| Sea ice is an important part of hydrological information and marine ecological cycle,and real-time,high-precision,wide-range sea ice detection can provide early warning for natural disasters,as well as observational data for studies on earth ecology and water cycle research.Traditional sea ice detection is based on manual or ground-based stations,with high measurement cost and small range,and limited by equipment and other problems,and fewer ground sea ice results are obtained.Based on the Global Navigation Satellite System-Reflectometry(GNSS-R)technology,sea ice detection is performed by using the reflection signals obtained from GNSS-R receivers,which provides a new detection method for sea ice detection.This method has significant advantages of all-weather,high-precision,real-time,etc.,and is of great importance for detecting global sea ice and its changes.This paper adopts a machine learning method for sea ice detection based on satellite-based GNSS-R data.To address the problems of detection accuracy,noise interference,and slow detection speed in sea ice detection,we propose a study of GNSS-R sea ice detection based on Alex Net convolutional neural network,a local linear discriminant(LDA)-local outlier factor(LDA-LOF)and a Local Linear Embedding-Support Vector Machine(LLE-SVM).The details are as follows:(1)In order to further improve the accuracy of GNSS-R sea ice detection using Delay Doppler Map(DDM),this paper proposes a GNSS-R sea ice detection method based on Alex Net convolutional neural network.This method has a deeper neural network structure and simpler activation function,which can further expand the data set to alleviate the overfitting phenomenon,and can extract more favorable features for sea ice detection and improve the accuracy of sea ice detection.Firstly,the DDM is interpolated to adapt to the Alex Net convolutional neural network,secondly,the GNSS-R sea ice detection model based on Alex Net convolutional neural network is built,and finally the sea ice detection method based on Alex Net convolutional neural network is trained and tested,and the sea ice detection accuracy can reach 98.44%.(2)In order to improve the utilization of sea ice data,reduce the influence of noise and ensure the accuracy of sea ice detection,this paper proposes a study of GNSS-R sea ice detection based on LDA-LOF.The Integral delay waveform(IDW)data has a smaller spatial dimension of DDM data,which reduces the impact dimension of noise.At the same time,the LDA algorithm has a significant advantage in processing one-dimensional row data,and the method can reduce the dimensionality of the data to maximize the difference between categories and minimize the difference within categories as much as possible to improve the processability of the noisy data.Firstly,the LDA sea ice detection model is built and tested,and secondly,the LOF algorithm is used to further optimize the LDA sea ice detection model.The accuracy of sea ice detection is improved,and finally the LDA-LOF sea ice detection model is used for sea ice detection research,and the detection accuracy can reach 99.11%.(3)In order to improve the GNSS-R sea ice detection speed,this paper proposes a GNSS-R sea ice detection method based on LLE-SVM.The method focuses on maintaining the local linear characteristics of the sample,which can reduce the complexity of the data and make the sea ice detection simpler and faster.idw data belongs to one-dimensional row matrix,and the LLE algorithm has significant advantages when dealing with this type of data.Firstly,the linear weights for data dimensionality reduction are constructed,and secondly,SVM has the ability to construct hyperplanes in low-dimensional space and is commonly used in feature classification research scenarios,so the SVM classifier for surface category detection is built using SVM,and the features after LLE dimensionality reduction are classified,and finally the sea ice detection research of LLE-SVM is conducted,and the detection accuracy reaches 99.51% and the detection efficiency improved by 37.11 times. |
