摘要
热轧带钢的表面图像往往存在氧化铁皮等伪缺陷的干扰与光照不均的问题,目前的识别方法存在着误识率高的问题。将分形维数作为特征量,用于对热轧带钢表面缺陷的自动识别。利用peleg毯覆盖法计算图像在不同尺度下的分形维数,并提出最优尺度概念,通过尺度-分形维数曲线图估计最优尺度。对麻面、氧化铁皮和夹杂等进行试验,分别计算不同尺度下的分形维数,作为特征量输入Adaboost分类器进行训练和测试。试验结果表明用最优尺度下的分形维数作为特征量,分类器得到的识别率是所有尺度下最优的,即87.96%。
False alarm is a main problem in classification of surface defects for hot-rolled steel strips, because there are a lot of scales on surface of hot-rolled steel strips, and un-homogeneous illumination is an another reason. Fraction dimensions were introduced as features, and applied to classification of surface defects for hot-rolled steel strips. Fraction dimensions under different scales were computed through Peleg covered carpet method, and the optimized scale was proposed. Curves of scale and dimensions were used to estimate the optimized scale. Fraction dimensions under different scales were inputted as features into a classifier based on Adaboost, which was trained and tested with samples of pimples, scales and shells. Results of tests showed that the classification rate of the classifier with features of fractal dimensions computed under the optimized scale was 87.96%, which was the best of all the scales.
出处
《冶金设备》
2008年第2期1-4,18,共5页
Metallurgical Equipment
基金
国家自然科学基金(60705017)
"十一五"国家科技支撑计划(2006BAE03A06)资助项目
关键词
分形维数
最优尺度
热轧带钢
表面检测
Fractal dimension Optimized scale Hot-rolled steel strip Surface inspection
