The Classification And Application Of Remote Sensing Image Based On GA-PSO Optimization And HDT-SVM

The quality of remote sensing image classification is directly related to the extraction of imageinformation. Because of its excellent classification performance, SVM has been gradually occupiedan important position in the remote sensing image classification.This paper mainly studies remote sensing image classification based on hierarchical decisiontree support vector machine (HDT-SVM). The main work of this paper is as follows:1. Proposed a feature extraction method based on Kmeans. In remote sensing image featureextraction, we analysed the classification results based on supervised classification and visualInterpret. The results showed that the feature extraction based on pre-Kmeans classification methodcan improve the accuracy of extracted features, thus improving the accuracy of remote sensingclassification.2. Proposed a particle swarm optimization algorithm based on genetic algorithm (GA-PSO) tooptimize the mixed kernel function. When optimizing the parameters, we used PSO which includesthe crossover and mutation characteristics of genetic algorithm. Experiments showed that thismethod can optimize the kernel parameters more effectively and improve the accuracy of remotesensing image classification.3. Proposed the hierarchical decision tree SVM (HDT-SVM) classification based on optimalfeature weighting combination. Normal tree and skewness tree has its own advantages, and theHDT-SVM Learned the advantages of both. The experiment proved the HDT-SVM with optimalfeature weighting combination can improve the classification accuracy.4. I used the landsat7EM+remote sensing image of Nanjing to do the application research.Firstly, I did some preprocessing for the remote sensing image; Then extracted the feature of theimage based on kmeans. and then used GA-PSO to optimize the mixed kernel parametres; Finally,HDT–SVM based on optimal feature weighting combination was used to classify the remotesensing image. Experimental results showed that, the comparison with the urban land utilizationratio, the classification accuracy is very similar.