Objective A space crack recognition method based on space domain and frequency domain feature fusion is proposed for the identification of cracks in concrete pile structures with complex backgrounds.Method In the spatial domain， local binary pattern （LBP） operator is used to extract texture information and edge information from the crack images. In the frequency domain， a local phase quantization （LPQ） feature extraction method is utilized to extract subtle features related to frequency-domain information.Result The LBP shape structural characteristics are fused with LPQ frequency characteristics to construct a convolutional neural network model， enabling multi-level identification and classification of cracks in complex regions while avoiding errors and accurately recognizing crack types.Conclusion Experimental results demonstrate that this method can effectively extract key features in fine and complex crack crossing regions， avoid errors， and accurately recognize various crack types.