Uncertainty estimation method of remote sensing image building recognition model

Abstract

The invention discloses an uncertainty estimation method of a remote sensing image building recognition model. A building recognition uncertainty estimation method considering semantic and shape features is provided on the basis of a Bayesian approximate reasoning theory, the core idea is that CNN approximate parameter distribution is obtained through a Bayesian approximate reasoning method, and then the method is applied to a building segmentation model and a shape optimization model to realize quantitative self-evaluation on the reliability of a building identification result from two aspects of pixel segmentation and vector generation, namely building segmentation uncertainty estimation based on an MC-dropout method and building vector optimization and uncertainty estimation based on shape modeling. According to the method, reliability quantitative self-evaluation is performed on the recognition result from two aspects of semantics and shape features, so that a user is helped to establish a result acquisition standard, and the method has important significance in promoting deep application of a deep learning technology in mapping practice and promoting intelligent development of related industries.

Description

technical field [0001] The invention relates to the field of surveying and mapping science and technology, in particular to an uncertainty estimation method of a remote sensing image building recognition model. Background technique [0002] At present, the deep learning model still has limitations in the recognition of buildings in high-resolution remote sensing images. It is foreseeable that the building recognition algorithm based on deep learning will still face a certain accuracy bottleneck for a long time, which means that the results automatically generated by machines Must be manually QA and revised before being submitted as a product. The workload of this quality inspection is very difficult, in large part because the inspectors have no way of knowing the reliability of the automatically generated results, and all buildings must be inspected one by one. If the algorithm itself can quantitatively describe the reliability when predicting the results, that is, how "con...

AI Technical Summary

Problems solved by technology

However, in general, the uncertainty estimation of deep learning models has not received sufficient attention in the field of remote sensing, and such research on building recognition is still rare, especially the lack of self-assessment of the reliability of building recognition results at the vector shape level. Related research

Images