A pattern recognition method based on a variable sample stack type self-coding network
A stacked self-encoding and self-encoding network technology, applied in the field of large-scale high-dimensional pattern recognition, can solve the problems of pattern recognition accuracy and efficiency decline, and achieve the effect of shortening the time required for recognition
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
specific Embodiment approach 1
[0022] A kind of pattern recognition method based on variable sample stack autoencoder network of the present embodiment, described method comprises the following steps:
[0023] Step 1. Filter out the noise in the high-dimensional space samples through the variable sample stack autoencoder network, and map it into a low-dimensional space denoising sample set;
[0024] Step 2, using the low-dimensional space denoising sample set obtained in step 1 to train the sample training classifier, and obtain a typical sample set in the low-dimensional space denoising sample set through an evolutionary learning process;
[0025] Step 3. Based on the typical sample set obtained in step 2, use inverse mapping to high-dimensional space to obtain a typical sample set in high-dimensional space, and use the similarity recognition method between the sample to be tested and the typical sample set in high-dimensional space to perform pattern recognition. Classification of test samples.
specific Embodiment approach 2
[0026] The difference from the first embodiment is that in this embodiment, a pattern recognition method based on a variable sample stacked autoencoder network, in the first step, the variable sample stacked autoencoder network is used to filter out the samples in the high-dimensional space Noise, the process of mapping into a low-dimensional space denoising sample set, specifically:
[0027] First of all, a stacked autoencoder network is established, and the samples at the edge of high-dimensional space samples are screened out layer by layer under the clustering method through the stacked autoencoder network, and a low-dimensional space sample set with a high degree of aggregation is obtained at the bottom of the stack to improve the low-dimensional space. typicality of the sample.
specific Embodiment approach 3
[0028] The difference from the second specific embodiment is that a pattern recognition method based on a variable-sample stacked autoencoder network in this embodiment, the stacked autoencoder network is established, and then the stacked autoencoder network is used under the clustering method The process of filtering out samples at the edge of high-dimensional space samples layer by layer, and obtaining a high-degree aggregation low-dimensional space sample set at the bottom of the stack, changes the low-dimensional space mapping of the sample stack autoencoder network. The high-dimensional space is mapped to the low-dimensional space through the stacked autoencoder network, such as image 3 shown. Due to the independent uncorrelation of random noise in high-dimensional space samples, outlier samples will be generated during the mapping process. In order to improve the representativeness of the sample set, unsupervised clustering based on unscented transformation is used to f...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More