FruitPhone: Detecting Sugar Content in Fruits Using Unmodified Smartphones with Spectral Imaging

Haiyan Hu, Yinan Zhu, Shanwen Chen, Qianyi Huang, Qian Zhang

Abstract: Developing an accessible fruit sugar content detection system for average users, particularly diabetic patients, is essential. Current fruit sugar detection solutions often fail to balance ease of use with high accuracy. In this study, we introduce FruitPhone, an innovative smartphone-based spectral imaging system that provides quantitative assessments of fruit sugar content. To enable the smartphone to capture the fine-grained spectral features necessary for accurate sugar content analysis, FruitPhone utilizes the smartphone screen to simulate multiple monochromatic light sources. This approach allows for the collection of spectral data that extends beyond the standard three-channel RGB format. To address the discrepancies between screen-generated and true monochromatic light, we propose a two-stage spectral reconstruction algorithm that translates the smartphone's pseudo-spectral data into real multi-spectral images for further analysis. Additionally, we tackle environmental variability by employing a black screen light for reference data collection, which helps mitigate ambient influences and enhance consistency. We also introduce a detrending algorithm for high-dimensional spectral images to correct spatial inconsistencies while preserving spectral trends. We evaluate FruitPhone using 335 spectral images from 37 different fruit types, achieving a normalized mean-absolute error of 8.48% °Bx, which is comparable to that of a laboratory spectral system and demonstrates a 19.98% reduction in reconstruction errors, along with a 45.58% R2 improvement over baseline schemes.

External IDs:doi:10.1145/3749470