Hybrid VGG model remote sensing image land class data training set making method

Abstract

The invention discloses a hybrid VGG model remote sensing image land class data training set making method. The method comprises the steps: S1, obtaining a high-resolution remote sensing image; S2, reading high-resolution remote sensing images, and cutting and classifying the high-resolution remote sensing images according to land types; S3, performing resolution unification on the cutting classification result according to the resolution of 0.8 m; S4, normalizing the unified result, normalizing all the classified images into images with 224 * 224 pixels, and generating a local 224 * 224 screenshot of the images; S5, performing color processing on the normalized image; and S6, outputting the mixed type land class data set. Compared with a traditional data set making method, the method provided by the invention has better feature extraction and classification effects on the VGG data model.

Description

technical field [0001] The invention relates to remote sensing technology, in particular to a method for making a hybrid VGG model remote sensing image terrain data training set. Background technique [0002] At present, the classification of high-resolution remote sensing images is one of the most important and core technologies in the application of remote sensing technology. hotspot. At this stage, especially after the neural network-based classification model is applied to remote sensing images, the preparation of the training set of terrain data based on high-resolution remote sensing images has become an important issue. In particular, the feature extraction and data training based on the VGG model require special requirements for the data set, such as input processing requires a 224*224 data set. Therefore, it is an important task how to make a widely representative and characteristic land type data set. [0003] The classic data set production method is mainly to ...

AI Technical Summary

Problems solved by technology

This kind of data set is more common in ordinary data set training, but in the land type data, there is a relatively high resolution of high-resolution remote sensing, and different remote sensing images have different resolutions, so the different separated land type pictures contain A large number of features will be lost with scaling, which will result in less representative feature extraction during training

Images