Artefact suppression for explainability

伪影抑制以提高可解释性

• 批准号: 2887793
• 金额: --
• 依托单位: University of Bristol
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2887793
• 关键词: Artefact; suppression; explainability

The objective of the project would be to develop a general machine learning method for removing benign artefacts (e.g. geometric reflections) from NDE data and demonstrate its use for specific inspection tasks, whilst also considering the explainability of the applied machine learning algorithm to real-world measurements in the Non-Destructive Evaluation (NDE) context. Therefore, I shall be seeking a more general approach to artefact suppression with applied machine learning methods that 1) Generalises easily to different modalities; and 2) Predicts artefacts with sufficient fidelity allowing them to be subtracted from a measurement (rather than simply masking that part of a measurement), thus allowing defect signals to pass even if they wholly or partly coincide with an artefact.

该项目的目标是开发一种通用的机器学习方法，用于从无损检测数据中去除良性伪影（例如几何反射），并演示其用于特定的检测任务，同时考虑应用机器学习算法在无损评估（NDE）环境中对实际测量的可解释性。因此，我将通过应用机器学习方法寻求一种更通用的伪影抑制方法，1)易于推广到不同的模态；2)以足够的保真度预测工件，允许从测量中减去它们（而不是简单地掩盖测量的那一部分），从而允许缺陷信号通过，即使它们全部或部分地与工件一致。