Methods and artificial neural networks for evaluating rice taste quality

Abstract

The present invention discloses an artificial neural network for evaluating the taste quality of rice, which is a feedforward network including an input layer, a hidden layer and an output layer. The input layer includes five input neurons: balance, elasticity, attenuation value, disintegration value and regeneration value, the hidden layer includes several hidden layer neurons, and the output layer includes an output neuron: comprehensive score; the hidden layer transfer function is an S-type logarithmic function, and the output layer transfer function is a purelin function. The artificial neural network randomly selects 80% of the data as a training set and 20% as a test set, and uses the training set data for training and learning starting from random initial weights and initial thresholds. The test set is used to evaluate the performance of the final model. The training function is a trainlm function, and the learning function is a learngdm function. The weights and thresholds are continuously corrected during the training and learning process, and the training is repeated until the mean square error meets the performance requirements. The training is completed and the artificial neural network is output.

Description

Technical Field

[0001] The invention relates to the technical field of rice taste quality evaluation, and more specifically, to a technology of using an artificial neural network model to assist in evaluating the taste quality of rice. Background Art

[0002] Rice is the staple food in my country. With the continuous improvement of living standards, people's requirements for its taste quality are gradually increasing.

[0003] Therefore, it is necessary to accurately and conveniently evaluate the taste quality of rice. The taste quality is a comprehensive expression of the appearance, smell, palatability, flavor and other characteristics of rice after cooking. Among them, palatability is the characteristic with the greatest difference and the most important to consumers. Palatability includes indicators such as hardness, viscosity, and elasticity.

[0004] Existing methods for evaluating rice taste quality mainly include physical and chemical indicators, sensory evaluation and instrumental measurement.

[0005] The physicochemical index method is to measure the cooking characteristic indexes that are correlated with the taste quality of rice, such as amylose content, protein content, gel consistency, etc., which indirectly characterize the taste quality of rice. For example, the method disclosed in the existing patent CN105067784A is a method for evaluating the taste quality of rice using physicochemical indexes. The disadvantages of the physicochemical index method are that it is time-consuming and labor-intensive. It takes several days to obtain the physicochemical indexes, and the operation method is complicated and labor-intensive.

[0006] The sensory evaluation method is manually evaluated by sensory evaluators. Since the sensory evaluation method is a manual evaluation, it is susceptible to subjective influences, inaccurate results, and unstable. In addition, sensory evaluators need long-term professional training, which takes a long time and is greatly affected by personal factors.

[0007] The instrumental method uses a taste meter to test specific indicators. The instrumental method has a fast detection speed, quantifiable results, and is relatively objective. The more famous instrumental method is Japan's Satake taste meter method. The Satake taste meter measures and evaluates hardness and viscosity. The development background of the Satake taste meter is based on the national conditions of Japan. Japanese rice is all japonica rice with fewer varieties and concentrated production areas, so the measurement indicators of the Satake taste meter are limited. China has many rice varieties and a wide range of production areas, especially there are large differences between indica rice varieties and japonica rice, which makes the Satake taste meter have deviations when measuring Chinese rice, especially for indica rice varieties.

[0008] According to the book "Application of MATLAB to Implement Neural Networks" by Wen Xin et al., compared with traditional linear or nonlinear regression methods, ANN (artificial neural network) has the following advantages: (1) ANN has strong learning ability; (2) ANN is a multi-input and multi-output system; (3) ANN is a complex nonlinear system; (4) ANN can operate in parallel, has extremely fast computing speed, short response time, and can meet the needs of online simulation and online optimization. However, the existing technology has not yet used the model established by ANN to predict the taste quality of rice. Summary of the invention

[0009] The present invention aims to propose a method for evaluating the taste quality of rice based on instrumental measurement indicators and with the aid of an artificial neural network model, comprising the following steps:

[0010] 1) Determination of sample parameters:

[0011] Determination of several parameters and sensory evaluation scores of rice samples;

[0012] 2) Establishment of artificial neural network model:

[0013] The artificial neural network is a feedforward network including an input layer, a hidden layer and an output layer; the input layer includes a plurality of input neurons, i.e., a plurality of parameters of the rice sample; the hidden layer includes a plurality of hidden layer neurons; the output layer includes an output neuron: a comprehensive score (the score is a predicted value of sensory evaluation);

[0014] 3) Training and testing of artificial neural network models to obtain effective artificial neural networks;

[0015] 4) Taste quality evaluation:

[0016] The parameters of the unknown new sample described in step 1) are measured and input into the effective artificial neural network described in step 3) as input values ​​to perform network simulation, and the output value is the comprehensive score; the higher the comprehensive score, the higher the taste quality of the rice.

[0017] Preferably, the several parameters in step 1) are balance, elasticity, reduction value, disintegration value and regeneration value.

[0018] In one embodiment, the method for determining the balance and elasticity is as follows: the rice sample is washed several times, soaked for 30 minutes at a rice-water ratio of 1:1.3, steamed into rice in an electric rice cooker, and then simmered for 20 minutes before breaking up the rice, taking out the inner pot of the electric rice cooker and placing it at room temperature, and when the rice is cooled to 55-60° C., the sample is weighed and pressed, and then the balance and elasticity of the rice are determined using a hardness viscometer; preferably, the hardness viscometer is a Satake hardness viscometer;

[0019] The method for determining the reduction value, disintegration value and recovery value is as follows: the rice sample is crushed to pass 50-100 mesh, and then the reduction value, disintegration value and recovery value are determined according to GB / T 24852-2010.

[0020] The sensory evaluation score is based on GBT15682-2008 "Sensory evaluation method for edible quality of steamed paddy and rice".

[0021] In one embodiment, the number of neurons in the hidden layer of step 2) is determined according to the following formula:

[0022]

[0023] Where l is the number of input neurons, m is the number of output neurons, a is a constant between [1,10], and N is the number of hidden layer neurons.

[0024] In one embodiment, the specific parameters of the artificial neural network model of the present invention are set as follows: training function: trainlm;

[0025] Learning function: learngdm;

[0026] Hidden layer transfer function: tansig;

[0027] Output layer transfer function: purelin;

[0028] Network performance target error: 1e-07;

[0029] Maximum number of training steps: 1000;

[0030] Other parameters were set by default in MATLAB software.

[0031] The data set of the artificial neural network model contains the input values ​​(balance, elasticity, reduction value, disintegration value, recovery value) and output values ​​(sensory evaluation score) of all rice samples. 80% of the data in the data set are randomly selected as training data, and the remaining 20% ​​of the data are used as test data. Training and learning are performed using the training data starting from random initial weights and thresholds. The weights and thresholds are continuously corrected during the training and learning process. One round is a training of all training data. After completing each round of training, the performance indicator mean square error is detected. If the mean square error does not meet the performance requirements, the training steps are updated, and the training data is reused for the next round of training. If the mean square error meets the performance requirements, the training ends and the artificial neural network model and results are output.

[0032] In one embodiment, each round of training of the artificial neural network includes:

[0033] Randomly initialize weights and thresholds;

[0034] Select a portion of the balance, elasticity, reduction value, disintegration value, recovery value and sensory evaluation scores from the training data as training samples;

[0035] Calculate the input and output of hidden layer neurons;

[0036] Calculate the input and output of the output layer neurons;

[0037] Calculate the output layer error;

[0038] Calculate the hidden layer error;

[0039] Modify the weights and thresholds of the output layer and hidden layer;

[0040] Select the next part of balance, elasticity, reduction value, disintegration value, recovery value and sensory evaluation score from the training data as training samples and repeat the above steps until all the training data are trained.

[0041] In one embodiment, the mapminmax function is used to normalize the data set so that all data samples fall within the neuron transfer function region. The normalization process includes:

[0042]

[0043] in,

[0044] x n is the normalized value of variable x;

[0045] x max 、x min are the maximum and minimum values ​​of x respectively;

[0046] y max ,y min are the maximum and minimum values ​​of the normalized index respectively.

[0047] In one embodiment, the mean square error (MSE) is used to evaluate network performance. If MSE < 1e-07, the mean square error is considered to meet the performance requirements. The mean square error (MSE) is calculated by the following formula:

[0048]

[0049] in,

[0050] T i , P i They represent the target value and predicted value of the artificial neural network respectively;

[0051] n represents the total number of input training samples;

[0052] In one embodiment, the effective artificial neural network in step 3) must meet the following conditions: the correlation coefficient R between the comprehensive score (the score predicted by the artificial neural network model) and the sensory evaluation score according to GB / T 15682-2008 satisfies R 2 >0.97. The correlation coefficient R represents the linear correlation between the target value comprehensive score of the artificial neural network and the sensory evaluation score. The higher the correlation, the higher the prediction accuracy.

[0053] The present invention also includes the application of the above-mentioned method for evaluating the taste quality of rice in compounding different varieties of rice. By inputting the balance, elasticity, reduction value, disintegration value and regeneration value of the compounded rice into the artificial neural network, an output value, i.e., a comprehensive score of the compounded rice, is obtained. By comparing the comprehensive score with the sensory evaluation score of the compounded rice according to GB / T 15682-2008, the correlation coefficient R satisfies R 2 >0.97, indicating that the artificial neural network of the present invention is suitable for application in rice compounding, that is, it can be used to evaluate the taste quality of compound rice.

[0054] The present invention also includes using the above-mentioned method for evaluating the taste quality of rice to evaluate the taste of rice. The higher the comprehensive score output by the artificial neural network, the better the taste.

[0055] The present invention also includes a system for evaluating the taste quality of rice, comprising:

[0056] machine readable storage; and

[0057] a processor configured to execute machine-readable instructions, the instructions being a number of input neurons of the input layer;

[0058] After reading the instruction, the system runs the instruction to execute the method for evaluating the taste quality of rice, and the output value is the comprehensive score of the rice.

[0059] The artificial neural network and method for evaluating rice taste quality of the present invention have the following advantages:

[0060] 1) Through correlation analysis of the data of all rice samples, it was found that the gelatinization characteristics of rice (reduction value, disintegration value, and recovery value), the texture characteristics of rice (balance, elasticity) and the sensory evaluation results were significantly correlated, and the P values ​​were all <0.01, indicating a very significant correlation; the Pearson correlation coefficient was between 0.6 and 1, indicating a strong correlation. Therefore, the above indicators can be used to accurately determine the taste quality of rice.

[0061] 2) The testing of rice gelatinization characteristics (reduction value, disintegration value, and recovery value) takes only 13 minutes, and the testing of rice texture characteristics (balance, elasticity) takes only 1-2 minutes, while the testing of amylose content and protein often takes 1-2 days. Therefore, compared with the method of using physical and chemical indicators to determine the taste quality of rice, this method greatly shortens the time and is more convenient and quick.

[0062] 3) At present, the balance and elasticity of cooked rice are measured after the cooked rice is cooled to room temperature. At this time, the rice has begun to age. Different rice ages at different rates, and its taste is inconsistent with the taste of rice sensory evaluation in accordance with GB / T15682-2008 "Sensory Evaluation Method for Cooked Rice and Rice". Therefore, placing the rice at 55-60℃ can maintain consistency with human sensory evaluation, and the judgment result is more accurate.

[0063] 4) This method uses artificial neural network to predict the taste quality of rice. Compared with traditional methods such as regression equation, response surface, principal component analysis, etc., artificial neural network has stronger learning ability, adaptability and higher prediction accuracy.

[0064] 5) Japanese rice is all japonica rice and there are fewer varieties. The Satake Taste Meter was developed based on Japanese conditions. Although a database of Chinese rice has been established, it still cannot accurately reflect the taste quality of samples when testing indica rice. Compared with the Japanese Satake Taste Meter, this model covers more comprehensive rice varieties, has more accurate prediction results, and is more suitable for Chinese rice. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A schematic diagram of the topological structure of an artificial neural network for evaluating rice taste quality according to an embodiment of the present invention is disclosed.

[0066] Figure 2 A schematic diagram of a training and testing process of an artificial neural network for evaluating rice taste quality according to an embodiment of the present invention is disclosed.

[0067] Figure 3 A flow chart of a method for evaluating rice taste quality according to an embodiment of the present invention is disclosed.

[0068] Figure 4 The training results of the artificial neural network of the training samples of Example 1 are revealed.

[0069] Figure 5 The training results of the artificial neural network of the training samples of Example 2 are revealed.

[0070] Figure 6 The training results of the artificial neural network of the training samples of Example 3 are revealed.

[0071] Figure 7The training results of the artificial neural network of the training samples of Example 4 are revealed.

[0072] Figure 8 The training results of the artificial neural network of the training samples of comparative example 1 are revealed.

[0073] Fig. 9 The training results of the artificial neural network of the training samples of comparative example 2 are revealed.

[0074] Fig.10 The training results of the artificial neural network of the training samples of comparative example 3 are revealed.

[0075] Fig.11 The training results of the artificial neural network of the training samples of Comparative Example 4 are revealed.

[0076] Fig.12 The correlation between the polished round-grain rice taste value measured by the Satake taste meter in Comparative Example 5 and the actual sensory evaluation score is revealed.

[0077] Fig.13 The correlation between the taste value of indica rice measured by the Satake taste meter in Comparative Example 5 and the actual sensory evaluation score was revealed.

[0078] Fig.14 The correlation between the actual sensory evaluation score of formulated rice and the model predicted score was revealed. DETAILED DESCRIPTION

[0079] The rice taste quality evaluation method of the present invention also belongs to the instrumental determination method. The selection of the determination index is based on the following considerations: the index needs to have a significant correlation with the sensory evaluation result, and the index needs to have an industry-recognized detection method and detection basis.

[0080] The inventors analyzed the five indicators of balance, elasticity, reduction value, disintegration value, and recovery value as shown in the table below. The results show that these five indicators are significantly correlated with the sensory evaluation results, and the P values ​​are all <0.01, indicating a very significant correlation; the Pearson correlation coefficient is between 0.6-1, indicating a strong correlation. Therefore, the above five indicators can be used to accurately determine the taste quality of rice. The relationship between the Pearson coefficients of balance, elasticity, reduction value, disintegration value, recovery value and sensory evaluation scores is as follows:

[0081] Pearson coefficient Sensory evaluation score Balance elasticity Reduction value Disintegration value Regeneration value Sensory evaluation score 1 -.939** .740** -.842** .813** -.674**

[0082] In the following examples, the balance and elasticity are tested using a reference method provided by the manufacturer of the hardness and viscosity instrument: Satake. The rice is washed and steamed according to the specified procedures, the cooked rice is covered and placed at room temperature, and after cooling, 8 g of the sample is weighed and pressed for 10 seconds to measure the balance and elasticity of the rice.

[0083] In the following examples, the reduction value, disintegration value and regeneration value were tested by using GB / T24852-2010 "Rapid Viscometer Method for Determination of Gelatinization Properties of Rice and Rice Flour". The rice was crushed to 90% and passed through a CQ23 sieve. The gelatinization properties of the rice were tested using an RVA rapid visco analyzer.

[0084] Figure 1 A schematic diagram of the topological structure of an artificial neural network for evaluating the taste quality of rice according to an embodiment of the present invention is disclosed. Figure 1 As shown, the artificial neural network is a feedforward network including an input layer 101, a hidden layer 102 and an output layer 103. The input layer 101 includes five input neurons x1, x2, x3, x4 and x5, which correspond to five indicators: balance, elasticity, reduction value, collapse value and regeneration value. The hidden layer 102 includes several hidden layer neurons H. The output layer 103 includes an output neuron o1, which represents: comprehensive score. The hidden layer transfer function is the tansig function. The output layer transfer function is the purelin function. The number of hidden layer neurons H included in the hidden layer 102 directly affects the performance of the neural network. If the number of hidden layer neurons is too large, it may lead to problems such as long learning time and poor generalization ability. If the number of hidden layer neurons is too small, the network may not be trained at all or the network performance may be poor. A successful artificial neural network (ANN) model should be able to summarize the rules from the input training data, rather than just remembering them. In one embodiment, the number of hidden layer neurons H included in the hidden layer 102 is determined according to the following formula:

[0085]

[0086] Where l is the number of input neurons, m is the number of output neurons, a is a constant between [1,10], and N is the number of hidden layer neurons.

[0087] For this embodiment, l=5, m=1, and N=[4,13] is calculated. Therefore, the number of hidden layer neurons H included in the hidden layer 102 is between 4 and 13.

[0088] The data set of the artificial neural network model contains the input values ​​(balance, elasticity, reduction value, disintegration value, recovery value) and output values ​​(sensory evaluation score) of all rice samples. 80% of the data in the data set are randomly selected as training data, and the remaining 20% ​​of the data are used as test data. Training and learning are performed using the training data starting from random initial weights and thresholds. The weights and thresholds are continuously corrected during the training and learning process. One round is a training of all training data. After completing each round of training, the performance indicator mean square error is detected. If the mean square error does not meet the performance requirements, the training steps are updated, and the training data is reused for the next round of training. If the mean square error meets the performance requirements, the training ends and the artificial neural network model and results are output.

[0089] Since the artificial neural network uses a data set containing balance, elasticity, reduction value, disintegration value, recovery value and sensory evaluation score as training data, the method of obtaining the training data is first introduced. After obtaining the rice sample, the balance and elasticity, as well as the reduction value, disintegration value, recovery value and sensory evaluation score are obtained according to the following process.

[0090] The method for determining the balance and elasticity is as follows: wash the rice sample 3 times, each washing includes rotating it clockwise and counterclockwise 5 times, soak it for 30 minutes at a rice-water ratio of 1:1.3, steam it into rice with a 0.8L Hongzhi rice cooker, simmer it for 20 minutes and then break up the rice, take out the inner pot of the rice cooker, put a filter paper with a diameter of 18cm between the inner pot and the lid, and place it at room temperature of 23℃±2℃. When the rice cools to 55-60℃, weigh 8g of the sample, place it in a rice press equipped by Satake and press it for 10s, and then use a Satake hardness viscosity meter to determine the balance and elasticity of the rice;

[0091] The method for determining the reduction value, disintegration value and recovery value is as follows: the rice sample is crushed to pass 50-100 mesh, and then the reduction value, disintegration value and recovery value are determined according to GB / T 24852-2010.

[0092] The sensory evaluation score is based on GBT15682-2008 "Sensory evaluation method for edible quality of steamed paddy and rice".

[0093] According to the above process, multiple rice samples are obtained, and multiple data sets including balance, elasticity, reduction value, disintegration value, recovery value and sensory evaluation score are measured. After obtaining a sufficient number of data sets, 80% of the data are randomly selected from the data sets as training data, and the remaining 20% ​​of the data are used as test data.

[0094] Since the input and output layer parameters of the artificial neural network model have different dimensions, in the network learning process, in order to facilitate training and better reflect the relationship between various factors, the input and output training sample data need to be normalized before starting training. In one embodiment, the mapminmax function is used to normalize the data set so that the data samples all fall within the neuron transfer function region. For the artificial neural network model of the present invention, the normalized data is required to be distributed in the interval [-1,1].

[0095] After completing the preparation of training data, the process of training and learning the artificial neural network can begin. Figure 2 A schematic diagram of the training and testing process of an artificial neural network for evaluating rice taste quality according to an embodiment of the present invention is disclosed. As shown in the figure, each round of training of the artificial neural network model includes:

[0096] Initialize the weights and thresholds. In this step, the weights and thresholds are initialized. The weights and thresholds will be assigned random values ​​by the MATLAB software.

[0097] Input training samples, select a part of the balance, elasticity, reduction value, disintegration value, recovery value and sensory evaluation score from the training data as training samples, and input them into the input layer. Balance, elasticity, reduction value, disintegration value and recovery value correspond to the 5 input neurons of the input layer, and the comprehensive score is the output neuron.

[0098] Calculate the input and output of the hidden layer neurons. When there are multiple hidden layer neurons, calculate the input and output of each hidden layer neuron. As described above, the number of hidden layer neurons of the artificial neural network of the present invention is 4 to 13.

[0099] Calculate the input and output of the output layer neurons. When there are multiple output layer neurons, calculate the input and output of each output layer neuron. As described above, the number of output layer neurons of the artificial neural network of the present invention is 1.

[0100] Compute the output layer error.

[0101] Calculate the hidden layer error.

[0102] Correct the weights and thresholds of the output layer and hidden layer. Correct the weights and thresholds of the output layer and hidden layer according to the calculated output layer error and hidden layer error.

[0103] Determine whether all training data have been trained. If not, select the next part of the balance, elasticity, reduction value, disintegration value, recovery value and sensory evaluation score from the training data as training samples and repeat the above steps. Training a part of the training samples is called a step, and using a new part of the training samples to train the artificial neural network is called starting a new step of training. Repeat the above process until all training data have been trained. When all training data have been trained, a round of training ends.

[0104] After a round of training is completed and all training data have been trained, the performance of the artificial neural network model is judged to determine whether the performance meets the requirements. In one embodiment, the mean square error (MSE) is used to evaluate the network performance. MSE is a common indicator for evaluating whether the network model is suitable. The smaller the MSE value, the better the performance of the neural network model, and vice versa.

[0105] In one embodiment, a performance index of 1e-07 is set for MSE, and if MSE<1e-07, it is considered that the MSE meets the performance requirement. Figure 2 As shown, after a round of training, it will be determined whether the MSE meets the requirement of <1e-07. If the requirement is met, it means that the performance of the artificial neural network is up to standard, and the artificial neural network is output. If the requirement is not met, it means that the performance of the artificial neural network is not up to standard and training needs to be continued. At this time, the number of training steps is updated, and a new round of training is started after the number of training steps is updated. Each round of training will use all the training samples in the training data for training. The process of each round of training refers to the content described above. In one embodiment, the maximum value of the number of training steps is 1000. When the number of training steps is updated, the number of training steps will not exceed the set maximum value of 1000.

[0106] Continue to refer Figure 2 As shown in the figure, after the artificial neural network meets the performance index of MSE<1e-07, the training ends and a trained artificial neural network is output. At this time, the test data consisting of the 20% of the data set left before will be used to evaluate the model performance with the test set. The test sample is input into the generated artificial neural network for network simulation to obtain the output result, which is the comprehensive score.

[0107] The artificial neural network established according to the embodiment of the present invention has good correlation, and the correlation is usually expressed by the correlation coefficient R. The correlation coefficient R represents the linear correlation between the predicted value (comprehensive score) of the artificial neural network and the sensory evaluation score obtained according to the GBT15682-2008 method. The larger the correlation coefficient R, the higher the prediction accuracy.

[0108] Through training and testing, an effective artificial neural network is obtained, which must meet the following conditions: the correlation coefficient R between the comprehensive score and the sensory evaluation score according to GB / T15682-2008 satisfies R 2 >0.97.

[0109] The invention also discloses a method for evaluating the taste quality of rice by using the artificial neural network. Figure 3 A flow chart of a method for evaluating the taste quality of rice according to an embodiment of the present invention is disclosed. Figure 3 As shown, the method comprises the following steps:

[0110] 1) Determination of sample parameters:

[0111] Determination of several parameters and sensory evaluation scores of rice samples;

[0112] 2) Establishment of artificial neural network model (such as Figure 1 shown):

[0113] The artificial neural network is a feedforward network including an input layer, a hidden layer and an output layer; the input layer includes 5 input neurons, namely the balance, elasticity, reduction value, disintegration value and regeneration value of the rice sample; the hidden layer includes several hidden layer neurons; the output layer includes an output neuron: comprehensive score;

[0114] 3) Training of artificial neural network models (such as Figure 2 As shown) and tested, an effective artificial neural network was obtained (the correlation coefficient R between the comprehensive score and the sensory evaluation score according to GB / T 15682-2008 satisfies R 2 >0.97); preferably, before starting training the artificial network neural network in step 3), the input and output training sample data need to be normalized: the mapminmax function is used to normalize the data set so that the data samples all fall within the neuron transfer function region.

[0115] 4) Taste quality evaluation:

[0116] The parameters of the unknown new sample described in step 1) are measured and input into the effective artificial neural network described in step 3) as input values ​​to perform network simulation, and the output value is the comprehensive score; the higher the comprehensive score, the higher the taste quality of the rice.

[0117] Several specific example data are listed below.

[0118] Example 1

[0119] The balance, elasticity, reduction value, disintegration value, recovery value and sensory evaluation score of 12 different varieties of japonica rice from different origins and 15 different varieties of indica rice from different origins were tested. The test results are shown in Table 1.

[0120] Determination of balance and elasticity: Wash the rice samples for 3 times, each washing including rotating clockwise and counterclockwise for 5 times, soak for 30 minutes at a rice-to-water ratio of 1:1.3, steam the rice in a 0.8L Hongzhi rice cooker, simmer for 20 minutes and then break up the rice, take out the inner pot of the rice cooker, put a filter paper with a diameter of 18 cm between the inner pot and the lid (the lid of the rice cooker is separate from the rice cooker), and place it at room temperature 23℃±2℃. When the rice is cooled to 55℃, weigh 8g of the sample, place it in a rice press equipped by Satake and press it for 10s, and then use Satake hardness viscosity meter to measure the balance and elasticity of the rice.

[0121] Reduction value, disintegration value, and recovery value: Each rice sample is crushed to 50-100 mesh (i.e., the rice sample can pass through a 50-mesh sieve, but cannot pass through a 100-mesh sieve; in this embodiment, the specific method is: a 50-mesh sieve and a 100-mesh sieve are placed up and down at the same time, and the crushed rice sample is between the 50-mesh and 100-mesh sieves after sieving), and the remaining steps are performed in accordance with GB / T24852-2010.

[0122] The sensory evaluation was carried out in accordance with GB / T 15682-2008 “Sensory evaluation method for edible quality of steamed paddy and rice”.

[0123] An artificial neural network was used to establish a model for predicting sensory evaluation scores. The input layer contained 5 neurons, namely balance, elasticity, reduction value, disintegration value, and regeneration value; the hidden layer was set to 6 neurons; and the output layer contained one neuron, namely the comprehensive score.

[0124] The specific parameters of the artificial neural network model are set as follows:

[0125] Training function: trainlm;

[0126] Learning function: learngdm;

[0127] Hidden layer transfer function: tansig;

[0128] Output layer transfer function: purelin;

[0129] Network performance target error: 1e-07;

[0130] Maximum number of training steps: 1000;

[0131] Other parameters were set by default in MATLAB software.

[0132] The artificial neural network training results of Example 1 are shown in Table 1:

[0133] Table 1 Artificial neural network training results of Example 1

[0134]

[0135] Figure 4 The training results of the artificial neural network of the training samples of Example 1 are revealed. Figure 4 The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy and Rice Cooking", and the vertical axis is the predicted score (comprehensive score) of the artificial neural network model. The dotted line and the point represent the correlation between the two. 2 =0.9857.

[0136] Example 2

[0137] Example 2: When measuring the balance and elasticity in Example 1, the rice temperature is changed from "cooled to 55° C." to "cooled to 60° C.", and the rest is the same as Example 1.

[0138] The artificial neural network training results of Example 2 are shown in Table 2:

[0139] Table 2 Artificial neural network training results of Example 2

[0140]

[0141] Figure 5 The training results of the artificial neural network of the training samples of Example 2 are revealed. Figure 5 The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy and Rice Cooking", and the vertical axis is the predicted score (comprehensive score) of the artificial neural network model. The dotted line and the point represent the correlation between the two. 2 =0.9852.

[0142] Example 3

[0143] In Example 3, the number of hidden layer neurons in the model in Example 1 is changed from "6" to "10", and the rest is the same as Example 1.

[0144] The artificial neural network training results of Example 3 are shown in Table 3:

[0145] Table 3 Artificial neural network training results of Example 3

[0146]

[0147] Figure 6 The training results of the artificial neural network of the training samples of Example 3 are revealed. Figure 6 The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy and Rice Cooking", and the vertical axis is the predicted score (comprehensive score) of the artificial neural network model. The dotted line and the point represent the correlation between the two. 2 =0.9796.

[0148] Example 4

[0149] In Example 4, the number of hidden layer neurons in the model in Example 2 is changed from "6" to "10", and the rest is the same as Example 2.

[0150] The artificial neural network training results of Example 4 are shown in Table 4:

[0151] Table 4 Artificial neural network training results of Example 4

[0152]

[0153] Figure 7 The training results of the artificial neural network of the training samples of Example 4 are revealed. Figure 7 The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy Rice and Rice Cooking", and the vertical axis is the predicted score (comprehensive score) of the artificial neural network model. The dotted line and the point represent the correlation between the two. 2 =0.9719.

[0154] Comparative Example 1

[0155] Comparative Example 1: When measuring the balance and elasticity in Example 1, the temperature of the rice was changed from "cooled to 55°C" to "cooled to room temperature (23±2°C) for 2h according to the conventional method", and the other standards were the same as in Example 1.

[0156] The artificial neural network training results of Comparative Example 1 are shown in Table 5:

[0157] Table 5 Artificial neural network training results of comparative example 1

[0158]

[0159] Figure 8 The training results of the artificial neural network of the training samples of comparative example 1 are revealed. Figure 8The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy and Rice Cooking", and the vertical axis is the predicted score (comprehensive score) of the artificial neural network model. The dotted line and the point represent the correlation between the two. 2 =0.9096.

[0160] Comparative Example 2

[0161] Comparative Example 2 In Example 1, when measuring the balance and elasticity, the temperature of the rice was changed from "cooled to 55° C." to "measured directly after the cooking was completed", and the other standards were the same as those in Example 1. The artificial neural network training results of Comparative Example 2 are shown in Table 6.

[0162] Table 6 Artificial neural network training results of comparative example 2

[0163]

[0164] Fig. 9 The training results of the artificial neural network of the training samples of comparative example 2 are revealed. Fig. 9 The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy and Rice Cooking", and the vertical axis is the score predicted by the artificial neural network model (comprehensive score). The dotted line and the point represent the correlation between the two. 2 =0.8996.

[0165] Comparative Example 3

[0166] Comparative Example 3: When measuring the reduction value, disintegration value and regeneration value in Example 1, the rice crushing particle size was changed from "50-100 mesh" to "less than 50 mesh (i.e., the crushed rice cannot pass through a 50-mesh sieve)", and the other standards were the same as in Example 1.

[0167] The artificial neural network training results of Comparative Example 3 are shown in Table 7:

[0168] Table 7 Artificial neural network training results of comparative example 3

[0169]

[0170] Fig.10 The training results of the artificial neural network of the training samples of comparative example 3 are revealed. Fig.10 The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy and Rice Cooking", and the vertical axis is the predicted score (comprehensive score) of the artificial neural network model. The dotted line and the point represent the correlation between the two. 2=0.9321.

[0171] Comparative Example 4

[0172] Comparative Example 4 When measuring the reduction value, disintegration value and regeneration value in Example 1, the rice crushing particle size was changed from "50-100 mesh" to "greater than 100 mesh (i.e., the crushed rice can pass through a 100 mesh sieve)", and the other standards were the same as in Example 1.

[0173] The artificial neural network training results of Comparative Example 4 are shown in Table 8:

[0174] Table 8 Artificial neural network training results of comparative example 4

[0175]

[0176] Fig.11 The training results of the artificial neural network of the training samples of Comparative Example 4 are revealed. Fig.11 The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy and Rice Cooking", and the vertical axis is the predicted score (comprehensive score) of the artificial neural network model. The dotted line and the point represent the correlation between the two. 2 =0.9352.

[0177] It can be seen from the above Examples 1-4 and Comparative Examples 1-4 that when the rice samples are crushed to 50-100 visual reduction value, disintegration value and regeneration value, and the balance and elasticity are measured after the rice temperature is cooled to 55-60°C, an effective artificial neural network can be trained, that is, the correlation coefficient R between the comprehensive score and the sensory evaluation score according to GB / T 15682-2008 satisfies R 2 >0.97.

[0178] Comparative Example 5

[0179] Comparative Example 5 is to directly use the Satake taste meter to detect the taste value of rice. The specific detection method is:

[0180] The rice sample was washed three times, each washing including rotating clockwise and counterclockwise for 5 times respectively, soaked for 30 minutes at a rice-to-water ratio of 1:1.3, steamed into rice in a 0.8L Hongzhi rice cooker, and simmered for 20 minutes before breaking up the rice. The inner pot of the rice cooker was taken out, and a filter paper with a diameter of 18 cm was placed between the inner pot and the lid. The rice was placed at room temperature of 23℃±2℃ for 2h. When cooled to room temperature, 7g of indica rice and 8g of japonica rice were weighed, placed in a rice press equipped by Satake and pressed for 10s, and then the taste value of the rice was determined using a Satake taste meter.

[0181] The test results of Comparative Example 5 are shown in Table 9:

[0182] Table 9 Taste value and sensory evaluation score of rice detected by using Satake taste meter in comparative example 5

[0183]

[0184]

[0185] Fig.12 and 13 These are the analysis results of the actual sensory evaluation scores and taste values ​​of japonica rice and indica rice. Fig.12 , 13 The correlation between the actual sensory evaluation score and the taste value in Comparative Example 5 is revealed. The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy and Rice Cooking", and the vertical axis is the taste value (comprehensive score) detected by the Satake taste meter. The dotted line and points represent the correlation between the two.

[0186] Correlation coefficient R between japonica rice and indica rice in comparative example 5 2 The correlation between the taste value of indica rice and the actual sensory evaluation score is extremely low, so the Satake taste meter is not suitable for the evaluation of the taste quality of indica rice.

[0187] Verification Example

[0188] The following four rice varieties are used in this embodiment: Longyang 16, Wuyou 4, Huanghuanian and Fengliangyou.

[0189] Table 10 reveals the results of the taste quality evaluation of compound rice using artificial neural network:

[0190] Table 10 Evaluation results of taste quality of compound rice using artificial neural network

[0191]

[0192] Fig.14 The correlation analysis results between the actual sensory evaluation score of the formula rice and the model predicted score (comprehensive score). The horizontal axis is the actual sensory evaluation score obtained according to GB / T 15682-2008 "Sensory Evaluation Method for Edible Quality of Paddy Rice and Steamed Rice", and the vertical axis is the sensory evaluation score (comprehensive score) predicted by the model. The dotted line and the point represent the correlation between the two. Correlation coefficient R 2 =0.973, it can be seen that the model has a high accuracy in predicting compound rice.

[0193] Table 11 reveals the evaluation results of single rice varieties:

[0194] Table 11 Evaluation results of taste quality of single rice varieties using artificial neural network

[0195] Serial number formula area Classification Sensory evaluation score Comprehensive score 1 100% Longyang 16 Harbin Japonica rice 77.9 78.1 2 100% Wuyou No. 4 Harbin Japonica rice 86.1 86.0 3 100% yellow flower sticky Wuhan Indica Rice 75.3 75.5 4 100% Fengliangyou Wuhan Indica Rice 70.7 70.3

[0196] Based on the above embodiments, the artificial neural network model of the present invention has small errors in predicting the taste quality of single-variety rice and the taste quality of formulated rice, and can be used to predict the taste quality of rice and directly screen out formulated rice with good taste quality.

[0197] The artificial neural network and method for evaluating rice taste quality of the present invention have the following advantages:

[0198] 1) The gelatinization characteristics of rice (reduction value, disintegration value, and recovery value) and the texture characteristics of rice (balance and elasticity) were significantly correlated with the sensory evaluation results, with P values ​​of <0.01, indicating a highly significant correlation; the Pearson correlation coefficient was between 0.6 and 1, indicating a strong correlation. Therefore, the above indicators can be used to accurately determine the taste quality of rice.

[0199] 2) The testing of rice gelatinization characteristics (reduction value, disintegration value, and recovery value) takes only 13 minutes, and the testing of rice texture characteristics (balance, elasticity) takes only 1-2 minutes, while the testing of amylose content and protein often takes 1-2 days. Therefore, compared with the method of using physical and chemical indicators to determine the taste quality of rice, this method greatly shortens the time and is more convenient and quick.

[0200] 3) At present, the balance and elasticity of cooked rice are measured after the cooked rice is cooled to room temperature. At this time, the rice has begun to age. Different rice ages at different rates, and its taste is inconsistent with the taste of rice sensory evaluation in accordance with GB / T15682-2008 "Sensory Evaluation Method for Cooked Rice and Rice". Therefore, placing the rice at 55-60℃ can maintain consistency with human sensory evaluation, and the judgment result is more accurate.

[0201] 4) This method uses artificial neural network to predict the taste quality of rice. Compared with traditional regression equation, response surface, principal component analysis and other methods, artificial neural network has stronger learning ability, adaptability and higher prediction accuracy. The prediction accuracy of the model established by artificial neural network can reach 99%.

[0202] 5) Japanese rice is all japonica rice and there are fewer varieties. The Satake Taste Meter was developed based on Japanese conditions. Although a database of Chinese rice has been established, it still cannot accurately reflect the taste quality of samples when testing indica rice. Compared with the Japanese Satake Taste Meter, this model covers more comprehensive rice varieties, has more accurate prediction results, and is more suitable for Chinese rice.

[0203] It should also be noted that the embodiments listed above are only specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and similar changes or deformations made therewith can be directly derived or easily associated with the contents disclosed by those skilled in the art from the present invention, and should all belong to the protection scope of the present invention. The above embodiments are provided to those familiar with the art to implement or use the present invention, and those familiar with the art can make various modifications or changes to the above embodiments without departing from the inventive concept of the present invention. Therefore, the protection scope of the present invention is not limited by the above embodiments, but should be the maximum scope of the innovative features mentioned in the claims.

Claims

1. A method for evaluating the taste quality of rice, characterized in that: The following steps are involved: 1) Determination of sample parameters: Determining several parameters and sensory evaluation scores of the rice sample; the several parameters are balance, elasticity, reduction value, disintegration value and regeneration value of the rice sample; The specific method for determining the sample parameters is as follows: Method for determining balance and elasticity: wash the rice sample several times, soak it for 30 minutes at a rice-water ratio of 1:1.3, steam it into rice with an electric rice cooker, simmer it for 20 minutes and then break up the rice, take out the inner pot of the electric rice cooker and place it at room temperature, and when the rice cools to 55-60°C, weigh the sample and press it, and then use a hardness viscosity meter to determine the balance and elasticity of the rice; Determination method of reduction value, disintegration value and recovery value: grind the rice sample to 50-100 mesh, and then determine the reduction value, disintegration value and recovery value according to the method described in GB / T24852-2010; The sensory evaluation score is based on GBT15682-2008 "Sensory evaluation method for the edible quality of rice and steamed rice"; 2) Establishment of artificial neural network model: The artificial neural network is a feedforward network including an input layer, a hidden layer and an output layer; the input layer includes a plurality of input neurons, i.e., a plurality of parameters of the rice sample; the hidden layer includes a plurality of hidden layer neurons; the output layer includes an output neuron: a comprehensive score; 3) Training and testing the artificial neural network model to obtain an effective artificial neural network; the effective artificial neural network must meet the following conditions: the correlation coefficient R between the comprehensive score and the sensory evaluation score according to GB / T 15682-2008 satisfies R 2 >0.97; 4) Taste quality evaluation: The parameters of the unknown new sample described in step 1) are measured and input into the effective artificial neural network described in step 3) as input values ​​to perform network simulation, and the output value is the comprehensive score; the higher the comprehensive score, the higher the taste quality of the rice.

2. The method for evaluating rice taste quality according to claim 1, characterized in that: The hardness viscometer is a Satake hardness viscometer.

3. The method for evaluating rice taste quality according to claim 1, characterized in that: The number of hidden layer neurons included in the hidden layer in step 2) is determined according to the following formula: Where l is the number of input neurons, m is the number of output neurons, a is a constant between [1,10], and N is the number of hidden layer neurons.

4. The method for evaluating rice taste quality according to claim 1, characterized in that: Step 3) The training of the artificial neural network includes: A data set containing several parameters of rice samples is used as training data, and training and learning are performed using the training data starting from random initial weights and thresholds; the weights and thresholds are modified during the training and learning process, and one round is defined as one round of training for all the training data; after completing each round of training, the performance indicator mean square error is detected, and if the mean square error does not meet the performance requirements, the number of training steps is updated, and the training data is reused for the next round of training; if the mean square error meets the performance requirements, the training is terminated, and the artificial neural network is output; wherein, the mean square error performance requirement meets MSE<1e-07.

5. The method for evaluating the taste quality of rice according to claim 4, characterized in that: Each round of training of the artificial neural network includes the following steps: S1. Initialize weights and thresholds; S2. Select a part of rice sample parameters from the training data as training samples and input them into the input layer; S3. Calculate the input and output of hidden layer neurons; S4. Calculate the input and output of the output layer neurons; S5. Calculate the output layer error; S6. Calculate the hidden layer error; S7. Modify the weights and thresholds of the output layer and hidden layer; S8. Select the next part of rice sample parameters from the training data as training samples and repeat the above steps until all the training data are trained.

6. The method for evaluating rice taste quality according to claim 1, characterized in that: In step 3), before training the artificial network neural network, the input and output training sample data need to be normalized: the mapminmax function is used to normalize the data set so that the data samples all fall within the neuron transfer function region.

7. The method for evaluating the taste quality of rice according to any one of claims 1 to 6, characterized in that: The method is applied in compounding of rice of different varieties.

8. A method for evaluating the taste quality of rice, characterized in that: The method for evaluating the taste quality of rice according to any one of claims 1 to 7 is used to evaluate the taste of the rice according to the obtained comprehensive score, wherein the higher the score, the better the taste.

9. A system for evaluating the taste quality of rice, characterized in that: include: machine readable storage; and a processor configured to execute machine-readable instructions, the instructions being a number of input neurons of the input layer; After reading the instruction, the system runs the instruction to execute the method for evaluating the taste quality of rice according to any one of claims 1 to 7, and the output value is the comprehensive score of the rice.

Images