Research on natural disaster target change detection method based on deep learning

In recent years, with the increasing global climate problems, our country has been relying on remote sensing technology for early warning and post-disaster assessment in the field of disaster prevention and control for some time. However, there are many kinds of natural disasters, especially sudden natural disasters, such as earthquakes, floods, typhoons and mudslides, which are often too late to prevent, and there is also a strong recurrence between disasters, which is very likely to cause massive data interference to remote sensing recognition. Compared with the traditional convolutional neural network-based remote sensing image recognition processing, this project proposes to use the Faster-RCNN target detection algorithm, which adds the Region Proposal Networks (RPN) module to the traditional Fast CNN to generate the algorithm, making the target change detection speed and accuracy greatly improved. The change detection algorithm is based on the traditional Fast CNN. This change detection model has the natural advantage of processing remote sensing data, which can automatically and multi-levelly extract the features of complex objects, learn to discover the geometric and spatial relationships of feature targets in images from massive data, and solve complex classification or regression problems through multi-layer nonlinear networks, thus significantly improving the accuracy of remote sensing image interpretation. After the model design is completed, in order to improve the detection accuracy of the model, it is proposed to train the Faster-RCNN change detection model by acquiring a large number of original samples of remote sensing images with manually expanded samples.