F. Martín and D. Borges (Spain)
Signal Analysis and Processing, Noise-reduction, outlierdetection, minimum distance classifier.
This paper describes an application of signal processing and data analysis techniques to the enhancement and classification of spectroscopic signals in non-destructive evaluation of fruit ripeness. Signals are provided by a suitably developed and configured system, realized after a careful examination of the commercial offer for each of its required components. The goal is to estimate, using the spectroscopic signal, the internal sugar content and the firmness of fruit, which are well grounded parameters for evaluating its maturity. We have compared the system responses with reference values: for the soluble solids content (sugar) of samples they have been measured using a refractometer while the reference values for firmness have been obtained on each side of the examined peaches using a penetrometer. These instruments are commonly used for sorting (destructively) fruits on a sampling base. The signals measured by the system have been first of all pre-processed using a noise-reducing method based on a packets-wavelet transform. Then, an outlier detection method has been used for identifying and removing irregular patterns inside each class before training the classifier. Finally, a minimum distance classifier has been applied for grading the experimental data. The results obtained in classification show that with this early version of the set up it is possible to discriminate correctly peaches with a percentage of 87%.
Important Links:
Go Back
