A wall-grinding robot's wall perception model for wall defect classification

The quality of wall surfaces is a core indicator in architectural decoration, and the detection of their defects directly affects construction efficiency and building quality. However, traditional detection methods perform poorly in complex lighting conditions on wall surfaces. This paper proposes a wall-grinding robot's wall perception model, which comprehensively utilizes the advantages of the DeFectSenseConv module, DHA module, and DSDF module. Firstly, the SFConv module is employed to extract multi-scale spatial features of wall defects; subsequently, the DualSpectra HyperAggregator module is used to further enhance the features of wall defects captured by the SFConv module; finally, through the attention mechanism of the DSDF module, precise identification of wall defects in complex environments is achieved. Experimental results demonstrate that the model exhibits high detection accuracy and robustness under strong light interference and complex texture backgrounds.