Predict changes of some quality parameters of black mulberry juice (Morusnigra L.) during ripening using machine vision and fractal analysis

Blackberry is a perennial woody plant native to warm, temperate, and subtropical regions of Asia, Africa, North America, and southern Europe. Blackberry fruit (Morus Alba Varnigra L.) is a rich source of anthocyanins. Furthermore, it has great many medicinal properties such as an antidiabetic (Asano et al., 2001), antihyperglycemic (Andallu &Varadacharyulu, 2003), antiviral (Du et al., 2003), antioxidative (Kim et al., 1998), hypolipidemic (El-Beshbishy et al., 2006), and neuroprotective (Kang et al., 2006). However, measuring some qualitative and nutrient parameters in this fruit such as anthocyanins, vitamin C and phenol directly has become a major issue (Pace et al, 2013). Therefore, researchers try to predict aforementioned parameters by mathematical models. One of these models is the fractal model which is widely used to study the properties of the images/objects (Welstead, 1999; Zhang, 2007). Recently, many researchers try to develop different methods to classify or predict the agricultural products quality (Langner, 2001). In a research Seng and Mirisaee (2009) designed a machine vision algorithm for classification of fruits (apple, lemon, strawberry and banana) based on color, shape and size. Li and He investigated the application of visible/near infrared spectroscopy (Vis/NIRS) for measuring the acidity of Chinese bayberry. The model for prediction the acidity (r=0.963), standard error of prediction (SEP) 0.21 with a bias of 0.138 showed an excellent prediction performance. Therefore, the aim of this study was to predict biochemical parameters (TSS, anthocyanins, browning compounds, total phenols, Ascorbic Acid, pH) of blackberry juice, nondestructively, during maturity process using machine vision and fractal analysis. To develop predictive models and data classification, artificial neural networks (ANN) and k-nearest neighbor (k-NN) were used.
Eighty blackberry fruits from four maturity stages were selected. The fruit samples were placed in airtight polyethylene bags, stored in an ice-filled cooler and transported to the laboratory to keep at cold temperature (4±1◦C).
Fresh fruits were squeezed by a household juicer, and immediately transported to the laboratory. Then, juice images were taken with a digital camera CASIO (Model Exilim EX-ZR700; 16 megapixels, Japan) and stored to the computer.
There are several ways to measure the fractal dimension. In this study, the proposed method by Addison (2005) was used to calculate the fractal dimension.
Feature selection is one of the issues that have been raised in the context of machine learning. In this study, floating search method feature selection was used (Pudil et al., 1994).
k-Nearest Neighbor (k-NN) is one of the simplest methods for information classification. In this study, the Euclidean distance between two points was used to determine the distance between the input data with the training patterns (Mucherino et al., 2009).
To train the neural network, Levenberg–Marquardt training algorithm was used. In this regard, the data were divided randomly into two parts (two-thirds for training (60) and one-third (20) for testing the network). Input parameters were Xa, Xb, X, Y and S and output parameters were TSS, ascorbic acide, acidity, polyphenols, anthocyanins, brown-causing substances and pH. Moreover, in this study, the number of neurons in the hidden layer was selected by trial and error method.
After selecting the best features extracted from the image processing with the highest correlation with chemical parameters (TSS, anthocyanins, total phenols, ascorbic acid, and pH), a machine vision system was designed and built to be able to determine the internal properties of black mulberry juice.
Total soluble solids (TSS) were determined by a hand refractometer device (model: MT03 Japan). The anthocyanin content was estimated following the procedure of Holecraft et al., (1998). Ascorbic acid of the juice was measured by titration with copper sulfate and potassium iodide based on the Barakat et al., (1973) procedure. Titratable acidity was measured according to the Eksi and Turkman, (2011) method. Waterhouse (2002) method was used for measuring the total phenol of juice.
Artificial neural network (ANN) and (k-NN) models were used to predict the changes of anthocyanin (AC), browning compounds, ascorbic acid (AA), total phenols (TP), acidity, TSS and pH in mulberry juice during ripening based on fractal analysis. Two features namely: maximum fractal and fractal curve area were selected from five extracted features and used for training neural network and k-NN classifier