A few shot semantic segmentation network for remote sensing images

Aiming at the problem of high cost of labeling samples of remote sensing images, a few-shot semantic segmentation method for remote sensing images based on migration learning is proposed. The method is based on the strategy of model fine-tuning, firstly, the deep residual network model is pre-trained on a large-scale image dataset, and then it is migrated to the remote sensing image dataset for model fine-tuning, and the criss-cross attention module (CCA) is introduced after the feature extraction network, in order to reduce the computational cost while obtaining rich contextual information; to address the issue of imbalanced sample size among different classes in the small sample, a weighted cross-entropy loss function is proposed to optimize the model.