基于Hough变换及卷积神经网络的工程图图线识别技术及应用研究

Research on Engineering Drawing Line Recognition Technology and Application Based on Hough Transform and Convolutional Neural Network

针对手绘工程图与设计自动化之间的矛盾、工程图图线语义较为丰富难以识别等问题,提出一种基于Hough变换及卷积神经网络的工程图图线识别技术。对扫描设备采集到的工程图预处理后,利用Hough变换检测工程图中直线,结合RANSAC(Random Sample Consensus)算法及最小二乘法进行直线拟合,检测出正确直线。根据曲线特征,使用形态学开闭运算及漫水填充算法检测并分离图像中的曲线。采用卷积神经网络对各工程图生成的图线ROI进行图线线型分类,其分类准确率达到98.96%。最后结合工程制图类课程试卷自动评阅需求,在图线识别算法的基础上添加坐标转换算法及曲线一致性判断,对30份试题答案进行自动评分。其评分结果与人工评分结果差距在10%以内,证明文章算法的可行性。

Aiming at the contradiction between hand-drawn engineering drawings and design automation,and that the semantics of engineering drawing lines are rich and difficult to recognize,a technology of engineering drawing line recognition based on Hough transform and convolution neural network is proposed.After pre-processing the engineering drawings collected by the scanning equipment,the straight lines in the engineering drawings are detected by Hough transform,and the correct straight lines are detected by line fitting based on RANSAC(Random Sample Consensus)algorithm and least square method.According to the characteristics of the curve,the morphological opening and closing operation and flooding filling algorithm are used to detect and separate the curve in the image.The convolutional neural network is used to classify the drawing lines ROI generated by each engineering drawing,and the classification accuracy is 98.96%.Finally,based on the requirement of automatic evaluation of engineering drawing course papers,the coordinate transformation algorithm and curve consistency judgment are added on the basis of the graph line recognition algorithm,and the 30 test questions are automatically graded.The difference between the scoring result and the manual scoring result is less than 10%,which proves the feasibility of the proposed algorithm.