摘要
为了提高机器学习中遥感图像样本的规范水平与采集效率,方便样本动态检索与动态采集,文章根据机器学习样本采样的规范性要求,构建了尺度可变、跨越平台的遥感图像样本采集方案,然后采用C#与数据库相结合的方式实现了样本图像分类、动态采集、结果可视、同时录入样本信息的样本采集模式,并完成了原型工具开发。研究结果表明:采集尺度程序化控制、与平台无关的截图方案可以根据学习要求任意改变图像截取标准高效地完成样本采集。样本信息分类化采集、管理、可视化浏览可以快速地实现机器学习样本库动态管理与问题排查,且为进一步划分训练集、测试集和验证集提供较好技术支撑。
In order to improve the standardization level and acquisition efficiency of remote sensing image samples in machine learning and facilitate dynamic retrieval and dynamic acquisition of samples,this paper constructs a scale-variable and platform-independent remote sensing image sample acquisition scheme according to the standardization requirements of sample sampling in machine learning,and then adopts C#and database to realize the image classification,dynamic acquisition,andresultvisualization of samples,as well as simultaneous entry of sample information.The results show that the programmed control of the acquisition scale and the platform-independent screenshot scheme can efficiently complete the sample acquisition by changing the image capture criteria arbitrarily according to the learning requirements.Categorized collection,management and visualization of sample information can quickly realize dynamic management and problem identification of machine learning sample base,and provide better technical support for further classification of training set,test set and validation set.
出处
《科技创新与应用》
2021年第28期32-34,38,共4页
Technology Innovation and Application
基金
吉林大学大学生创新创业训练计划项目(编号:201910183092)。
关键词
机器学习
遥感图像样本
动态采集
分类管理
machine learning
remote sensing image samples
dynamic acquisition
classified management
