The invention provides a pulmonary nodule automatic segmentation method based on deep learning. The method is used for judging the type of a pulmonary nodule and performing system segmentation, and comprises the following steps: step 1, acquiring CT image data from LIDC, and intercepting a pulmonary nodule region in the CT image data as a picture with the size of 50 * 50 pixels as training picturedata; step 2, establishing a pulmonary nodule CNN classification model, inputting the training image data into the CNN classification model to optimize and train the CNN classification model, and storing the CNN classification model under the highest classification accuracy for use; step 3, establishing 3DU-net segmentation model of each type of pulmonary nodules, taking image data of pulmonary nodule areas of the same type are used as input image of the 3DUnet segmentation model to perform optimization training on the 3DUnet segmentation model, and storing the 3DU-net segmentation model under the highest segmentation accuracy rate; and step 4, inputting the pulmonary nodule region image to be segmented into a CNN classification model, and then completing pulmonary nodule region segmentation by the 3DU-net segmentation model.