Dong Yang, Anxiang Lu, and Jihua Wang
Hyperspectral imaging; machine learning; cooked meat; texture; classification
The purpose of this study is to investigate the feasibility of an NIR hyperspectral imaging (HSI) technique coupled with machine learning methods for the classification of cooked meat. A total of 237 hyperspectral images were acquired of cooked beef, lamb, and pork, and then preprocessed for the current investigation. Mean spectra were extracted from the regions of interest and adjusted using the second derivative. Twelve optimal wavelengths were identified using a random frog algorithm (RFA) method. Furthermore, principal component analysis (PCA) was used to determine the optimal fusion images from 12 characteristic images, and a total of 8 texture feature variables were extracted by an integration of local binary patterns and Tamura features from fusion images. Using the spectral data, texture information, and the fusion of them, three classification models were established employing different machine learning algorithms: fuzzy neural networks (FNN), random forests, and k-nearest neighbours. The results of tests on the fusion data showed that the RFA-FNN model performed the best, with a high correct classification rate of 96.2% and 94.93% for the calibration sets and prediction sets, respectively. These results demonstrate that the integration of HSI techniques and machine learning methods is an effective approach to the classification of cooked meat products.
Important Links:
Go Back
