Advanced image analysis of semiconductor structures

算法和计算能力的最新进展产生了新的方法，对于某些问题，这些方法可以在数据分析，图像处理和模式识别方面超越人类的专业知识。这些技术已被用于检查和分析单个和多个图像以及视频片段中的特征。这些机器学习方法开始彻底改变包括天文学，医学和机器视觉在内的领域。本项目的目的是应用这些新技术中的一些技术来快速和准确地检查由半导体和光谱学（SSD）小组使用纳米级扫描电子探针和半导体纳米结构的显微镜测量产生的实验数据。本项目将包括开发新的标准技术来处理单个和多个图像数据集。这将与现有的高光谱分析工具和现代机器学习相结合，以提取有关先进材料纳米级特征的形态和光谱行为的有意义的信息。具体而言，通过减少在寻找重要图像特征和提取有意义的结构-性质相关性和成因特征方面的人为干预水平，将加速氮化物半导体纳米结构的结构和相关光谱性质的分析。该项目的主要目的是：1）规范SSD组图像的对齐和处理，开发一个一致的管道，用于多图像和多技术数据，可以与CHIMP高光谱工具在组中开发的集成。2）发展学生在应用监督和非专业知识，监督机器学习，从英国和国际合作伙伴的氮化物结构测量数据中提取重要特征。应用最近开发的统计测试来区分测量数据中的因果关系，以发现导致短波长发光结构中高效发光的原因。

Recent advances in algorithms and computing power have produced new methods which, for some problems, can outperform human expertise in data analysis, image processing and pattern recognition. These techniques have been used to examine and analyse features in single and multiple images, as well as video footage. These machine learning approaches are beginning to revolutionise areas including astronomy, medicine and machine vision. This project aims to apply some of these new techniques to rapidly and accurately examine experimental data generated by the Semiconductor and Spectroscopy (SSD) Group using nanoscale scanning electron probe and microscopy measurements of semiconductor nano-structures.This project will consist of developing new standard techniques for image processing of single and multi-image data sets. This will be coupled with existing hyperspectral analysis tools and modern machine learning to extract meaningful information on the morphology and spectral behaviour of nanoscale features of advanced materials. Specifically analysis of the the structural and correlated spectral properties of nitride semiconductor nanostructures will be accelerated by reducing the level of human intervention in finding the significant image features and extracting meaningful structure-property correlations and causative features.Primary aims of the project are to:1) Standardise the alignment and processing of images in the SSD group, developing a consistent pipeline for use on multi-image and multi-technique data that can be integrated with the CHIMP hyperspectral tools developed in the group.2) Develop student expertise in applying supervised and un-supervised machine learning to extract the significant features from measured data on nitride structures from UK and international partners.3) Apply recently developed statistical tests that distinguish cause and effect in measured data to discover what leads to efficient light production in short wavelength light emitting structures.