The invention discloses a curvelet filter and convolutional structure learning-based SAR image segmentation method, and mainly solves the problem of inaccurate SAR image segmentation in the prior art. The method is implemented by the steps of (1) performing SAR image sketching to obtain a sketch; (2) according to a regional chart of an SAR image, dividing pixel sub-spaces of the SAR image; (3) constructing a curvelet filter set; (4) building a convolutional structure learning model; (5) by adopting a curvelet filter and convolutional structure learning model-based SAR image segmentation method, segmenting pixel sub-spaces of a ground object adopting a mixed aggregation structure; (6) performing sketch line collection feature-based independent target segmentation; (7) performing visual semantic rule-based line target segmentation; (8) segmenting pixel sub-spaces of a homogeneous region by adopting a polynomial-based logic regression priori model; and (9) combining segmentation results to obtain an SAR image segmentation result. According to the method, good segmentation result of the SAR image is obtained; and the method can be used for semantic segmentation of the SAR image.