A hyperspectral classification detection method for medical foreign matter

Abstract

The invention discloses a hyperspectral classification detection method for foreign matter in medicine. Firstly, the hyperspectral image of foreign matter in medicine is input; secondly, a hyperspectral denoising method for foreign matter by polynomial smoothing filtering is proposed to suppress spectral noise interference; The feature extraction method combining dimensionality and semi-supervised LDA first uses band clustering and grouping PCA dimensionality reduction to reduce the dimensionality of the preprocessed image, and extracts spectral features through semi-supervised LDA, and then uses two-dimensional Gabor filter to extract spatial features. The above features are combined as the classification features of the image; finally, the support vector machine is used to realize the detection of foreign objects in medicine and output the categories of foreign objects. The present invention proposes a secondary dimensionality reduction method of PCA and LDA in order to extract spectral features that are more conducive to subsequent classification operations; at the same time, semi-supervised LDA is introduced to reduce the dependence on label data and achieve high accuracy of foreign objects under a small number of label data samples rate detection.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for hyperspectral classification and detection of medical foreign matter. Background technique [0002] In recent years, my country's pharmaceutical market has developed rapidly, and the market size has increased from 1,220.7 billion yuan in 2015 to 1,633 billion yuan in 2019, with a compound annual growth rate of 7.5%. It is expected to further grow at a compound annual growth rate of 6.8% from 2020 to 2021 Maintain growth and reach 1,305.7 billion yuan in 2021. At the same time, medical safety, as the cornerstone of the development of the pharmaceutical industry and an important guarantee for the health of the people, has also attracted more and more attention from inside and outside the industry. In the process of pharmaceutical production, various and weak foreign objects such as glass, debris, stones, hair, and rubber crumbs often appear, which bring potent...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the present invention proposes a hyperspectral classification detection method for medical foreign matter, which is based on band clustering and grouping PCA dimensionality reduction + semi-supervised LDA spectral feature extraction to solve the problem that the existing PCA algorithm cannot extract more separable features At the same time, the semi-supervised LDA algorithm solves the problem of relatively few label samples

Images