Hyperspectral imaging (HSI) is emerging as a potential tool for non-contact, non-destructive high-throughput analysis of products and processes. The main benefit of HSI over point spectroscopy techniques is the possibility to explore both spatial and chemical information present in samples. However, there are challenges associated with using HSI as an automated tool for high-throughput applications.This study involves the application of near-infrared (NIR) HSI and data analysis methodologies for assessment of tea products. Assessment of tea products involved the development of support vector machine classification models for different tea products sourced from Unilever, UK. The study further involves covering challenges related to handling, managing and processing of HSI data in an automatic sense.Methodologies related to automatic pre-processing, compression and data fusion for assessment of tea products are presented.The findings showed that NIR HSI can be used for the classification of tea products with an accuracy of more than 97%. The pre-processing methodologies developed provided automatic removal of noise from the HS images. The shearlet-based de-noising method improved the classification accuracy by 11% and 19% compared to Savitzky Golay and median filtering, respectively.The total variation based de-noising method removed the dead strips from HS images, which is not achievable with traditional methods. The compression scheme devised, which utilised 2D wavelets and variance decomposition methods, gave a reduction in the size of HS images by a factor of 40. Finally, a methodology based on grey-level co-occurrence matrices is presented for extracting textural information from HS images. The fusion of textural and NIRinformation showed an improvement in classification accuracy compared to use of spectral or textural information alone.The methodologies presented will support implementation of high-throughput HSI forassessment of tea products and processes.
|Date of Award||3 May 2019|
- University Of Strathclyde
|Supervisor||Alison Nordon (Supervisor) & Anthony Morris (Supervisor)|