Satellite image cloud detection method based on improved D-LinkNet

The invention discloses a satellite image cloud detection method based on an improved D-LinkNet. The method comprises the following steps: generating a binary image label by applying an artificial cloud hooking sample; secondly, improving an encoder of D-Link Net50, and replacing the original ResNet50-Block with the ResNeSt50-Block with a channel attention mechanism; besides, weighting on the loss function, performing test analysis to find that when only cross entropy loss is used as the loss function, the detection precision is higher; and finally using a conditional random field (CRF) for carrying out post-processing on a predicted result. Experimental results show that the mIoU of the improved D-LinkNet model on a test set is improved by 1.93%, the precision is improved by 2.45%, good cloud edge information is kept, the improved D-LinkNet model can be well used for cloud detection, and the effect of the improved D-LinkNet model is obviously higher than that of an original D-LinkNet model. 本发明公开了一种基于改进D-Lin