Convolutional dendritic method for extracting feature logic for classification

Abstract

The invention discloses a convolutional dendritic method for extracting feature logic for classification, and belongs to the field of artificial intelligence. Classification is carried out by relying on the feature logic of input data, and a logical relationship between the features can be obtained while data features are extracted; comprising the following steps: (1) firstly, performing convolution operation on input image signal or time sequence signal data by using a weight matrix to extract input data features to obtain feature data; (2) summing the feature data and the offset matrix, and introducing a constant term to obtain intermediate data; (3) performing Hadamm product construction on the intermediate data containing the feature data and the constant term and the intermediate data itself to construct a logic relationship between features; and (4) iteratively optimizing the weight matrix and the bias matrix by adopting an error back propagation algorithm so as to reserve a feature logic relationship contributing to classification precision as output data of the layer, and the output data of the layer is input data of the next layer. The invention has the advantages of being small in calculation amount, high in classification precision, good in model stability, high in convergence speed and high in transportability.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a convolution dendritic method applied to image signals or time series signals to extract feature logic for classification. Background technique [0002] Classification is a fundamental problem in several domains such as fault diagnosis, automation, computer vision (CV), and natural language processing (NLP). Machine learning has always been a useful tool for solving classification problems. Classification is the task of dividing data according to sample characteristics. Therefore, it is natural to think that this problem can be solved by finding suitable classification curves or surfaces. However, machine learning algorithms using this strategy only generate a black-box model. Traditional classification models mostly find appropriate classification curves or surfaces to divide data sets according to sample features, but ignore the logical relationship between sample fea...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is to overcome the problems that the traditional dendritic network cannot be used for feature extraction, it is difficult to embed into other networks, and the logical relationship extracted by the previous network is lost, and a feature extraction logic for classification is designed for the classification field. The product dendritic method, which can be applied to image signal processing and time series signal processing, considers the logical relationship between the features while extracting the features of the input data, which is completely different from the traditional network principle of finding classification curves and surfaces for classification , which has the advantages of high classification accuracy, fast convergence speed, good stability, and good portability.

Images