specular points was considered. Afterwards, histogram features were extracted to express the distribution of scattered energy. The random forest (RF) model was developed to distinguish sea ice from open water. The performance of the method by using TechDemoSat-1 (TDS-1) dataset was evaluated with the sea ice concentration products provided by OSISAF. The results show that the overall accuracy is 98.17%, which outperforms traditional observation methods.

Underground coal fires are global catastrophes that result in energy waste, carbon emission and eco-environment pollution. Remote sensing (RS) detection is essential for underground coal fire extinguishing engineering, and the mostly used is thermal infrared (TIR) RS. It can well obtain the thermal anomalies of land surface temperature (LST) which is the most direct surface feature of underground coal fires. However, most studies using TIR RS simply delineate underground fire sources vertically according to LST anomalies, which has relatively little impact when initially determining coal fire area locations on the large scale. As for the precise location of small-scale subsurface fire sources, the deviation between subsurface fire source locations inferred and real locations could lead to errors or even mistakes to fire extinguishing engineering. There is a lack of subsurface fire source evolution model reconstruction method, and the spatiotemporal correlations characteristic of LST thermal anomalies and underground fire sources have not yet been discussed. To this end, taking Miquan coalfield (western China) as an example, a 3D Empirical Bayesian Kriging (EBK3D) method is firstly proposed to reconstruct the 4D temperature fields of underground fire sources. Then the feasibility of the vertical correspondence approach to inferring small-scale subsurface fire sources through LST thermal anomalies detected by unmanned aerial vehicle TIR RS and satellite TIR RS is analyzed. Finally, the spatiotemporal correlation characteristic of LST thermal anomalies and subsurface fire sources is analyzed. As the results show that, it is feasible to reconstruct the underground fire source evolution model by EBK3D method. The reconstructed 4D temperature fields can dynamically reflect the evolutionary states of underground fire sources in three time periods, with cross-validated root mean square errors of 52.2°C, 49.6°C, and 37.1°C and the R ² of linear regressions of 0.925, 0.9145, and 0.8429, respectively. The LST thermal anomalies show a significant spatiotemporal delay with respect to the subsurface fire source evolution. This makes the locations of the underground fire sources traced by the vertical correspondence method deviate from the real ones. The offsets of underground fire sources relative to surface thermal anomalies in the coal seam strike and dip directions for different time periods at depths of (T1: -44.43 m, T2: -27.72 m, T3: -20.04 m) are (T1: 73.80 m, T2: 52.33 m, T3: 45.06 m) and (T1: 16.79 m, T2: 17.27 m, T3: 24.82 m), respectively. The R ² for linear regression model of the offset averages in three directions versus time and fire source size are (0.9247, 0.7949, and 0.9564), (0.8739, 0.85 and 0.9152) respectively.

High spatial resolution soil moisture (SM) mapping is essential for a wide range of applications, especially for precision irrigation and crop management. This work proposes an SM estimation method combined with time series of L-band fully polarimetric synthetic aperture radar (PolSAR) and passive SM products over crop areas. Regarding the challenge of eliminating vegetation canopy scattering on SM estimation, model-based polarimetric decomposition is implemented as a pretreatment step in which the surface scattering component in both HH and VV channels are extracted. Afterward, dual-pol surface scattering information normalization is dealt with the cosine-squared incidence angle normalization method, which makes it possible for SM inversion with multiple tracks and multi-incidence SAR observations. With the time series of normalized surface scattering information, the alpha approximation-based change detection algorithm (AACD) is used for SM estimation. Since the AACD algorithm is reported with an underdetermined problem of parameter solution and the underestimation issue of soil moisture inversion, an extended AACD which incorporates dual-pol (HH and VV) SAR observations, namely the Dual-pol AACD algorithm, is proposed in this study. Besides, the minimum and maximum values of passive microwave soil moisture data of the whole study area and the entire study period are introduced as constraints in Dual-pol AACD when solving the unknown parameters of the real part of the soil dielectric constant. Finally, the obtained time series of soil dielectric constants are converted to volumetric soil water content using dielectric mixing model. 56 sets of collected UAVSAR L-band data with 4 different flight lines (#31603, #31604, #31605, #31606) of Winnipeg, Manitoba, Canada in 2012 (SMAPVEX12) are used to validate the Dual-pol AACD algorithm. Passive microwave SM constraints are collected from Soil Moisture and Ocean Salinity (SMOS) and Advanced Microwave Scanning Radiometer 2 (AMSR-2) products. The performance of the proposed method is evaluated by comparing the in-situ measurements against the soil moisture estimates of wheat, corn, soybeans, bean, and canola fields at different phenological stages. Results show that the proposed method provides an accuracy of RMSE ≤ 6.5 cm3•cm-3 over all the selected crop fields, which is better than that without the introduction of constraints from passive microwave SM products. This work also compares the SM estimation performance using constraints from SMOS and AMSR-2. In addition, SM estimates in different crop fields and growth stages are also provided regarding the variation of crop morphological characteristics and biophysical properties. It concludes that the proposed SM estimation method has great potential for local and global SM mapping in a high resolution with existing and upcoming L-band SAR data, such as ALOS-2 (Japan), LT-1 (China), NISAR (America and India) and Tandem-L (Germany), etc.

Soil moisture monitoring is widely used in agriculture, water resource management, and disaster prevention, which is of great significance for sustainability. The global navigation satellite system interferometric reflectometry (GNSS-IR) technology provides a supplementary method for soil moisture monitoring. However, due to the quality of the signal-to-noise ratio (SNR) measurements and the complex surface environment, inevitable outliers in multipath interference signal metrics (amplitude, frequency, and phase) were used as modeling variables to inverse GNSS-IR soil moisture. Besides, it is hard to use the univariate model to comprehensively analyze the relationship between the various factors, due to the poor fitting effect and weak generalization ability of the model. In this paper, the minimum covariance determinant (MCD) robust estimation and machine learning algorithms are adopted. The MCD robust estimation can eliminate outliers of the multipath signal metrics and machine learning algorithms, including the back propagation neural network (BPNN), Gaussian process regression (GPR), and random forest (RF), and can comprehensively establish nonlinear GNSS-IR soil moisture inversion models using multipath interference signal metrics. Moreover, the study of the modeling parameter selection for the three machine learning algorithms and the inversion results for single satellite and all satellites are also carried out to make the algorithms more generalizable. The results show that the correlation coefficients (R) and the root mean square error (RMSE) of the machine learning models for all satellite tracks are increased by 4.3~86.6% and reduced by 2.8~30%, respectively, compared with the MCD multiple regression model. The RF model with 80 decision trees and 1 node shows the clearest improvement. The total model using all satellite data has more generalization ability than the single satellite model but causes some loss of accuracy.

Continuous and accurate acquisitions of surface water distribution are important for water resources evaluation, especially high-precision flood monitoring. During surface water extraction, optical imagery is strongly affected by clouds, while synthetic aperture radar (SAR) imagery is easily influenced by numerous physical factors; thus, the water extraction method based on single-sensor imagery cannot obtain high-precision water range under multiple scenarios. Here, we integrated the radar backscattering coefficient of ground objects into the Normalized Difference Water Index to construct a novel SAR and Optical Imagery Water Index (SOWI), and the water ranges of five study areas were extracted. We compared two previous automatic extraction methods based on single-sensor imagery and evaluated the accuracy of the extraction results. Compared with using optical and SAR imagery alone, the accuracy of all five regions was improved by up to 1–18%. The fusion-derived products resulted in user accuracies ranging 95–99% and Kappa coefficients varying by 85–97%. SOWI was then applied to monitor the 2021 heavy rainfall-induced Henan Province flood disaster, obtaining a time-series change diagram of flood inundation range. Our results verify SOWI’s continuous high-precision monitoring capability to accurately identify waterbodies beneath clouds and algal blooms. By reducing random noise, the defects of SAR are improved and the roughness of water boundaries is overcome. SOWI is suitable for high-precision water extraction in myriad scenarios, and has great potential for use in flood disaster monitoring and water resources statistics.