The current method mostly adopts a single sensing temperature parameter for compensation, without distinguishing the different temperature parameters of each sensor, resulting in a poor compensation effect. In this regard, a temperature drift compensation method for laser gyroscopes with improved radial basis functions is proposed. Firstly, wavelet decomposition is used to decompose the drift signal of the laser gyroscope containing noise, and the wavelet coefficients of each scale are calculated and corrected to obtain the reconstructed non-interference temperature sensing signal. Then, by using values such as pulse vector and assembly error matrix, the laser gyroscope accelerometer is measured to distinguish the temperature parameters of each sensor. The self-organized feature mapping network is used to classify the laser gyro output parameters, and the radial basis function neural network transformation function is selected to calculate the error reduction rate of each temperature sensor parameter. Finally, the radial basis function neural network is trained with the orthogonal least squares algorithm to complete the temperature drift compensation of the laser gyroscope. Through experiments, it is proved that the proposed method for temperature drift compensation of laser gyroscopes has good performance, stable output, and can enhance the performance of high-precision navigation, positioning, orientation and other equipment.