NIR-BASED SENSING TO MEASURE SOLUBLE SOLIDS CONTENT OF FLORIDA CITRUS

Abstract

A test program was undertaken to assess the accuracy of near infrared (NIR) technologies to measure SolubleSolids Content (SSC) of intact Florida grapefruit and Honey tangerines. Two units were utilized: (1) an on-line prototypeunit (Mitsui QScope, New York, N.Y.) and (2) a hand-held unit developed for fresh fruit quality assessment (Inst. of Agric.Engr.-Bornim, Potsdam, Germany). Integration time was set to approximately 150 ms in both systems. Experiments withgrapefruit were undertaken to evaluate influence of size and temperature pre-conditioning at 5.C and 20.C. Comparativemeasurements were obtained for both on-line and hand-held NIR units with the fruit subsequently juiced to obtain a referencebased on refractive index measurement of the fruits SSC. Linear regression relationships were developed between resultantnon-destructive NIR measurements and the laboratory SSC measurements. The highest coefficient of determination, r2 =0.67**, was found for Indian River red grapefruit of small fruit (9.5- to 11.0-cm diameter) tested on-line. However, r2 valuesranged from 0.15 to 0.67. Percent correct classification analysis was ascertained at 9Brix for Interior grapefruit and 10.Brixfor Indian River red grapefruit. Comparable results were obtained for the on-line and hand-held prototypes with slightlyhigher correlation coefficients for the on-line equipment. For both devices higher classification rates in sorting the fruit wereobtained for ambient temperature tests at ~20.C, while decreased correct classification was obtained on fruit pre-conditionedat 5.C. Merging NIR data and data gathered with an automatic sorting unit was evaluated with respect to fruit classificationregarding fruits SSC. A neural network (NN) model with combined inputs of physical and color attributes, plus predicted SSC,was constructed. An additional classification experiment was carried out on Indian River red grapefruit throughout an entireseason. In this second year test program, percent correct classification averaged 85% for SSC at 10.Brix and 79% for an11.Brix setpoint.