Development of a Deep Ultraviolet Hyperspectral Camera for Pollen Detection on Insects

Abstract

The deep ultraviolet (DUV) spectral region, from 200-400 nm, can reveal unique information about biological samples, where proteins and biomolecules exhibit distinct spectral signatures in this wavelength region. However, because of technical limitations, the absence of stable DUV light sources has left the research in this spectral range largely unexplored. Recent advancements in laser-driven plasma light sources (LDLS) have made extensive DUV spectroscopy studies a practical possibility.

In this project, I developed a DUV push-broom line-scanning hyperspectral camera, featuring a custom DUV spectral analyser DUVI (47 effective bands, 200-400 nm) and integrated with a UV-sensitive camera. Parallel to this, custom illumination optics were developed to collimate the LDLS light source and shape the LDLS output into a focused line that matches the pixel footprint of the hyperspectral camera. The design of the DUV spectrum analyzer DUVI and collimation optics for the light source employs ray-tracing (Zemax) and is mounted within a CAD-designed, 3D-printed sandwich structure. The entire assembly is integrated with a motorized translation platform, which translates the imaging subject across the scan area.

To ensure the spectral accuracy of the DUV hyperspectral camera and eliminate 'smile' (spectral bending) and 'keystone' (spatial scaling) distortions, a two-step calibration was performed. This involved wavelength assignment using a mercury lamp and spatial slit-position correction. Following this, a series of hyperspectral images of a bumblebee with pollen loads was captured. Subsequent analysis of the hyperspectral data cubes allowed for the extraction of DUV reflectance signals from various biological structures. We identified distinct spectral reflectance patterns corresponding to the bumblebee's bare exoskeleton, the Yew tree pollen, and the natural pollen load. Furthermore, false-color RGB images were generated to visually highlight these spectral differences, which correspond to variations in biochemical composition, distinguishing the chitin-rich exoskeleton from the pollen. To provide deeper classification, principal component analysis (PCA) techniques were employed to statistically distinguish between the different pollen and visualize their specific distribution across the bumblebee.

These results provide a foundation for future research, as the hyperspectral camera will later be used to characterize a diverse array of pollen species, constructing a comprehensive spectral reference library. This database will subsequently support the development of a deep-ultraviolet Scheimpflug lidar system, designed to remotely detect and identify pollen on insects in flight.

Popular Abstract

Seeing Pollen on Insects using a Deep Ultraviolet Hyperspectral Camera

In this project, I developed a specialized camera to capture the deep ultraviolet spectrum, which is invisible to the human eye. This device is designed to detect spectral signatures of microscopic pollen particles, revealing invisible biological information that is otherwise hidden under visible light.

Pollen is almost everywhere in nature. It drifts through the air we breathe and clings to various insects. Insects such as bees serve as their carriers, and the pollen attached to them is vital for plants reproduction. Understanding these pollination networks is crucial for agriculture, as identifying which plants serve as pollen sources for different pollinators helps farmers secure better crop yields. However, monitoring pollen distribution and identifying its source has always been challenging. The difficulty arises not only from their microscopic size but because pollen grains often share similar colors and shapes, making it nearly impossible to trace their origin in the field through standard visual inspection. The solution lies at the short-wavelength side of the spectrum: deep ultraviolet (DUV) light. Just like satellite sensors use infrared bands to distinguish between different types of vegetation on Earth, biological molecules inside the pollen, such as proteins, possess unique "spectral fingerprints" when analyzed through DUV reflectance (200-400 nm). By capturing these signatures, the camera can identify pollen grains that are otherwise indistinguishable under visible light.

In this degree project, I tackled the challenge of capturing these signals working as both a scientist and an engineer. I built the DUV hyperspectral imaging system from scratch. The process began with optical design and ray tracing simulations to optimize the light path, followed by mechanical modeling in CAD, and finally 3D printing of the structure for assembly. Its core component is a powerful laser-driven light source (LDLS) that emits exceptionally bright, stable ultraviolet light. Alongside this is the DUV spectral analyser DUVI, which can discern spectral features from targets as small as pollen grains. This DUV hyperspectral camera captures "data cubes" containing both spatial and spectral dimensions. This allows us to extract a detailed spectral spectrum for every pixel, enabling us to characterize substances based on both visual appearance and spectral characteristics.

To test this system, we chose a bumblebee with a pre-existing pollen load as our primary test subject. We also applied yew tree pollen to the bumblebee's body. Using this system, we were able to capture distinct reflectance signatures from the yew pollen, natural pollen load already present in the specimen pollen basket, and the insect’s chitin-rich body. We then performed Principal Component Analysis (PCA) to identify unique spectral patterns that distinguish these different materials. Furthermore, these spectral variations were mapped into a false-color image. By assigning specific DUV bands to visible color channels, we created a spatial map that translates invisible spectral differences into high-contrast visual data, and allowing for a clear discrimination of the pollen grains from the insect body.

The next phase of this research is to document different pollen types to build a comprehensive "pollen fingerprint library". In the future, this technology can be redirected for remote hyperspectral lidar system operating in the field to detect pollens in the air or on insects in flight. By "making the invisible visible," we are redefining the boundaries of environmental perception. This DUV hyperspectral camera transcends the limits of visible light perceivable by the human eye. It allows us to glimpse the ultraviolet environment invisible to the naked eye and capture and characterize the microscopic particles silently flowing around us.

• author: Sun, Shijun
• supervisor: MEng Li; Mikkel Brydegaard
• organization: Combustion Physics; Department of Physics
• course: PHYM03 20252
• year: 2026
• type: H2 - Master's Degree (Two Years)
• subject: Physics and Astronomy; Biology and Life Sciences; Technology and Engineering
• language: English
• id: 9220692
• date added to LUP: 2026-01-23 14:27:20
• date last changed: 2026-01-23 14:27:20