Vibration mode shape based damage detection is a research hotspot in the structure health monitoring field. However, the existing methods are commonly suffered from several limitations such as low immunity to noises and insensitivity to the weak damage. This study aims to develop a new damage detection method for plate-like structures to overcome the limitations. First, multi-scale mode space concept is put forward based on the 2-D Gaussian wavelet transform in view of the different scale characteristics of the noise, trends and damage information in mode shape signal. By this way the components of the mode shape can be independently expressed, and the damage information is highlighted. And then, complexity of a point is defined and by means of the sliding window continuous scanning, multi-resolution complex spectrum is formed in the multi-scale mode. Multi-resolution complexity spectrum can make the damage more apparent and can reveal the damage location, size, shape and damage quantification for plate-like structures in noisy conditions. Numerical simulation of an actual engineering hydraulic gate and two model tests verify the correctness and effectiveness of the proposed theory and method in this paper.