TY - GEN
T1 - Development of Scanning and Snapshot Hyperspectral Imaging Systems for Industrial Sorting Machines
AU - Peters, Mads
PY - 2024/10/31
Y1 - 2024/10/31
N2 - This industrial Ph.D. project, a collaboration between Newtec Engineering A/S
and SDU NanoSyd at the University of Southern Denmark, focuses on advancing hyperspectral imaging (HSI) technology to overcome the limitations of traditional optical sorting methods. Supported by Innovation Fund Denmark, the research primarily aims to develop, optimize, and implement hyperspectral snapshot and scanning cameras, with a particular emphasis on the snapshot computed
tomography imaging spectrometer (CTIS).The project successfully designed, fabricated, and tested a custom snapshot
CTIS prototype capable of real-time hyperspectral datacube reconstruction in the
600 to 850 nm range. Significant enhancements were made to the system’s mechanical and optical design, including the integration of a custom near-infrared
(NIR) diffractive optical element, advanced 3D-printed enclosures, and the incorporation of a GSENSE2020 image sensor with improved NIR sensitivity. Quality
control procedures for assembly, alignment, and validation were also developed
to ensure reliable performance.To improve reconstruction accuracy and processing speed, iterative algorithms,
including the Expectation Maximization (EM) algorithm, were implemented. Additionally, artificial neural network (ANN)-based reconstruction models were developed, including convolutional neural networks (CNNs) and hybrid iterativeCNN models. These models provided enhanced accuracy and enabled video-rate
snapshot hyperspectral imaging. A CTIS simulator was also created, allowing for
the simulation of CTIS images based on real system parameters, facilitating the
generation of training data for ANN models and the exploration of new CTIS
designs.Throughout the project, Newtec’s line-scan hyperspectral imaging systems
served as reference points, particularly in a comparative study on predicting °Brix
and pH values in grapes. While the line-scan system outperformed in predictive
accuracy, the CTIS system delivered reasonable estimates, especially for °Brix,
demonstrating its potential for in-field applications.Beyond industrial applications, the CTIS system was applied to cultural heritage studies, revealing previously unknown features in Vilhelm Lundstrøm’s
paintings and enabling pigment identification. Furthermore, a new hyperspectral microscopy system was developed using Newtec’s Vis-SWIR Oculus camera,
offering detailed spatial and spectral analysis across a spectral range of 450 to
1700 nm. In summary, this thesis establishes a solid foundation for future research and
development in snapshot hyperspectral imaging, contributing valuable advancements to both theoretical knowledge and practical applications across various
scientific, industrial, and cultural domains.
AB - This industrial Ph.D. project, a collaboration between Newtec Engineering A/S
and SDU NanoSyd at the University of Southern Denmark, focuses on advancing hyperspectral imaging (HSI) technology to overcome the limitations of traditional optical sorting methods. Supported by Innovation Fund Denmark, the research primarily aims to develop, optimize, and implement hyperspectral snapshot and scanning cameras, with a particular emphasis on the snapshot computed
tomography imaging spectrometer (CTIS).The project successfully designed, fabricated, and tested a custom snapshot
CTIS prototype capable of real-time hyperspectral datacube reconstruction in the
600 to 850 nm range. Significant enhancements were made to the system’s mechanical and optical design, including the integration of a custom near-infrared
(NIR) diffractive optical element, advanced 3D-printed enclosures, and the incorporation of a GSENSE2020 image sensor with improved NIR sensitivity. Quality
control procedures for assembly, alignment, and validation were also developed
to ensure reliable performance.To improve reconstruction accuracy and processing speed, iterative algorithms,
including the Expectation Maximization (EM) algorithm, were implemented. Additionally, artificial neural network (ANN)-based reconstruction models were developed, including convolutional neural networks (CNNs) and hybrid iterativeCNN models. These models provided enhanced accuracy and enabled video-rate
snapshot hyperspectral imaging. A CTIS simulator was also created, allowing for
the simulation of CTIS images based on real system parameters, facilitating the
generation of training data for ANN models and the exploration of new CTIS
designs.Throughout the project, Newtec’s line-scan hyperspectral imaging systems
served as reference points, particularly in a comparative study on predicting °Brix
and pH values in grapes. While the line-scan system outperformed in predictive
accuracy, the CTIS system delivered reasonable estimates, especially for °Brix,
demonstrating its potential for in-field applications.Beyond industrial applications, the CTIS system was applied to cultural heritage studies, revealing previously unknown features in Vilhelm Lundstrøm’s
paintings and enabling pigment identification. Furthermore, a new hyperspectral microscopy system was developed using Newtec’s Vis-SWIR Oculus camera,
offering detailed spatial and spectral analysis across a spectral range of 450 to
1700 nm. In summary, this thesis establishes a solid foundation for future research and
development in snapshot hyperspectral imaging, contributing valuable advancements to both theoretical knowledge and practical applications across various
scientific, industrial, and cultural domains.
U2 - 10.21996/630e-gr84
DO - 10.21996/630e-gr84
M3 - Ph.D. thesis
PB - Syddansk Universitet. Det Tekniske Fakultet
ER -