Non-Destructive Monitoring of Fruit Ripeness. Is My Avocado Ready-to-eat?

Introduction

Indeed, skin colour can be a misleading or confusing indicator of fruit ripeness, as late-season skin darkening can occur also before harvest, while low temperatures storage can produce fruits with reduced skin darkening. At the same time, squeezing the fruit to assess its firmness not only damages the fruits but even raises serious hygiene concerns, especially in this Covid19-era.

The growing market of tropical fruits and, in general, the demand for food products compliant with high-quality standards and freshness, is rapidly boosting the development of novel “intelligent packaging”. Beyond the usual containment, protection, and preservation functions - the latter solutions embed new sensing capabilities to provide the whole supply and value chain and the consumers with augmented, real-time information about the fruit ripening stages and its quality parameters.

Triggered by the challenging request of our customer ILIP - an Italian manufacturer of thermoformed food packaging made from plastic and compostable bioplastic - we have over the past three years been exploring, testing, and implementing low-cost and non-invasive solutions for estimating the ripening stage of tropical fruits, with particular focus on avocados.

The rationale – chemistry and RF dielectric properties

Avocado ripening generates mechanical, colorimetric, and chemical variations of the fruit. The main physical modification is the fruit softening due to the decrease in cell-to-cell adhesion, as a consequence of the activities carried out by different enzymes (cellulase, pectin methylesterase, polygalacturonases ...) that work synergistically. Physical changes contribute to fruit texture, particularly, both viscosity and elasticity in fruit tissues change as a function of fruit ripening, the fruit becoming creamier and less watery in texture.

As avocados soften along with time, the Dry Matter (DM) is a commonly adopted maturity index, but it requires a destructive test to be measured (Fig.1). Moreover, avocado ripening also produces physiological modifications of chemical properties, and in particular an increase in oil concentration and a decrease of sugar during the storage related to post-harvest dehydration.

Such time-variant chemical and physical variations are expected to produce a macroscopic and measurable, effect on the radiofrequency (RF) dielectric properties (permittivity 𝛆r and conductivity 𝛔 [S/m]) of the constitutive materials of the fruit, especially into the pulp.

So, at first, we demonstrated the existence of a dielectric contrast among differing ripening states (unripe, ripe, overripe) through an experimental campaign performed by characterizing the fruit pulp by means of an open-ended coaxial probe connected to a Vector Network Analyzer (VNA). The measurements revealed a time-variant, yet non-monotonic, the profile of the variation along with the time, as shown in Fig.2

As a consequence, if an antenna is placed in the close proximity of the fruit e.g. directly attached to the peel or integrated within its plastic packaging, any variations of the fruit properties will, in turn, produce modifications of its performance in terms of impedance and radiation gain.

By properly mastering this phenomenon, a self-sensing, battery-less and wireless device - namely a simple antenna! - can be turned into an (electromagnetic) transducer of the target process to be monitored.

The Dewesoft solution – electromagnetic monitoring 

The implementation of such an electromagnetic-based solution in a realistic application and processes relies on Radio Frequency Identification [RFID] in the UHF band (840-960 MHz), which is currently recognized as one of the enabling technologies for the digital Industry 4.0 revolution. 

Starting from commercially available components, we designed and optimized a totem-like system (Fig.3) comprising:

• A small-size loop tag, resembling conventional fruit stickers, is applied on a PET shell shaped around the fruit,

• An RFID reader connected to the near-field antenna focuses and excites a strong and uniform electromagnetic field distribution within the tag detection zone, and

• An embedded computing unit running a custom processing algorithm.

The electromagnetic (EM) signals backscattered and transmitted by the tag toward the reader during communication steps define the electromagnetic fingerprint of the ripening state of fruit.

The intelligence of the monitoring system can hence be moved from the sensor level - which is extremely simple and reduced down to an IC with a small antenna costing a few cents - to the processing level. 

Indeed, several analog and digital metrics (activation power, amplitude, and phase of the backscattered signal...) can be extracted from the raw signals and used to feed a supervised machine-learning algorithm - here a Binary tree classifier - to predict the fruit ripening status over a discretized scale (unripe, ripe and overripe).

In collaboration with the Department of Agricultural, Forest, and Food Sciences of the University of Turin (DISAFA) we performed a pioneering experimental campaign involving more than 300 fruits. 

Based on the results, the maximum accuracy achievable by processing RF features rarely exceeds 65-70%, due to the non-monotonic evolution of the electromagnetic parameters, above all, the high variability of fruits (in terms of shape, size, and spatial progression of the ripening process within the pulp)

Our classifier model was then enforced accounting for the temporal evolution of the ripening process, meaning that additional information - such as the storage temperature, the initial condition at packaging time, and the elapsed time from packaging, was input to the classifier. 

These indicators can be inherently managed by any RFID platform and automatically stored inside the IC memory during the different phases of the fruit distribution, from orchard to retail, thus being available at any time along the chain for feeding the classification algorithm.

By combining the RF and temporal metrics, the accuracy of the binary tree classification can be increased up to 85-95% provided that.

Is that all? Of course, not! 

There is enough room - i.e. the residual misclassification percentage - to work on to further empower the solution. 

A multiphysics approach can be explored since - similarly to the EM features, the fruit response to an imposed acoustic or vibrational stimulus that excites the structural resonances has been demonstrated in the scientific literature to be related to its consistency and composition and, accordingly subjected to the ripening process.

On the wave of what we found in the literature, we arranged a proof-of-concept for assessing the feasibility of our idea and showing to our customers the potential of combining the analysis of heterogeneous multiphysics signals.

The early set-up included a Dewesoft vibration shaker and an accelerometer sensor touching the fruit peel to record the response of the fruit to the solicitation. 

When the vibration shaker is used for the excitation of structures, single or multiple shakers are driven by either analog output or external shaker systems. The force is induced in the structure and the response of the structure is measured by the accelerometers or other vibration transducers.

Dewesoft produces a complete set of vibration shakers, permanent magnet shakers, modal shakers, and inertial shakers, perfect for R&D applications like ours.

Using the DewesoftX data acquisition software it was possible to drive the shaker with any sort of signal, thanks to the embedded functionalities of the Dewesoft Function Generator. Indeed, the Dewesoft Function Generator allows reproducing in the analog output different kinds of waveforms, such as sine, triangular, noise, or arbitrary signal, with the desired signal amplitude and frequency.

We found out that the FRF software package was very useful for this application. Thanks to the Modal Test functionalities it is possible to directly drive the shaker and calculate the mechanical transfer function of any mechanical body - or the fruit in our case. Thanks to the graphic user interface provided for the FRF software package it was easy to view the real-time data.

DewesoftX offers auto-generated displays, which already come with the most often used instruments and an arrangement that makes sense for them according to the type of application. For the signal conditioning and data acquisition, we used a SIRIUS data acquisition system, which includes all needed functionalities in one box.

The same system is able to provide high-quality signal conditioning for IEPE Accelerometers, excellent signal-to-noise ratio ADC converter and analog outputs. With a simple USB connection, our lab was ready to go.

A - very preliminary - laboratory test was arranged using a stepwise amplitude-modulated signal (10 Hz-1kHz) and two fruits having different ripeness stages. Early results suggest that the peak of the measured transfer function could be related to fruit status, with a detectable downward frequency shift, about 5Hz, along with the maturation process (Fig. 4).

Motivated by this encouraging evidence, we are currently scheduling a richer experimental campaign to tune the variables of the proposed method - e.g. which is the best position for the accelerometer transducer? which is the best excitation signal? - and to assess its reproducibility on a statistically significant fruit dataset. 

Then, the ambitious goal will be to develop a RADIO-Mechanical Totem based on Dewesoft which integrates two signal sources (RFID reader and a shaker), two transducers (a tag and accelerometers), and a synergic visualization and processing of the measured data, which can be streamed toward a unique interface thanks to OPC UA protocol. The capability of such integration is another added value provided by the Dewesoft DAQ hardware and software facilities.

We expect to significantly improve the accuracy of the classification system when using both radio and mechanical signals. Indeed, even in this case, the mechanical analysis would be performed only at some crucial points of the supply chain - due to the higher cost of the solutions - the gathered meaningful information about the fruit status will be stored inside the IC memory of the electronic label. Accordingly, the latter data will be available as additional input for the simpler RFID totem throughout the whole lifespan of the product. 

Conclusion – less food waste and plastic packaging

Intelligent packaging for food continuously generates digital/analog informative content about the inner products during their entire lifespan, thus becoming one of the enabling bricks of the modern data-driven economy. 

Reducing the waste of perishable food, optimizing the shelf exposition, suggesting when it is the perfect time to eat and engaging the customers with enhanced user experiences are just a few examples of the expected benefits from the platform

We started from “premium” fruit and vegetable products, such as tropical fruit (avocado, mango, and papaya) but the principle is straightforwardly applicable to any perishable food whose chemical/physical features change significantly along with time, provided that the classifier is properly trained over the corresponding datasets.

Our customer was really enthusiastic about the study and presented the RFID system - named SmartRipe - during one of the most important international trade shows for the food sector, Macfrut 2019, gathering very positive feedback from the business operators of the sector.

Smart Ripe is a novelty of which we are very proud because it goes in the direction of packaging that is not only protection but also an element of value creation for the supply chain. In a period in which plastic packaging is the subject of even critical discussions, we propose a vision whereby the sustainability of packaging also passes through advanced functions that create value and improve supply chain management.

Though described the “Multiphysics” approach is in the infancy stage, the customer caught the potential of the solution and commissioned us a new contract for further investigating it.

Overall, we had a lot of fun during the characterization of avocados. We had so many for lunch - and used the overripe for guacamole sauce … lucky they did not ask us to test chocolate!

References

• Radio-Frequency Identification-Based Intelligent Packaging: Electromagnetic classification of tropical fruit ripening. C. Occhiuzzi, N. D’Uva, S. Nappi, S. Amendola, C. Gialluca, V. Chiabrando, L. Garavaglia, G. Giacalone, and G. Marrocco. Antennas and Propagation Magazine, October 2020.