Worm time sequence classification typical sample confirmation method

Abstract

The invention discloses a worm time sequence classification typical sample confirmation method, which comprises the steps of firstly, performing intra-class Euclidean distance calculation of a time sequence to obtain an alternative time sequence group smaller than a distance average value and the distance average value; the new time sequence enters intra-class training, determining whether the time sequence meets the time sequence group distance requirement or not according to the distance result, and if yes, entering the time sequence group; secondly, calculating an inter-class distance between the alternative time sequence group and other types of time sequences; calculating the new time sequence and typical time sequence groups of other classes, determining the relationship between thenew time sequence and the inter-class distance, and if the relationship is greater than the inter-class distance, the new time sequence completely meets the requirements of typical time sequence samples; and finally, checking the classification accuracy, and making a corresponding mark. An acknowledged time sequence database is adopted for experimental verification, and a result proves that the worm time sequence classification typical sample confirmation method can obtain each type of typical sample, so that the effectiveness of the samples is ensured, and the type division is clearer.

Description

technical field [0001] The invention relates to the technical field of incremental learning in pattern classification, in particular to a method for confirming typical samples of worm time series classification. Background technique [0002] In the modeling process of pattern classification, training samples are often generated by artificial division. The number of training set samples has a clear effect on obtaining sufficient information. If the number of training samples is too small, the amount of information obtained cannot be completed to represent the number of samples and cannot be completed. learning tasks; similarly, too large a number of samples will also cause overfitting caused by redundant samples, resulting in a decline in generalization ability. Therefore, how to select appropriate training samples from the original time series samples of known time series types, Removing duplicate information and samples that lead to overfitting has become one of the key fac...

AI Technical Summary

Problems solved by technology

However, the typical samples obtained based on the above methods are affected by the final sample size. When the sample set contains clear interfering samples, the data noise is too large, which will affect the confirmation of the typical sample set;

[0006] The goal of using the generic function is to ensure that as many training samples with the largest generic value as possible are retained in the final typical training sample set, but since the number of training samples with the closest distance to the individual to be evaluated is set artificially, the definition of the generic category belongs to manual judgment It does not follow the calculation results; at the same time, when obtaining training samples with significant correlation, only one of them is finally used as a typical sample, which limits the projection of data in each dimension from the perspective of high-dimensional attributes; the obtained above samples The assembly deviates from the real data position, which eventually leads to a decrease in the accuracy of the classification results;

[0007] In the process of genetics research, it is an important method to distinguish worm types according to the movement and crawling traces. The movement traces of worms on the plate are expressed as time series, and these time series have corresponding type marks. How to ensure the time The role of sequence samples in the classification process requires confirmation of the typicality of samples, but in the prior art, there is no research on the typicality confirmation of worm movement samples

Images