Research Of Non-destructive For Watermelon Maturity Based On Its Acoustic Properties

Watermelon is an important cash crop in our country, which has become one of the preferred cash crops for rural industrial structure adjustment and for farmers to increase wealth, with its short growing period, high yield, and good economic returns. Currently, Watermelon production has been developping toward the direction of large-scale, base, industrialization and export-oriented, and the situation of big market and circulation has taken shape. In this situation, watermelon production should be based on market distance, transport distance, and storage time to determine the appropriate harvest period. On the other hand, market is increasingly demanding good quality than ever before, so it's advisable to accurately their mature degree during the processes of watermelon classification, storage, transportation, and sales.Presently, however, the estimate of maturity of watermelon is still using the traditional identification of calculating, watching, listening, measuring and other methods.These methods, depending on personal experience in most time, are difficult to make accurate estimation of maturity of watermelons, meanwhile, it is a large-scale-labor-and-time-consuming testing. With the diversification of watermelons as well as their cultivated forms, these methods can hardly meet the current needs of producting and marketing. Therefore, an accurate, rapid and simple non-destructive testing of watermelon maturity technology has a great theoretical and practical significance for properly harvesting, avoiding immatured fruit saling, grading post-harvest storage, transportation, as well as sales links, etc.This paper extractes several kinds of characteristics closely related with water-melon maturity, and finds out a characteristic with supreme relevance to achieve the non-destructive detection of the water-melon maturity, through analyzing the acoustic signals taken by hitting the water-melon samples of different maturity, using advanced technology and experience both at home and abroad, . Main research contents and research results are as following:1.Set up a simple and convenient, swift, and costing little audio frequency gathering device by one portable PC , with a 5mm electret microphones;2.Use Matlab programming to build a GUI audio recording and analyzing systems based on Windows. Collect audio signals by hitting 23 watermelon samples (of the variety widely cultivated in northern China with thin pericarp and bearing seeds, named Jingxin Watermelon) for three times in succession, and save them in computer as'.vaw'form;3.Research on the Wavelet de-noising and end-point detection. Cut out the single-hitting signal in each sample as the smallest unit of signal analyzing by the methods of short-term energy and zero-crossing rate. In this way, the 23 samples are divided into three groups, with each group corresponding to 23 signals taken from a hitting signal in every samples;4.Extract six kinds of acoustic feature by each analytical unit for testing. four commonly used features mentioned in other papers, including resonant frequency, damping coefficient, symmetry of wave and band magnitude, was extracted. For getting a highter correlation results of maturity, this paper also presents two new features: Mel-Frequency Cepstral Coefficients and Band Magnitude Vector;5.Evaluate the six features by F-ratio(or D-ratio) method. In this paper, the D-radio result of band magnitude vector (BMV),which is 2.0456, is better than other features. The result illustrates the highly correlation between the BMV characteristics and the maturity stages of watermelon;6.Detect the stages of maturity by Probabilistic Neural Network. The BMV features from 23 watermelon samples were used to train PNN. To test the effect of detection, another BMV features from 46 new melons were calculated and classified into ripeness stages by trained PNN, and then destructive detection was performed to check the classification results later. As a result, the percentage of well-classified watermelons was 86.96%, compared to 71.5% of human experts.