摘要
电池金属表面浅划伤、浅凹陷等微弱缺陷在传统二维图像中对比度低、与背景纹理难区分等问题,降低了缺陷的检出率;为解决以上问题,提出了一种基于多视图融合的微弱缺陷检测增强方法;针对微弱缺陷在不同光源方向下合成的三维信息存在缺失问题,提出通过多方向光源装置采集八张不同光源角度图像增加金属表面的光度信息;通过改进的八方向光度立体简化模型获取金属表面三维信息,凸显缺陷的三维特征;针对微弱缺陷在深度图像中存在的图像模糊、对比度低下等问题,通过分析微弱缺陷高度特征呈现角度敏感性特点,拆分抽取深度相关性高的三维信息分量图,由融合系数融合得到增强图像,提高了微弱缺陷的对比度;实验结果表明,该方法应用于实际金属表面缺陷图像检测中,检测精确率提升了19.8%,召回率提升了18.9%,能够较好地解决金属表面微弱缺陷图像检测对比度低下的问题。
There are faint defects such as shallow scratches and dents on the surface of battery metals,resulting in low contrast and difficulty to distinguish them from background texture in traditional 2D images,and reducing detection accuracy.To address these issues,an enhanced detection method based on multi-view fusion for faint defect identification is proposed.Aimed at the missing 3D information synthesized by faint defects in different lighting resource directions,eight images with different lighting source angles are used to increase the photometric information of the metal surface through multiple lighting resource devices.The 3D information of the metal surface is obtained through the improved eight-directional photometric stereo simplification model,highlighting the three-dimensional characteristics of the defects.To address the issues of image blurring and low contrast of faint defects in depth images,the angle sensitivity of height features of faint defects is analyzed.The depth-related 3D information component maps with high correlation are extracted and fused by using the fusion coefficients to generate an enhanced image,thereby improving the contrast of faint defects.Experimental results show that the proposed method increases the detection accuracy by 19.8% and the recall rate by 18.9% in the detection of actual metal surface defects,effectively achieving the low contrast issue in the detection of faint defects in metal surface images.
作者
邵天成
吴静静
SHAO Tiancheng;WU Jingjing(School of Mechanical Engineering,Jiangnan University,Wuxi 214122,China;Jiangsu Key Laboratory of Advanced Food Manufacturing Equipment and Technology,Wuxi 214122,China)
出处
《计算机测量与控制》
2024年第8期86-92,共7页
Computer Measurement &Control
基金
国家自然科学基金项目(62072416)
国家自然科学基金项目(61873246)。
关键词
金属表面
机器视觉
光度立体视觉
图像融合
图像增强
metal surface
machine vision
photometric stereo vision
image fusion
image enhancement
