Long-tail image recognition method based on causal calibration

The invention relates to a long-tail image recognition method based on causal calibration. The method comprises the following steps: obtaining and preprocessing a long-tail distribution image data set; constructing a long-tail image recognition model based on causal calibration on the basis of the deep learning neural network model; training a long-tail image recognition model based on causal calibration; obtaining a to-be-recognized long-tail image and performing preprocessing; and inputting a to-be-recognized long-tail image into the trained long-tail image recognition model based on causal calibration, inputting a finally output logic calibration value into the Softmax classifier, and finally obtaining a long-tail image recognition result. According to the method, the training and reasoning process of the model can be effectively intervened, the recognition accuracy of the model for each category is improved, and the generalization ability of the long-tail image recognition model based on causal calibratio