Unmanned aerial vehicle image building roof extraction method based on full convolutional neural network

Abstract

The invention discloses an unmanned aerial vehicle image building roof extraction method based on a full convolutional neural network. The method comprises the following steps: in a first part, establishing an aerial image building roof sample library; in the second part, designing a full convolutional neural network to carry out feature learning on a building roof sample; performing building roofdetection by using the trained network, and obtaining a more accurate building roof result through post-processing of an extraction result in a third part. The method is different from a traditionalextraction method, and makes full use of rich unmanned aerial vehicle image resources in the aspect of data acquisition. In an algorithm design aspect, a specific full convolutional neural network based on skip layer connection is designed, gradient diffusion and gradient explosion are prevented while building roof features are fully extracted. In the aspect of post-processing, a conditional random field and a D-S evidence theory are utilized to carry out building roof extraction result post-processing, and the extraction precision of the unmanned aerial vehicle image building roof is improvedthrough post-processing.

Description

technical field [0001] The invention relates to the technical field of UAV image processing, in particular to a method for extracting building roofs from UAV images based on a fully convolutional neural network. Background technique [0002] With the advancement of my country's urbanization process and the rapid development of economic construction, the automatic extraction of buildings has become more and more important for the public and various industries. The rapid extraction and update of building elements has become the basic geographic information construction of our country. a very important content. At present, the comprehensive internal adjustment based on high-resolution remote sensing images is the main means of updating basic geographic information elements. Compared with remote sensing images for building interior judgment, UAV image acquisition is less difficult, data production is more flexible, and is less restricted by external conditions. The advantages of...

AI Technical Summary

Problems solved by technology

Although the current number-driven method has better results in comparison, it still does not make full use of the building features and has poor robustness.

Based on the model-driven method, there are various types of buildings, and the establishment of the target model relies heavily on prior knowledge. At present, only part of the extraction problem is solved in a limited environment, and it is difficult to find a universal model to describe

Images