Application of Spectroscopy and Spectral Image to Detect Chicken Carcasses and Their Lesions

碩士 === 國立屏東科技大學 === 生物機電工程系所 === 101 === In this study, the visible/near infrared spectroscopy detection technique is nondestructive development of fairly mature detection technology, and the wavelength range of 400-1000 nm and interval of 1 nm. The laboratory had developed hyperspectral imaging...

Full description

Bibliographic Details
Main Authors: Jau Ting Wu, 吳昭葶
Other Authors: Ching Lu Hsieh
Format: Others
Language:zh-TW
Published: 2013
Online Access:http://ndltd.ncl.edu.tw/handle/36116052119173470994
Description
Summary:碩士 === 國立屏東科技大學 === 生物機電工程系所 === 101 === In this study, the visible/near infrared spectroscopy detection technique is nondestructive development of fairly mature detection technology, and the wavelength range of 400-1000 nm and interval of 1 nm. The laboratory had developed hyperspectral imaging system with wavelength range 400-1000 nm and interval of 0.3 nm . The research purpose of a comparison was made between visible/near infrared and hyperspectral imaging system.In the future, the chicken carcasses samples spectral method for automated on-line inspection . It had been used as a research to identify key wavelengths for detected 70 wholesome chicken carcasses and 70 unwholesome chicken carcasses ,and classified analysis for wholesome and unwholesome chicken samples .To used principal component analysis (PCA) and classified to wholesome, unwholesome were differentiated from each other with high accuracy. key wavelengths were identified intensity of abdomen loading weights at 442.4、580、582、800.3 nm for PC1and PC2.And key wavelengths were identified intensity of back loading weights at 441.8、480、580、917.6 nm PC1and PC2, in conclusion were wavelengths at visible wavelength region. Three pre-processing steps include smoothing, offset, normalization methods and the results to showed obvious difference in classification accuracy spectroscopy. In general, the classification accuracy was improved by increased the number of scores of input data, including score 2, score 4, score 6, score 8, score 10 to achieved 90%. The training achieved 90% classified wholesome and unwholesome. The testing achieved 90% classified wholesome and unwholesome 90%.