Detection And Recognition For Radar Ground Target In Complex Environments

The radar seeker for air-to-ground missile detects,recognizes and tracks target of interest by understanding the complex battlefield environment.It can guide missiles to strike targets precisely,and has played an important role in modern warfare.Detecting targets from complex backgrounds and identifying target attributes are two practical problems which restrict the large-scale application of radar seeker for air-to-ground missile.Therefore,study on radar target detection and recognition in complex environments is of great value in military and engineering applications.Based on the military requirements,this dissertation studies the key technologies of detection and recognition for radar ground target in complex backgrounds,aiming at improving the strike accuracy of radar seeker.The main content is summarized as the follows.1.For the problem of detection for extended distance targets,a novel target detection method is proposed based on the online estimation of strong scattering points.First,the number of strong scattering points and the first detection threshold are estimated online by the unsupervised clustering algorithm.Then,the second threshold is computed using the false alarm rate.Finally,the two thresholds are used to judge if a target exists.Without the prior knowledge of target scatter distribution,the proposed method exhibits robust detection performance under various scatter distribution.In the experiments,the detection results on the simulated data and measured data verify the superiority of the proposed method over the compared methods.2.For target detection under low signal-to-clutter ratio(SCR)conditions,a novel target detection method is proposed based on energy-polarimetric entropy features.Different from the traditional radar detection method,the proposed method does not use the energy as the unique feature.On the contrary,the target scattering mechanism is considered,and the energy-polarimetric entropy features are used.Furthermore,the target detection problem is transformed as anomaly detection model learned online.That is,the presence of target is judged by the abnormality of detecting unit compared to its surrounding units.The experiments conducted on the measured data demonstrate effectiveness of the proposed method under low SCR conditions.Moreover,the proposed method provides a new thinking for radar target detection.3.To tackle the HRRP target recognition task under relatively strong clutter environment,a novel clutter robust recognition method is proposed based on double anomaly detection.First,a new anomaly detection method based on spherical hypothesis clustering is adopted to suppress clutter in small clutters.Then,the isolated clutter points are suppressed by the anomaly detection method based on parameter-independent local outlier factor.Finally,the HRRPs whose clutter has been suppressed are used for feature extraction and target recognition.The proposed method has the following advantages: 1)it has no requirement of target motion state;2)its parameters are set automatically;3)it is computationally efficient.The experiment results on the measured data show that the proposed method is more noise/clutter robust than the state-of-the-art methods.4.To deal with the problem of group target recognition in a beam,a novel HRRP multi-target recognition method is proposed based on the prior-independent density-based spatial clustering of applications with noise(PI-DBSCAN).In the training phase,various features of training samples are extracted,among which the distribution of strong range cells is utilized to obtain the parameters of PI-DBSCAN,while the others are used to train SVM classifier.In the test phase,PI-DBSCAN algorithm is exploited at the radial distance-azimuth distance plane to segment the test multi-target HRRP.Afterwards,the features of segmented HRRPs are extracted and fed to the SVM classifier to be recognized.The experiment results of measured data show that the proposed method is robust against noise level and the increasing target HRRP overlap ratio compared with traditional methods.5.For the problem of high computational complexity of the group target recognition algorithm,a new multi-target recognition method is proposed based on the weighted mean shift(Weighted-MS)clustering algorithm.First,the weighted-MS clustering algorithm is utilized at the radial distance-azimuth distance plane to extract the HRRP of each sub-target.Then,the features of the sub-target HRRP are extracted and fed to the SVM classifier to be recognized.The proposed method weights the cluster samples by their echo amplitude,so it is more robust to noise.Moreover,the initial value of the weighted-MS clustering algorithm is designed according to the maximum amplitude criterion,thereby improving the iteration efficiency.The experiment results of the measured data show that the proposed method exhibits better performance when the overlap rate of each sub-target HRRP is low in the distance dimension,and has time efficiency.