针对训练样本不足情况下的煤岩图像识别问题,提出了一种局部约束的自学习(LCSL)煤岩识别方法。该方法首先从辅助数据中通过局部约束的字典优化模型获取高层结构特征,这些辅助数据是无标签的非煤岩自然图像,与煤岩图像的特征分布不同,且更容易获取;然后利用学习的高层结构特征结合局部约束线性编码提取煤岩图像特征;最后利用SVM算法对煤岩图像进行分类识别。实验表明:通过该方法得到的特征能够有效地表征煤岩图像,具有很强的鉴别性和鲁棒性,达到了很好地识别效果,相比于原有煤岩识别方法平均识别率提高了1%～3%。

In terms of the coal-rock image recognition problems in case of insufficient training samples,a new method of locality-constrained self-taught learning for coal-rock recognition was proposed. This method first obtains the diction- ary matrix from the unlabeled random images. These training samples are unlabeled non-coal-rock images from a com- mon data set,which are different from coal-rock images. Then the method uses the local linear constraint coding meth- od to get spare coding of coal-rock image samples,which are used as a feature of coal-rock image. Finally,the SVM classification algorithm is adopted to identify coal-rock images. Experiments show that the image high-level visual structures can be obtained from a large number of unlabeled random image samples through self-taught learning local constraint linear coding method,and the image features can be expressed concisely and effectively with these struc- tures. This algorithm has a wide range of applicability. Compared with the original coding algorithm, the proposed method is more efficient,and the average coal-rock image recognition accuracy increases by 1% -3% .