Video image segmentation stability improving method based on loss function

The invention discloses a target segmentation stability improvement method based on a loss function. The method comprises the following steps: S1, simulating a video sequence; S2, training the model; S3, performing fine adjustment on the pre-training model; and S4, introducing a stability loss adjustment model. According to the embodiment of the invention, a detector constructed for a deep learning scheme not only improves the precision, but also remarkably improves the consistency of the same target in continuous frames at the boundary under visual verification, reduces a large amount of wrong segmentation and hole missing segmentation, improves the segmentation stability of a static image training model under video data, and improves the segmentation efficiency. And the jitter problem of video segmentation is effectively relieved. The stability loss is introduced into the loss function, the segmentation precision is improved, and the optimization of the video data segmentation precision is realized. 本发明公开了基