摘要
随着地理信息系统的广泛应用,线划图面状地物的自动识别对于城市规划、土地利用监测等领域具有重要意义。致力于开发一种基于U-Net神经网络的线划图面状地物自动识别方法,旨在提升识别的准确性和效率。研究中基于U-Net神经网络架构,引入CBAM注意力机制与ReLU6激活函数,并结合数据增强与迁移学习技术,对部分无人机影像进行了训练和预测。实验结果表明,该方法在数据集上取得了较高的识别精度,平均准确率可以达到90%以上。并最终使用Canny算子对预测结果进行边缘提取,获得一份简易线划图。
With the widespread application of geographic information systems(GIS),the automatic recognition of areal features in line drawings is of significant importance for urban planning,land use monitoring,and other related fields.This study is committed to developing an automatic recognition method for areal features in line drawings based on the U-Net neural network,aiming to improve the accuracy and efficiency of recognition.In the research,the U-Net neural network architecture was used,and the CBAM attention mechanism and ReLU6 activation function were introduced.Combined with data augmentation and transfer learning techniques,some drone images were trained and predicted.The experimental results show that this method has achieved high recognition accuracy on the dataset,with an average accuracy rate of over 90%.Finally,the Canny operator was used to extract edges from the prediction results to obtain a simplified line drawing.
作者
王鹤
吴长悦
刘政
Wang He;Wu Changyue;Liu Zheng(School of Mining Engineering,North China University of Science and Technology,Tangshan,China)
出处
《科学技术创新》
2025年第18期5-8,共4页
Scientific and Technological Innovation
关键词
深度学习
语义分割
面状地物
自动识别
deep learning
semantic segmentation
areal features
automatic recognition
