基于深度学习图像配准的燃调表面缺陷检测算法

Research on Surface Defect Detection Algorithm of Fuel Pump Regulator Based on Deep Learning Image Registration

燃油泵调节器的外观检测目前仍采用人工目检方式,存在效率低、效果不稳定的问题。为此,开展表面缺陷检测算法研究,提出一种基于Lite-HRNet的深度学习图像配准网络模型和基于改进图像差分法的表面缺陷检测算法。使用整体和局部的图像增强和故障模拟方法制作数据集,模型采用无监督方式训练。实验结果表明,提出的图像配准算法具有更快的推理速度,单张图片耗时约为传统算法的1/3;表面缺陷检测算法准确率达97.1%,误检率为4.7%。在光照条件变化的实际检测环境下,算法仍具备良好的鲁棒性和适应性,满足燃油泵调节器表面缺陷检测的实时性和准确性要求。

The appearance inspection of the fuel pump regulator still adopts the manual visual inspection method,which has the problems of low efficiency and unstable effect.Therefore,the research on surface defect detection algorithms is carried out.A deep learning image registration model based on Lite-HRNet and a surface defect detection algorithm based on an improved image difference method were proposed.Datasets are made by using global and local image enhancement and fault simulation methods,and the model is trained in an unsupervised manner.The experimental results show that the proposed image registration algorithm has a faster inference speed,the time is about 1/3 of the traditional algorithm;the accuracy of the surface defect detection algorithm is 97.1%,and the false detection rate is 4.7%.In the actual detection environment of changing light conditions,the algorithm still has good robustness and adaptability,which meets the requirements of real-time and accuracy of surface defect detection of fuel pump regulators.