基于图像处理与卷积神经网络的零件识别  

Part Recognition Based on Image Processing and Convolutional Neural Network

作　　者：朱文博[1] 余琦 ZHU Wen-bo;YU Qi(School of Mechanical Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学机械工程学院,上海200093

出　　处：《计算技术与自动化》2022年第1期106-111,共6页Computing Technology and Automation

摘　　要：为了提高零件识别的正确率和效率,提出了一种基于图像处理与机器学习的零件识别算法。首先对图像进行基于饱和度的灰度化;接着通过显著性增强、最大类间方差法(OTSU)的二值化和形态学闭运算求得二值图像;再以改进的种子填充法提取零件区域;最后通过图像关键点的尺度不变特征转换(SIFT)特征与卷积神经网络(CNN)模型相结合的方法识别零件。实验对减速箱、柱塞泵等其中的19种零件进行识别,结果显示零件识别算法的正确率可达98.95%,识别速度约5 fps。通过实验对比与分析,证明方法快速有效,具有较高的正确率和良好的鲁棒性。In order to improve the accuracy and efficiency of parts recognition,a parts recognition algorithm based on image processing and machine learning is proposed.First,the image is grayed out based on the image saturation;then the binary image is obtained through saliency enhancement,maximum between-class variance(OTSU)binarization and morphological closing operations;next,the improved seed filling method is used to extract the part area;Finally,the parts are identified through the combination of the key point Scale-invariant Feature Transform(SIFT)feature of the image and the convolutional neural network(CNN)model.The experiment identified 19 parts of the gearbox,plunger pump,etc.The results showed that the accuracy rate of the part recognition algorithm can reach 98.95%,and the recognition speed is about 5fps.Through experimental comparison and analysis,the method is fast,effective,and has a high correct rate and good robustness.

关 键 词：零件识别 图像饱和度 种子填充法 尺度不变特征转换 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]