A transfer learning picture classification method and device based on principal component analysis

Abstract

The invention discloses a transfer learning picture classification method and device based on principal component analysis, and the method comprises the following steps: S100, carrying out the featurevector dimension reduction of each sample in a small sample data set through employing a principal component analysis method, and obtaining a feature vector after dimension reduction; S200, traininga full connection layer neural network in the deep convolutional neural network by using the feature vectors after dimension reduction to obtain a classifier; S300, extracting feature vectors of the to-be-classified pictures, projecting the feature vectors of the to-be-classified pictures to a low-dimensional space to obtain a dimension reduction result, inputting the dimension reduction result into a classifier, and outputting the classification result; and the total number of each type of samples in the small sample data set is less than or equal to 15. According to the method, the sample feature vectors are projected to the low-dimensional space, so that the sample density is increased, the influence of noise on image recognition is reduced, and the classification accuracy is improved.On the other hand, the invention further provides a transfer learning picture classification device based on principal component analysis.

Description

technical field [0001] The present application relates to a transfer learning image classification method and device based on principal component analysis, belonging to the field of image classification. Background technique [0002] Image classification refers to the process of identifying objects in images with only one primary object. In the prior art, a deep convolutional neural network is commonly used for image classification. Since the training of the deep convolutional neural network requires a large number of samples, but the training library that can be formed by the target to be recognized has a small number of samples, the training result of the deep convolutional neural network is easy to overfit. [0003] To solve this problem, the existing methods mostly use the "fine-tuning" transfer learning method to solve it, but the existing "fine-tuning" classification methods have poor recognition ability and low classification accuracy. Contents of the invention ...

AI Technical Summary

Problems solved by technology

[0003] To solve this problem, the existing methods mostly use the "fine-tuning" transfer learning method to solve it, but the existing "fine-tuning" classification methods have poor recognition ability and low classification accuracy

Images