摘要
针对传统识别方法对输变电工程图纸的分类效果较差且精确度偏低的问题,在注意力机制和改进动态ReLU基础上,提出了一种基于深度学习的工程图纸智能评审方法。利用Xception基础网络与动态ReLU函数优化小样本数据的分类效果,进而完善样本数据的ReLU参数分配。通过引入改进注意力机制模块,深化神经网络算法中特征图的权重分配,进一步提升了工程图纸的分类效果。仿真结果表明,与传统工程图纸识别方法相比,基于深度学习的工程图纸智能评审方法具有更优分类效果。
Aiming at the problems of poor classification effect and low accuracy of traditional recognition methods for power transmission and transformation engineering drawings,an intelligent evaluation method based on depth learning for engineering drawings was proposed according to attention mechanism and improved dynamic ReLU.By using the basic Xception network and dynamic ReLU function,the classification effect of small sample data was optimized,the ReLU parameter allocation of sample data was optimized,the weight allocation of feature map in neural network algorithm was deepened by introducing improved attention mechanism module,and the classification effect of engineering drawings was further enhanced.The simulation results show that the intelligent evaluation method based on deep learning for engineering drawings has better classification effect than the traditional methods.
作者
陈晨
薛文杰
董平先
翟育新
齐桓若
CHEN Chen;XUE Wenjie;DONG Pingxian;ZHAI Yuxin;QI Huanruo(Economic and Technological Research Institute,State Grid Henan Electric Power Company,Zhengzhou 450052,Henan,China)
出处
《沈阳工业大学学报》
CAS
北大核心
2024年第6期772-778,共7页
Journal of Shenyang University of Technology
基金
河南省科技计划项目(S2022CXCPB0465)。
关键词
输变电工程图纸
改进SE模块
ReLU函数
深度学习
Xception网络
图像识别
图像分类
卷积神经网络
power transmission and transformation engineering drawing
improved SE module
ReLU function
deep learning
Xception network
image recognition
image classification
convolutional neural network
