以機器視覺與可見光／近紅外光應用於畜肉之分類

Classification of Meat Using Machine Vision and Visible/Near-Infrared Spectroscopy

周震煌(Chen-Huang Chou) ； 劉昌群(Chan-Chung Liu) ； 蕭介宗(Jai-Tshung Shaw)

《農業機械學刊》 14卷3期 (2005/09) Pp. 39-49

https://doi.org/10.30062/JAM.200509.0004

摘要

Near infrared transmittance and reflectance spectra on the tenderloins of black and white hair hog samples and on the loins of beef samples were analyzed and calibrated for the three measured pork and beef classification indexes including gray level intensities of green, blue, and red. If the relationship between spectra and the three measured pork and beef classification indexes was established by principal component analysis (PCA) model and SIMCA classification, the calibration model could be used to classify the tenderloins of black- and white-hair hogs and the loins of beef with the advantages of saving time and zero-polluted environment. With 25 black- and 15 white- hair hogs samples, and 15 beef samples, the average intensities of red existed no big difference except a little low intensities of green and blue on beef samples. It would be difficult to classify these samples by RGB intensities from machine vision. With total of 15 wavelengths in the range of 652~668 nm, 804~836 nm and 1320 nm to validate 25, 15 and 15 extracted samples, the classification rates by SIMCA had 83.6% at 5% level of significance with PCA calibration model from extracted samples of 45 black-hair hogs, 30 white-hair hogs and 30 beefs. The three important wavelengths with the corresponding absorbance on this classification were 664 nm, 804 nm and1320 nm.

關鍵字

近紅外光 ； 分類 ； 豬肉 ； 牛肉 ； 機器視覺

並列摘要

Near infrared transmittance and reflectance spectra on the tenderloins of black and white hair hog samples and on the loins of beef samples were analyzed and calibrated for the three measured pork and beef classification indexes including gray level intensities of green, blue, and red. If the relationship between spectra and the three measured pork and beef classification indexes was established by principal component analysis (PCA) model and SIMCA classification, the calibration model could be used to classify the tenderloins of black- and white-hair hogs and the loins of beef with the advantages of saving time and zero-polluted environment. With 25 black- and 15 white- hair hogs samples, and 15 beef samples, the average intensities of red existed no big difference except a little low intensities of green and blue on beef samples. It would be difficult to classify these samples by RGB intensities from machine vision. With total of 15 wavelengths in the range of 652~668 nm, 804~836 nm and 1320 nm to validate 25, 15 and 15 extracted samples, the classification rates by SIMCA had 83.6% at 5% level of significance with PCA calibration model from extracted samples of 45 black-hair hogs, 30 white-hair hogs and 30 beefs. The three important wavelengths with the corresponding absorbance on this classification were 664 nm, 804 nm and1320 nm.

並列關鍵字

Near infrared ； Classification ； Pork ； Beef ； Machine vision