基于PyTorch深度学习框架的武汉市森林资源变化监测模型研究  

Research of a Forest Resources Change Monitoring Model for Wuhan City Based on the PyTorch Deep Learning Framework

作　　者：邓春成 孙巧峰 唐星 李艳丽 卢秉俊 陶章思 DENG Chuncheng;SUN Qiaofeng;TANG Xing;LI Yanli;LU Bingjun;TAO Zhangsi(Wuhan Municipal Forest Fire Monitoring and Early Warning Center,Wuhan 430023,Hubei,China;Beijing Dilin Weiye Forest Technology Corp.,Ltd.,Beijing 100043,China)

机构地区：[1]武汉市森林防火监测预警中心,湖北武汉430023 [2]北京地林伟业科技股份有限公司,北京100043

出　　处：《中南林业调查规划》2025年第1期45-49,共5页Central South Forest Inventory and Planning

基　　金：武汉市园林和林业局科技计划项目“基于遥感+AI森林资源变化监测模型研究”(No.WHGF2023A08)。

摘　　要：基于武汉市2期遥感影像数据,采集变化识别样本,采用深度学习框架PyTorch对样本模型进行训练、效果预测,构建变化监测模型,利用变化监测模型进行遥感影像变化信息的提取,形成变化提取成果。结果显示:武汉市森林资源变化识别准确率达88.1%,召回率达88.5%,且随着样本数量的增加和影像质量的提升,基于深度学习框架的变化识别提取准确率和召回率也得到了提升。构建的武汉市森林资源变化监测模型可应用于森林资源变化识别,能够为森林资源监测监管提供疑似变化图斑,及时准确地发现森林违法活动、林业灾害等信息,具有较大的实际应用价值和潜力。Based on two-phase remote sensing image data from Wuhan City,this study collected change detection samples and employed the PyTorch deep learning framework for model simulation.The process involved training the sample model and predicting its effectiveness in constructing a change detection model.This model was subsequently used to extract change information from remote sensing images,resulting in change extraction outcomes.The results demonstrated that the accuracy of forest resource change detection in Wuhan City reached 88.1%,with a recall rate of 88.5%.As the number of samples increased and image quality improved,the accuracy and recall rate of change detection extraction based on the deep learning framework were also improved.The constructed change detection model for monitoring forest resources provides suspected change patches for forest resource monitoring and supervision.It enables the timely and accurate detection of illegal forest activities,forestry disasters,and other issues,thereby exhibiting substantial practical application value and potential.

关 键 词：深度学习 遥感影像 森林资源 变化监测模型 PyTorch 

分 类 号：S771.8[农业科学—森林工程]