Research On Retrieval Of Sea Surface Wind Speed From Marine Radar Images Under Rainfall Interference

As a major maritime country,China’s marine environmental monitoring is a top priority for national development.Through marine radar image retrieval technology,we can extract a lot of marine environment information,of which the information of sea surface wind speed is particularly important.However,rainfall interference seriously affects the accuracy of extracting sea surface wind speed information.On this basis,this thesis carries out the research on retrieval of sea surface wind speed from marine radar images under rainfall interference.The full text is mainly divided into three parts: research on marine radar image rainfall recognition,research on marine radar image rainfall correction and research on retrieval of sea surface wind speed.In the research of marine radar image rainfall recognition methods,in view of the low recognition accuracy of traditional methods,this thesis combines the advantages of bit plane hierarchical image processing methods in local and global feature extraction,noise immunity,timeliness,and simplicity,Its zero intensity percentage decreases with the increase of the interference of rainfall on marine radar images.This method effectively determines the zero intensity percentage threshold for rainfall image recognition,with a recognition accuracy of over 95%,greatly improving the accuracy and reliability of rainfall image recognition.In the research of marine radar image rainfall correction method,this thesis combines the advantages of complex Gaussian wavelet and Daubechies wavelet wavelet transform in time-frequency domain with good resolution and hierarchical nature,and proposes an automatic correction method for marine radar rainfall image based on wavelet transform.Firstly,two kinds of wavelet function transforms are used to extract the feature points and feature curves in the marine radar image interfered by rainfall,and then the extracted feature information is weighted fusion corrected to maximize the advantages of the two kinds of wavelet functions in extracting key information.Finally,the fitted feature data is restored by wavelet reconstruction.The experimental research results indicate that this method has good correction effect and provides strong support for subsequent sea surface wind speed inversion.In the research of sea surface wind speed retrieval method,on the basis of the previous two steps,this thesis proposes an improved RBF neural network sea surface wind speed retrieval method based on clustering algorithm.In this method,k-means clustering algorithm is used to remove the uncorrected heterogeneous data of rainfall interference in marine radar images,so as to obtain the marine radar data after identification and correction by the above method.The inversion process of sea surface wind speed is based on a model trained by a single hidden layer RBF neural network.It is proposed to use a subtractive clustering algorithm to determine the conditions for determining the number of hidden layer units and the center and extension constant of the basis function of the neural network.Recursive least squares are used to obtain the network output layer connection weights.Finally,the actual radar data was applied for verification.The correlation coefficient between the inversion results of this method and the actual wind speed reached 0.87,with a standard deviation of2.1m/s and a deviation of-0.08m/s,indicating good inversion accuracy and stability.