| 基于优化的ICNet高分遥感影像城市建成区分类 |
| |
| 引用本文： | 夏国静,黄凯,郑庆雅,夏萍,田波,周婷,牛鑫鑫.基于优化的ICNet高分遥感影像城市建成区分类[J].安徽农业大学学报,2023,50(2):303. |
| |
| 作者姓名： | 夏国静 黄凯 郑庆雅 夏萍 田波 周婷 牛鑫鑫 |
| |
| 作者单位： | 安徽农业大学工学院，合肥 230036;安徽工程大学人工智能学院，芜湖 241000;安徽农业大学工学院，合肥 230036; 安徽省智能农机装备工程实验室，合肥 230036 |
| |
| 基金项目： | 国家自然科学基金（11802003）, 安徽省自然科学基金（2008085ME158）和安徽省国际科技合作计划项目（1604b0602029）共同资助。 |
| |
| 摘 要： | 城市建成区是一类具有大面积的组合型目标群体，该区域地物丰富，光谱特征复杂多变，且具有大量的同物异谱与地物像素单元交错等现象，影像分类难度显著增加。针对图像级联网络（image cascade network，ICNet）计算复杂、分类精度低的问题，采用优化的ICNet对高分辨率遥感影像城市建城区地物分类进行研究，通过添加高效通道注意力机制（efficient channel attention，ECA）和联合金字塔上采样模块（joint pyramid upsampling，JPU）替换空洞卷积来获得ICNet改进网络，采用总体分类精度（overall accuracy，OA）、Kappa系数与F13个指标对分类结果进行精度评估，并与随机森林（random forest，RF）、ENet和ICNet3种方法进行对比分析。结果表明，优化的ICNet网络模型能够更准确的进行地物分类，总体分类精度为75.12%，相较于其他分类方法分别提高16.56%、10.48%和4.81%。后用开源数据集进一步验证了优化模型的有效性，说明优化的ICNet网络可用于城市建成区的分类研究。
|
| 关 键 词： | 卷积神经网络 ICNet语义分割模型 分类 高分遥感影像 城市建成区 |
Classification of urban built-up area from high-resolution remote sensing image based on the optimized ICNet |
| |
| XIA Guojing,HUANG Kai,ZHENG Qingy,XIA Ping,TIAN Bo,ZHOU Ting,NIU Xinxin.Classification of urban built-up area from high-resolution remote sensing image based on the optimized ICNet[J].Journal of Anhui Agricultural University,2023,50(2):303. |
| |
| Authors: | XIA Guojing HUANG Kai ZHENG Qingy XIA Ping TIAN Bo ZHOU Ting NIU Xinxin |
| |
| Institution: | School of Engineering, Anhui Agricultural University, Hefei 230036;School of Artificial Intelligence, Anhui Polytechnic University, Wuhu 241000;School of Engineering, Anhui Agricultural University, Hefei 230036; Anhui Intelligent Agricultural Machinery Equipment Engineering Laboratory, Hefei 230036 |
| |
| Abstract: | Urban built-up area is a kind of combined target group with large area, which is rich in features, complex and variable spectral characteristics, and has a large number of phenomena such as different spectra of the same object and the interleaving of ground object pixel units, making image classification significantly more difficult. To address the problems of complex calculation and low classification accuracy of image cascade network (ICNet), the optimized semantic segmentation model (ICNet) is used to classify the ground objects in the urban built-up area of high-resolution remote sensing images, by adding the efficient channel attention (ECA) and joint The classification results are evaluated by adding efficient channel attention (ECA) and joint pyramid upsampling (JPU) to replace dilated convolution to obtain the improved ICNet network. The overall accuracy (OA), kappa and F13 were used to evaluate the accuracy of the classification results, and were further compared with random forest (RF), ENet and ICNet methods. The results showed that the optimized ICNet network model could classify ground objects more accurately with an overall classification accuracy of 75.12%, which was 16.56%, 10.48% and 4.81% higher than that with RF, ENet and ICNet method, respectively. Then, the open source data set was used to further verify the effectiveness of the optimized model. The experiment results showed that the optimized ICNet network can be used for the classification of urban built-up areas. |
| |
| Keywords: | convolutional neural network ICNet semantic segmentation model classification high-resolution remote sensing images urban built-up area |
|
| 点击此处可从《安徽农业大学学报》浏览原始摘要信息 |
| 点击此处可从《安徽农业大学学报》下载免费的PDF全文 |
|
