基于局部特征加权编码的带钢表面缺陷分类

Strip Surface Defect Image Classification based on Local Feature Weighted Coding Algorithm

针对带钢表面缺陷的几何与纹理分布情况复杂多变、类内缺陷分布形式不固定等特点,提出了一种基于局部特征加权编码的带钢表面缺陷特征提取算法。首先,利用SURF算子提取带钢表面缺陷图像的局部特征。然后,对这些特征进行统计分析后提取能够表征缺陷的关键特征。最后利用TFIDFIG算法和硬量化编码算法对这些关键特征进行加权编码,构造各种不同类型的缺陷特征表示形式。通过对实际带钢生产线上采集的缺陷数据进行验证分析,相比于已有分类算法,该算法提取的缺陷特征能够有效地提高缺陷的识别率,平均识别率达到了99.45%。

Allowing for the strip surface defect images with the typical characteristics of complex geometry and texture distribution, unfixed distribution form of intra-class defects, a strip surface defect feature extraction method based on local feature weighted coding algorithm is proposed. Firstly, the local features of the strip surface defect images are extracted by using the SURF operator. Secondly, the key features are calculated to represent the defects by counting and analyzing these features. Lastly, these key features are coded by using TFIDFIG algorithm and hard vector quantization coding algorithm to establish the feature representation of the various types of defects. Through verifying and analyzing the proposed method on the defect data from the actual strip production line, compared with the existing classification algorithm, the feature extraction method effectively improves defect recognition rate, and the average recognition rate reaches to 99.45%.