基于卷积神经网络的焊接装配特征识别研究

Recognition Algorithm of Welding Assembly Characteristics Based on Convolutional Neural Network

为实现高铁白车身焊接拼装技术的智能化与自动化,解决焊接过程中特征区域小、背景干扰多等问题,提出了基于迁移学习和卷积神经网络的焊接装配特征快速识别算法。首先采用二值化等传统图像处理算法确定待提取特征的粗略位置,在此基础上再使用sobel、腐蚀、霍夫线段检测确定特征区域的精确位置。其次,考虑到不同环境下,精确定位后特征区域表现不同,故采用基于卷积神经网络的分类模型以增强预测模型的鲁棒性和准确性。最后,选择基于迁移学习的的视觉几何群网络(VGG16)来解决样本量不足以训练整个模型参数的问题。实验结果表明,本文所提的识别算法能够准确识别型材的状态,且在识别检测速度上优于YOLOV3,在准确率上劣于YOLOV3,算法满足使用场景下的实时性要求。

In order to realize the intellectualization and automation of welding and assembling technology for high-speed white body,the problems of small feature area and multi-background interference in welding process are solved,a novel fast recognition algorithm of welding assembly based on migration learning and convolution neural network is proposed.Firstly,the traditional image processing algorithms such as binarization are used to determine the rough position of the feature to be extracted.On this basis,Sobel,corrosion and Hough line detection are used to determine the precise position of the feature area.Secondly,considering the different performance of feature regions in different environments,a classification model based on convolution neural network is adopted to enhance the robustness and accuracy of the prediction model.At last,Visual Geometry Group Network(VGG16)based on transfer learning is selected to solve the problem that the number of the samples is not enough to train the parameters of the whole model.The experimental results show that the recognition algorithm proposed in this paper can accurately identify the state of profile,and the detection speed is better than YOLOV3,and the accuracy is inferior to YOLOV3.The algorithm can meet the real-time requirements in the use scene.