A Remote Sensing Supervised Classification Method Based on Subclass Training Samples

Abstract

The invention discloses a remote sensing supervised classification method based on subclass training samples. It includes the following steps: S1: According to the land use map, automatically extract the large-category samples of the target feature images to be classified; S2: perform unsupervised classification on the large-category samples of the feature categories, and obtain the large categories of features subclass and subclass samples belonging to each subclass; S3: train the subclass samples of each subclass to obtain the subclass classification standard of each subclass; S4: treat the classification target according to the subclass classification standard of each subclass Supervised classification of ground object images to obtain sub-category classification results based on sub-category classification standards; S5: Merge the sub-category classification results belonging to the same ground feature category to obtain a classification result image based on ground feature categories. The image is the result of remote sensing supervised classification. The classification result of the present invention has high precision and complete classification result, does not require a large amount of manual intervention, and saves human resources.

Description

technical field [0001] The invention relates to the technical field of remote sensing, in particular to a remote sensing supervised classification method based on subclass training samples. Background technique [0002] The existing remote sensing supervised classification methods mark samples according to the classification target category, and then carry out training classification. Such sample label classification methods have the following defects: [0003] (1) For the samples marked according to the classification target category, due to the large differences in the characteristics of some ground objects, it is easy to cause discriminant bias when classifying some uniform types of ground object samples with different attribute characteristics; [0004] (2) Directly establish a large class sample classification library, and through training classification samples, it is easy to cause fragmented classification results; [0005] (3) Modifying the fragmentation and plaque ...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to overcome the technical problems of the existing remote sensing supervised classification method, such as the low accuracy of the classification results, the fragmentation of the classification results, and the need to spend a lot of manpower to modify, and provides a remote sensing supervised classification method based on subclass training samples. The classification results High precision, complete classification results, no need for a lot of manual intervention, saving human resources

Images