Extraction And Representation Of Cyclone Image Feature And Typhoon Center Location

With the increasing application of Doppler radar as an effective tool of detecting severe convective weather, domestic and overseas researchers found that the mesocyclone has a close relationship with catastrophic weathers, such as storm, strong wind, hail, and so on. Therefore, the information of mesocyclone which is provided by Doppler radar has an obvious advantage in pre-alarming of the severe weather and tornado accurately and in time.Of all natural disasters, typhoon poses the most dangerous threat to life and property. It is very important to recognize typhoon and locate the center through satellite image. But the work is still in research due to the complex nature of cloud shape and structure of typhoon.The paper mainly includes these parts:1.extraction and representation of spiral band;.2. center location for developing typhoon;3. the improvements of mesocyclone recognition .First of all, spiral band extraction is performed on the basis of cloud image segmentation.. The paper stresses on self adapt threshold segmentation to radar reflection image combined with direction multiscale growth method. The outcome of spiral band extraction is very valid.A principal curve is used to represent the shape feature of every region. Then Hough transform is applied to these curves to detect a logarithm spiral equation which can match the curves well. To achieve the goal, coordinates transform is employed to reduce the parameter dimension of Hough transform. After the spiral is obtained, the typhoon center is located on the center of the spiral.As to mesocyclone recognition, the valid segmentation method to reflection image is improved. Rankine pattern is applied to filter false velocity pair, furthermore three people's voting method is put forward and used. The accuracy and the calculation speed of mesocyclone recognition are enhanced.Based on theories of image processing and pattern recognition, and with expert's experience, all algorithms are coded with C++ language. The paper optimizes the algorithm with OpenCV. It is stable and effective after testing.