Font Size: a A A

The Research On Sar Image Target Recognition Technology Based On Feature Fusion And Extreme Learning Machine

Posted on:2017-03-31Degree:MasterType:Thesis
Country:ChinaCandidate:S Q WangFull Text:PDF
GTID:2348330491463041Subject:Electronic and communication engineering
Abstract/Summary:PDF Full Text Request
Synthetic aperture radar (SAR) is widely used in military and civilian field. How to design a more intelligent SAR automatic target recognition system (ATR) becomes a pressing demand. Especially SAR ATR system is a technology from mutual connection of multi-subject, a number of new theories’ emergencies further promote its’ development. Feature extraction, optimization and classification are the key steps in SAR ATR.An effort is made to research the feature extraction, feature fusion and classification of SAR images. A new scheme to solve the problem of SAR image classification and object matching is proposed. On the one hand, feature fusion and extreme learning machine model are implemented in classification after SAR image texture features extracted; on the other hand, a new similarity function based on the point group is proposed to measure the SAR image similarity and realizes matching between the target and template library. The work includes the following three aspects:1、The results of SAR image feature extraction can directly affect the recognition correct rate of the subsequent classification algorithm. The texture feature has been the focus of research and application for the object classification of SAR image. According to feature fusion algorithm, different kinds of texture features are combined by serial and parallel method and removed the redundancy information by linear dimension reduction, which improves the recognition correct rate and stability of classifier.2、Extreme learning machine (ELM) with high efficiency and fast convergence speed is widely used to solve classification. The effect of the number of hidden layer nodes on the convergence speed and training time is studied based on the incremental extreme learning machine, and a new network construction mode named variable length incremental extreme learning machine (VI-ELM) is proposed. The algorithm is validated on the UCI standard test set and applied to SAR image classification, which has achieved good classification results.3、Peak characteristic is an important characteristic of SAR image, according to the strong scattered point group characteristics in the SAR image, a new similarity function is proposed based on the point group. The point-to-point matching is extended between point and point groups. The similarity measurement is defined between point groups. The experimental results show that this method can meet the image similarity regularity, and there is certain tolerance to distortions such as noise, partial occlusion, deformation, which could also get more reasonable similarity and great improvement in efficiency.
Keywords/Search Tags:Synthetic Aperture Radar, Feature Fusion, Extreme Learning Machine, Variable Length Incremental Extreme Learning Machine, Point Group Matching
PDF Full Text Request
Related items