Research on Scatter Classification of Small Sample

Abstract

Pattern recognition is a popular frontier research topic in the world. Scatter point classification is an important branch of pattern recognition, which combines image processing and so on. Now it should be given corresponding attention. There are three types of scatters. The training sample points are classified by classifiers of the nearest neighbor method, the class means method, the 1-r strategy algorithm, the fixed incremental algorithm and the multi-class discriminant algorithm. Find the lowest classification error rate method and use this method to classify the test sample points. Finally, the classification method is compared and analyzed according to the classification error rate and iteration time. The classification of the test samples is as follows: 1-r strategy Test points 1 2 3 Error 1 1 0 Fixed increment Test points 1 2 3 Error 0 0 0 Nearest neighbor Test points 1 2 3 Error 0 1 0 Class mean Test points 1 2 3 Error 0 1 0 Multiple categories Test points 1 2 3 Error 0 0 0 It is shown that the fixed increment and multiple categories are better than others. The iteration time of fixed increment is 6.453277 seconds and iteration time of multiple categories is 3.345234 seconds. The multiple categories is better than fixed increment. And the calculation method has generalization ability, no Hughes phenomenon.