[Study on Multi-Bands Recognition for Varieties of Mutton by Using Hyperspectral Technologies].

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Accurate identification of mutton varieties is crucial for quality control and traceability in the meat industry.
• Traditional methods for distinguishing sheep breeds can be time-consuming and subjective.
• Hyperspectral imaging offers a non-destructive and objective approach for analyzing food products.

Purpose of the Study:

• To develop and evaluate a hyperspectral imaging method for identifying and classifying three mutton varieties: Tan-han hybrid sheep, Yanchi Tan-sheep, and small-tailed sheep.
• To compare the effectiveness of different spectral preprocessing techniques and wavelength selection algorithms.
• To assess the performance of linear discriminant analysis (LDA) and support vector machine (SVM) models for mutton variety classification.

Main Methods:

• Visible/near-infrared (400-1000 nm) and near-infrared (900-1700 nm) hyperspectral imaging were employed.
• Spectral data underwent preprocessing, including baseline correction and Savitzky-Golay (SG) convolution smoothing.
• Successive Projection Algorithm (SPA) was used for characteristic wavelength extraction.
• Linear Discriminant Analysis (LDA) and Radial Basis Function Support Vector Machine (RBFSVM) models were applied for classification.

Main Results:

• The RBFSVM model achieved higher identification accuracy than the LDA model across both spectral ranges.
• In the 400-1000 nm range, the Baseline-Fullwave-RBFSVM model with 12 characteristic wavelengths yielded 100% accuracy.
• In the 900-1700 nm range, the Baseline-Fullwave-RBFSVM model with 7 characteristic wavelengths achieved 96.25% accuracy.
• Classification accuracy was higher in the 400-1000 nm band, indicating that color and texture differences are more significant than component content variations among the breeds.

Conclusions:

• Combined hyperspectral imaging technology and RBFSVM models provide an effective method for recognizing mutton varieties.
• The visible/near-infrared spectrum (400-1000 nm) is more effective for distinguishing these mutton varieties than the near-infrared spectrum (900-1700 nm).
• This approach offers a rapid, non-destructive, and accurate tool for mutton variety identification in the agricultural and food industries.