Complex background building crack detection method based on machine vision

The invention discloses a complex background building crack detection method based on machine vision. The method comprises the following steps: acquiring an image by using a camera; taking the resnet101 network as a backbone network, and inputting the image into the backbone network for feature extraction; performing rough semantic segmentation on the middle-layer features extracted by the backbone network by adopting an OCR structure, and calculating K groups of vectors, namely object region representation, by combining a rough semantic segmentation result and the deepest-layer features extracted by the backbone network; calculating and obtaining a relation matrix of pixels and object region representation by combining the object region representation and the deepest layer feature extracted by the backbone network, and obtaining values of each pixel and object region representation; performing weighted summation according to the value and the object region representation to obtain a final object context feat