No abstract

People usually assess fruit qualities from external features such as color, shape, size, and texture. However, it is quite common that we select fruits with perfect appearances but rotten inside, especially for fruits with thick pericarps. Thus the accurate measurement is desirable to evaluate the internal conditions of fruits. As two key features of fruit internal qualities, existing methods on measuring fruit moisture and soluble solid contents (SSC) are either destructive or costly, limiting their adoption in daily life. In this paper, we propose Wi-Fruit, a non-destructive and low-cost fruit moisture and SSC measurement system leveraging Wi-Fi channel state information (CSI). First, to cope with the fruit structure dependency challenge, we propose a double-quotient model to pre-process CSI on adjacent antennas. Second, to address the fruit size and type dependency challenges, a lightweight artificial neural network (ANN) model with visual information fusion is proposed for fruit moisture and SSC estimations. Extensive evaluations are conducted on 6 types of fruits with both thick (i.e., watermelon and grapefruit) and thin pericarps (i.e., dragon fruit, apple, pear, and orange) over a month in either an empty laboratory room or a library with massive books. Results demonstrate that Wi-Fruit achieves an acceptable estimation accuracy (RMSE=0.319). It is independent of various fruit structures, sizes, and types, while also robust to time and environmental changes. The fruit internal sensing capabilities of Wi-Fruit can help fruit saving and safety in both pre-harvest and post-harvest applications.

No abstract

No abstract