The embodiment of the invention relates to a method and device for training an image recognition model and related equipment. The method comprises the following steps: extracting subgraph feature datafrom a detection frame subgraph of an input image; determining element feature data matching the subgraph feature data from an index element library, the index element library being pre-constructed based on a plurality of element feature data extracted from the plurality of alternative images. Further, the method may include outputting an image related to the element feature data as a training image to train the image recognition model. According to the technical scheme disclosed by the invention, enough labeled images with the similar detection frame contents can be quickly and accurately obtained by providing or automatically labeling a small amount of images, so that the working efficiency is remarkably improved.