Research on classification method of unknown objects based on support vector machine

• 1、College of Telecommunications and Information Engineering, Nanjing University of Posts and Telecommunications, nanjing 210003

Abstract：For the MNIST database of handwritten digits. In this paper uses linear discriminant analysis to extract feature vectors of handwritten digital images, and design feature patterns according to relevant characteristics, in order to make the bucket detector values of the feature patterns illumination more discriminating. And then the training set image illuminated by feature patterns and collecting light field signals with barrel detectors. Later, input the collected signals into the SVM for training. Finally, we only need to collect the light field signals, which from the feature patterns illuminates the unknown image, to complete the classification. Through the simulation of 10,000 handwritten digital images, this method of classification accuracy reached 91.62%. And only 9 ghost imaging feature patterns are needed to effectively classify unknown handwritten digit sets.

Keywords： ghost imaging linear discriminant analysis machine learning support vector machine classification