联合上下文感知与边界约束的遥感影像林地提取方法  

Extraction method of remote sensing images forest land by joint context awareness and boundary constraints

作　　者：胡永进[1] Hu Yongjin(Jiangsu Vocational College of Agriculture and Forestry,Jurong,212400,China)

机构地区：[1]江苏农林职业技术学院,江苏句容212400

出　　处：《中国农机化学报》2024年第12期162-167,共6页Journal of Chinese Agricultural Mechanization

基　　金：江苏农林职业技术学院科技项目(2024kj29)。

摘　　要：准确地提取林地分布信息是林地管理、保护和可持续利用的关键步骤。针对现有林地提取方法精度较低且边缘不够精细的问题,提出一种联合上下文感知与边界约束(CABC-Net)的遥感影像林地提取方法。首先,设计上下文感知(CA)模块,用于探索空间像素信息之间的联系,并通过层间特征传递提取足够的全局上下文信息,以降低上下文差异和复杂背景对检测结果的干扰。其次,为进一步优化边缘细节,提出边界约束(BC)模块,将边界特征与深层特征结合作为模型的辅助特征,缩小定位林地边界的波动范围,校准不确定性区域以提高边界判别能力。最后,为验证方法的有效性,建立一个新的数据集并开展案例试验分析。结果表明,该方法交并比提高0.55%~9.45%,像素准确率提高0.19%~7.53%,本方法在面对复杂场景时具有更好的边界完整度,能够更好地用于林地提取。Accurate extraction of forest resources and their distribution is crucial for the management,protection,and sustainable utilization of forests.In order to address the issues of low accuracy and poor edge definition in existing methods of forest extraction,this paper introduces a novel approach by using remote sensing images called the Context Awareness and Boundary Constraints Network(CABC-Net).Initially,a Context Awareness(CA)module is developed to explore the interconnections among spatial pixel data,extracting ample global contextual information through the transfer of features across layers.This reduces the impact of contextual variations and complex backgrounds on the detection outcomes.Secondly,to further enhance the edge precision,a Boundary Constraints(BC)module is proposed that integrates boundary features with deeper network features to stabilize the localization of forest boundaries and refine areas of uncertainty,thus boosting edge discernment.Finally,to verify the validity of the method,a new dataset is created and case test is analyzed.The results indicate that the Intersection over Union(IoU)and pixel accuracy(PA)of this method are improved by 0.55%to 9.45%,and by 0.19%to 7.53%,respectively.The analysis demonstrates that the proposed method has better boundary integrity in the face of complex scenes,and can be better applied to forest land extraction.

关 键 词：林地提取 上下文信息 边界约束 遥感影像 语义分割 

分 类 号：S771.8[农业科学—森林工程]