Apple quality assessment by fusion three sensors

Three different sensors (a near-infrared spectrophotometer \´NIR\´, a machine vision system \´MV\´ and an electronic nose system \´EN\´) are combined for nondestructive quality detection of "Fuji" apples. The intention is to take advantage of the three sensors, one performing a local measurement of one physical property of the fruit (sugar content), the others performing a global assessment of other physical properties of the fruit (color, size, shape and aroma), and to combine those types of measurement (local and global) to improve the accuracy of quality assessment. The EN is used to assess rotting stage of apple based on ANN (artificial neural network). A relationship is also found between sugar content and different NIR wavelengths by using MLR (multiple linear regression). The surface color, shape and size of apples were accessed by MV technique. The three sensors are working at the same time. 104 "Fuji" apples were detected by the three-sensor combination system, and were divided into two sets, with 84 in set A and 20 in set B. By combining the three different kinds of sensors, it is shown that the accuracy of quality assessment of apple can be improved with the high-level fusion technique. For sugar content assessment, the classification error rate drops from around 17%, using only NIR spectra, to around 6% when the three sensors are combined through ANN. Finally, the three sensors are combined to evaluate the quality of apples through a decision tree, and only 6 apples in set A, 1 apple in set B were misclassified