Target-based multiple-scale change detection from time series remotely sensed environmental data

遥感数据已广泛应用于环境监测和健康领域。将多时相遥感数据集的光谱和模式变化与地物和生物物理变量的变化联系起来一直是遥感技术的重要应用。虽然已经发展了各种变化检测方法，但它们主要是基于对从相同或相似传感器获取的多时相图像进行图像分析。随着遥感技术的快速发展和大型多尺度、多传感器对地观测数据集的日益可用性，变化检测已经被迫纳入这些数据集，需要更高效的变化检测方法和强大的工具来应对这一挑战。本研究基于我们提出的基于目标的变化检测的创新概念，旨在通过以下方法提高变化检测的精度：(i)评估不同光谱解混和亚像元方法在不同尺度上绘制目标土地覆盖及其变化的性能；（ii）通过数据挖掘和深度学习，开发智能变化检测程序，从不同景观的不同目标的多光谱、多尺度、多时相遥感数据立方体中提取数据；（iii）在涉及多尺度和多传感器时间图像时，开发用于混合变化检测的评估框架。本研究的长期目标是为更好地理解变化检测中的误差和不确定性以及从大量多传感器时间序列RS图像和产品中检测模式和变化提供理论框架和强大的工具。本研究提出的创新变化检测方法和评估框架将极大地促进RS数据和产品在许多应用中的最大利用。在建议的预算和研究范围内，将培养至少10名研究生和10名本科生。

Remotely sensed (RS) data has been widely used in environmental monitoring and health applications. Linking the spectral and pattern changes among multi-temporal RS data sets to changes in surface features and biophysical variables has long been an important application of remote sensing technology. Although various change detection methods have been developed, they are mainly based on image analysis on multi-temporal images acquired from the same or similar sensor. With the rapid development of remote sensing technology and the increasing availability of large multi-scale and multi-sensor earth observation data sets, change detection has been forced to incorporate those data sets, and more efficient change detection methods and robust tools are needed to meet the challenge. This research is based on the innovative concept of target-based change detection we have proposed and is aimed to improve the accuracy of change detection by (i) evaluating the performance of different spectral unmixing and sub-pixel methods for mapping targeted land cover and its change across different scales; (ii) developing intelligent change detection procedures through data mining and deep learning from multi-spectral, multi-scale, multi-temporal RS data cubes for different targets on different landscapes; and (iii) developing an evaluation framework for hybrid change detection when multi-scale and multi-sensor temporal images are involved. The long-term goal of the proposed research is to provide theoretical frameworks and robust tools for better understanding the error and uncertainty in change detection and detecting patterns and changes from the large amount of multi-sensor time series RS images and products. The innovative change detection methods and evaluation framework from the proposed research will significantly contribute to the maximum use of RS data and products in many applications. Within the proposed budget and research scope at least ten graduate students and ten undergraduate students will be trained.