Global Navigation Satellite System Reflectometry(GNSS-R)remote sensing has demonstrated broad application potential in marine oil spill monitoring due to its advantages such as all-weather capability,wide spatial cove...Global Navigation Satellite System Reflectometry(GNSS-R)remote sensing has demonstrated broad application potential in marine oil spill monitoring due to its advantages such as all-weather capability,wide spatial coverage,and high spatiotemporal resolution.However,the morphological features of Delay-Doppler Map(DDM)provided by GNSSR are simultaneously influenced by factors such as wind speed and surface oil films,leading to high false positive rates in traditional oil spill detection methods.To address this challenge,this study proposes a multimodal recognition framework that integrates both DDM imagery and wind speed information.A dual-branch Residual Network(Res Net)-based feature extraction network is designed to extract high-dimensional features from DDM and wind speed data,which are then fused at the feature level to enable accurate identification of oil-contaminated areas.To alleviate the scarcity of labeled real-world samples,a large-scale synthetic multimodal dataset is generated based on the ZavorotnyVoronovich(Z-V)scattering model for pretraining,followed by transfer learning using a small number of real GNSS-R observations to enhance the model's adaptability to real marine environments.Experimental results show that the proposed method achieves 91%classification accuracy on the test set,significantly outperforming baseline models such as Convolutional Neural Networks(CNNs).Furthermore,the incorporation of wind speed as an auxiliary modality effectively reduces false positives caused by wind-wave interference and enhances the model's sensitivity to oil film signatures.This study provides a feasible theoretical foundation and technical pathway for the application of GNSS-R in marine oil spill detection.展开更多
星基全球导航卫星系统反射测量(Global Navigation Satellite System Reflectometry,GNSS-R)技术已经成为大范围监测地表土壤湿度的一个有效手段.旋风全球导航卫星系统(Cyclone Global Navigation Satellite System,CYGNSS)以高时空分...星基全球导航卫星系统反射测量(Global Navigation Satellite System Reflectometry,GNSS-R)技术已经成为大范围监测地表土壤湿度的一个有效手段.旋风全球导航卫星系统(Cyclone Global Navigation Satellite System,CYGNSS)以高时空分辨率的优势广泛应用于土壤湿度反演研究.2024年9月,土壤湿度主被动(soil moisture active passive,SMAP)遥感卫星首次公开了GNSS-R反射率数据.本文首先对多源GNSS-R地表反射率、SMAP卫星土壤湿度产品和第五代全球气候和天气再分析数据集(fifth generation ECMWF atmospheric reanalysis of the global climate,ERA5)土壤湿度产品进行时空匹配;其次讨论不同地理纬度、不同土地类型、不同植被光学厚度情况下CYGNSS和SMAP的星基GNSS-R地表反射率在全球范围内的差异性,并提出了基于幂律函数的经验公式模型对它们的差异进行了修正;最后分析了星基GNSS-R地表反射率对土壤湿度的响应.结果表明:CYGNSS在38°S~38°N数据量充足且分布较均匀,有利于开展地表参数反演,而SMAP数据量偏少,但能覆盖中高纬度地区,二者具有互补性;在不同地理纬度、土地类型、植被光学厚度情况下,CYGNSS和SMAP的地表反射率在数值上存在非线性差异,这与二者接收到的信号频率和极化方式不同有很大关系,前者接收的是GPS L1频段的左旋圆极化反射信号,后者接收的是GPS L2C频段的水平和垂直线性极化反射信号,可用幂律函数很好地修正它们之间的差异;CYGNSS和SMAP的地表反射率与土壤湿度整体上存在较好的相关性.研究结果有利于未来开展多源星基GNSS-R反射率联合反演地表环境参数.展开更多
基金The Shandong Key Laboratory of Marine Ecological Environment and Disaster Prevention and Mitigation under contract No.202408the Key Program of Joint Fund of the National Natural Science Foundation of China and Shandong Province under contract No.U22A20586the National Natural Science Foundation of China under contract No.42274159。
摘要Global Navigation Satellite System Reflectometry(GNSS-R)remote sensing has demonstrated broad application potential in marine oil spill monitoring due to its advantages such as all-weather capability,wide spatial coverage,and high spatiotemporal resolution.However,the morphological features of Delay-Doppler Map(DDM)provided by GNSSR are simultaneously influenced by factors such as wind speed and surface oil films,leading to high false positive rates in traditional oil spill detection methods.To address this challenge,this study proposes a multimodal recognition framework that integrates both DDM imagery and wind speed information.A dual-branch Residual Network(Res Net)-based feature extraction network is designed to extract high-dimensional features from DDM and wind speed data,which are then fused at the feature level to enable accurate identification of oil-contaminated areas.To alleviate the scarcity of labeled real-world samples,a large-scale synthetic multimodal dataset is generated based on the ZavorotnyVoronovich(Z-V)scattering model for pretraining,followed by transfer learning using a small number of real GNSS-R observations to enhance the model's adaptability to real marine environments.Experimental results show that the proposed method achieves 91%classification accuracy on the test set,significantly outperforming baseline models such as Convolutional Neural Networks(CNNs).Furthermore,the incorporation of wind speed as an auxiliary modality effectively reduces false positives caused by wind-wave interference and enhances the model's sensitivity to oil film signatures.This study provides a feasible theoretical foundation and technical pathway for the application of GNSS-R in marine oil spill detection.
摘要星基全球导航卫星系统反射测量(Global Navigation Satellite System Reflectometry,GNSS-R)技术已经成为大范围监测地表土壤湿度的一个有效手段.旋风全球导航卫星系统(Cyclone Global Navigation Satellite System,CYGNSS)以高时空分辨率的优势广泛应用于土壤湿度反演研究.2024年9月,土壤湿度主被动(soil moisture active passive,SMAP)遥感卫星首次公开了GNSS-R反射率数据.本文首先对多源GNSS-R地表反射率、SMAP卫星土壤湿度产品和第五代全球气候和天气再分析数据集(fifth generation ECMWF atmospheric reanalysis of the global climate,ERA5)土壤湿度产品进行时空匹配;其次讨论不同地理纬度、不同土地类型、不同植被光学厚度情况下CYGNSS和SMAP的星基GNSS-R地表反射率在全球范围内的差异性,并提出了基于幂律函数的经验公式模型对它们的差异进行了修正;最后分析了星基GNSS-R地表反射率对土壤湿度的响应.结果表明:CYGNSS在38°S~38°N数据量充足且分布较均匀,有利于开展地表参数反演,而SMAP数据量偏少,但能覆盖中高纬度地区,二者具有互补性;在不同地理纬度、土地类型、植被光学厚度情况下,CYGNSS和SMAP的地表反射率在数值上存在非线性差异,这与二者接收到的信号频率和极化方式不同有很大关系,前者接收的是GPS L1频段的左旋圆极化反射信号,后者接收的是GPS L2C频段的水平和垂直线性极化反射信号,可用幂律函数很好地修正它们之间的差异;CYGNSS和SMAP的地表反射率与土壤湿度整体上存在较好的相关性.研究结果有利于未来开展多源星基GNSS-R反射率联合反演地表环境参数.