Abstract: Ultra-wideband (UWB) radar is widely used in the field of penetration detection. In the disaster scenarios such as mine collapse and building collapse caused by natural or human factors, the propagation medium is non-uniform and discontinuous, and it is difficult to obtain the micro-motion echo of buried life in the sheltered space. The difference of human micro-Doppler characteristics between mine and ground building penetration detection is a valuable exploration direction. In order to explore the difference of weak echo characteristics of human body detected by radar in complex sheltered scenes, a self-built ultra-wideband radar life detection system was used to carry out radar detection experiments in different scenes such as air medium, mine buildings and ground buildings. The anchor net support wall, concrete wall, block wall and composite laminated structure were used as shelters. The original radar signal data set was established by obtaining the micro-motion signal of human arm. Based on the principle of Doppler effect and multi-pulse accumulation method, a weak echo feature recognition framework of human life information based on frequency domain micro-variation information was constructed to realize the feature separation of human weak echo signal. Systematically characterize and compare the characteristics evolution mechanism of standardized human motion signals after crossing different media. The results showed that the framework can effectively extract and quantify the weak echo signal, and the shielding medium has obvious influence on the recognition difficulty and image clarity. The time-distance map positioning accuracy of the life weak echo signal is ± 0.5 m. The detection difficulty of the ground scene is generally better than that of the underground scene, and the harmonic intensity decreases with the increase of the order. It is clear that the micro-motion frequency range of the arm detected in different shielding environments is 5.8-6.8 Hz. Compared with the unshielded environment, the positive drift of 3.45%-17.24% is presented. The echo intensity shows an overall attenuation trend, with a large attenuation range of 2.08%-76.67%. According to the specific attenuation, the energy follows the order of unshielded > laminated structure > block wall > concrete wall > anchor net support wall. The attenuation of electromagnetic waves in complex shadowing scenes is the coupling result of multiple physical mechanisms. The attenuation is mainly manifested in amplitude reduction and energy loss. The mechanism can be roughly summarized into three categories: reflection loss, absorption loss and scattering loss. The feature evolution mechanism of standard action signals in different complex masking scenes is revealed. The experimental results lay a theoretical and experimental foundation for the establishment of life weak echo correction database and rescue action instruction set for mine and building collapse rescue.