基于GF-2遥感影像和改进后PSPNet模型的丘陵地区耕地图斑提取方法  

Extracting patches of arable land in hilly areas based on GF-2 remote sensing images and an improved PSPNet model

作　　者：颜玲 李少达[2] 李彩瑛 陈薇 刘林 宋承远 杨莉 吴若楚 冉培廉 YAN Ling;LI Shaoda;LI Caiying;CHEN Wei;LIU Lin;SONG Chengyuan;YANG Li;WU Ruochu;RAN Peilian(Sichuan Sheng Budongchan Dengjizhongxin,Chengdu 610072,China;College of Earth and Planetary Science,Chengdu University of Technology,Chengdu 610059,China)

机构地区：[1]四川省不动产登记中心,成都610072 [2]成都理工大学地球与行星科学学院,成都610059

出　　处：《成都理工大学学报（自然科学版）》2024年第2期269-280,共12页Journal of Chengdu University of Technology: Science & Technology Edition

基　　金：四川省自然资源科研项目(Kj-2022-19)。

摘　　要：针对丘陵地区耕地地块具有边界模糊、覆盖物种类多样、大小和空间位置分布不规则等特点,传统分类方法难以快速准确提取耕地信息的问题,本文以四川省金堂县竹篙镇和高板镇为研究区域,利用高分二号卫星影像和改进后的PSPNet语义分割网络模型进行耕地图斑提取研究。在模型训练中,引入CBAM注意力模块以提高整个网络的特征提取和表达能力,采用余弦退火学习率以加快模型的收敛速度。结果表明,改进后的PSPNet模型在丘陵地区耕地提取精度方面取得了显著提高,耕地识别精度达到了95.69%,比标准PSPNet模型提高了1.07%,比Unet++,DeepLabv3+和支持向量机方法方法提高了1.32%,1.75%和6.33%。基于改进后的PSPNet模型具有更强的特征提取和表达能力,可以更准确地提取丘陵地区的耕地信息,为农业决策提供更准确的数据支持,促进农业智能化和精准化,提高农作物产量和质量,推动农业现代化进程。Traditional methods of classification encounter challenges in extracting accurate and timely information on cultivated land in hilly areas owing to blurred boundaries,diverse forms of land cover,and irregular spatial distribution in the relevant images.In this study,we used Zhugao Town and Gaoban Town in Jintang County of China as the objects of research,and proposed an improved model of the PSPNet semantic segmentation network to extract cultivated land patches from high-resolution satellite images of hilly areas taken by the GF-2 satellite.The CBAM attention module was introduced to the network to enhance its capabilities of feature extraction and expression during model training,and the cosine annealing learning rate was used to accelerate the convergence of the model.The results of tests showed that the improved PSPNet model could extract cultivated land from images of hilly areas with an overall accuracy of 95.69%,1.07%higher than that of the standard PSPNet model,and 1.32%,1.75%and 6.33%higher than the those of the models of semantic segmentation UNet++,DeepLabv3+,and the support vector machine,respectively.This showed that the improved PSPNet model had strong capabilities of feature extraction and expression that could be used to accurately identify cultivated land in hilly areas.This provided important support for decision-making in agriculture,promotes intelligence and precision in the field,and helped improve the yield and quality of crops.

关 键 词：丘陵耕地 PSPNet模型 CBAM注意力模块 余弦退火学习率 GF-2遥感影像 

分 类 号：P237[天文地球—摄影测量与遥感]