Attention mechanism-embedded iterative aggregation neural network high-resolution remote sensing scene classification method

Abstract

The invention discloses an attention mechanism-embedded iterative aggregation neural network high-resolution remote sensing scene classification method. The method comprises the steps of firstly performing convolution, channel attention screening and fusion on a high-resolution remote sensing image by adopting an iterative aggregation module with attention to obtain bottom-layer features of the image; secondly, carrying out convolution on the obtained bottom-layer features, and then sending the bottom-layer features to a next iterative aggregation module with attention to extract middle-layerfeatures of the image; carrying out convolution on the obtained middle-layer features, and then sending the middle-layer features to the last iterative aggregation module with attention to extract high-layer features of the image; finally, classifying the remote sensing scene images by the feature map through a pooling layer and a full connection layer. According to the method disclosed in the invention, the remote sensing image is subjected to feature extraction and fusion by using the iterative aggregation module with attention, the extracted features cover relatively strong semantic information, and meanwhile, a structure with an attention mechanism is embedded in the module, so that channels with useful information can be effectively screened out for fusion, and the recognition capability and classification performance of a classifier are improved.

Description

technical field [0001] The invention belongs to the field of image processing, in particular to an iterative aggregation neural network high-resolution remote sensing scene classification method embedded in an attention mechanism. Background technique [0002] Scene classification of high-resolution remote sensing images is based on the contents of remote sensing images, which are labeled with a set of semantic categories. In recent years, scene classification of high-resolution remote sensing images has become more and more popular due to its wide application in natural disaster detection, land cover and land use classification, geospatial object detection, geographic image retrieval, urban planning, environmental monitoring, etc. important. Among earlier works, hand-crafted features are most widely used in this work and have been intensively studied, such as color histogram, scale-invariant feature transform (SIFT), and histogram of oriented gradients (HOG). These method...

AI Technical Summary

Problems solved by technology

[0007] (1) The feature extraction process is complex: the traditional classification method adopts the idea of ​​​​divide and conquer, and it is necessary to artificially design an image feature extraction algorithm to extract various types of features of the image. The feature extraction process algorithm is complex, and the extracted features may contain information. Lost redundancy and other phenomena, resulting in low classification accuracy;

[0008] (2) The feature expression ability is not strong: the traditional remote sensing image scene classification method usually only uses one or two types of features as the input of the classifier. When the scene image has distortion, blur, and different categories, the feature expression ability is affected. limit, thus weakening the classification performance

[0009] (3) The selection of classifier parameters is difficult: commonly used image classifiers, such as SVM, KNN (K-Nearest Neighbor) and other parameter settings have a great impact on the performance of classification. In order to obtain better classification results, a lot of manual participation is required. The selection of the optimal parameters leads to poor versatility of the algorithm

Images