Grantee Research Project Results
Final Report: A High-speed, Low Cost, Machine Learning Enhanced, Hyperspectral Imaging System for Improved Identification of Microplastics
EPA Contract Number: 68HERC23C0011Title: A High-speed, Low Cost, Machine Learning Enhanced, Hyperspectral Imaging System for Improved Identification of Microplastics
Investigators: Harsh, Kevin
Small Business: Sporian Microsystems Inc.
EPA Contact: Richards, April
Phase: I
Project Period: December 1, 2022 through May 31, 2023
Project Amount: $99,981
RFA: Small Business Innovation Research (SBIR) Phase I (2023) RFA Text | Recipients Lists
Research Category: SBIR - Homeland Security , SBIR - Sustainability , SBIR - Air and Climate , SBIR - Water
Description:
Microplastics are environmental pollutants ubiquitous in marine and freshwater environments and have been established to have deleterious impacts on aquatic life and human health. Microplastics (broadly defined as plastic particles less than 5mm in diameter) originate from a variety of sources either from degradation and fragmentation of larger plastics or by direct release into the environment. As the production of plastics continues to grow, there is an emerging need for innovative technologies to efficiently detect and quantify microplastics in support of environmental monitoring and removal processes. Chemical analytical methods and FTIR/Raman Spectroscopy are the traditional lab-scale methods for providing chemical information and classification of microplastics, but such methods are slow and very costly due to equipment and labor costs. An ideal technology would rapidly and inexpensively characterize environmental samples of microplastics in environmental matrices, such as water, wastewater, or soil, and utilize artificial intelligence or other methods to automate rapid (sub-second) classification, identification, and quantification. A detection technology that supports low-cost integration with a number of platforms for various use cases, including remote field monitoring, handheld point-of-use assessments, and lab-scale rapid-throughput sample measurement, would be highly advantageous.
Hyperspectral imaging (HSI), where imaging and spectral scanning are combined to provide spatially represented spectral information, is emerging as a non-destructive, real-time detection tool for industrial and environmental sensing and inspection processes. It would be a transformational step if inexpensive (<$10K), high-speed, continuous-range hyperspectral (350-2600nm) measurement systems could be implemented in microplastic measurement form-factors such as field-portable handheld implementations and high-speed automated sample throughput measurement systems while supporting easy integration with conventional imaging and microscopy optical front ends. The long-term goal of the proposed work is to realize such a hyperspectral measurement system by utilizing a novel optical system designed using a set of low-cost components to achieve the cost, size, and performance goals as described above. The novel optical design originates from Sporian’s prior development efforts on low-cost, compact, hyperspectral measurement systems, combined with embedded advanced machine learning (ML) based materials classification algorithms to perform on-hardware classification.
Figure 1: Sporian’s current commercial BroadSpec™ HSI product for small UAS deployment leveraged for this effort
Summary/Accomplishments (Outputs/Outcomes):
The Phase I effort focused on: an in-depth study of how the technology can and should be adapted for portable and high throughput scanning microscopy in target use cases; image and spectral ML algorithm evaluation; detailed design development; and experimental testing/demonstration of modified measurement hardware/firmware with relevant materials targets.
All Phase I proposed objectives and milestones were successfully completed. Results are summarized below by proposed task/technical objective.
Task/Technical Objective 1: Work with technical partners and potential end users to guide the development of useful implementations of the proposed technology and facilitate transition efforts.
Efforts to define the technical and market application space through stakeholder identification and engagement resulted in a clear definition of near-term application use-cases and needs within research and water/wastewater industries. Identifying stakeholder needs and technical challenges associated with those applications led to the definition of technical and functional requirements for application-specific systems, subsequently used to guide system design efforts (Task 3). Market drivers and high value/high impact commercial points of entry were identified for research and water/wastewater industries, as well as other adjacent industries.
Task/Technical Objective 2: Use benchtop-scale prototype of the HSI system with relevant materials to test/demonstrate measurement and ML-based classification performance.
A first-generation, optical bench-top scale version of a hyperspectral microscopy system was built and used for demonstrating the feasibility of the proposed technology. The prototype hardware was used to generate spectral data for a range of microplastics (materials and sizes) as well as other relevant materials, and combined with ML-trained models to successfully demonstrate:
- Micron scale particle size limits of detection.
- The ability to correctly discern/classify microplastic particles from non-plastic particles (dirt, sand, glass, and biological) without issue.
- The ability to correctly discern/classify the specific type of microplastic particles from a range of 17 common plastics types without issue.
- The ability to correctly discern/classify sub types of specific plastic materials with a high degree of confidence.
- That while the size of the particles can affect the measured spectra, with proper training data these differences do not adversely impact the ability to classify the type of material.
While these were preliminary examples of detection/classification capability, these demonstrations clearly showed the feasibility and value of the proposed technology for microplastics detection and characterization.
Figure 2. Example of how, using the reflectance microscope-based setup, HS measurement spot was moved to several different locations (particles and algae) and spectra generated for subsequent processing/classification.
Figure 3. Example data from 17 different types of microplastics, showing how experimental data was used to train ML models and demonstrate ability to correctly classify particles. (Left) Raw measured spectral data from microplastics samples, (middle) the data transformed for ML algorithm training and classification, and (right) the results of testing model’s ability to accurately classify material types, where values represent % samples by label true (actual) vs predicted. The model was 100% accurate for every class/type of material.
Task/Technical Objective 3: Leveraging prior work and the results of experimental efforts, develop optimized application-specific measurement hardware, electronics, and firmware designs for Phase II realization.
Experimental results, technical requirements, and use case details were then used to evaluate and define application-specific HSI hardware designs that could meet target technical and functional specifications, including detailed designs for:
- The overall system architecture.
- All key mechanical/optical subcomponents and subsections, including spectroscopic, image scanning, and illumination hardware designs.
- Electrical systems and firmware/software architecture. This included the design and prototyping of initial system electronics to verify feasibility of high-data-rate throughput in a low-cost format.
- An overall full-system hardware packaging/integration design that can be realized early in subsequent efforts.
This also included initial hardware cost estimates, which were found to be within cost targets defined by stakeholders.
As a result of this Phase I effort, Sporian is well positioned for the Phase II efforts focused on full system prototyping and relevant environment testing/demonstration
Conclusions:
Sporian defined potential end markets as researchers and scientists focused on environmental studies (early adopters) and stakeholders within the industrial water distribution/treatment industry (primary initial commercial market). Through prior commercial efforts related to water/wastewater monitoring applications. Sporian has significant stakeholder relationships and market information for end-market assessment. Through industry stakeholder interviews, Sporian identified technology gaps hindering the understanding of microplastics’ effects on the environment and human health and the ability of water treatment operators and their technology providers to respond to potential regulations. Beyond these initial applications, the proposed low-cost materials inspection/characterization product has application in wide range of additional industries with similar technical and economic considerations, including food inspection/safety, pharmaceuticals, biotechnology, industrial machine vision, and semiconductor QC/QA.
By supporting the monitoring and active reduction of microplastics in the environment, the environmental benefits of the proposed technology’s use are substantial. Microplastics have significant environmental impacts because they migrate throughout marine and land ecosystems, are ingested either directly or indirectly by all levels of life forms, and have been found to have negative impacts on those life forms. In addition, other toxic pollutants tend to collect on the surface of microplastics and are then ingested along with the microplastics, accumulating in the consuming organism and slowly making their way up the food chain including into humans. Microplastics make their way into drinking water as well as foods like salt, honey, and sugar. Research suggests that humans are consuming more than 100,000 microplastics particles a year and there are still many unanswered questions about the impacts of microplastics on human health. Further studies in this area will require effective microplastic detection and characterization. In addition to monitoring, the proposed technology could support plastics/microplastics removal and recycling processes. The environmental costs associated with the lifecycle of this technology are minimal. All aspects of the hyperspectral suite design are tailored to be minimally harmful and inert in use case scenarios. With no use of reagents or toxic chemicals, as well as with its ability to identify materials in real-time with improved efficiency and effectiveness, this technology will offer more environmental benefits than costs.
The perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.