Label-free pathological image feature extraction method based on spatial position information

The invention discloses a label-free pathological image feature extraction method based on spatial position information, which does not need to manually label lesions, constructs a data set and labels by means of a self-supervised algorithm and by using information such as spatial positions and zoom magnifications of different pathological image blocks, and is used for extracting pathological image features. And a correlation model is trained, so that extraction of pathological image features and subsequent tasks are completed. Compared with an existing deep learning algorithm based on supervision and weak supervision, the method is easier to implement, on one hand, similar difference data pairs are automatically constructed by utilizing a self-supervised learning mode and spatial position information, manual annotation is not needed, the data set obtaining difficulty is greatly reduced, and a large amount of cost is saved; and on the other hand, a novel training framework is constructed, the similarity diffe