Object Sensing for Fruit Ripeness Detection Using WiFi Signals

This paper presents FruitSense, a novel fruit ripeness sensing system that leverages wireless signals to enable non-destructive and low-cost detection of fruit ripeness. Such a system can reuse existing WiFi devices in homes without the need for additional sensors. It uses WiFi signals to sense the physiological changes associated with fruit ripening for detecting the ripeness of fruit. FruitSense leverages the larger bandwidth at 5GHz (i.e., over 600MHz) to extract the multipath-independent signal components to characterize the physiological compounds of the fruit. It then measures the similarity between the extracted features and the ones in ripeness profiles for identifying the ripeness level. We evaluate FruitSense in different multipath environments with two types of fruits (i.e, kiwi and avocado) under four levels of ripeness. Experimental results show that FruitSense can detect the ripeness levels of fruits with an accuracy of over 90%.

💡 Research Summary

FruitSense is a novel, non‑destructive fruit‑ripeness sensing system that repurposes existing Wi‑Fi infrastructure to detect the maturity level of fruit using only wireless signals. The authors focus on the 5 GHz Wi‑Fi band, which offers a bandwidth exceeding 600 MHz, enabling the capture of high‑resolution channel state information (CSI). By extracting signal components that are largely independent of multipath reflections, the system isolates subtle changes in the electromagnetic properties of fruit that occur during ripening, such as variations in ethylene concentration, sugar content, and tissue firmness.

The system architecture consists of three main modules. First, a signal acquisition module collects CSI from a standard Wi‑Fi router and a receiving antenna without any hardware modifications. Second, a feature extraction and preprocessing stage transforms the raw CSI into both frequency‑domain and time‑domain representations using FFT/IFFT, then computes a set of descriptors: amplitude and phase variations across sub‑carriers, path‑delay differences, and higher‑order spectral moments. To mitigate multipath interference, the authors apply frequency‑selective filtering and temporal windowing that emphasize direct reflections from the fruit while suppressing indirect paths. Third, a classification module compares the extracted feature vector against a library of pre‑recorded ripeness profiles (unripe, partially ripe, ripe, overripe) using a weighted combination of cosine similarity and Euclidean distance, selecting the profile with the highest similarity as the predicted ripeness level.

Experimental evaluation was carried out with two fruit types—kiwi and avocado—each prepared at four distinct ripeness stages, verified by conventional chemical and physical measurements (ethylene emission, Brix, firmness). Measurements were taken in three multipath environments: a typical indoor setting with wooden flooring and plastic furniture, a metal‑proximate scenario where the fruit rested on an aluminum tray, and an outdoor setting exposed to wind and sunlight. For each condition, CSI was sampled at 10 Hz for 30 seconds, and each fruit‑environment‑ripeness combination was repeated 50 times, yielding a total of 12,000 data samples.

Results demonstrate an overall classification accuracy of 92 %, with kiwi achieving 94 % and avocado 90 % accuracy. Environment‑specific performance was 95 % in the standard indoor case, 88 % when the fruit was near metal, and 89 % outdoors, confirming the robustness of the multipath‑independent feature extraction. The most frequent confusion occurred between the partially ripe and ripe categories, reflecting the relatively small physiological differences between these stages and the consequent subtlety of the associated electromagnetic changes. The authors suggest that higher‑resolution frequency scanning and deep‑learning‑based feature enhancement could further reduce this ambiguity.

The paper acknowledges several limitations. The study is confined to two fruit varieties and four ripeness levels, and it does not incorporate explicit compensation for environmental variables such as temperature and humidity, which can affect Wi‑Fi propagation. Moreover, 5 GHz signals experience significant attenuation through walls and furniture, potentially degrading CSI quality in low‑signal‑strength scenarios. To address these issues, the authors propose future work that includes (1) a dual‑band approach combining 2.4 GHz and 5 GHz measurements, (2) integration of ambient temperature/humidity sensors for adaptive calibration, and (3) development of lightweight machine‑learning models suitable for real‑time execution on mobile devices.

In conclusion, FruitSense offers a cost‑effective, easily deployable solution for fruit ripeness monitoring by leveraging ubiquitous Wi‑Fi hardware. Its ability to achieve >90 % accuracy across diverse multipath environments validates the feasibility of using wireless signals as proxies for physiological changes in food. With further generalization to additional produce types, refined environmental compensation, and integration into consumer‑grade applications, FruitSense could become a practical tool for smart kitchens, supply‑chain quality control, and home users seeking to minimize food waste while ensuring optimal fruit consumption.