基于YOLOv2网络模型的手机镜片缺陷实时检测方法

Real Time Detection Method of Mobile Phone Lens Defects Based on YOLOv2 Network Model

针对手机镜片缺陷检测采用人工目视法存在效率低、误检率高、易受主观影响;且现有的机器视觉法存在成本高和应用场景为静态图片等问题,提出一种基于YOLOv2网络模型的手机镜片缺陷实时检测方法。首先,对手机镜片缺陷检测数据集样本进行归一化处理;然后,调整网络模型的训练参数,并将数据集送入YOLOv2网络模型进行迭代训练,分析训练过程的性能曲线,得到最优权重;最后,对验证集样本进行检测分析。实验结果表明:该检测方法的实时检测速度可达14帧/s;缺陷定位精度高,检测准确率为96.13%;可满足低成本条件下,手机镜片缺陷实时检测的需求。

For the defect detection of mobile phone lens,the manual visual method has the advantages of low efficiency,high false detection rate and easy to be affected by subjectivity;and the existing machine vision methods have the problems of high cost and static pictures.A realtime detection method of mobile phone lens defects based on YOLOv2 network model is proposed.Firstly,the samples of mobile phone lens defect detection data set are normalized.Then,the training parameters of the network model are adjusted,and the data set is sent to the YOLOv2 network model for iterative training.The performance curve of the training process is analyzed,and the optimal weight is obtained.Finally,the verification set samples are tested and analyzed.The experimental results show that the real-time detection speed of this method can reach 14 frames/s;the defect location accuracy is high,and the detection accuracy is 96.13%.It can meet the needs of real-time detection of mobile phone lens defects under low-cost conditions.