Towards Model-Ready AI for Apple Shelf-Life Estimation: A Fruit-Proximal Sensing Framework
Abstract Reliable apple-shelf-life AI requires exposure histories linked to defined biological outcomes. This paper presents a serviceable, three dimensional printed smart apple and a framework for preparing fruit proximal measurements for predictive evaluation. The device combines environmental and inertial sensing, timestamped local storage, and in termittently connected data transfer. A farm-to-packhouse trial recorded 470 BME280, 468 BME680, and 466 inertial observations across a nomi nal 79-minute journey without gaps exceeding 60 s. Cold-storage analyses include a 10.8-hour Device 1 cooling segment, a continuous 24.8-hour Pi1 record, and a duty-cycled 15.1-hour Pi2 record with 27.3–27.5% observed-interval coverage. These records quantify acquisition perfor mance and channel-specific environmental responses. A six-point model readiness gate specifies outcome definition, exposure linkage, temporal integrity, independent units, technical quality, and group-level validation. Firmness provides the quantitative endpoint for the proposed remaining life workflow, with cultivar- and protocol-specific acceptance criteria. The empirical contribution is deployment and data-quality evidence; linked biological prediction is the next evaluation stage. The framework pro vides a reproducible route from field sensing to grouped comparison against static and temperature-only baselines
Authors
- Clement Nyirenda (ORCID: https://orcid.org/0000-0002-4181-0478)
- Jean Frederic Isingizwe Nturambirwe (ORCID: https://orcid.org/0000-0002-1794-7343)
- Pascal Nsunba Kanuayi (ORCID: https://orcid.org/0009-0005-7995-5487)
Institutions
- University of the Western Cape (ZA)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-17
- DOI
- https://doi.org/10.21203/rs.3.rs-11742556/v1
- Primary Topic
- Postharvest Quality and Shelf Life Management
- Type
- preprint