An image classification method and system

Abstract

The present application discloses an image classification method and system, wherein the method includes: inputting the image to be classified into a trained convolutional neural network to obtain the features corresponding to each pixel of the image, and each pixel corresponds to The feature of each pixel is regarded as the predicted probability that each pixel belongs to each preset category; the feature corresponding to each pixel and the original image of the image are processed by the conditional random field CRF model to adjust the predicted probability to obtain Image classification results. The invention can realize automatic, rapid and high-precision classification of high-resolution remote sensing images, and greatly improves classification efficiency.

Description

technical field [0001] The present application relates to the field of image processing, in particular to an image classification method and system. Background technique [0002] With the continuous use of high-resolution remote sensing satellites and micro-satellites, more and more massive high-definition remote sensing data have begun to attract the attention of various industries and fields, hoping to quickly and accurately extract the required information from them, turning it into a novel information acquisition way, and even generate huge commercial value. However, the main bottleneck restricting its realization is the automatic classification technology from data to information. [0003] The classification of remote sensing images can be mainly divided into pixel-oriented and object-oriented methods. The pixel-oriented method pays too much attention to the spectral information on each pixel, and does not consider enough the semantic information formed by the pixel i...

AI Technical Summary

Problems solved by technology

Obviously, the above methods have many disadvantages: 1) rely on superpixel segmentation results, and often need to choose different segmentation methods and thresholds to process different data; Corresponding knowledge: construct one or more effective classification features; 4) It is necessary to manually select samples to learn feature parameters, thresholds, etc.; 5) Poor universality: each time data processing has to repeat the tedious work of the whole process

Images