基于可见光与红外数据融合的地形分类

针对单传感器地形分类效果不佳的问题，提出一种基于可见光与红外数据融合的地形分类方法。分别对可见光图像与红外图像提取特征，使用最近邻分类器和最小距离分类器进行后验概率估计，将来自不同特征、不同分类器的后验概率加权组合，通过散度计算得到特征的权重，实验确定分类器的权重，并在最小距离的后验概率估计中，使用马氏距离代替欧氏距离。实验结果表明，该方法对水泥路和沙子路的识别率分别达到99.33%和96.67%，均高于同类方法。

Terrain Classification Based on Visible Light and Infrared Data Fusion

Aiming at the bad performance of terrain classification based on one sensor, a terrain classification method based on visible light and infrared images is proposed. Extracting features from visible and infrared images, nearest neighbor classifier and minimum distance classifier are adopted to estimate the posterior probabilities. The probabilities from different features and classifiers are weighted composed. The features’ weights are computed by scatter while the classifiers’ weights are obtained by experiments. In the estimate of posterior probabilities based on minimum distance classifier, Mahalanobis distance is used instead of Euclidean distance. Experimental results show that the recognition accuracy of cement road and gravel road are 99.33% and 96.67%, which are higher than similar algorithms.