| The detection of small targets on the sea surface under sea clutter waves has been a major problem in the field of radar detection.This technology is an important way to obtain sea surface information and plays an important role in military and civilian fields.However,due to the complex sea conditions and radar parameter configuration,the target signal in the radar echo signal is very easy to be interfered by the sea clutter signal,which makes the target signal difficult to be detected.To address this problem,this thesis firstly investigates the sea clutter suppression technology,which aims to reduce the interference of sea clutter on the target signal.Then,we study the radar echo signal feature extraction technique,aiming to build a feature space that can accurately capture the data information,in which the sea clutter signal can be distinguished from the target signal more effectively.Finally,the detection problem in the radar echo signal is transformed into a binary classification problem from the perspective of machine learning.The details are as follows:1.According to the difference on the sparse domain between the sea clutter signal and the target signal,the idea of sparse representation is introduced and sequential generalization of k-means(SGK)dictionary learning algorithm is used to suppress the sea clutter.The algorithm replaces k-singular value decomposition(K-SVD)algorithm for dictionary updating and replaces the singular value decomposition with the least squares method.It ensures that the sea clutter suppression effect is not weakened and effectively reduces the complexity of the algorithm.In addition,the dictionary training method is improved to directly train the signals in the form of sea clutter plural,which makes fuller use of both I and Q signals,avoids the loss of effective information,and enhances the practicality of the algorithm.Finally,the simulation results show that the sea clutter suppression algorithm proposed in the text effectively reduces the interference of sea clutter on the target signal,improves the signal-to-noise ratio,and lays the foundation for the later feature extraction and target detection.2.In order to improve the detector performance,this thesis constructs the feature space of radar echo signal from the perspective of statistical characteristics and autoencoder,and compares the differences between them.Among them,the use of autoencoder to extract the features of radar echo signal avoids the tediousness of investigating the sea clutter-related characteristics in advance,and only needs to focus on the neural network model design.In addition,using ice multi-parameter imaging X-band(IPIX)radar data as the test set,the stability of the features constructed by the two methods is studied under different sea conditions and different observation times.The simulation results show that the features extracted by the autoencoder still maintain good classification effects under different sea states and different observation durations,with the characteristics of adaptive observation time and adaptive sea state.3.To address the problem that existing machine learning algorithms cannot directly fix the false alarm rate.In this thesis,a detector with controllable false alarm rate is designed based on the feature space constructed by the autoencoder and using a decision tree as a classifier.Finally,the performance of the detector is tested under different polarization methods,different false alarm rates and different observation times using IPIX radar data as the test set.The simulation results show that the detection framework proposed in this thesis still maintains good classification performance in the case of short radar observation time,even if the observation time is reduced by half the difference between the two detections is kept within 5%,and when the observation time is reduced from 256 ms to 128 ms,the detection probability decreases by less than 3% on average,which has good stability. |
PDF Full Text Request