Prediction of Dust Storm Direction from Satellite Images by Utilized Deep Learning Neural Network

A dust storm is a bad weather phenomenon described by dust-filled air and strong winds across an extensive area. It is the most significant weather phenomenon in the Iraq region, which can cause massive damages both in natural environments and human lives. This work is aimed at predication dust storms direction from satellite images by using a convolutional neural network. The proposed method is different from most previous approaches, that in our method we utilized image processing to detect storm and find the moving direction of dust storm through prediction the minimal bounding rectangle of the dust storm and determine the orientation angle. The proposed method principles are depending on a series steps that employed desired algorithms to be able to predict dust storm direction, which are: Segmentation by used Feature Pyramid Network (FPN), Registration and Classification by used Region Proposal Networks Learning and location and Angle detection by used ground truth estimation algorithm. The proposed system has been testing by using real storm satellite images, in which the proposed method will detect storm from the first storm image and predict it is direction then compared with the real storm direction. FPN and ROI are used to recognized the storm area and then extracted the sand storm area. Next, the starts to identify the ground truth of the object to find both location and angle of direction. From the results, the proposed method successfully detects storm and predict its direction with good accuracy.