Semi-supervised iterative training method for remote sensing interpretation in complex scene

The invention discloses a semi-supervised iterative training method for remote sensing interpretation in a complex scene, and the method comprises the following steps: accumulating preliminary annotations in a semi-automatic collection or manual annotation mode in the early stage of model training, constructing a training sample set, carrying out the training, obtaining an annotation model, and expanding an interpretation model library; a labeling model is constructed in a self-training mode of cross pseudo label constraint, and non-label data is gradually introduced, so that more sample labels of representative scenes are obtained, and sample expansion and refinement are realized. According to the method, a model set in a large-area complex scene is obtained by utilizing a multi-modal network and based on an iterative training technology of semi-supervised learning such as self-training and cross pseudo-tag, and a model matching and integration technology is constructed based on a scene rule set and model se