基于机器学习的装配质量图像识别研究

Research on Assembly Quality Image Recognition Based on Machine Learning

针对柔性化生产线上人工装配过程中装配质量的检测问题,提出基于机器学习的装配图像识别方法。首先设计整体方案,采用普通工业相机结合图像识别软件的方式,构建硬件系统并开发软件模块。接着利用监督算法通过训练分类器实现图像识别。测试各算法方案并比较图像识别准确率、训练和分类耗时以及训练所需数据量,选取效果较好的算法方案。然后研究训练数据扩展方法,以降低训练图片数量并提高识别正确率。结果表明:采用合适算法方案的装配图像识别系统能满足工业应用需要。

In order to detect assembly quality during manual assembly process on the flexible production line,this paper presents an assembly image recognition method based on machine learning.Firstly,the overall scheme is designed,the hardware system is built and the software module is developed by using the common industrial camera combined with the image recognition software.Image recognition is then performed by training classifier using supervised algorithms.According to the test results of each algorithm scheme,the image recognition accuracy,training and classification time-consuming and the amount of data required for training are compared,and the best scheme is selected.Then the training data expansion method is studied to reduce the number of training pictures and improve the recognition accuracy.The results show that the assembly image recognition system with suitable algorithm scheme can meet the needs of industrial use.