High Confidence Visual Recognition of Persons by a Test of Statistical Independence
IEEE Transactions on Pattern Analysis and Machine Intelligence
Pattern Classification (2nd Edition)
Pattern Classification (2nd Edition)
| Hi-index | 0.00 |
Pork quality assessment is important in the pork industry application. However, traditional pork quality assessment is conducted by experienced workers and thereby is subjective. In this paper, a computerized system scheme based on hyperspectral imaging technique is proposed for objective pork quality evaluation. This hyperspectral imaging technique used texture characteristics to develop an accurate system of pork quality evaluation. Hypercube, which is a set of spectral images over all wavelengths, were filtered by oriented Gabor filters to obtain texture characteristics. Spectral features were extracted from Gabor-filtered cube as well as from hypercube directly and then compressed by the principal component analysis (PCA) over the entire wavelengths (400-1000 nm) into 5 and 10 principal components (PCs). ‘Hybrid' PCs were created by combining PCs from hypercube and from Gabor-filtered cube. Linear discriminant analysis (LDA) was employed to classify pork samples based on hybrid PCs as well as pure PCs. The cross-validation technique was applied on LDA to produce the unbiased statistical results. The overall average accuracy was 72% by 5 pure PCs and reached 84% by 5 hybrid PCs. The highest accuracy, 100% classification for all samples, was obtained when using 5 hybrid PCs. Thus, a statistical significant improvement was achieved using image texture features. Results showed that the proposed computerized system worked well on pork quality evaluation and has potential for on-line pork industry application.