Polarization image identification and pixel level classification method

The invention belongs to the technical field of polarization imaging, image classification, pixel-level classification and deep learning, and particularly relates to a polarization image recognition and pixel-level classification method. According to the method, a polarized image and a non-polarized image can be distinguished, and five types of polarized images including an I polarized image, a Q polarized image, a U polarized image, a DOP polarized image and an AOP polarized image with different polarization characteristics can be further identified. Firstly, due to the fact that a Q polarization image and a U polarization image are similar in feature, the Q polarization image and the U polarization image are classified into a QU polarization image, and I, QU, DOP, AOP and unbiased image five-classification training is carried out through a polarization image five-classification ResNet50 residual network model. Secondly, image sharpening processing is carried out on the QU polarization image, the polarizatio