Probability least squares support vector machine with L1 norm for remote sensing image retrieval

摘要

This paper proposes a probability least squares support vector machine (PLSSVM) classification method that remotely senses image data, like high-dimension, nonlinearity, and massive unlabelled samples. Hybrid entropy was designed by combining quasi-entropy with entropy difference, which was then used to select the most 'valuable' samples from a larger set to be labelled. An L1 norm distance measurement was then used to further select and remove outliers and redundant data. Finally, based on the originally labelled samples and the screened samples, the PLSSVM method was implemented through training, and it is also more efficient in than the tradition SVM in both accuracy and speed. The experimental results of the classification of ROSIS hyperspectral remote sensing images show that the overall accuracy and Kappa coefficient of the proposed classification method are more accurate than existing methods. The proposed method obtains a higher classification accuracy with fewer training samples, allowing it to be applicable to current problems of classification.