Center Location Of Typhoons In SAR Images Based On Visual Saliency And Feature Learning

Typhoon is a kind of disastrous weather system.It brings strong wind,torrential rain and storm surge,causing huge loss of life and property.To strengthen the monitoring and forecasting of typhoons is an important means to reduce the disaster caused by typhoons.The location of the typhoon center is a key factor for the prediction of typhoons.For its large coverage and high scanning frequency,in the past satellite cloud images were widely used in the study of the location of the typhoon center.However,due to the influence of cloud cover and rainfall,satellite cloud images can not show the interaction between the air and the sea water and the internal structure of the typhoon.Compared with visible light and infrared sensor,Synthetic Aperture Radar(SAR)works in the microwave band,has the penetrating cloud capability,can detect the actual rainfall regions,displays the internal structure of the typhoon and has very intuitive objective reflection for the spatial distribution of the spiral cloud bands and the center of strong rainfall.Besides,it can work all day and night under all kinds of weather with high spatial resolution.So it has been widely used in the study of the typhoon.The use of SAR in the monitoring and forecasting of typhoons will be one of the important research topics in the ocean and meteorological science in the future.Although there is a large number of research work about SAR image processing,it is still relatively lacking for the automatic extraction of the typhoon information using the SAR images.The work in this paper focuses on the center location of typhoons in the SAR images.We analysis the characteristics of the typhoon SAR images,combined with the theory and methods of SAR image segmentation and computer vision,and propose center location methods for both the typhoons with eyes and the typhoons without eyes.The main work and innovation of this paper are as follows:1,Usually the center of the typhoon with its eye is thought in the eye area or near the eye area,so the key of the location of the typhoon with its eye is extraction of the typhoon eye.The eye area in the typhoon SAR image appears as black or dark grey area for there being no rain and little wind in the eye area.There may be no complete and clear eye when a typhoon is in the development period or the recession period.The eye area in the typhoon SAR image may appears as light grey area at these periods.The gray level contrast is not always obvious.So it is necessary to enhance the gray level contrast before image segmentation.Besides,denoising the speckle noise is also necessary for the following image processing.We proposed a typhoon eye extraction method based on non-local means and labeled watershed algorithm in the second chapter.We firstly use PPB filter to denoise the speckle noise.Then we use top-hat transform to enhance the contrast.At last we extract the typhoon eye using labeled watershed algorithm.At last we calculate the eye area extracted with our method and the eye area extracted manually and compare them,showing the accuracy and the effectiveness of our method.2,The typhoon eye in the SAR image is shown as dark grey with unconspicuous orientation feature.The surrounding cloud bands are light grey with obvious orientation feature.The typhoon is developed on the ocean surface,and it will gradually decline after landing for the huge energy consumption and the lack of supply of energy.So the typhoon SAR images are usually imaged on the ocean surface,making the background of the images simple with little texture information.That is helpful for construct the saliency map of the typhoon SAR image,accentuating the structure of the typhoon.We proposed a typhoon eye extraction method based on visual saliency and morphology.The visual saliency map is constructed by the use of gray level information and direction information.Then the typhoon eye is extracted using winner-take-all and region growing algorithm on the saliency map.3,Generally speaking,the center of the typhoon without its eye is located with template matching method for a single image.In the past,the image segmentation methods were used to extract the spiral cloud band.The segmentation results are processed by by morphological method and select useful information based on statistical analysis.Then the skeleton lines of cloud bands are fitted by Hough transform and so on.The calculation complexity is high,and the result is not accurate and the procedure is complex.In the fourth chapter we propose a center localization method of the typhoon without its eye based on the particle swarm optimization algorithm and the logarithmic spiral model.We use canny edge detection method to extract the edge line with spiral geometry of the spiral cloud bands,which greatly simplifies the procedure and calculation complexity before template matching.Then fit the spiral line with logarithmic spiral model.The fitting problem is transformed into a multi-objective optimization problem,using particle swarm optimization algorithm to find the optimal solution,which can be seen as the estimation of the typhoon center.4,The spiral cloud band of the typhoon without its eye is the information can be fully used in the typhoon SAR image.Take the advantage of simple background with little texture information,in the fifth chapter we propose a typhoon center of the typhoon without its eye localization method based on feature learning and visual saliency detection.Spiral cloud bands appear as light and dark spiral structure in the typhoon SAR image,containing rich directional information.Therefore we can build salient region map taking advantage of the gray contrast feature and orientation feature.The salient region map makes the spiral cloud bands outstanding and the irrelevant clouds excluded.Then we use morphology method to extract the spiral bands in the salient region map.5,Limited by the radar scanning range and so on,not every typhoon SAR image contains the whole spiral structure.Besides,limited by its scanning range,sometimes the typhoon SAR image does not contain the complete structure of the typhoon.It may only contain part of the typhoon or the typhoon center is outside the image.These will limit the use of logarithmic spiral model.Therefore we propose a center of typhoon without its eye location method based on visual saliency and inflow angle model in the sixth chapter.We extract the spiral cloud band using the extraction method proposed in the fifth chapter,and extract the skeleton lines of spiral cloud bands.Then estimate the center with the inflow angle model and the particle swarm optimization algorithm.And the estimation results are compared with the Best Track Data,confirming the validity of the algorithm.