基于神经网络的壁板铆接视觉测量研究

Study on Visual Measurement of Riveting Panel Based on Neural Network

铆接制孔直径是影响制孔质量的重要参数之一,针对人工测量存在的人为偏差和效率低的问题,提出了一种基于神经网络的机器视觉测量方法,该方法通过图像预处理、ROI提取和图像识别及尺寸测量四个阶段对产品进行检测。图像预处理阶段的目标是通过基于神经网络的镜头补偿模型获得更好的原始图像。检测到的边界可作为滤波图像ROI提取的基础,压缩后的ROI存储在数据库中并由图像识别模块发送,图像识别模块是基于反向残差块构建的,在保持识别精度的同时减少了模型大小和计算时间,孔的图像识别准确率达到94.67%。利用参照物的像素直径得到壁板铆接被测孔的像素直径。实验结果表明,在不同位置的孔直径测量中,本方法的直径测量精确率平均在95%左右。

The diameter of the riveting hole is one of the important parameters affecting the quality of the riveting hole.This paper proposes a machine vision measurement method based on neural network,which detects products through image preprocessing,region of interest(ROI)extraction,image recognition,and size measurement.The objective of image preprocessing is to obtain a better representation of the original image through the lens compensation model based on a neural network.The detected boundary can be used as the basis for filtering image ROI extraction,and the compressed ROI is stored in the database and sent out by the image recognition module.The image recognition module is constructed based on reverse residual blocks,which reduces the model size and calculation time while maintaining the recognition accuracy,and the image recognition accuracy of holes reaches 94.67%.The pixel diameter of the panel riveted to the measured hole is obtained by using the pixel diameter of the reference.Experimental results show that the diameter measurement accuracy of the proposed method is about 95%on average.