The traumatic penumbra refers to a repairable region surrounding the central area of brain trauma. The efficient and precise segmentation of this region is crucial for the effective treatment of brain traumatic and thus reducing the rate of disability. In this paper, we propose an improved UNet++ model, which can be used to segment the traumatic penumbra in rats quickly and accurately by substituting the convolution module of the original model with the Ghost convolution and using some inexpensive operations to obtain more feature images. This model segmented the traumatic penumbra region with improved Intersection over Union (IOU) and Differential Scanning Calorimetry (DSC) coefficients, with the IOU coefficient increasing to 0.6 and the DSC coefficient increasing to 0.75, verifying the validity of the method.