Grantee Research Project Results

Final Report: Real-time red meat freshness assessment using multi-mode spectroscopy and fusion-based machine learning

EPA Contract Number: 68HERC24C0021
Title: Real-time red meat freshness assessment using multi-mode spectroscopy and fusion-based machine learning
Investigators: Vasefi, Fartash
Small Business: SafetySpect Inc
EPA Contact: Richards, April
Phase: I
Project Period: December 1, 2023 through May 30, 2024
Project Amount: $99,093
RFA: Small Business Innovation Research (SBIR) - Phase I (2024) RFA Text |  Recipients Lists
Research Category: Small Business Innovation Research (SBIR)

Description:

This Phase I SBIR project, sponsored by the EPA, addresses the critical issue of food waste prevention in the food supply chain, focusing specifically on real-time food freshness assessment. By leveraging multimode spectroscopy measurements acquired through a handheld scanner, the project aims to estimate the freshness and shelf-life of red meat. This capability enables better management of storage, transportation, and pricing strategies. The combination of spectroscopy and machine learning makes the automatic freshness assessment technology suitable for unskilled workers. The project integrates custom-designed hardware, user interface software, and fusion AI to deliver this innovative solution.

Summary/Accomplishments (Outputs/Outcomes):

Though effective, current meat quality assessment methods are time-consuming and require specialized personnel. Throughout Phase I, we developed hardware that includes three modes of spectroscopy (reflectance in Visible and Near-Infrared ranges, and fluorescence spectroscopy) with a user-friendly interface for reliable data collection. This device was calibrated using radiometrically calibrated light sources to ensure measurement accuracy. Exploratory measurements were conducted on beef, pork, and lamb samples from a trusted local shop.

Spectra were captured using three illuminations covering visible and near-infrared (VisNIR) and short-wave infrared (SWIR) ranges. Various precautions were taken to ensure data consistency, including white and dark signal recordings, avoiding fat and bone areas during measurements, and maintaining consistent storage conditions. Studies revealed that plastic wraps over meat samples had minimal impact on measured spectra, validating their use in retail or restaurant settings. The collected data were analyzed for consistency and compared with literature data, identifying significant spectral peaks corresponding to various biomarkers that aid in assessing meat freshness and quality. Machine learning algorithms were employed for classification based on spectral data, achieving promising results. Linear Discriminant Analysis (LDA) emerged as the top-performing model, exhibiting high accuracy and efficiency across various spectral ranges.

Data fusion and normalization techniques were utilized to enhance classification performance, with the SVM model demonstrating superior accuracy for VisNIR and SWIR spectral ranges. ANOVA analysis identified significant wavelength bands associated with biochemical changes in meat, providing valuable insights for future research.

The mechanical design features a measuring head equipped with UV LEDs and micro incandescent lamps, with light emitted from samples captured by optical fibers and directed to micro-spectrometers. The device's components are housed in a black anodized aluminum enclosure, ensuring stability and ease of use. The electrical design includes readout circuit boards connected to micro-spectrometers and LEDs, powered by rechargeable Li-ion batteries for prolonged operation. Software developments include a Python-based program with functionalities for controlling light sources, reading spectrometers, displaying spectra, and storing data, all accessible through a graphical user interface. Experimental results have shown successful reflectance measurements of mineral oil and water emulsions, with distinct features identified in the NIR range. Future improvements will focus on optimizing the circuit board architecture, implementing Bluetooth Low Energy (BLE) for wireless communication, and enhancing the user interface for better usability.

Conclusions:

These updates have also been reflected in our business requirements document (BRD), which describes the product and includes a list of necessary functionality features for the device's development. The commercialization aspect of this project involved several key initiatives to establish a solid foundation for bringing the technology to market. Participation in the George Washington University Regional I-Corps program facilitated valuable insights through 20 different interviews, confirming the demand for technology-driven solutions in assessing red meat freshness. Retailers expressed interest in predicting freshness degradation, highlighting the need for advanced tools beyond sensory inspection methods. Conversations with meat processing facilities and regulatory bodies revealed industry practices and standards, presenting opportunities for non-destructive, rapid, and cost-effective technologies. Discussions with academic experts showcased the potential for machine learning-enhanced prediction models.

Market analysis identified existing competitors offering multi-purpose spectroscopy devices but lacking all the functional capabilities of our QAT device. SafetySpect’s engagement with industry leaders such as a leading meat producer in Brazil may provide valuable partnerships for future pilot studies and market penetration strategies. They can also provide insights into the device's usability in a real-world environment. The foundation for commercializing our product was significantly bolstered by services from technical and business assistance (TABA). We utilized services for a commercialization plan review, IP landscape review, and website and social media presence review. The IP review provided a comprehensive overview of our IP portfolio, highlighting our strong and unique position in the market for meat freshness indication. The commercialization plan review helped us draft an early version, defining our target markets and estimating market sizes, which informed us of our financial projections. Finally, the website and social media review offered valuable recommendations for improving our online presence, emphasizing the benefits and practical applications of our technology to engage visitors and industry stakeholders effectively. These insights will be integrated into our marketing strategy, setting a clear path for successful commercialization.