Remote Sensing Image Object Recognition Based On Sparse Dictionary Learning

Remote sensing images are playing an increasingly important role in urban planning and construction,road traffic,and other aspects.However,traditional remote sensing image processing techniques are insufficient to support the classification and recognition of complex image information due to the complexity and openness of the surface structure system.With the rapid development of technology and the emergence of high-performance sensors,it has become particularly important to select appropriate methods for image classification and recognition.Sparse dictionary learning,as an emerging data processing technology,can deeply explore the intrinsic features of images and has enormous application potential in solving image classification and recognition problems.To address the issues of complex features of remote sensing images that are difficult to classify and recognize,and poor classification performance of classifiers when the data volume is large,the following remote sensing image recognition methods are constructed by combining machine learning and deep learning models such as sparse dictionary learning,convolutional neural networks,and attention mechanisms:（1）Combining the idea of maximizing the extraction of favorable information from images through image fusion and synthesizing high-quality images,a sparse dictionary learning model-based image reconstruction algorithm was constructed,and the reconstructed images were classified using a classifier.The joint dictionary matrix and the sparse matrix are calculated based on sparse representation and dictionary learning.In order to maintain the stability and uniqueness of dictionary matrix and sparse matrix,the joint application of7)₁norm and7)₂ norm approach.Based on UC Merced Land Use,Sentinel-2 surface images,and RSI-CB remote sensing image datasets,experimental analysis was conducted.The experimental results show that the constructed model can effectively classify and recognize remote sensing images.The constructed model solves the problem of difficult classification of remote sensing images due to small inter group differences.（2）Combining the idea of image feature extraction,attention mechanism method was selected to extract features from remote sensing images based on sparse dictionary reconstruction,and the extracted features were classified using support vector machines.After adding the attention mechanism,more categories were not missed in the evaluation index of misclassification error when classifying remote sensing images based on sparse dictionaries.Additionally,more categories were correctly classified in the mapping accuracy evaluation index.This indicates that adding an attention mechanism is more conducive to the classification of reconstructed remote sensing images.（3）To address the issue of difficulty in classifying classifiers when there is a large amount of data,a convolutional neural network was selected to extract features from remote sensing images based on sparse dictionary reconstruction.The extracted features were classified through the network,and the classification accuracy of the constructed model was obtained through experiments.The three models were validated on the vehicle datasets of three intersections in Chongqing and four intersections in Changchun,China.The validation results showed that when classifying the vehicle dataset,the classification accuracy of the reconstructed images from the sparse dictionary model based on convolutional neural networks was the highest.The classification accuracy on the vehicle datasets of three intersections in Chongqing and four intersections in Changchun reached 90%and 92%,respectively.The recognition of vehicle images has achieved excellent results,providing support for traffic scene monitoring at intersections.