In order to solve the problem of RFID communication security, this paper proposes a new physical layer tag ant-counterfeiting technology. This technology separates the original signal to obtain the expected, noise and normalized signals, and extracts the spectral entropy features of the three signals.Finally, feature selection is used to achieve true and false label classification. In addition, this paper proposes a new cross-validation to objectively test the performance of the physical layer method. The results show that the accuracy of the method in this paper is nearly 4% higher than that of the traditional method. Under the new cross-validation, the classification accuracy of the physical layer method will drop by 8~10 percentage points. From this, we get an important conclusion, it is of great significance to use the spectral entropy feature in the physical layer identification method to realize the label anti-counterfeiting.