激光与光电子学进展, 2019, 56 (15): 151202, 网络出版: 2019-08-05
基于改进Faster R-CNN的子弹外观缺陷检测 下载: 1007次
Bullet Appearance Defect Detection Based on Improved Faster Region-Convolutional Neural Network
测量 目标检测 子弹外观缺陷 卷积神经网络 measurement target detection bullet appearance defect convolutional neural network
摘要
为了实现子弹外观缺陷的自动检测,解决传统机器视觉方法在缺陷检测方面手工设计目标特征耗时和泛化能力差的问题,针对子弹外观缺陷数据集,采用K -means++算法改进锚框的生成方法,提出了Faster R-CNN子弹外观缺陷检测模型。该模型采用卷积神经网络,可以自动提取目标特征,泛化能力强。将该检测模型分别与ZFNet、VGG_CNN_M_1024和VGG16结合,结果表明,与VGG16结合的检测模型的检测精度高于其他两种模型方案,并且在所提算法的基础上,精度提升到了97.75%,速度达到28 frame·s
-1。
Abstract
To realize automatic detection of bullet appearance defects and to overcome the limitations associated with traditional machine vision methods, i.e., excessive time required to manually design a target feature and generalization ability is poor in defect detection, we use the K-means++ algorithm to improve the anchor frame generation method and propose a bullet appearance defect detection model based on the improved faster region-convolutional neural network (R-CNN). The proposed model uses a CNN that can automatically extract target features and has strong generalization ability. The detection model is combined with ZFNet, VGG_CNN_M_1024, and VGG16, respectively. Results demonstrate that the detection accuracy of the detection model combined with VGG16 is higher than the others. The results show that that of the improved model demonstrates 97.75% accuracy and the speed reaches 28 frame·s
-1.
马晓云, 朱丹, 金晨, 佟新鑫. 基于改进Faster R-CNN的子弹外观缺陷检测[J]. 激光与光电子学进展, 2019, 56(15): 151202. Xiaoyun Ma, Dan Zhu, Chen Jin, Xinxin Tong. Bullet Appearance Defect Detection Based on Improved Faster Region-Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2019, 56(15): 151202.