Method of Fusion of Binary Classification Semantic Segmentation Map into Multi-Classification Semantic Map Based on High-resolution Remote Sensing Image

Abstract

The invention provides a method for fusing binary classification semantic segmentation maps into multi-classification semantic maps based on high-resolution remote sensing images. The image data set is sent to the binary classification semantic segmentation network to train binary classification model groups, and the confidence degree of each binary classification model is calculated. ; Input the high-resolution remote sensing image to be predicted into the binary classification model group, and obtain the binary classification results output by each binary classification model and the initial prediction probability matrix; multiply the confidence of the binary classification model and the corresponding initial prediction probability matrix to The final discriminant probability matrix of the high-resolution remote sensing image to be predicted in the binary classification model; when classifying the pixels according to the binary classification results, for the pixel points with classification conflicts, the final discriminant probability of the pixel points in the binary classification model group The category with the largest probability value corresponding to the position of the matrix is ​​used as the classification category of the current pixel point. The invention can effectively improve the overall classification accuracy and IoU and kappa coefficient indexes of the final composite classification map.

Description

technical field [0001] The invention relates to computer image processing and high-resolution No. 2 satellite remote sensing image classification technology, in particular to a technology for fusing binary classification semantic segmentation prediction maps generated by deep learning semantic segmentation into high-resolution remote sensing images into multi-class semantic maps. Background technique [0002] High-resolution remote sensing image interpretation is widely used in land classification, land survey, regional investigation, ground feature change detection and other fields. In the early days, it mainly relied on interpretation professionals for manual interpretation, which required a lot of manpower and material resources. At present, the traditional method is mainly used to directly extract ground feature information directly according to the texture features and multi-dimensional information of the image. [0003] For example, the normalized difference vegetation...

AI Technical Summary

Problems solved by technology

In the early days, it mainly relied on interpretation professionals for manual interpretation, which required a lot of manpower and material resources. At present, the traditional method is mainly used to directly extract ground object information directly according to the texture characteristics and multi-dimensional information of the image.

Images