Loess plateau terrace automatic identification method based on deep transfer learning

The invention provides a loess plateau terrace automatic identification method based on deep transfer learning, and the method comprises the steps: firstly, taking a U-Net network as a basis, adding a Dropout layer and a BN layer, constructing an IEU-Net network in order to improve the training speed and robustness of the U-Net network, training the IEU-Net network through employing a WorldView-1 source domain data set, and obtaining a source model suitable for the terrace identification of a plurality of types of data sources; taking the source model as a pre-training model to obtain the weight of the pre-training model; then, a new convolutional layer and a Softmax layer are added on the basis of an IEU-Net network to design a migration network, and the obtained weight is migrated to a new network; and finally, training the migration network by using a GF-2 target domain small sample data set to realize accurate terrace identification. According to the method, the completion capability of deep transfer lear