Research On Pattern Recognition Based On Imaging Processing And Spectral Analysis Techniques

Pattern recognition refers to the processing and analysis of the various forms information on the characterization of things or phenomena (numeric, text or logical relations) for the interpretation and identification of things or phenomena. Since the]940s with the invention of computer, pattern recognition has been applied in the field of food industry, biotechnology, energy processing, medical diagnostics, environmental monitoring, and pollution control. At present, image processing, spectral analysis, and hyperspectral imaging techniques are used for the fast and nondestructive acquisition of information. With the continuous progress of the fabrication technology of the instrument, the acquisition technology of various data such as digital image, spectral, hyperspectral images has been widely used in various fields. With the increase of the number and complexity of the data, people faced with huge challenge of these data comprehension. Therefore it is necessary to extract useful information from the data and to support decision-model establishment. This study presents the methods of data mining and modeling establishment for one-dimensional spectral information, two-dimensional image information, and three-dimensional hyperspectral image using imaging processing and spectral analysis technique. The main research contents and results are shown as follow:(1)The quality determination algorithms of diesel fuels were analyzed based on spectroscopy technique. The A new spectral calibration algorithm, Laplacian regularized least squares (LapRLS), was proposed. LapRLS is a semi-supervised machine learning algorithm which makes use of both labeled and unlabeled data for training. In this study, LapRLS was used to establish the quantitative relationship between near infrared (NIR) spectra and cetane number (CN) and total aromatics of diesel fuels. Near infrared (NIR) spectroscopy is a widely used technique for monitoring chemical compounds in petroleum industry. A total of381obtained samples were randomly split into two sets under different proportion. One set was used as calibration set (labeled data) whereas the remaining samples were used as the prediction set (unlabeled data). LapRLS, LS-SVM, and PLS were used to establish determination models based on NIR spectra. Results show that the best performance of determination was achieved by LapRLS, which indicates that LapRLS can utilize unlabeled data effectively on NIR spectral data for the determination of the chemical compounds of diesel fuels.(2)The capacity of color features for quality determination of Vitis labrusca was analyzed. The aim was to evaluate the feasibility of using color features for the rapid determination of soluble solids content (SSC) and pH in Vitis labrusca. A total of120color features were obtained from the mean and standard deviation of the interested area’s pixel’s intensity values at each color channel of RGB, HIS, NTSC, YCbCr, and HSV color spaces. Least-squares support vector machines (LS-SVM) models were established based on color features. The range of correlation coefficients of SSC prediction was0.870-0.886and that of pH prediction was0.695-0.727. The maximum correlation-coefficient method was used to select the most important color features. The standard deviation of pixel’s intensity value in Y channel and mean of pixel’s intensity in Y/Cb channel of YCbCr color space were chose as the best features for pH. The mean of pixel’s intensity value in V/S channel of HSV color space and mean of pixel’s intensity value in Cb channel of YCbCr color space showed best performance for SSC. Their established models obtained the root mean square error for prediction (RMSEP) of0.0611and0.737for pH and SSC prediction, respectively. It is concluded that the color features can be used to fast and non-invasively measure pH and SSC of Vitis labrusca.(3)Based on the hybrid image features, the fast and non-invasive classification algorithm of raisin was analyzed. A new approach for the non-invasive classification of raisins is presented based on the hybrid image features, namely morphological, color and texture features. A total of74features (8morphological,30color, and36textural) were extracted from RBG images. Seven kinds of models were established based on different feature sets. They were three kinds of models established based on single feature set, three kinds of models established based on the combination of two feature sets, and one kind of model established based on the combination of all feature sets. Five kinds of classifiers, namely partial least squares (PLS), linear discriminant analysis (LDA), soft independent modeling of class analogy (SIMCA), and least squares support vector machine (LS-SVM) with linear and radial basis function (RBF) kernels were used for the model establishment based on different feature sets. The best correct answer rates (CAR) of99%was obtained when LDA was used to establish the classification model based on the combination of all feature sets, which was higher than those of the models established based on single feature set or the combination of two feature sets. The results show that the feature combination is helpful to improve the accuracy of raisin classification. It was concluded that the varieties of raisin could be accurately classified based on RGB image features and the combination of morphological, color and texture features was an accurate way to improve the accuracy of classification.(4)The determination algorithm of beef moisture content was analyzed based on hyper-spectral imaging technique. The visible and near infrared spectrum were extracted from hyperspectral images of beef using two methods (manual using ENVI software, imaging processing technique using Matlab software). The results show that the spectrum data extracted by imaging processing technique could give better prediction results of beef moisture content. Then, with the spectrum data extracted by imaging processing technique, least squares support vector machine (LS-SVM) and partial least squares (PLS) were used to establish the models for the prediction of moisture content, and LS-SVM shows better capability for obtained information form spectrum data. Because there much invalid information in the spectrum data, it is necessary that successive projections algorithm (SPA), uninformative variable elimination (UVE), and UVE-SPA was using to extract effective features from spectrum data. With extracted features. LS-SVM was used to establish prediction models. Compared with other algorithms, UVE-SPA shows more capacity to extract effective features for moisture content prediction. It could be concluded that LS-SVM with UVE-SPA obtained best prediction result of beef moisture content. Finally, based on the LS-SVM and UVE-SPA algorithm, a visualized image of moisture distribution of beef was obtained by using imaging processing technique.