Abstract: Abstract: In order to improve the generality of the water content hyperspectral calibration model for different varieties of cold fresh pork, a new hyperspectral signal correction algorithm called a variety sensitive wavelength selection combining with piecewise direct standardization (VSWS-PDS) is proposed. The variety sensitive wavelengths were first selected, based on a regression coefficient of the partial least square regression (PLSR) model, then the piecewise direct standardization(PDS) algorithm was utilized to correct the selected wavelengths and to eliminate the impact of variety difference on model prediction results. The detailed process of the VSWS-PDS algorithm is illustrated as follows: (1) The samples of the "master" variety were divided into calibration set mC and prediction set mP by utilizing the Kennard and Stone(KS) algorithms. (2) A PLSR model named PLSR1 was built with a calibration set mC. (3) A small quantity of representative "slave" variety samples selected from "slave" variety samples by utilizing sample set partitioning based on a joint X-Y distances (SPXY) algorithm were added to mC，and then a new PLSR model named PLSR2 was built. (4) The variety sensitive wavelengths were selected on the basis of the relative difference between the regression coefficients of model PLSR1 and PLSR2. (5) The standardization samples mTC were selected among the calibration samples mC by utilizing the KS algorithm, then the Euclidean distances between the physical or chemical reference values of the sample set mTC and all of the "slave" variety samples were calculated one by one, the "slave" variety samples with the minimum distance from mTC were chosen as standardization samples, yielding sTC, and the rest, yielding sP. (6) The average spectrum of mTC and sTC was calculated, yielding mTmean and sTmean. If wavelength i was variety sensitive, a new spectrum matrix Zi was reconstructed from sTmean in a small window from i-k to i+k, a multiple linear regression (MLR) model was built between the spectral intensity mTmeani and the corresponding spectrum matrix Zi, then the regression coefficient vector bi was obtained. If wavelength i was not variety sensitive, then bi was equal to 1. (7) The regression coefficient vectors were placed in a banded diagonal transfer matrix F. (8) The "slave" variety samples sP are chosen to evaluate the correction effect of the VSWS-PDS algorithm. First of all, spectrum matrix was obtained from sP, yielding XStest. Subsequently, matrix XStest was multiplied by the transfer matrix F, yielding XStesttrans, and then the "master" variety calibration model PLSR1 was utilized to predict the physical or chemical values corresponding to XStesttrans directly. The prediction results of PLSR1 were used to evaluate the correction effect of the VSWS-PDS algorithm. The No. 0 indigenous pork was selected as the "master" variety and the Enshi mountain pork was selected as the "slave" variety. The spectra of the "slave" variety samples sP were corrected by the VSWS-PDS algorithm, water content was then predicted by the "master" variety calibration model PLSR1 with the cross validation determination coefficients () of 0.91, cross validation root mean square error (RMSECV) of 0.29% ,and cross validation residual prediction deviation (RPD) of 3.3, with prediction determination coefficients () of 0.84, prediction root mean square error(RMSEP) of 0.50%, and prediction RPD of 2.58. Meanwhile, the experimental results without the VSWS-PDS correction process were of 0.20, RMSEP of 1.42%, and prediction RPD of 0.91. Obviously, the VSWS-PDS algorithm can significantly improve the prediction ability of the model built with No. 0 indigenous pork samples to predict water content of the Enshi mountain pork samples. Finally, a comparative study between direct standardization (DS), PDS, and the VSWS-PDS is conducted. It is shown that the transfer results obtained with the proposed method VSWS-PDS were better than those obtained by DS or PDS.