To quickly and effectively identify water injection meat, the spectral characteristics of water injection and normal meat were analyzed through the hyperspectral analysis technology. The spectrum characteristic parameters of every sample were obtained by using the Fourier transform and statistical calculation. Then the support vector machine (SVM) and neural network (BP) classification models were developed based on the full spectrum, characteristic spectrum and spectrum characteristic parameter, respectively. Finally, the two models were validated by an independent validation set and three indicators such as product’s accuracy (PA), user’s accuracy (UA) and overall accuracy (OA) were used to test the model performance. The results showed that the neural network classification recognition model based on spectrum characteristic parameters had optimal classification recognition rate for water injection pork, with the correct recognition rate of 98.8%. The neural network classification and recognition model based on the characteristic spectrum had the second best recognition, with the correct recognition rate of 96.4%. The classification recognition model of SVM based on full spectrum had the worst classification and recognition for water injection pork, and its correct recognition rate was only 84.5%. These results suggest that hyperspectral technique can be used for rapid and effective detection and identification of watered-down pork.