RUI: Collaborative Research: CDS&E: A Modular Multilayer Framework for Real-Time Hyperspectral Image Segmentation

对图像的分析已经被科学界用来解决具有挑战性的问题，并深入了解各种自然、社会和技术现象。不同类型的图像被用于不同的研究领域。其中一个例子是高光谱图像，与传统相机图像相比，它具有更高的分辨率。分析这样的图像有其挑战。例如，它的计算量很大，传统的方法有一定的局限性。该项目通过利用高性能计算工具和机器学习技术，为分析此类图像提供了有效的解决方案。将所得方法应用于基于图像的大气云检测。该项目为高光谱图像开发了一个实时、多层和模块化的分割框架。开发的框架通过对图像的每个像素进行分类并将其与类段相关联来自动识别高光谱图像中的各个区域。开发的系统是多层的，其中每一层的职责是对其输入执行操作，生成区域分类数据，并将结果输出传递给下一层。重要的是，每一层都从不同的角度分析其输入，利用光谱和空间数据，形成多层框架，其中各层相互补充。此外，本项目旨在为实时高光谱图像分析提供优化的高性能（加速和准确性）计算工具。这是通过调整模型不同部分中使用的算法来实现并行处理的。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The analysis of images has been used by the scientific community to solve challenging problems and to get insight into diverse natural, social, and technical phenomena. Different types of images have been employed in various areas of study. One example is the hyperspectral images, which have higher resolution when compared to conventional camera images. Analyzing such images has its challenges. For instance, it is computationally demanding, and traditional methods have some limitations. This project provides an efficient solution to analyze such images, by exploiting high-performance computing tools and machine learning techniques. The resulting methods are applied to image-based atmospheric cloud detection.The project develops a real time, multi-layer, and modular segmentation framework for hyperspectral images. The developed framework automatically identifies various regions within a hyperspectral image by classifying each pixel of the image and associating them to class segments. The developed system is multi-layer, where each layer’s responsibility is to perform an operation on its input, generate region classification data, and pass the resultant output to the next layer. Importantly, each layer analyzes its input from distinct viewpoints, utilizing spectral and spatial data, resulting in a multi-layer framework where the layers complement each other. Also, this project aims to provide an optimized high-performance (speed-up and accuracy) computational tool for real-time hyperspectral image analysis. This is achieved by adapting the algorithms used in the different parts of the model for parallel processing.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.