Small Target Detection And Identification On The Sea

With the rapid development of modern electronic information technology,materials technology and many other technologies,the modern warfare has gradually developed in the direction of scientific intelligence and electronic warfare.This means that radars should have the ability as required to detect and identify targets in strong clutter and noise.Well,as an important branch of radar signal processing,the target detection and identification has farreaching significance in military and civilian applications.When the radar completes the detection and tracking of target on the sea,it will face the problem of effective detection in the ocean background.In addition,the modern warfare needs higher and higher requirements on information acquisition,and a new field of radar target recognition has emerged.Therefore,the detection and recognition of small targets in the ocean background has become a research hotspot.In this paper,the target size is limited to 2 meters to detect and identify small targets.In this context,the paper studies the following: 1.The target detection algorithm in the background of specific sea clutter is researched,mainly to study the extended target detection problem under Weibull distribution and compound K distribution.It is found that the Weibull distribution becomes a standard exponential distribution through a nonlinear transformation,and so the target detection problem under Weibull distribution is simplified to that under standard exponential distribution;usually,there is a certain relative motion between the radar and the target.When the coherent accumulation time is long,the target echo will move across the distance unit.If the traditional detection sliding window estimation method is used for estimation,the detection window will not be accurate enough.Therefore,this section introduces short time coherent integration into the two-level binary detector to study the target detection under the compound K distribution,and proposes a two-level binary detection algorithm based on short time coherent accumulation to optimize the two-level binary detection algorithm.2.The target detection algorithm based on feature is studied.When the model of sea clutter is not matched properly,the performance of the parameterized target detector will decrease rapidly.Around this problem,the feature detection algorithm of the extended target is developed.To analyze the frequency domain characteristics of the wideband radar echo,the frequency domain discrete spectrum of the radar target echo(hereinafter referred to as the discrete spectrum)is composed of multiple harmonics.And the energy in the Wigner-Ville distribution of the target discrete spectrum is concentrated in the high frequency band.But the energy in the Wigner-Ville distribution of the clutter discrete spectrum is evenly distributed in each frequency band,so the high frequency energy in the Wigner-Ville distribution of the discrete spectrum can be used as the basis for distinguishing the target from the clutter;The discrete spectral amplitudes between adjacent echoes are the same,only differ in phase.The correlation coefficient is high after modulo,and the correlation coefficient of clutter discrete spectrum after modulo is low.Therefore,the correlation coefficient between echoes can be used as the distinguishing target and clutter,too.So,after combining these characteristics of the discrete spectrum and with the convex hull algorithm,the clutter sequence and the clutter sequence containing the target can be well distinguished.3.The radar target recognition algorithm based on high resolution range profile(HRRP)is studied.The difficulty of radar target recognition based on HRRP is mainly the target sensitivity problem.The locality preserving projection(LPP)algorithm is a classic local manifold learning algorithm.In the case of linearity,there is still a good dimensionality reduction effect.When the attitude angle of the target changes,the LPP algorithm can maintain the invariance of the reconstruction weight of the HRRP sequence samples.Therefore,the LPP algorithm can solve the sensitivity problem of HRRP to a certain extent.This part is based on the object recognition problem of LPP algorithm,and improves the problem of traditional LPP algorithm which is easily affected by noise.A locality preserving projection based on metric learning(MLLPP)algorithm is proposed and the kernel function is introduced into MLLPP algorithm.In this,a Kernel Locality Preserving Projection(KMLLPP)algorithm based on distance metric is obtained.