一种原理图美观度等级评价算法  

An algorithm for evaluating aesthetic quality level of schematic diagrams

作　　者：孙辉 丁延峰 王刚 李桢荣 SUN Hui;DING Yanfeng;WANG Gang;LI Zhenrong(College of Computer Science,Nankai University,Tianjin 300350,China;Empyrean Technology,Beijing 100102,China)

机构地区：[1]南开大学计算机学院,天津300350 [2]北京华大九天科技股份有限公司,北京100102

出　　处：《华中科技大学学报（自然科学版）》2024年第11期8-14,共7页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(62141412,62272253,62272252)。

摘　　要：为解决电路布局布线原理图生成算法缺乏从电路设计人员角度评估生成原理图优劣程度的问题,提出原理图美观度等级评价指标(SAE).为量化求解SAE,首先对原始布局布线原理图采用滑动窗口策略分割为多张子图,并使用拉普拉斯卷积核自动提取子图特征;然后引入主成分分析法对子图特征向量自动降维;其次引入遗传聚类算法根据子图特征向量自动寻找中心子图;最后采用预训练的残差神经网络对中心子图执行美观度等级预测,且根据不同聚合规则得到最终美观度等级.实验结果表明:在4种不同聚合规则下所提出的DeepSAE算法准确率最高分别为0.677,0.953,0.734和0.953.To address the problem of the lack of quantitative indicators for evaluating the quality of generated schematic diagrams from the perspective of circuit designers,a schematic aesthetics evaluation(SAE)metric was proposed.To quantify the SAE,the original layout wiring schematic diagram was divided into multiple subgraphs using a sliding window strategy,and the subgraph features were automatically extracted using the Laplacian convolution kernel.Then,the principal component analysis method was introduced to automatically reduce the dimensionality of the subgraph feature vectors.Furthermore,the genetic clustering algorithm was employed to automatically search for the central subgraph based on the subgraph feature vectors.Finally,a pre-trained residual neural network was used to predict the aesthetic level of the central subgraph,and the final aesthetic level was obtained based on the different aggregation rules.Experimental results show that under four different aggregation rules,the proposed DeepSAE algorithm achieves the highest accuracy ratios of 0.677,0.953,0.734,and 0.953,respectively.

关 键 词：电路自动化设计技术 电路布局布线原理图 原理图美观度评价 卷积神经网络 遗传聚类算法 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]