Development Of Copnuterized Spectral Pattern Recognition System For Spectrometer Based On Computer Vision

Spectrometer is mainly used to quickly detect ferrous metal and nonferrous metal, so that, different metal brands can be classified and specified. But it is detected mostly by eyes at present, and depends so much on operators'skill, so misjudging happens frequently. Aiming at the situation above, the computerized spectral pattern recognition system is developed.Main technology of the system is that understanding how men detect different metal brands which means simulating analytical method of spectral pattern by computer.First, image collecting system is designed. CCD is used to do this work and the coupling mechanism which bonds CCD and spectrometer together is also designed to lay image on CCD sensitive surface.Second, analyzing the collected image and fixing on pretreatment arithmetic. Median filtering is chosen to modify the fuzzy edge by using the smoothing arithmetic, Sobel operator is used to detect edge because of its filtering characteristic. Image is changed from gray-scale to two-valued according to the threshold which is automaticly selected by Otsu.Third, arithmetic for spectral pattern intelligent-identified is designed in consideration of the way men detecting different metal brands. Basing on Hough transform, two-valued spectral pattern image is fitted to linear array single-element wide and different-distance apart, and then position information of every line in the array is noted by order. Searching the fitted linear array by initialized value and the identifying goal is accomplished. Clipping the image and judging local threshold are put forward by a lot of experimental analysis, and it works to compensate the influence lead-in by gradient and bended spectral pattern.At last, application software and single unit system are finished. Measuring console is also designed to assemble every part to a whole body.