Poster: MobiChem: A Ubiquitous Smartphone-Based Toolkit for Practical Fruit Monitoring and Analysis

This paper introduces MobiChem, a low-cost, portable, practical, and ubiquitous smartphone-based toolkit for fruit monitoring. The key idea is to leverage the light emitted from a smartphone's screen and front camera, coupled with a custom-built screen cover, to perform comprehensive hyperspectral analysis on targeted objects. Specifically, we designed a zero-powered screen cover that selectively filters wavelengths essential for hyperspectral sensing. We then incorporate a CNN-based algorithm and a novel ranking-based learning technique that manipulates the latent space to classify maturity stages and characterize their chemical and physical factors. We showcased its application in tomato, banana, and avocado sensing. Our system examines the maturity, chlorophyll, lycopene content, free sugar levels, and firmness, enabling various dietary assessments and food safety applications.