Discrimination of beef muscle based on visible-near infrared multi-spectral features: Textural and spectral analysis

Figures & data

Figure 1. A: An example of a cube image of beef muscles (y = number of pixel; x = number of pixels, z = number of LED; in our study x = 1200 pixels, y = 1200 pixels, and z = 19) and image of Biceps femoris: B; Longissimus thoracis: C and Semimembranosus: D muscles recorded after excitation at 405 nm for Limousine breed.

Figure 2. Process of GLCM (Gray level co-occurrence matrix) features extraction from the Region of interest (ROI; A, B, and C: the three steps necessary for the extraction of the GLCM texture features; X: number of pixels in the x axis, 575; Y: number of pixels in the y axis, 700; Z: number of LED, 19; Moy: Mean; Var: Variance; ASM: Angular second moment; CONT: Contrast, CORR: Correlation; CORR_N: Non normalized correlation; ENT: Entropy; HOM: Homogeneity; DIAG: Diagonal moment; CSHADE: Cluster shade; CPROM: Cluster prominence).^[1]

Figure 3. Process of GLHD (gray level histogram difference) features extraction from the region of interest (ROI; A, B, and C: the three steps necessary for the extraction of the GLHD texture features; X: number of pixels in the x axis, 575; Y: number of pixels in the y axis, 700; Z: number of LED, 19; V: image obtained after translation; U: initial image having the same number of pixel as the image V; W: image obtained after subtraction of pixel of image V from pixel of image U; MOY: Mean, VAR: Variance, ASM: Angular second moment, CONT: Contrast, ENT: Entropy).

Figure 4. Process of multispectral image acquisition and data (image mean spectra, GLHD: Gray level histogram difference and GLCM: Gray level co-occurrence matrix) extraction.

Figure 5. Example of mean spectral data of the three muscles based on limousine breed (BF: biceps femoris; LT: longissimus thoracis; SM: semimembranosus).

Figure 6. Mean spectral data performed on the three muscles types for each animal types (AA: Aberdeen angus, LI: Limousine, BA: Blonde d’Aquitaine).

Table 1. PLSDA results of beef meat samples for calibration and external validation based on the animal breed.

						Confusion matrix for calibration			Confusion matrix for validation
Method	Preprocess	Direction	LED (nm)	PLSDA factor (validation)	Classes	CC (%)	Precision (%)	Error (%)	CC (%)	Precision (%)	Error (%)
SPECT	RAW	—	—	10	BA	87.0	86	14	88.0	88.0	12
		—	—		LI	60.0	85	32	56.0	83.0	40
		—	—		AA	77.0	89	16	87.0	86.0	13
					Mean	75	87	21	77	86	22
GLCM	—	0°	505	5	BA	84.0	75	24	84.0	74.0	19
	—				LI	42.0	81	32	37.0	73.0	36
	—				AA	70.0	83	24	67.0	86.0	25
					Mean	65	80	27	63	78	27
GLHD	—	135°	525	5	BA	62.5	74	27	73.0	72.0	23
	—				LI	51.0	72	37	39.0	60.5	40
	—				AA	77.0	84	19	62.0	85.5	25
					Mean	63.5	77	28	58	73	29

BA: Blonde d’Aquitaine; LI: Limousine; AA: Angus Aberdeen; CC: percentage of correct classification; SPECT: image mean spectra, GLHD: gray level histogram difference; GLCM: gray level co-occurrence matrix; RAW: un-preprocessed spectra; SNV: standard normal variate spectra, MSC: multiple scattering correction spectra; ARE: area under curve = 1; Direction: best angle identified for predictive models; CC: percentage of correct classification; PLSDA factor: number of loading vectors for the best discrimination.

Table 2. PLSDA results of beef meat samples for calibration and external validation based on the muscle type.

						Confusion matrix for calibration			Confusion matrix for validation
Method	Preprocess	Direction	LED (nm)	PLSDA factor (validation)	Classes	CC (%)	Precision (%)	Error (%)	CC (%)	Precision (%)	Error (%)
SPECT	AREA	—	—	10	BF	78	88.5	16	69.0	95.5	16
		—	—		LT	96	95.0	6	97.5	97.6	8.5
		—	—		SM	83	89.0	12	80.5	86.2	14
					Mean	83	91	11	82	93	13
GLCM	—	135°	940	5	BF	61.0	79.0	25	62.0	79.0	24
	—				LT	97.5	94.0	4	96.0	94.0	5.0
	—				SM	60.0	74.0	24	57.0	80.0	30
					Mean	73	82	18	72	84	20
GLHD	—	45°	940	4	BF	51.0	73.0	28	54.0	75.0	22
	—				LT	97.0	96.0	3	96.0	94.0	5.0
	—				SM	63.0	83.0	24	58.0	73.0	20
					Mean	70	84	18	69	81	16

BA: Blonde d’Aquitaine; LI: Limousine; AA: Angus Aberdeen; CC: percentage of correct classification; SPECT: image mean spectra, GLHD: gray level histogram difference; GLCM: gray level co-occurrence matrix; RAW: un-preprocessed spectra; SNV: standard normal variate spectra, MSC: multiple scattering correction spectra; ARE: area under curve = 1; Direction: best angle identified for predictive models; CC: percentage of correct classification; PLSDA factor: number of loading vectors for the best discrimination.

Figure 7. PLS-DA score plot of the two first loading vectors (LV); A: illustrating the separation of the three beef muscles for the different animal types (BF: biceps femoris; LT: longissimus thoracis; SM: semimembranosus), and loading plot; B: depicting the contribution of the wavelengths to the muscles discrimination.

Cozzolino, D.; Murray, I. Identification of Animal Meat Muscles by Visible and Near Infrared Reflectance Spectroscopy. LWT–Food Science and Technology 2004, 37, 447–452.

 Web of Science ®Google Scholar