基于Gaussian-yolov3的铝型材表面缺陷检测

Surface Defect Detection of Aluminum Profile Based on Gaussian-yolov3

摘要:

在铝型材的实际生产过程中,由于各方面因素的影响,铝型材表面会产生碰伤,刮花,凸粉等瑕疵,这些瑕疵会严重影响铝型材的质量。目前主要采用人工检测,由于铝型材表面自身含有纹路,与瑕疵区分度不高,传统人工肉眼检查十分费力,质检的效果难以控制。为解决上述问题,以铝型材表面缺陷为研究对象,使用Gaussian-yolov3为基础目标检测网络,针对铝型材表面部分条状缺陷的特性,使用变形卷积技术增强卷积的适应性。针对小缺陷检测问题,使用密集连接技术。使用数据增强扩展数据。通过对比Faster R-CNN、SSD实验,结果表明,基于Gaussian-yolov3的检测方法准确率可以达到96%,检测速度可以满足实时性要求,具有较强的实用性。

Abstract:

In the production process of aluminum profiles, due to various factors, defects such as bumps, scratches, and dirty spot will occur on the surface of aluminum profiles. These defects will seriously affect the quality of aluminum profiles. At present, manual inspection is mainly used. Because the surface of the aluminum profile itself contains texture and is not highly distinguishable from defects, the traditional manual visual inspection is very laborious and the quality inspection effect is difficult to control. In order to solve the above problems, the surface defects of aluminum profiles are used as research objects, and Gaussian-yolov3 is used as the target detection network. For strip defects in aluminum profiles, deformable convolution technology is used to enhance the adaptability of convolution. For small defect detection, we used dense connection technologyEnhance extended data. By comparing Faster R-CNN and SSD experiments, the results show that the accuracy of the detection method based on Gaussian-yolov3 can reach 96%, the detection speed can meet the real-time requirements, and it has strong practicability.