[目的]基于图像分割模型SAM高效获取田块信息,分析评价高标准农田建设田块平整的实际成效.[方法]文章选取安徽省六安市裕安区江家店镇龙门村为示例区,采用两期高分辨率遥感影像,基于"先分割—后分类"的思路,利用SAM和残差神经网络(ResNet)相结合的方法,在无样本的条件下快速准确地提取示例区内田块的信息,并从田块的数量、规模和形状三个方面选择指标,对示例区田块在平整前后的变化情况进行分析.[结果]田块平整后,示例区内田块数量由205个降为81个,减少了60.5%;田块平均面积由0.166 hm2增大到0.435 hm2,增加了162.0%;田块密度相较于平整前减少了61.9%;形状指数下降了43.9%;面积加权形状指数降低了15.0%,表明田块平整效果明显.[结论]通过SAM分割结合ResNet分类的方法可以快速获取精准的田块信息,进而为高标准农田块平整成效评价提供精准的基础数据,推动田块平整评价分析工作的现代化和智能化进程.
[Objective]Efficiently obtaining field information based on the image segmentation model SAM and analyzing and evaluating the actual effectiveness of high-standard farmland construction in terms of field leveling is an ongoing process.[Methods]The article selected Longmen Village,Jiangjiadian Town,Yu'an District,Lu'an City,Anhui Province as the example area and used two high-resolution remote sensing images.Based on the idea of"segmentation first and classification later",the combination of SAM and residual neural network(ResNet)was used to quickly and accurately extract the information of the fields in the example area under the condition of no samples.Three indicators were selected from the aspects of the number,scale,and shape of the fields to analyze the changes in the fields before and after leveling in the example area.[Results]After leveling,the number of fields in the example area decreased from 205 to 81,a decrease of 60.5%;the average area of the fields increased from 0.166 hm2 to 0.435 hm2,an increase of 162.0%;the field density decreased by 61.9%compared to before leveling;the shape index decreased by 43.9%;and the area-weighted shape index decreased by 15.0%,indicating a significant leveling effect of the fields.[Conclusion]The method of combining SAM segmentation with ResNet classification can quickly obtain accurate field information,thereby providing accurate basic data for the evaluation of field leveling effectiveness in high-standard farmland,and promoting the modernization and intelligentization of field leveling evaluation and analysis work.