SPIE DCS Digital Forum: multispectral detector assesses quality of fruit

06 May 2020

University of Waterloo develops low-cost device to monitor ripeness of kiwifruit.

A substantial proportion, by some estimates over 30 percent, of commercially grown fruit becomes rotten before reaching the customer, representing a significant problem for growers and retailers.

"Currently, retailers do not have low cost tools to ascertain the expected life of fruit, and they have to use destructive testing methods on a very small sample size," said Hardit Singh of Centennial Public School in Waterloo, Canada. "The retailers we talk to are expressing an urgent need for non-destructive sensors which can measure taste attributes and predict shelf life."

Speaking as part of the SPIE DCS Digital Forum in the Sensing for Agriculture and Food Quality and Safety conference, Singh presented research undertaken with the University of Waterloo into a low-cost multispectral detector, developed to measure internal quality parameters of kiwifruit.

Although near-IR systems to examine fruit have become widely used in large-scale packing and production facilities, the cost of the technology has prevented wide adoption of the same approach by the retail sector. Equally, low cost consumer spectrometers such as the SCiO product from Consumer Physics have been applied to the testing of domestic food quality, but the measurement accuracy from such units is still considered to be unacceptably low for such commercial applications.

"Our goal is to develop a very low cost sensor, ideally below $100, which is portable and easy to use, and easy to construct for use at the store level," said Singh. "Since the absorption peaks of the chemical bonds in fruit are quite broad, we questioned whether full high-resolution spectroscopy is required, or if the required information and can be found with multispectral detection at few wavelength bands."

The project envisaged a device with LEDs emitting at selected wavelengths, with reflected light from the fruit detected with a photodetector and analyzed in a microcontroller. Initial bench-top tests on pears, selected for their high chlorophyll content, employed a rotating stage to change the angle of LED illumination and measure the reflection at eight wavelengths, from 527 to 1050 nanometers. Trials showed that illumination at an angle of 31 degrees provided consistent results for different fruit curvatures.

"Reflections at 680 nanometers increased with time as pears become less fresh due to a decrease in chlorophyll absorption at that wavelength," said Singh. "However, the reflection values around 770 nanometers do not change appreciably, suggesting the use of the 770 nanometer wavelength band for self-referencing the measurements. By looking at the ratio of the reflection at a specific wavelength to the reflection at 770 nanometers, shelf life and optimal time to consume can be predicted."

Cost reduced by two orders of magnitude

The project designed a hand-held device to test these findings, manufacturing a 3D-printed housing in which a set of LEDs arranged in a circle are kept at a set distance from a fruit brought into contact with the device. The design also keeps the angle of incidence for the LED illumination at 31 degrees, while filtering out background light.

In use, the LEDs are switched on for 100 milliseconds, and the signal returning to a photodetector is integrated over 50 milliseconds with background light subtracted to measure the reflectance at a specific wavelength. A complete scan of all the 12 wavelengths employed takes 1.5 seconds.

The team tested its device on kiwifruit, to allow comparison with existing data from non-destructive spectroscopic studies of that fruit, and also because of known quality issues for kiwifruit expressed by retailers.

Parallel measurements of dry matter in sections of kiwifruit using a full bench-top spectrometer allowed a calibration model for the handheld device to be created, before the team then tested its system on kiwifruit samples. A total of 378 fruits were tested in the experiment.

"Although the accuracy of the multispectral detector is slightly reduced as compared to the spectrometer, the cost is reduced by two orders of magnitude," commented Singh. "If the accuracy of measurement is considered to be plus or minus 1 percent, then 78 percent of the kiwifruits were correctly measured by the multispectral detector, against 89 percent of the kiwifruit correctly characterized by the bench top spectrometer measurements."

The team believes that its prototype unit demonstrates the potential value of multispectral detection for fruit quality assessment, and that an extended baseline range of wavelengths provides better assessment of dry matter content than other comparable units. Use of reflectance at 770 nanometers as a self-reference wavelength to reduce errors in the assessment has also been clearly demonstrated, according to the project.

"The performance of the device is only marginally degraded compared to high-resolution spectrometer measurements, but it provides large cost savings," concluded Singh.