Development of Discriminant Models for Wooden Breast Based on Visible and Near Infrared Hyperspectral Information and Their Fused Data

Abstract

Wooden breast barriers the development of broiler industry, and traditional detection methods are time-consuming and inefficient. To investigate the feasibility of the hyperspectral imaging (HSI) technique for the detection of wooden breasts, four different grades of white feather chicken breast were selected and their HSI information of 400~1000 and 1000~2000 nm was collected. After spectral preprocessing and spectral variable selection, partial least squares discriminant (PLS-DA) models and support vector machine (SVM) models were developed based on full wavelength and characteristic spectral variables, as well as their fused HSI data. The results showed that SVM models showed better results than PLS-DA models to discriminate woody grades of chicken breasts. The overall discrimination rates based on the full HSI bands and selected spectral variables in 1000~2000 nm were higher than those of models in 400~1000 nm. Besides, the discrimination models based on fused HSI data of HSI bands and selected spectral variables provided the best results, with the overall discrimination rate of 96.7% for four different woody grades, and the accuracy of the four grades could reach more than 90%. The research results provided technical support for HSI to achieve rapid and non-destructive detection of wooden chicken breasts.

Keywords

• wooden breast
• visible-near infrared hyperspectral
• short-wave infrared hyperspectral
• spectral data fusion
• discriminant model