Remote sensing image road information extraction method and device based on multi-task learning

The embodiment of the invention provides a remote sensing image road information extraction method and device based on multi-task learning, and is applied to the technical field of big data information extraction of digital traffic. The method comprises the following steps: acquiring remote sensing image data to be processed; calculating the remote sensing image data according to a random forest algorithm to obtain a road type; inputting the remote sensing image data into a pre-constructed remote sensing image road surface mask extraction model, outputting a road surface mask, and performing intersection extraction on the road surface mask to obtain intersection data; for the road surface mask, the cross point data and the road type, introducing a road specification as a constraint, and adopting a Bayesian decision to obtain a road level; and inputting the remote sensing image data into a pre-trained road PCI index extraction model, and outputting a road PCI index as a road integrity result. A plurality of se