基于决策融合的南方复杂地区覆膜农田信息快速提取研究  

Rapid Extraction of Plastic-Mulched Farmland Information in Complex Southern Regions Based on Decision Fusion

作　　者：林娜[1] 陈宏[1] 谢骞 赵健[1] LIN Na;CHEN Hong;XIE Qian(Institute of Digital Agriculture,Fujian Academy of Agricultural Sciences,Fuzhou,Fujian 350003)

机构地区：[1]福建省农业科学院数字农业研究所,福建福州350003

出　　处：《安徽农业科学》2025年第3期229-235,242,共8页Journal of Anhui Agricultural Sciences

基　　金：福建省农业高质量发展协同创新工程项目(XTCXGC2021015);福建省农业科学院科技创新团队项目(CXTD2021031);福建省农业科学院自由探索科技创新项目(ZYTS2023022)。

摘　　要：为实现快速且准确地获取南方复杂地区覆膜农田信息,探索一种基于决策融合规则的单时相遥感提取方法。首先基于Sentinel-2影像数据,在南方丘陵山区这一典型地表混杂区域,应用特征提取算法与最小距离、最大似然、支持向量机、BP神经网络、随机森林5种单分类器进行遥感影像分类,在此基础上依据各分类器分类结果与分类性能,构建一种结合层次分析与投票机制的自适应决策融合规则,进行了覆膜农田信息的提取,并评估其精度。对比5种单分类器与决策融合模型的分类性能,结果表明决策融合模型在精度评价指标上均显著优于单一分类器,总体精度达到91.82%,Kappa系数达到0.89,对覆膜农田的提取识别能力也表现优异,其生产者精度、用户精度和F_(1)Score分别达到92.68%、81.74%和0.87。提出的方法有效提高了覆膜农田的提取准确率、复杂度和计算成本较低,具有较强的泛化性与可操作性,适用于南方复杂农业环境,为实际生产应用提供了可靠的解决方案。To achieve rapid and accurate plastic-mulched farmland information in the complex regions of southern China,this study explores a single-temporal remote sensing extraction method based on decision fusion rules.Utilizing sentinel-2 imagery data in the southern hilly areas,a typical region with mixed land surfaces,the Jeffries-Matusita(JM)distance feature extraction method along with five individual classifiers including minimum distance,maximum likelihood,support vector machine,BP Neural Network,and random forest were employed for remote sensing image classification.Based on the classification results and performance of these individual classifiers,an adaptive decision fusion rule combining the Analytic Hierarchy Process(AHP)and voting mechanism was developed for the extraction of plastic-mulched farmland information,and its accuracy was evaluated.By comparing the classification performance of the five individual classifiers and the decision fusion model,results indicate that the decision fusion model significantly outperforms the individual classifiers in terms of accuracy metrics,achieving an overall accuracy of 91.82%and a Kappa coefficient of 0.89.The model also demonstrated superior capability in identifying plastic-mulched farmland,with producer accuracy(PA),user accuracy(UA),and F 1Score reaching 92.68%,81.74%,and 0.87,respectively.The proposed method fully leverages the advantages of multiple classifiers and decision fusion,enabling more accurate classification decisions for plastic-mulched farmland.It not only reduces complexity and computational cost but also exhibits strong generalization and operability,making it suitable for complex agricultural environments in southern regions.This work provides a reliable solution for practical production applications.

关 键 词：决策融合 多分类器 覆膜农田 Sentinel-2 复杂地区 

分 类 号：S127[农业科学—农业基础科学]