基于Q-relief的图像特征选择算法

Image feature selection algorithm based on Q-relief

针对特征选择算法——relief在训练个别属性权值时的盲目性缺点,提出了一种基于自适应划分实例集的新算法——Q-relief,该算法改正了原算法属性选择时的盲目性缺点,选择出表达图像信息最优的特征子集来进行模式识别。将该算法应用于列车运行故障动态图像监测系统(TFDS)的故障识别,经实验验证,与其他算法相比,Q-relief算法明显提高了故障图像识别的准确率。

Image feature selection is the significant part in pattern recognition, image understanding and so on. The relief algorithm has a blind deficiency in training feature weight, Q-relief was a new algorithm which was based on dividing instance set in self-adapting. Q-relief was proposed to solve the blind selection problem in the original relief algorithm. The presented algorithm was applied in Trouble of Moving Freight Car Detection System （TFDS）. The classification results show that the Q- relief algorithm can improve the accuracy of recognition compared with other algorithms.