*2.4. Selecting the Optimal Wavelength*

One HSI dataset contained the spectral information of samples from 400 nm to 1000 nm, simultaneously. However, certain wavelengths had redundant data, resulting in the timeconsuming processing of HSI data [28]. Therefore, it is necessary to eliminate wavelengths containing redundant and irrelevant information to optimize the texture profiles for data analysis samples using the wavelength/variant selection of hyperspectral data [23]. Regarding wavelength selection for HSI analysis, the regression coefficient (RC) is commonly utilized [29,30]. We utilized the RC to determine the optimal wavelength that contributed the most to the prediction of TPA values in common carp muscle. In the calculation of RC, the optimal wavelength is chosen by computing the β-coefficient from the full-wavelength PLSR model. The wavelength with the highest absolute value of the β-coefficient is considered to be the optimal wavelength [30]. The program for RC was operated in MATLAB 2021a software (The MathWorks Inc., Natick, MA, USA).

#### *2.5. Measuring the Texture Indicators of Common Carp Muscle*

We extracted the muscle corresponding to each of the four skin regions. The muscle size was 20 mm × 20 mm × 15 mm. Eight texture indicators of the muscle, including gumminess, springiness, cohesiveness, resilience, hardness, brittleness, adhesiveness, and chewiness, were measured using a texture analyzer (TA.XTC-18, Baosheng, Shanghai, China) and a TA/36 cylindrical probe. The measurement speed was 2 mm/s, and the

trigger force was 5 N. The compressive deformation of one sample was set to 60%. These eight texture indicators were derived from the TPA curves of each sample, and the TPA parameters listed above were calculated using Bourne's technique [31].

For each texture indicator, to examine whether there were significant differences among four muscle regions, we measured the distances with PCA analysis using Tassel 5.0 [32]. The Spearman correlation coefficient of the contents of any two indicators in four samples was calculated using the R 'cor. test' function in the R software (version 4.0.2).
