耕地是保障人类粮食安全的根本，它为农作物的生长提供了基本条件。只有耕地条件得到了妥善的保障与提升，才能稳定地保障粮食的供应、确保其优良品质，进而为广大民众筑牢粮食安全的防线。然而，我国在改革开放以来，随着经济快速发展、劳动力大量转移，中西部农村地区的大片耕地遭到撂荒，对我国的粮食安全带来了巨大挑战，同时也阻碍了农业的持续发展。目前撂荒耕地的调查方式主要是各地政府部门组织人员实地勘查，不仅耗时耗力、数据更新缓慢，而且难以描述撂荒耕地的时空变化规律。因此，基于遥感技术开展撂荒耕地的高精度提取具有重要研究意义和实践价值。

遥感技术的应用目前分为基于像元和面向对象两条路线，基于像元的影像分类技术，浪费了遥感影像中的大量信息，且对撂荒耕地的提取精度较差。基于此，本研究以2017-2023年Sentinel-2遥感影像为数据源，采用面向对象的分类方法构建出一种新的撂荒耕地提取指标，应用该指标获取渭南市撂荒耕地分布信息，并结合GIS（Geographic Information System）技术分析其时空动态变化特征。主要研究及成果如下：

（1）阐述影像分割的技术原理，使用多尺度分割算法完成影像对象层的建立。基于预处理后的渭南市Sentinel-2影像，通过ESP（Estimation Scale Parameter）模块评估法与最大面积法，得到本研究区内多尺度分割算法的最优分割尺度为35、70、125。随后应用最优尺度将影像分割为三个对象层，每个对象层由若干个对象单元构成，每个对象单元具有光谱、方差、形状、纹理等特征，不同特征用于不同地物的分类、提取，其中35分割尺度的对象层用于撂荒耕地的提取。

（2）利用不同地物在不同特征上的取值差异构建撂荒耕地的提取指标，并应用该指标按照二分决策树的分类流程，完成渭南市2017年、2019年、2021年、2023年的土地利用类型分类及撂荒耕地信息的提取。之后建立混淆矩阵计算分类精度和Kappa系数，各年的总体分类精度在86.7%-89.9%之间，Kappa系数在0.82-0.88之间；撂荒耕地提取结果的用户精度在86.5%-90.3%之间，生产精度在87.1%-91.8%之间。精度验证结果表明本研究所采用的面向对象方法提取精度高，分类效果好。

（3）绘制渭南市2017年、2019年、2021年、2023年的撂荒耕地空间分布图，并计算各年的撂荒率，结合莫兰指数、标准差椭圆、近邻分析等GIS技术，阐明了渭南市2017年-2023年间的撂荒耕地时空动态变化规律。结果表明：渭南市的撂荒耕地面积逐年缓步减少，平均撂荒率从2017年的9.17%降至2023年的4.65%；全市撂荒耕地存在空间正相关聚集的特征，整体分布呈现出中部较为密集、四周分散的格局；耕地距离河流较远、坡度较大时，撂荒率偏高。

Arable land is fundamental to human food security, providing the basic conditions for the growth of crops. Only good arable land conditions can ensure a stable supply and good quality of food, and thus guarantee food security for the general public. However, since China's reform and opening up, with the rapid development of the economy and the massive transfer of labour, a large area of arable land in the rural areas of central and western China has been abandoned, which has posed a great challenge to China's food security, and also impeded the sustainable development of agriculture and rural areas. At present, the survey of abandoned arable land is mainly carried out by field surveys organised by local government departments, which is not only time-consuming and labour-intensive, but also slow in updating data, and difficult in describing the pattern of change in space and time. Therefore, the high-precision extraction of abandoned arable land based on remote sensing technology has important research significance and practical value.

Existing studies mainly use image classification techniques based on image elements, which wastes a lot of information in remote sensing images and lacks the analysis of spatial and temporal changes of abandoned arable land. Based on this, this study takes the 2017-2023 Sentinel-2 remote sensing images as the data source, and adopts the object-oriented classification method to construct a new index for the extraction of abandoned arable land, applies this index to obtain the distribution information of abandoned arable land in Weinan City, and analyses its spatio-temporal dynamic characteristics by combining with the GIS technology. The main research results are as follows:

(1) The image segmentation algorithm was summarised systematically, and the image object layer was established using the multi-scale segmentation algorithm. Based on the preprocessed Weinan City Sentinel-2 image, the optimal segmentation scales of the multi-scale segmentation algorithm were obtained as 35, 70, and 125 through the ESP module evaluation method and the maximum area method, and the original image was then divided into three object layers by applying this segmentation scale, and each object layer was composed of several object units. According to the spectral, variance, shape, texture and other features of each object unit, they are used for the classification of different features, in which the object layer of 35 segmentation scale is used for the extraction of abandoned farmland.

(2) In this study, the extraction index of abandoned arable land was constructed by using the differences in the values of different features of different features, and the index was used to complete the classification of land use types and the extraction of abandoned arable land information of Weinan City in 2017, 2019, 2021, and 2023 according to the classification process of dichotomous decision tree. Afterwards, the confusion matrix was used to calculate the classification accuracy and Kappa coefficient, and the overall classification accuracy for each year ranged from 86.7% to 89.9%, and the Kappa coefficient ranged from 0.82 to 0.88; the user accuracy of the abandoned arable land extraction results ranged from 86.5% to 90.3%, and the production accuracy ranged from 87.1% to 91.8%. The accuracy verification results show that the object-oriented method adopted in this research has high extraction accuracy and good classification effect.

(3) The spatial distribution maps of abandoned arable land in Weinan City in 2017, 2019, 2021 and 2023 were drawn, and the abandonment rate of each year was calculated, and the spatial and temporal dynamics of abandoned arable land in Weinan City during the period of 2017-2023 was elucidated by combining GIS techniques such as Moran's index, standard deviation ellipse, and nearest-neighbour analysis. The results show that the area of abandoned arable land in Weinan City has been decreasing year by year, and the average abandonment rate has decreased from 9.17% in 2017 to 4.65% in 2023; there is a positive spatial correlation between the aggregation of abandoned arable land in the city, and the overall distribution shows a pattern of denser in the central part of the city and dispersed in the surrounding area; when the arable land is far away from the river and has a larger slope, the abandonment rate is high.