基于多分类器融合的高光谱图像分类研究

Hyperspectral image classification based on multi classifier fusion

针对由于高光谱图像存在数据量大、数据相关性强、图谱合一等特点导致高光谱图像分类难度较大的问题,构建一种基于多分类器融合的高光谱图像分类模型.该模型首先使用双边滤波算法进行去噪处理,然后使用LDA算法与PCA算法相结合、单独PCA算法、Gabor滤波与PCA算法相结合三种方式分别对数据进行降维与特征提取,并分别使用SVM分类器、LightGBM分类器与AdaBoost分类器进行分类.最后设计一种AHP-投票法,将三个分类器的分类结果进行融合.结果表明,模型融合后效果显著提高,总体精度(OA)可达97.59%,平均精度(AA)可达98.95%以上,Kappa系数可达97.32%以上,OA、AA、Kappa系数比单个模型分类器平均提高2.30%、1.13%、2.54%.

Aiming at the problem of huge amount of data,strong data correlation,and integration of maps and spectrum in hyperspectral images,which might cause difficulty to classify hyperspectral images,a hyperspectral image classification model based on multi-classifier fusion is constructed.The model first uses bilateral filtering algorithm for denoising,and then uses the combination of LDA algorithm and PCA algorithm,separate PCA algorithm,and Gabor filter and PCA algorithm to perform dimensionality reduction and feature extraction on the data respectively,and use SVM classifier,LightGBM classifier and AdaBoost classifier for classification.Finally,an AHP-voting method is designed to merge the classification results of the three classifiers.The results show that the effect of the fusion model is significantly increased,the overall accuracy(OA)up to 97.59%,the average accuracy(AA)up to 98.95%,Kappa coefficient can reach more than 97.32%,the OA,AA,and Kappa coefficients are improved by 2.30%,1.13%,and 2.54%on average compared with a single model classifier.