Method and system for directed transfer of cross-domain data based on high-resolution remote sensing images

A system for directed transfer of cross-domain data based on high-resolution remote sensing images, wherein the directed transfer system comprises:

• a module for establishing a training data sample set, configured to acquire labeled source-domain images and unlabeled target-domain images, to establish a training data sample set, wherein training data samples comprise source-domain images, true labels of the source-domain images, and target-domain images;
• and the source-domain images, the true labels of the source-domain images, and the target-domain images are respectively formed in a source-domain image set, a source-domain image label set, and a target-domain image set;
• a module for establishing an image translation network model, configured to establish an image translation network model for directed transfer of cross-domain data of high-resolution remote sensing images;
• an objective loss function establishing module, configured to establish an objective loss function which combines an image translation loss and a model adaptive loss of the image translation network model;
• a training module, configured to train the image translation network model by using the training data sample set, to obtain a trained image translation network model;
• a module for establishing a data sample set, configured to input the labeled source-domain images into the trained image translation network model to translate the labeled source-domain images into target-domain transferred images;
• by using labels of the labeled source-domain images as labels of the transferred images, establish a data sample set for fine tuning;
• and classify the data sample set for fine tuning into a training set and a test set;
• a fine tuning module, configured to fine-tune a semantic segmentation model of the trained image translation network model by using the training set such that the model adaptive loss is less than a threshold of the model adaptive loss, to obtain a fine-tuned image translation network model;
• an objective loss function calculation module, configured to input the test set into the fine-tuned image translation network model, and calculate a value of the objective loss function of the fine-tuned image translation network model;
• a first determining module, configured to determine whether the value of the objective loss function is less than a threshold of the objective loss function, to obtain a first determining result;
• a returning module, configured to:
• if the first determining result is no, return to the step of “training the image translation network model by using the training data sample set, to obtain a trained image translation network model”, to train and fine-tune the fine-tuned image translation network model once again;
• a model output module, configured to:
• if the first determining result is yes, output the fine-tuned image translation network model as the trained image translation network model;
• and
• a directed transfer module, configured to input the high-resolution remote sensing images to be processed in the source domain into the trained image translation network model, to obtain target-domain translated images;
• and transfer the target-domain translated images in a directed manner.