Research Of Jewelry Classification And Recognition Based On Hyperspectral Matching Technology

Hyperspectral remote sensing technology is a new type of map technology developed in recent years.Hyperspectral technology makes people know the world and things have improved qualitatively.The hyperspectral data of "atlas integration" not only contains the spatial information of the sample,but also can extract the spectral information of the sample.In this way,hyperspectral images have the characteristics of large amount of information,high spectral resolution,so it can not only provide rich surface information,but also provide more abundant spectral information reflecting the composition characteristics of the material.Therefore,the research on the potential characteristics of hyperspectral image data has attracted wide attention from scholars at home and abroad.Its applications include precision agriculture,atmospheric research,environmental monitoring,medicine and so on.At present,the application of hyperspectral technology in the identification of jewelry species and authenticity identification is at the stage of research,which has the application prospect and technical advantages of spectrum.In this paper,17 kinds of standard jewelry samples and jewelry samples to be tested were collected and extracted by Beijing Zolix GaiaSorter hyperspectral sorter.The image preprocessing,image feature extraction and feature selection,and spectral matching of jewelry were discussed.In this paper,we operate and discuss according to the following flow: Firstly,various pretreatment methods of hyperspectral image are discussed and compared in detail.Savitzky-Golay smoothing method is used to smooth the original spectral curve.Several methods of feature extraction and feature selection are discussed in detail.The spectral data of jewelry samples after pretreatment and feature extraction are used to build the jewelry spectrum database.Finally,common and combined spectral matching methods are used to match color jewelry,crystal jewelry and agate jewelry.The results show that the combination measure algorithm can significantly improve the recognition accuracy when matching different ethnic jewelry samples:(1)The Euclidean distance method(ED)can be used to calculate the spectral difference of spectral similarity of the same kind of jewelry.However,the matching accuracy is poor,because it is only sensitive to the spectral amplitude,and can not characterize the shape characteristics of spectral curves.Therefore,Euclidean distance is often unable to distinguish different groups of jewelry,and is usually used to combine other similarity measures to form a combination matching method.The spectral angle cosine method(SAM)considers the angle between the spectral vectors,and has a good matching effect when the spectral curve features are different.Correlation coefficient method(SCC)takes into account the overall shape of the spectral curve.After modifying the range of values,it can magnify the difference of different types of jewelry.It is not ideal for identical jewelry with similar overall shape of the spectral curve.Spectral Information Dispersion(SID)describes the similarity of spectral vectors with information entropy.The more jewelry samples are,the better the discrimination is.(2)Spectral Angle Cosine-Euclidean Distance(SAM-ED): Because the spectral similarity of Euclidean Distance Measure is only sensitive to the curve amplitude characteristics,but not good at spectral shape feature processing,the precision of hyperspectral matching recognition of jewelry by this distance similarity measure alone is often low;Spectral Angle Cosine Measure is more sensitive to the shape of spectral curve.However,the ability to process spectral distance features is weak,and it is often difficult to deal with jewelry categories with similar spectral curve shape features and large amplitude differences.The spectral angle cosine-Euclidean distance(SAM-ED)method considers the shape and amplitude information of the spectral curve comprehensively.Compared with the single matching method,it can significantly improve the matching accuracy.The Euclidean Distance-Correlation Coefficient(EDSCC)considers the shape and amplitude information of the spectral curve comprehensively.After modifying the range,the difference of different jewelry types can be enlarged.Spectral angle cosine-spectral information divergence(SAM-SID)combination algorithm compensates for the defect that spectral angle algorithm can not measure spectral amplitude characteristics,and measures spectral probability distribution at the same time.