Heterogeneous convolutional network remote sensing image land utilization parallel classification method based on ground feature dependency relationship

The invention discloses a semantic-space fusion heterogeneous convolutional neural network remote sensing image land utilization classification method based on a ground feature dependency relationship. According to the method, semantic features of different levels in images captured by a deep convolutional neural network are designed. Then, graph construction is carried out by using a super-pixel segmentation algorithm, a weighted adjacent matrix is constructed by measuring the degree of dependence between adjacent objects, then spatial information is extracted by using a graph convolutional network, and finally, a multi-head attention mechanism is used to identify the multi-head attention of the objects. And weighting and fusing the output features of the two branches according to the contribution of semantic and spatial features of each stage in the classification process, thereby realizing accurate classification of land utilization. The provided classification method is verified on the basis of experiment