基于迁移学习的机械制图智能评阅方法

Intelligent proofreading method of engineering drawing based on transfer learning

针对机械图样几何特征种类多、线条线型易混淆、人工制图风格多样导致校对效率低、误检、漏检等问题,提出基于迁移学习的机械制图智能评阅方法.对机械图样进行预处理,采用改进的阀值迭代算法去除背景、噪点和干扰,完成图样图像的分割,提取机械图样的特征投影图像.通过训练源领域图片的特征提取器,将特征提取器的网络权值迁移到机械图样评阅模型中,完成相似领域的知识迁移.训练逻辑回归分类器,建立基于神经网络权重参数自适应的智能评阅模型,对几何特征、投影特征、图线、剖面符号等机械图样的制图标准要素进行识别.实验结果表明,所提出的机械制图智能评阅方法具有较高的错误识别率和鲁棒性能,单个测试样本平均评阅时间为0.95 s,机械图样的平均评阅正确率为98.82%;与人工评阅相比,所提方法能够在提高评阅效率的同时具有较高准确率.

There are many problems in mechanical drawings,such as thousands of geometric features,confusing line styles and various manual drawing styles,which lead to low proofreading efficiency,false picking and missing inspection.Aiming at the above problem,an intelligent proofreading method of mechanical drawings based on transfer learning was presented.Firstly,the mechanical drawings were preprocessed,an improved threshold iterative algorithm was used to remove the background,noise and interference,and then the projections of mechanical drawings were obtained by segmentation operations.Secondly,by training the feature extractor of the source domain image,the network weight of the feature extractor was transferred to the mechanical drawing evaluation model and the knowledge transfer of similar domain was completed.Finally,the logistic regression classifier was trained and the intelligent evaluation model was established based on self-adjustment of neural network weight parameter.The standard elements of a mechanical drawing such as geometric feature,projection feature,line and section plane symbols were recognized.Experimental results show that the average proofreading time was 0.95 s,and the average judgement accuracy of engineering drawing was 98.82%.Compared with human proofreading of mechanical drawings,the proposed method has higher efficiency and accuracy.