中文摘要：

列车长时间运行后，车轮侧面出现许多裂纹，数量和尺寸达到一定标准后需进行更换，以保证安全。该标准定制的前提是对大量的车轮裂纹进行测量统计，而由人工进行这些工作需要大量的人力和物力，因此采用基于图像模式识别的自动检测能够节省大量的资源。文中主要研究列车车轮裂纹的检测定位算法，针对车轮表面裂纹的图像特征，提出一种基于LoG梯度加权Haar-like特征，能够有效地描述裂纹周围图像特性，从而更精确地进行裂纹描述，并提出改进的基于阈值限制LUT的Real—Adaboost机器学习算法训练裂纹检测分类器，进行快速精确的裂纹目标检测，实验部分验证了算法的有效性。

英文摘要：

There will be generated a lot of cracks on wheel surfaces after long time running of trains. When the amount and size of the cracks reach certain thresholds, the wheels should be replaced with new ones so as to secure train safety. In this paper, a new method was proposed to detect the cracks on wheel surfaces to save labor costs and avoid uncertainty of manual measurement. The LoG gradient based method was used to weight the pixels within the Haar-like feature mask region. With this, a more precise description of crack images was achieved. The Boosting based machine learning method, the Threshold Constrained Look Up Table based RealAdaboost, was presented for classifier training. The improved Real-Adaboost algorithm performs robustly with noise training samples and converges more rapidly. Finally, a cascaded classifier with rapid and a high rate of crack detection was put forward. Experiment serve to verify the effectiveness of proposed method.