针对深度学习模型易出现灾难性遗忘的关键难点,提出了一种基于原型对比的合成孔径雷达(synthetic aperture radar,SAR)图像增量小样本目标检测算法-InFSAR(prototype contrast based incremental few-shot SAR object detection)。首先...针对深度学习模型易出现灾难性遗忘的关键难点,提出了一种基于原型对比的合成孔径雷达(synthetic aperture radar,SAR)图像增量小样本目标检测算法-InFSAR(prototype contrast based incremental few-shot SAR object detection)。首先,采用基础数据集对检测器进行预训练,以构建初步的特征提取能力;其次,设计一种类原型表征生成模块,以构建一组能够代表数据内在特征的类原型。在增量学习阶段,设计一种混合类原型对比编码模块,以有效学习新类别与基础类别之间的区分性特征。此外,为缓解灾难性遗忘问题,引入类原型校准策略,使模型在类原型上的预测分布逐步逼近真实分布,从而保持对基础类别识别的稳定性。在小样本目标检测数据集SRSDD-v1.0上的实验结果表明,在5-shot设置下,InFSAR对船舶细粒度目标的检测精度达到46.5%。同时,该方法能够在无需访问基础类训练数据的情况下,实现对少量标注新类别的增量检测与识别。展开更多
Landslide susceptibility mapping(LSM)is an essential tool for mitigating the escalating global risk of landslides.However,challenges such as the heterogeneity of different landslide triggers,extensive engineering acti...Landslide susceptibility mapping(LSM)is an essential tool for mitigating the escalating global risk of landslides.However,challenges such as the heterogeneity of different landslide triggers,extensive engineering activities exacerbated reactivation,and the interpretability of data-driven models have hindered the practical application of LSM.This work proposes a novel framework for enhancing LSM considering different triggers for accumulation and rock landslides,leveraging interpretable machine learning and Multi-temporal Interferometric Synthetic Aperture Radar(MT-InSAR)technology.Initially,a refined fieldinvestigation was conducted to delineate the accumulation and rock area according to landslide types,leading to the identificationof relevant contributing factors.Deformation along the slope was then combined with time-series analysis to derive a landslide activity level(AL)index to recognize the likelihood of reactivation or dormancy.The SHapley Additive exPlanation(SHAP)technique facilitated the interpretation of factors and the identificationof determinants in high susceptibility areas.The results indicate that random forest(RF)outperformed other models in both accumulation and rock areas.Key factors including thickness and weak intercalation were identifiedfor accumulation and rock landslides.The introduction of AL substantially enhanced the predictive capability of the LSM and outperformed models that neglect movement trends or deformation rates with an average ratio of 81.23%in high susceptibility zones.Besides,the fieldvalidation confirmedthat 83.8%of newly identifiedlandslides were correctly upgraded.Given its efficiencyand operational simplicity,the proposed hybrid model opens new avenues for the feasibility of enhancement in LSM at urban settlements worldwide.展开更多
Bathymetric measurement of shallow water is of fundamental importance to coastal environment research and resource management.However,there are still great challenges in estimating water depth using satellite observat...Bathymetric measurement of shallow water is of fundamental importance to coastal environment research and resource management.However,there are still great challenges in estimating water depth using satellite observations in turbid coastal waters.In this paper,we developed a physicsenhanced deep neural network to estimate bathymetry of highly turbid waters of the Changjiang(Yangtze)River estuary from dual-polarized synthetic aperture radar(SAR)images.Sentinel-1A/B SAR images with a spatial resolution of 20 m×22 m were collected and matched with water depth data from nautical charts during 2017-2023.For the input parameters of the model,in addition to the normalized radar backscatter cross section(NRCS)at single polarization and incidence angle,the impacts of both polarimetric characteristics and physical environmental factors on model performance were discussed in detail.Results of feature importance analysis and sensitivity experiments indicate that the polarization ratio and NRCS after removing the influence of background sea surface wind field make significant contributions to the bathymetry retrieval model.The root mean square error(RMSE)of SAR derived water depth decreases from 1.44 to 0.78 m within 0-30-m depth,and the mean relative error(MRE)is reduced from 15.6%to 8.6%.Compared with other machine learning models such as ResNet,XGBoost,and Random Forest,the MRE is reduced by 3.9%,5.7%,and 7.4%,respectively.The spatial distribution of SAR derived water depth also exhibits a high degree of consistency with observations,demonstrating the great potential of the model in estimating the depth of turbid shallow waters.展开更多
文摘针对深度学习模型易出现灾难性遗忘的关键难点,提出了一种基于原型对比的合成孔径雷达(synthetic aperture radar,SAR)图像增量小样本目标检测算法-InFSAR(prototype contrast based incremental few-shot SAR object detection)。首先,采用基础数据集对检测器进行预训练,以构建初步的特征提取能力;其次,设计一种类原型表征生成模块,以构建一组能够代表数据内在特征的类原型。在增量学习阶段,设计一种混合类原型对比编码模块,以有效学习新类别与基础类别之间的区分性特征。此外,为缓解灾难性遗忘问题,引入类原型校准策略,使模型在类原型上的预测分布逐步逼近真实分布,从而保持对基础类别识别的稳定性。在小样本目标检测数据集SRSDD-v1.0上的实验结果表明,在5-shot设置下,InFSAR对船舶细粒度目标的检测精度达到46.5%。同时,该方法能够在无需访问基础类训练数据的情况下,实现对少量标注新类别的增量检测与识别。
基金supported by the National Key R&D Program of China(Grant No.2023YFC3007201)the National Natural Science Foundation of China(Grant No.42377161)the Opening Fund of Key Laboratory of Geological Survey and Evaluation of Ministry of Education(Grant No.GLAB 2024ZR03).
文摘Landslide susceptibility mapping(LSM)is an essential tool for mitigating the escalating global risk of landslides.However,challenges such as the heterogeneity of different landslide triggers,extensive engineering activities exacerbated reactivation,and the interpretability of data-driven models have hindered the practical application of LSM.This work proposes a novel framework for enhancing LSM considering different triggers for accumulation and rock landslides,leveraging interpretable machine learning and Multi-temporal Interferometric Synthetic Aperture Radar(MT-InSAR)technology.Initially,a refined fieldinvestigation was conducted to delineate the accumulation and rock area according to landslide types,leading to the identificationof relevant contributing factors.Deformation along the slope was then combined with time-series analysis to derive a landslide activity level(AL)index to recognize the likelihood of reactivation or dormancy.The SHapley Additive exPlanation(SHAP)technique facilitated the interpretation of factors and the identificationof determinants in high susceptibility areas.The results indicate that random forest(RF)outperformed other models in both accumulation and rock areas.Key factors including thickness and weak intercalation were identifiedfor accumulation and rock landslides.The introduction of AL substantially enhanced the predictive capability of the LSM and outperformed models that neglect movement trends or deformation rates with an average ratio of 81.23%in high susceptibility zones.Besides,the fieldvalidation confirmedthat 83.8%of newly identifiedlandslides were correctly upgraded.Given its efficiencyand operational simplicity,the proposed hybrid model opens new avenues for the feasibility of enhancement in LSM at urban settlements worldwide.
基金Supported by the National Natural Science Foundation of China(Nos.T2261149752,41976163,42476172)。
文摘Bathymetric measurement of shallow water is of fundamental importance to coastal environment research and resource management.However,there are still great challenges in estimating water depth using satellite observations in turbid coastal waters.In this paper,we developed a physicsenhanced deep neural network to estimate bathymetry of highly turbid waters of the Changjiang(Yangtze)River estuary from dual-polarized synthetic aperture radar(SAR)images.Sentinel-1A/B SAR images with a spatial resolution of 20 m×22 m were collected and matched with water depth data from nautical charts during 2017-2023.For the input parameters of the model,in addition to the normalized radar backscatter cross section(NRCS)at single polarization and incidence angle,the impacts of both polarimetric characteristics and physical environmental factors on model performance were discussed in detail.Results of feature importance analysis and sensitivity experiments indicate that the polarization ratio and NRCS after removing the influence of background sea surface wind field make significant contributions to the bathymetry retrieval model.The root mean square error(RMSE)of SAR derived water depth decreases from 1.44 to 0.78 m within 0-30-m depth,and the mean relative error(MRE)is reduced from 15.6%to 8.6%.Compared with other machine learning models such as ResNet,XGBoost,and Random Forest,the MRE is reduced by 3.9%,5.7%,and 7.4%,respectively.The spatial distribution of SAR derived water depth also exhibits a high degree of consistency with observations,demonstrating the great potential of the model in estimating the depth of turbid shallow waters.
