Progress 07/01/24 to 02/28/25
Outputs
Target Audience:This SBIR Phase I project was primarily an internally focused research and development (R&D) effort, and as such, it did not include specific outreach or educational objectives targeting external audiences during its execution. The primary audience throughout the duration of the project was our internal technical and development teams. However, we maintained regular communication with our industry partner, Good Farms, collecting strawberries from their fields andkeeping them informed of our progress and key milestones. While not structured as a formal educational engagement, this collaboration ensured that relevant stakeholders were aware of the project's development and potential applications. Additionally, following the conclusion of the project, we demonstrated the system publicly at the 2025 Vine Connect Field Day held at the University of California Hansen Agricultural Research and Extension Center. This event served as an informal outreach opportunity, allowing us to share our findings and innovations with a broader agricultural community, including growers, researchers, and extension professionals. Changes/Problems:The challenges we encountered during this project did not prevent us from achieving our objectives and no significant changes needed to be made to our approach.Most of the difficulties were of the type often encountered doing research and development projects and were worked through in the normal course of the grant's activities. The data labeling activities did take more effort and time than expected, but this was mitigated by shifting resources and makingadjustments to our internally planned sequence of events. What opportunities for training and professional development has the project provided?The Project Director spent dozens of hours mentoring junior technical personnel concerning the objectives and considerations of properly collecting and curating a research-grade dataset as well as the design and fabrication of the prototype assessment station. Furthermore, the team was exposed to and gained familiarity with three-dimensional scanning technology as well as state-of-the-art machine learning algorithms and data labeling strategies. How have the results been disseminated to communities of interest?This project was primarily an internally focused research and development (R&D) effort, and as such, it did not include specific outreach or educational objectives targeting external audiences during its execution. Nonetheless, in addition to keeping our partner apprised on our progress, we demonstrated the prototype system at the 2025 Vine Connect Field Day at the Hansen Agricultural Research and Extension Center, showcasing its capabilities to industry stakeholders. This demonstration marked an important step, exhibitingthe system's capabilitiesto a broader audience. What do you plan to do during the next reporting period to accomplish the goals?
Nothing Reported
Impacts
What was accomplished under these goals?
During Phase I of this SBIR project, we successfully achieved the major goal of researching, designing, and prototyping a field-operable strawberry assessment system capable of estimating ripeness, damage, shape, and size from multiple camera views. To meet our technical objectives: Labeled Dataset Creation: We developed a comprehensive, labeled dataset ofstrawberries, capturing a wide range of berry conditions encountered in the field from multiple farms and cultivars grownin Californa. The samples weremeticulously measured and annotated with key characteristics such as ripeness level, presence of damage, shape, and size. This dataset enabled the training of machine learning models that achieved accuracy levels comparable to or exceeding those of human inspectors. Prototype Development: We designed and built a portable, field-operable prototype system that captures multiple views of each strawberry. The system integrates imaging hardware and trained algorithms to assess berry characteristics. It meets the portability and performance requirements necessary for field deployment and aligns with grower business needs and USDA inspection standards. The system exceeded 95% accuracy when detectingdefects and assigning a sale category (fresh market or juice) to assessed berries. It calculates a numeric ripeness score per berry and achieves human-level performance when sortingstrawberries accoringto ripenesslevel. It provides extremely accurate shape and size measurementsfrom berries assessed in the prototype. The knowledge and technology developed in this phase lay a strong foundation for future work in PhaseII and beyond, where we aim to extend the system tointegrate the assessment capability on-board the harvester,support pick/no-pick decisions when harvesting, and further improve and extend the defect detection andquality-based categorization.
Publications
|