基于经典机器学习的墙柱施工图图像识别

提出了基于经典机器学习的、旨在利用计算机高效的识别图纸能力的、提高审图效率的施工图图像识别方法. 首先, 使用 $K$ 均值($K$-means)算法对墙柱施工图进行聚类, 在聚类基础上搜索局部边界并划分字符; 然后, 使用 $k$-邻近($k$-nearest neighbor, kNN)算法对字符进行识别, 得到墙柱施工图的墙柱名称和配筋信息. 此外, 提出了一种高效计算剪力墙施工图中墙柱面积的影射求和算法. 分别任意取 2 张墙柱施工图进行实验, 研究结果表明: 影射求和算法对墙柱施工图字符识别准确率分别达到 99.60% 和 99.40%, 对面积的识别误差率分别为 0.66% 和 0.34%, 可见该算法对墙柱施工图识别有较好的效果, 可以将墙柱施工图中的主要信息一次性输出为可编辑的文本格式, 具有一定的工程应用价值.

Classic machine learning for image recognition in wall column construction drawings

Drawing review is important in the engineering design field, and efficient and accurate drawing review is a direction for future development. In recent years, the field of artificial intelligence has developed rapidly. This paper proposes an image recognition method based on classic machine learning that aims to use a computer to identify drawings efficiently and improve the efficiency of drawing review. First, the $K$-means algorithm is used to cluster wall column construction drawings, and based on the clustering, the local boundaries are searched, and the characters are divided. Subsequently, the $k$-nearest neighbour (kNN) model is used to identify the characters, and the reinforcement information of the wall column construction drawings is obtained. In addition, this paper proposes the Shadow-Sum method, an algorithm that can efficiently calculate the wall column area in shear wall construction drawings. Experiments are performed by randomly taking two wall column construction drawings. The results reveal that the accuracy of character recognition for the two wall column construction drawings is 99.6% and 99.4%. The recognition error rates for the areas are 0.66% and 0.34%. Column construction drawing identification is beneficial because the main information in wall column construction drawings can be output as editable text, which has certain engineering application value.