Data fusion for fruit quality authentication: combining non-destructive sensing techniques to predict quality parameters of citrus cultivars

Effective fruit quality inspection and authentication methods and techniques are the essential requirement at various stages of fruit chain cycle (from field to market) to maintain the quality of end products. Several destructive (impedance type, capacitive type etc.) and non-destructive (electronic nose, optical type, ultrasonic based etc.) type techniques and methods have been explored from various researchers for fruit quality measurement and authentication. Most of them uses single sensing technique to extract various quality parameters while as combining two or more complementary sensing techniques will provide better insight about various fruit samples. This manuscript reports four different non-destructive sensing techniques such as electronic nose, UV–Vis–NIR spectroscopy, and ultrasound and machine vision to acquire internal and external attributes of citrus samples. 100 citrus samples have been collected from local market and nearby citrus orchard. Data acquisition process has been performed episodically using developed four independent handheld systems (electronic nose, UV–Vis–NIR spectroscopy, ultrasonic and machine vision). Standard systems such as refractometer, Vernier calipers and weighing machine have been used to generate target predictors such as total soluble solid (%Brix), volume and weight. While as different chemical experiments have been used to generate dataset of chlorophyll and sugar content in the collected citrus samples. Different statistical and ANN based modelling methods have been used to correlate sensory panel response of each developed system separately. Further, various data fusion architectures such as low, mid and high level have been explored to combine the data coming from different implemented sensing techniques. Similar modelling methods have been used to correlate combined dataset of multiple sensing techniques with standard quality parameters. Results obtained with single sensing techniques have been compared with the results generated using multiple sensing technique fusion. Best obtained result for each quality parameter have been used to develop new hybrid data fusion model to predict various fruit quality parameters quantitatively. Predicted quality parameters have been further used to correlate with various important fruit quality decision judgement such as maturity, ripeness and flavor and further predicted quality parameters have been used to judge the optimal harvesting time of on-tree citrus sample.