基于级联卷积网络的紧固件异常检测

紧固件广泛应用于日常生活和工业生产制造中,其异常状态在许多场景中会导致严重的安全隐患。目前紧固件异常检测仍依赖人工排查,很难通过常规无损检测技术自动识别。针对该问题,提出了一种基于级联卷积网络的自动检测方案,能够快速的检测固定场景下的紧固件异常情况。首先采集紧固件图像,使用目标检测网络确定所有紧固件区域;接着使用所提出的紧固件关键点回归网络预测关键点特征信息;最后通过对比同一紧固件不同时刻的关键点特征信息实现紧固件异常检测。在自制的重庆市轻轨轨道梁指型板紧固件数据集进行了测试,实验结果显示该方法在准确率达到96. 5%时,对于异常紧固件的召回率高达99%,结果表明该方法具有可行性,在类似场景中具有实际应用价值。

Cascade convolutional neural network based abnormal detection of fasteners

Fasteners are widely used in daily life and industrial manufacturing. In many scenarios,the its abnormal state could cause security risks. For now,it’s hard to automatically detect the abnormal state of fastener by non-destructive testing,and still relies on manual checking. In this paper,a solution based on cascade convolutional neural network is presented to quickly detect abnormal state of fasteners in fixed scenes. The workflow is first to collect images and use objective-detection network to detect the fastener regions. Then,the proposed key-point regression network is used in prediction of key points on fasteners and getting the feature of key point. Finally,the abnormal detection of fasteners is conducted by comparing the key point feature information of the same fastener through two measurements at different times. The proposed approachisevaluated on dataset of fasteners on finger plate in Chongqing railway transportation,experimental results reach performance with accuracy of 96. 5% and recall of 99% which shows the great value of practical application of the proposed approach.