Abstract: The optical fiber inertial navigation of coal mine excavation equipment is easy to be disturbed by multi-source heterogeneous noise such as vibration and electromagnetism in the underground excavation face, which leads to the reduction of the accuracy of position and attitude detection of excavation equipment. Therefore, in order to eliminate the influence of noise interference on the positioning and detection of integrated inertial navigation, a multi-source heterogeneous noise reduction method of optical fiber inertial navigation based on improved empirical mode decomposition (EMD) is proposed. Firstly, according to the environmental noise characteristics of underground heading face, the composition and characteristics of multi-source heterogeneous noise are analyzed. Then, the improved EMD noise reduction method is used to reduce the multi-source heterogeneous noise of optical fiber inertial navigation. This method adaptively selects the effective eigenmode function (IMF) by calculating the continuous mean square error (CMSE), optimizes the effective IMF component of each layer by setting the threshold function, and reconstructs the optimized effective IMF component, so as to obtain the signal after noise reduction. In order to verify the noise reduction effect of the improved EMD method, the comparative experiments of wavelet noise reduction method, variational modal decomposition (VMD) noise reduction method and improved EMD noise reduction method are carried out. The results show that the noise reduction effect of the improved EMD noise reduction method is better than that of wavelet noise reduction and VMD noise reduction method. In order to verify the improvement effect of the improved EMD Method on the pose detection accuracy of tunneling equipment, the combined positioning results of optical fiber inertial navigation and displacement sensor before noise reduction and after noise reduction by three noise reduction methods are compared and analyzed. The results show that the pose error of the combined positioning after noise reduction of optical fiber inertial navigation signal by the improved EMD method is the smallest, and it is greatly reduced compared with that before noise reduction, The Pitch angle error decreased from 0.1131° to 0.0286°, the Roll angle error decreased from 0.0101° to 0.0038°, the Yaw angle error decreased from 2.7051° to 0.0272°, the North position error decreased from 36.8872 to 0.0822 m, the East position error decreased from 0.4990 to 0.0161 m, and the Upward position error decreased from 0.9141 to 0.0143 m. The improved EMD noise reduction method proposed in this paper can effectively filter the multi-source heterogeneous noise of optical fiber strapdown inertial navigation signal of coal mine excavation equipment, which lays a good foundation for the accurate positioning of coal mine excavation equipment.