Classification Prediction Of Gem Greenbased On The SVM Nder Uniform Color Space

With the approaching of “colored stones Era” in China, standardizing the marketand promote its sound development is of great significance. The color of gemstoneshas a major impact to the appraisal and quality grading. The traditional way of visualdetection for the color by the expert is often easily influenced by objective factorssuch as environment and subjective factors like the experience of the evaluator withlarge deviation to the result which isn’t convincing.On account of the above reason, this thesis starts with researching greengemstones. Four common gemstones have been chosen-jade, turquoise, emerald andolivine (all green), and these four types of natural colored gemstone have been putunder D65light with10°angle for color measures by color photometer on the basis ofCIE1976L*a*b*uniform color space. In total156、212、253and138data sets havebeen measured for emerald, jade, turquoise and olivine respectively. Scatter diagramin three-dimensional space is shaped for those data for statistical analysis anddistinguishability research thereafter. And then this article continues using Fisherdiscrimination function for green gemstone’s classification forecasting with thefoundation of predecessors’ work, and turns out the classification forecastingaccuracy-98.5%, is better comparing to previous experimental results. However,similar classification forecasting method using traditional statistics is limited. Thus, anew and small sample size method-Support Vector Machine (SVM) is adopted in thisdissertation, which can overcome the dimensional curse, local minimum point and astudy hard difficulties and is good at generalization.712sets of data have beenrandomly distributed,500former sets are chosen to be training set and the other212sets are test set. By using [0,1] normalization method and radial basis function, themodel of Matlab simulation is able to be built with the optimum parameters of c=4,g=16respectively. Corresponding classification and prediction accuracy is99.1%by this method, whose result is much more ideal.In view of the SVM’s high accuracy to gemstone color’s classificationforecasting as well as its acute identification ability to color, a parameter database of the gemstones’ color can be built based on increasing gemstones’ types and quantitiesduring practical application, and then appraisal of gemstone can be done，and we alsocan combine learning the characteristic absorption spectrum with color,i.e. bymeasuring the color parameter, the gemstone type can be automatically shown,therefore appraisal efficiency can be raised, convenience can be bought to thecustomers, and the sound and orderly development of gemstone market can bepromoted...