After billions of years of evolution, creatures in nature have formed complicated, elaborate and reasonable structures, which enable lighter materials to have great properties of high strength and tenacity, resistance to bend, reverse and buckling etc. And these structures have properties which beyond the synthetic counterparts a lot. While traditional bionics structure design simulates them in the macro perspective, the newly emerging micro-3D-imaging technology enables researchers explore inside of materials. The new method offers a novel way to break the bottleneck of traditional bionics design. For example, traditionally methods like FEM are used to analyze and optimize the simple ordered porous structure, but for complicated disordered porous structures, they do not have too much more application. Here we use bamboo and aspen samples as examples, try to obtain the correlation between the porous structure and their mechanical properties of them, and intend to direct the structural optimization. Firstly, we got the X-Ray CT images of bamboo and aspen samples and reconstructed their 3D porous models; secondly, based on the porous media theory we extracted pore parameters, such as pore-size distributions, connectivity function curves, tortuosity and so on; Then with the help of 3D models, we simulated their mechanical properties and obtained the correlation of each pore structure parameter and mechanical properties, which were used to find the optimum porous parameters combination; Finally, we used these parameters to reconstruct a new 3D porous structure with machine learning method, and validated its mechanical properties by experiments and the FEM method. Results showed that the specifically designed porous structure has extraordinary resistant bend, reverse and buckle etc.