|
| 基于农作物物候特征的干旱区撂荒耕地识别——以甘肃省凉州区为例 |
| Identification of abandoned farmland in arid areas based on crop phenological characteristics：a case study of Liangzhou District，Gansu Province |
| 投稿时间：2024-03-29 |
| DOI：10.13254/j.jare.2024.0205 |
| 中文关键词: 撂荒耕地，MaxEnt模型，特征优选，空间分布，凉州区 |
| 英文关键词: abandoned farmland, MaxEnt model, feature optimization, spatial distribution, Liangzhou District |
| 基金项目:甘肃省自然科学基金项目（24JRRA176，22JR5RA247） |
| 作者 | 单位 | E-mail | | 张秀霞 | 兰州理工大学土木工程学院, 兰州 730050
兰州理工大学甘肃省应急测绘工程研究中心, 兰州 730050 | | | 邓灵芝 | 兰州理工大学土木工程学院, 兰州 730050
兰州理工大学甘肃省应急测绘工程研究中心, 兰州 730050 | | | 李旺平 | 兰州理工大学土木工程学院, 兰州 730050
兰州理工大学甘肃省应急测绘工程研究中心, 兰州 730050 | lwp_136@163.com | | 张喜来 | 兰州理工大学土木工程学院, 兰州 730050
兰州理工大学甘肃省应急测绘工程研究中心, 兰州 730050 | | | 汪孝贤 | 兰州理工大学土木工程学院, 兰州 730050
兰州理工大学甘肃省应急测绘工程研究中心, 兰州 730050 | | | 林庆润 | 兰州理工大学土木工程学院, 兰州 730050
兰州理工大学甘肃省应急测绘工程研究中心, 兰州 730050 | | | 程小强 | 兰州理工大学土木工程学院, 兰州 730050
兰州理工大学甘肃省应急测绘工程研究中心, 兰州 730050 | |
|
| 摘要点击次数: 162 |
| 全文下载次数: 190 |
| 中文摘要: |
| 针对西北干旱区撂荒耕地面积小、分布分散、背景复杂及光谱异质性大等特点，基于MaxEnt模型，提出了一种基于多时相多特征的撂荒耕地识别方法。本研究以甘肃省凉州区为例，基于Sentinel-2A影像数据构建光谱特征、纹理特征及物理特征共42个特征因子，优选特征因子后，利用MaxEnt模型进行撂荒耕地的精细提取，并分析了凉州区2021年和2022年撂荒耕地的时空分异特征。结果表明：5月缨帽变换特征的亮度分量、7月纹理特征的相关性统计量和光谱特征的归一化植被指数及 9月光谱特征的归一化差值裸地与建筑用地指数对研究区撂荒耕地识别精度贡献较大；基于农作物物候特征的MaxEnt模型提取撂荒耕地具有较高的可靠性，ROC曲线下面积（AUC）值均大于0.90，处于“极好”水平；混淆矩阵评估总体精度两年均在90.00%以上，Kappa系数均高于0.80；2021、2022年撂荒面积分别为8 605.29、5 642.42 hm2，撂荒率分别为8.80%、5.65%，2022年较2021年撂荒面积有所减少，两年连续撂荒面积共 3 637.62 hm2，撂荒率为 3.64%。凉州区撂荒耕地呈整体零散稀疏、局部集中的分布特征。研究表明，基于农作物多时相多特征的MaxEnt模型撂荒耕地的提取方法，能够在干旱区少样本的情况下获得较高的提取精度，对实现河西地区大区域尺度的撂荒耕地提取具有重要意义。 |
| 英文摘要: |
| In view of the characteristics of abandoned farmland in the arid region of northwest China, such as small area, scattered distribution, complex background and large spectral heterogeneity, based on the MaxEnt model, a method for identifying abandoned farmland considering multiple temporal characteristics was proposed. Taking Liangzhou District of Gansu Province as an example, 42 features such as spectral features, texture features and physical features were constructed based on Sentinel-2A image data, and the MaxEnt model was used to extract the abandoned farmland in Liangzhou District in 2021 and 2022. The results showed that the brightness component of the May tassel cap transformation feature, the correlation statistics of the July texture feature, and the normalized vegetation index of the spectral feature, as well as the normalized difference between the bare land and building land index of the September spectral feature, contribute significantly to the accuracy of identifying abandoned farmland in the study area. The MaxEnt model based on crop phenological characteristics had high reliability in extracting abandoned farmland, and the area under the ROC curve（AUC）values were all greater than 0.90, which was at an "excellent" level; The overall accuracy of the confusion matrix evaluation had been above 90.00% for two years, and the Kappa coefficients were all above 0.80. The abandoned land area in 2021 and 2022 was 8 605.29 hm2 and 5 642.42 hm2, the abandonment rate was 8.80% and 5.65%, and the abandonment area in 2022 had decreased compared with 2021. The total area of continuous abandonment in two years was 3 637.62 hm2, the abandonment rate was 3.64%. The abandoned farmland in Liangzhou District exhibits a scattered and sparse distribution pattern, with localized concentration. The MaxEnt model based on crop multi temporal and multi feature extraction method for abandoned farmland can achieve high extraction accuracy in arid areas with few samples, which is of great significance for achieving large-scale extraction of abandoned farmland in the Hexi region. |
| HTML
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
