基于机器视觉和深度神经网络的零件装配检测

Component Assembly Inspection Based on Deep Neural Network

针对工业生产中存在着大量的零件识别定位以及装配检测等,传统人工检测效率低、劳动强度大、识别不准确。文章提出了一种基于深度神经网络的零件装配检测方法。首先该方法对零件图像和装配图像进行采集,选择Mask-RCNN网络进行训练,对装配零件进行分类以及定位,通过已识别的零件类型判定装配件是否存在漏装;然后将分割后的零件图像进行二值化处理,利用Canny算子提取零件图像轮廓;最后利用图像的Hu矩特征与正确的零件图像轮廓进行对比,判断装配是否正确。通过实验验证可得,该方法在零件装配中的漏装和换装检测中效果较好,并表现出较高的鲁棒性。

Aiming at the existence of a large number of parts identification and assembly inspection in industrial production,traditional manual detection has low efficiency,high labor intensity and inaccurate recognition.This paper proposes a method of part assembly detection based on deep learning.Firstly,this method collected the single part image and the assembly image,and the selected Mask-RCNN network for training,classifies and locates the assembled parts,determines whether the assembly is missing through the identified part type;Secondly,this method binarizes the segmented part image and extracts the part image outline using the Canny operator.Finally,the Hu moment feature of the image is compared with the correct contour of the part image to determine whether the assembly is correct.It is proved by experiments that the method has better effect in the missing and misaligned detection of parts assembly and shows higher robustness.