论文部分内容阅读
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%。
The surface images of hot-rolled strip often have the problems of interference of false defects such as scale and uneven illumination. The current identification methods have the problem of high false-positive rate. The fractal dimension as a feature quantity for the automatic identification of the surface defects of hot rolled strip. The peleg blanket coverage method was used to calculate the fractal dimension of the image at different scales. The concept of optimal scale was proposed and the optimal scale was estimated by using the scale-fractal dimension curve. Experiments were carried out on the surface of mahogany, scale and inclusions. The fractal dimensions at different scales were calculated respectively and then input into Adaboost classifier for training and testing. The experimental results show that using the fractal dimension of the optimal scale as the feature quantity, the recognition rate obtained by the classifier is the best under all scales, that is, 87.96%.