Hyperspectral imaging for cocoa material classification on conveyor belts
This technology leverages hyperspectral imaging (HSI) to accurately classify cocoa bean fractions like nibs and shells on conveyor belts, using their unique spectral signatures. It aims to enhance precision in sorting processes through advanced imaging and machine learning.
Overview
The proposed solution utilizes hyperspectral imaging (HSI) to accurately classify and differentiate cocoa bean fractions such as nibs, shells, and fine particles on conveyor belts. By exploiting their unique spectral signatures, this technology enhances precision in sorting processes. The capability to detect subtle differences among visually similar small objects enables high-precision classification and optimization of cocoa processing.
Technical specifications
Key features:
- Utilizes short-wave infrared (SWIR) hyperspectral imaging (950-1700nm) for high spatial and spectral resolution.
- Employs machine learning algorithms like Competitive Adaptive Reweighted Sampling (CARS) for significant spectral feature identification.
- Integrates YOLO series light-weight deep learning models optimized for improved classification accuracy, precision, and speed.
- Conducts cross-validation and utilizes new datasets to enhance model robustness, generalizability, and reduce latency.
- Testing in a controlled conveyor system environment ensures performance reliability and identifies potential bottlenecks.
Technology readiness level
This technology is at TRL 2, indicating that its basic principles have been observed and reported, with ongoing efforts to validate functionalities in a controlled environment.
About University of Minnesota
The University of Minnesota is a flagship, comprehensive public research university spanning multiple campuses, with a large research enterprise and clinical integration. Industry engages through co-located labs on the Twin Cities campuses, access to an academic health system for clinical translation, and pilot and field-testing facilities that speed scale-up. A statewide extension network and outreach centers provide real-world sites and data partnerships across Minnesota, while proximity to a dense medtech and Fortune 500 corridor enables frequent collaboration. Research is supported by competitive federal funding, including NIH, NSF, DOE, USDA, and DoD. A dedicated technology transfer office manages IP, licensing, sponsored research agreements, and startup incubation to speed commercialization.
delivered to your inbox.