Abstract: In order to solve the problem of precise positioning, semantic association and type recognition of the small-scale and complex multi-target power equipment in digital twin 3D inspection, a complex space fusion model of two-dimensional plane vision, three primary colors and depth dimension (RGB-D) is constructed. By extracting the pixel gray value features of the multiple elements such as the three primary colors and the texture of the two-dimensional image, and the depth of the binocular three-dimensional image in the dense power scenario, the three characteristic elements of the small-scaled, occlusive and tilt equipment are characterized in the RGB-D complex space. The linear superposition and fusion of elements is adopted to realize the expression of the spatial structure of the target body, and the normalization and equalization of complex spatial elements is used to enhance the depth characteristics of the target body, so as to achieve the purpose of spatial positioning and quantification of the small-scaled, complex multi-target bodies. On this basis, with the optimized non-maximum suppression (NMS) algorithm, the target location mapping results are optimized and screened. After robust analysis, the method in this paper has strong anti-occlusive robustness. The experimental results of the disconnecting switches, circuit breakers and other equipment show that: Compared with the traditional detection methods, the proposed method here significantly improves the accuracy of target positioning and recognizing. Also, this method has the ability to deal with the complex occlusion of the equipment in dense power scenarios. These advantages of the method can effectively promote the development of the theory and technology of power digital twinning.