Multispectral image classification method based on adaptive feature fusion residual network

Abstract

The invention discloses a multispectral image classification method based on an adaptive feature fusion residual network, which mainly solves the problems of low universality and insufficient utilization of multi-level features in the prior art. The multispectral image classification method comprises the specific steps of (1) inputting a multispectral image; (2) performing normalization processingon the multispectral image; (3) selecting training samples and testing samples; (4) generating a training data set; (5) constructing a basic residual network; (6) constructing an adaptive feature fusion network; (7) generating an adaptive feature fusion residual network; (8) training the adaptive feature fusion residual network; (9) generating a testing data set; and (10) performing classification on the testing data set. The multispectral image classification method can adaptively fuse the multi-level features, extracts features with better discrimination and richer semantic information andhas the advantages of simple training and testing process and sufficient feature utilization.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a multispectral remote sensing image classification method based on an adaptive feature fusion residual network in the technical field of image classification. The invention can be used to classify ground objects in multispectral remote sensing images. Background technique [0002] In the field of image processing technology, the deep learning method shows strong feature representation ability, which reduces the uncertainty of artificially designed feature extraction. The deep learning method mainly builds a deep model, uses the model to extract the deep features of the multispectral remote sensing image, and uses the features to classify the data. However, due to the large number of network layers currently used, the extracted features are not very good. to fit the characteristics of multispectral images. [0003] Beijing University of Aeronautics and Astronauti...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that the spectral and texture features used in the extraction of ground object features are artificially designed for the experimental data, and the universality is not high, and the training and testing process includes multi-level block processing of the image 1. Extracting spectral and texture features, classifying image blocks, and processing edge areas. The training and testing process is relatively complicated, which affects the efficiency of multispectral image classification.

The disadvantage of this method is that when using the deep residual network to extract the features of the two data sets, the semantic information contained in the low-level features is ignored, the fusion between the low-level features and the high-level features is not considered, and the depth is not fully utilized. The multi-level features in the residual network, the discriminative and robustness of the extracted features are not good, which affects the classification accuracy

Images