Ultra-Low-Cost Self-Referencing Multispectral Detector for Non-Destructive Measurement of Fruit Quality

Abstract

A very low-cost sensor is developed to non-destructively test fruits by sequentially turning on light-emitting diodes at 12 different wavelengths and measuring the reflectance in the interactance mode. The detector is tested on kiwifruit to measure soluble solids content (SSC) and dry matter (DM) non-destructively, while the performance is compared with a benchtop spectrometer. For SSC and DM measurements, a total of 378 and 200 samples of kiwifruits were measured respectively. Non-parametric regression was used for the multi-spectral detector, while a partial least squares regression model was used for the benchtop spectrometer. Different regression techniques were used as they provided the best prediction for the two different measurements. For SSC measurements, coefficient of determination (R2) of 0.83, root mean square error for prediction (RMSEP) of 0.85%, bias of −0.004%, and SDR of 2.45 were observed using the multispectral detector. The corresponding values achieved with the benchtop spectrometer were 0.98, 0.37%, 0.01%, and 5.60, respectively. For dry matter measurements with 200 kiwifruits, R2, RMSEP, bias, and SDR of 0.83, 0.61%, −0.02, and 2.44 were achieved with the prototype multispectral detector compared with the values of 0.96, 0.31%, −0.11, and 4.80, respectively, achieved with the benchtop spectrometer. The efficacy of the multispectral detector was also tested by measuring 30 navel oranges, wherein an R2 and root mean square error for calibration (RMSEC) of 0.79 and 0.47% were achieved for SSC measurements. While the performance of the multispectral detection is lower than a benchtop spectrometer, its performance is still better than low-cost spectrometers and can be used to inexpensively measure the quality of fruits in a non-destructive manner.