Traveling ionospheric disturbances (TIDs) may significantly change the ionospheric properties in the region, which in turn affects the radio propagation process, especially on short-wave communication, satellite navigation and positioning. TIDs detection and propagation parameter calculation are essential foundations for ionospheric disturbance monitoring and early warning. Comparisons between Cascade Mask R-CNN and the classical Mask R-CNN models in instance segmentation results were conducted using global LSTIDs and European MSTIDs data. The results indicate that Cascade Mask R-CNN outperforms Mask R-CNN in image processing accuracy and training convergence speed, with an improvement of approxi- mately 4.7% in bounding box precision and about 3.6% in mask accuracy. The model achieved mask accuracies of 79.34% and 73.37% in the European region and globally, respectively. Subsequently, irregular disturbances were normalized using a least squares ellipse fitting method, and isolated disturbances were filtered and eliminated using filtering criteria and a nonlinear programming solver. When the filtering threshold T1 was set to 40, isolated disturbances could be effectively fil- tered out while retaining wave disturbance components. The method yielded TIDs propagation parameters in DTEC maps in different regions that closely matched actual results.