CN105548268A - Method for fast predicting processing time of pecan based on electronic nose - Google Patents

Abstract

The invention discloses a method for fast predicting processing time of pecan based on an electronic nose. The method specifically comprises the following steps of obtaining pecan samples of different processing time after boiling, drying and baking fresh pecans; directly detecting headspace gas generated by the pecans by using the electronic nose; extracting a characteristic value of a sensor array response curve by using a stability value method; building predicting models based on electronic nose signal characteristic values by using a plurality of regression methods; selecting the predicting model with a big coefficient of determination and a small root-mean-square error to be a final processing time predicting model; performing electronic nose detection, characteristic value extraction and the like on the samples, under detection, with unknown processing time; leading the extracted characteristic values into the built predicting model so as to obtain the exact processing time; identifying the processing quality of the samples under detection according to the predicting results of processing time. The method is significant in pecan processing field, so that the method is worth widely popularizing.

Description

A kind of method based on Electronic Nose fast prediction hickory nut process time

Technical field

The present invention relates to quality of agricultural product detection field, particularly relate to a kind of method based on Electronic Nose fast prediction hickory nut process time.

Background technology

Hickory nut is the distinctive nut fruits product of China, owing to containing the nutriments such as abundant grease and protein, and has special nutty flavor and mouthfeel, is more and more subject to liking of consumer.Special aroma and the mouthfeel of hickory nut produce in bake process, and different baking times has a great impact hickory nut overall quality, and baking time deficiency causes hickory nut cannot produce its distinctive fragrance component; Baking time long causing occurs being charred taste and astringent taste, and hickory nut Total Product quality declines.Now, whether hickory nut processing enterprise often relies on the subjective appreciation of the individual of operator sufficient to pass judgment on baking time.But subjective appreciation has the shortcomings such as subjectivity is strong, poor repeatability, makes judged result lack accuracy and reliability, therefore, is badly in need of a kind of method can differentiating hickory nut processing quality and fast prediction hickory nut baking time quickly and accurately.

Electronic Nose is a kind of instrument of simulated animal olfactory system, comprise sensor array and pattern-recognition two parts, its principle of work is: the smell that sample volatilizes acts on sensor array, produce response signal, this signal is called pattern or " the smell finger-print " of sample smell, utilize the method establishment such as multivariate statistical analysis and neural network to classify and forecast model, thus distinguish the hickory nut sample of different baking time fast and predict baking time accurately.

Summary of the invention

For the deficiencies in the prior art, the invention provides a kind of method based on Electronic Nose fast prediction hickory nut process time, the method can distinguish the hickory nut sample Accurate Prediction baking time simultaneously of different baking time fast, has higher actual application value, should be widely promoted.

In order to achieve the above object, the technical solution adopted in the present invention is as follows: a kind of method based on Electronic Nose fast prediction hickory nut process time, specifically comprises the steps:

(1) hickory nut detects sample processing procedure: choose fresh hickory nut, boils and remove astringent taste in 3-4 hour in boiling water; Take out well-done hickory nut, at 35 DEG C, the climatic chamber inner drying 3.5-4.5 hour of 30% relative humidity; Dried hickory nut is placed in hot-air oven and is divided into some batches and toasts, each batch of baking time difference, thus obtain process time different hickory nut and detect sample;

(2) detection by electronic nose process: detect in sample by obtaining process time different hickory nut in step (1), the hickory nut of Identical Processing has 3 groups of different detection samples at least, detect sample according to the ratio of 1/50ml headspace gas by each group and put into tasteless closed container respectively, the volume of described closed container is not less than 500ml; Leave standstill 45-75 minute at room temperature, the smell that hickory nut is distributed reaches capacity; Before each detection by electronic nose starts, use dried and clean air to clean electric nasus system, arranging cleaning flow velocity is 500ml/min-700ml/min, and scavenging period is 60-80 second; After having cleaned, extract the headspace gas in closed container, arranging the flow velocity extracting headspace gas is 150ml/min-300ml/min, and detection time is 70-90 second; The hickory nut of detection record sensor array to different process time detects the response signal of sample, and then obtains the response curve that the hickory nut of sensor array to different process time detects sample;

(3) characteristics extraction process: the eigenwert on the sensor array response curve adopting stationary value method extraction step (2) to obtain, described stationary value method is specially: the response signal extracting the stabilization sub stage on sensor array response curve;

(4) forecast model process of establishing: using the eigenwert that extracts in step (3) as independent variable, the process time of the detection sample that process time is known is as dependent variable, set up forecast model process time, select wherein to determine that coefficient is large and root-mean-square error is little forecast model is as final forecast model process time;

(5) process time forecasting process: the eigenwert being obtained the detection sample response curve of the unknown process time by step (2) and step (3), by forecast model process time set up in the eigenwert steps for importing (4) obtained, thus obtain the process time of the hickory nut of the unknown process time.

Further, in described step (1), the baking temperature of hot-air oven is 150 DEG C, and each batch of baking time all must not more than 35 minutes, and each batch of baking time difference must not be less than 1 minute.

Further, the determination coefficient in described step (4) and the computing formula of root-mean-square error as follows:

$R^2=\frac{\left(\underset{i=1}{\overset{N}{\Σ}}\right(X_i-\stackrel{\‾}{X}\left)\right(Y_i-\stackrel{\‾}{Y}\left)\right)}{\underset{i=1}{\overset{N}{\Σ}}{(X_i-\stackrel{\‾}{X})}^2\underset{i=1}{\overset{N}{\Σ}}{(Y_i-\stackrel{\‾}{Y})}^2}$

$RMSE=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{(X_i-{\stackrel{\‾}{Y}}_i)}^2}$

Wherein, R ²for determining coefficient; RMSE is root-mean-square error; N is process time of using in forecast model process of establishing of known number of samples; X _ifor the actual value of i-th sample process time in forecast model process of establishing; for the mean value of the actual value of all samples process time in forecast model process of establishing; Y _ifor the predicted value of i-th sample process time in forecast model process of establishing; for the mean value of sample predicted values process time all in forecast model process of establishing.

Further, described step (4) forecast model method for building up middle process time is data statistical analysis method or neural network algorithm.

Beneficial outcomes of the present invention is, by adopting Electronic Nose directly to detect hickory nut product, simple to operate, do not need complicated front sample preparation and the instrument of costliness; Adopt in method and determine coefficient and the performance of hickory nut forecast model process time constructed by root-mean-square error assessment, select wherein to determine that coefficient is large and root-mean-square error is little model is as final forecast model; Constructed process time, forecast model can fast prediction hickory nut process time accurately, thus accurately controlled the quality of hickory nut in process, had higher actual application value, should be widely promoted.

Accompanying drawing explanation

Fig. 1 is Electronic Nose response signal in the embodiment of the present invention;

Fig. 2 is the LDA classifying quality figure of different baking time hickory nut sample in the embodiment of the present invention;

Fig. 3 is the impact of performance figure based on partial least square method forecast model in the embodiment of the present invention;

Fig. 4 is the forecast model based on BP neural network in the embodiment of the present invention;

Fig. 5 is the impact of performance figure based on BP neural network prediction model in the embodiment of the present invention.

Embodiment

Below in conjunction with embodiment, the present invention is further illustrated.

Based on the method for Electronic Nose fast prediction hickory nut process time, its step is as follows:

(1) hickory nut detects sample processing procedure: choose fresh hickory nut, boils and remove astringent taste in 3-4 hour in boiling water; Take out well-done hickory nut, at 35 DEG C, the climatic chamber inner drying 3.5-4.5 hour of 30% relative humidity; Be divided into five batches in the hot-air oven dried hickory nut being placed in 150 DEG C to toast, each batch of baking time difference, is respectively 18 minutes, 22 minutes, 25 minutes, 28 minutes and 32 minutes, thus the hickory nut obtaining different process time detects sample;

(2) detection by electronic nose process: the hickory nut obtaining different process time in step (1) is detected in sample, the hickory nut of Identical Processing has 3 groups of different detection samples at least, according to the ratio of 1/50ml headspace gas, each detection sample is put into tasteless closed container respectively, the volume of described closed container is not less than 500ml; After putting into hickory nut, use tasteless preservative film to seal, and leave standstill 45-75 minute at room temperature, the smell that hickory nut is distributed is full of whole closed container and reaches capacity, thus obtains headspace gas; Before each detection by electronic nose starts, use dried and clean air to clean electric nasus system, arranging cleaning flow velocity is 500ml/min-700ml/min, and scavenging period is 60-80 second; After having cleaned, Electronic Nose extracts the headspace gas in closed container, and arranging the flow velocity extracting headspace gas is 150ml/min-300ml/min, and detection time is 70-90 second; According to the speed record sensor response signal of 1 time/second in testing process, the ratio of conductivity G0 when conductivity G when response signal is sensor array detection sample gas and detection pure air, thus obtain the response curve of sensor array to different hickory nut detection process time sample;

(3) characteristic extraction procedure: the response of gas sensor stabilization sub stage can represent the odor characteristics of detected sample, therefore the eigenwert on the sensor array response curve adopting stationary value method extraction step (2) to obtain, wherein stationary value method is specially: the response signal extracting the stabilization sub stage on sensor array response curve;

(4) forecast model process of establishing: using the eigenwert that extracts in step (3) as independent variable, the process time of the detection sample that process time is known is as dependent variable, set up forecast model process time, select wherein to determine that coefficient is large and root-mean-square error is little forecast model is as final forecast model process time;

(5) process time forecasting process: the eigenwert being obtained the detection sample response curve of the unknown process time by step (2) and step (3), by forecast model process time set up in the eigenwert steps for importing (4) obtained, thus obtain the process time of the detection sample of the unknown process time; According to the process time of the detection sample of the unknown process time of prediction, the processing quality detecting sample can be known.In specific implementation process, the hickory nut product of definition baking time in 23-27 minute is that product are of fine quality, and the hickory nut product of baking time in 19-23 minute or 27-31 minute is that quality is good, and the hickory nut product being greater than 31 minutes or being less than 19 minutes is poor quality; The estimated performance of aid illustration forecast model is carried out by the discriminating accuracy (W) of processing quality;

Further, the computing formula of the discriminating accuracy in the determination coefficient in described step (4) and root-mean-square error and step (5) is as follows:

$R^2=\frac{\left(\underset{i=1}{\overset{N}{\Σ}}\right(X_i-\stackrel{\‾}{X}\left)\right(Y_i-\stackrel{\‾}{Y}\left)\right)}{\underset{i=1}{\overset{N}{\Σ}}{(X_i-\stackrel{\‾}{X})}^2\underset{i=1}{\overset{N}{\Σ}}{(Y_i-\stackrel{\‾}{Y})}^2}$

$RMSE=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{(X_i-{\stackrel{\‾}{Y}}_i)}^2}$

Wherein, R ²for determining coefficient; RMSE is root-mean-square error; W is for differentiating accuracy; N is process time of using in forecast model process of establishing of known number of samples; X _ifor the actual value of i-th sample process time in forecast model process of establishing; for the mean value of the actual value of all samples process time in forecast model process of establishing; Y _ifor the predicted value of i-th sample process time in forecast model process of establishing; for the mean value of sample predicted values process time all in forecast model process of establishing; N _correctlyfor correct number of samples of classifying.

Embodiment:

In the present embodiment, select Linan City of Zhejiang Province hickory nut as detected object; Select in the same size, the undamaged fresh hickory nut of external standard, in boiling water, boil 4 hours, remove astringent taste; Well-done hickory nut is put in 35 DEG C, the climatic chamber inner drying of 30% relative humidity 4 hours, removes excessive moisture in hickory nut; Be divided into five batches in the hot-air oven finally dried hickory nut being placed in 150 DEG C to toast, each batch of baking time difference, baking time is respectively 18 minutes, 22 minutes, 25 minutes, 28 minutes and 32 minutes; After baking, hickory nut at room temperature places 1 hour, makes its cool to room temperature, then carries out detection by electronic nose.

In the present embodiment, the PEN2 type Electronic Nose of German Airsense company is adopted to be described in detail as detecting instrument.The sensor array that PEN2 Electronic Nose adopts comprises 10 dissimilar metal oxide sensors, and sensor model number and principal character describe as shown in table 1.

Table 1PEN2 type Electronic Nose sensor model number and key property thereof

Detection by electronic nose process is carried out in accordance with the following steps: the hickory nut sample through different time baking is numbered 18min respectively, 22min, 25min, 28min, 32min, difference random selecting 30 groups from each sample, often organize 10 hickory nuts, be put in 500ml beaker, in room temperature (20 ± 1 DEG C), 1 hour is left standstill with after preservative film sealing, the gas given out until hickory nut is full of whole headspace and after reaching balance, extract headspace gas and carry out detection by electronic nose, the detection by electronic nose time is 80 seconds, the speed extracting headspace gas is 200ml/min, each second records primary transducer response, scavenging period is 70 seconds, the speed of extracting air is 600ml/min.

In the present embodiment, Electronic Nose response curve as shown in Figure 1.Experiment proves, the eigenwert that different feature extracting methods extracts, its classifying quality is also different, wherein in Electronic Nose response curve, the response signal of stabilization sub stage has best classifying quality, therefore select the response at 75 seconds places to carry out follow-up modeling as eigenwert, within the 75th second, locate the LDA classifying quality of response as shown in Figure 2 based in response curve.The eigenwert that the method is extracted is the data matrix of 150 × 10, and the row wherein in matrix represents 150 hickory nut samples, and the row in matrix represent 10 sensors.30 samples in often kind of hickory nut sample are divided at random checksum set (25 samples) and test set (5 samples), use the Electronic Nose data in checksum set to set up regression model, the estimated performance of the Electronic Nose data in test set to institute's Modling model is tested.

In the regression model of all foundation, select wherein to determine coefficients R ²the little forecast model of large and root-mean-square error RMSE is as final forecast model process time; In addition, the hickory nut product of definition baking time in 23-27 minute is that product are of fine quality, and the hickory nut product of baking time in 19-23 minute or 27-31 minute is that quality is good, and the hickory nut product being greater than 31 minutes or being less than 19 minutes is poor quality; The estimated performance of aid illustration forecast model is carried out by the discriminating accuracy (W) of processing quality; Above-mentioned mentioned determination coefficients R ², root-mean-square error RMSE and differentiate that the computing formula of accuracy W is as follows:

$R^2=\frac{\left(\underset{i=1}{\overset{N}{\Σ}}\right(X_i-\stackrel{\‾}{X}\left)\right(Y_i-\stackrel{\‾}{Y}\left)\right)}{\underset{i=1}{\overset{N}{\Σ}}{(X_i-\stackrel{\‾}{X})}^2\underset{i=1}{\overset{N}{\Σ}}{(Y_i-\stackrel{\‾}{Y})}^2}$

$RMSE=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{(X_i-{\stackrel{\‾}{Y}}_i)}^2}$

Wherein, R ²for determining coefficient; RMSE is root-mean-square error; W is for differentiating accuracy; N is process time of using in forecast model process of establishing of known number of samples; X _ifor the actual value of i-th sample process time in forecast model process of establishing; for the mean value of the actual value of all samples process time in forecast model process of establishing; Y _ifor the predicted value of i-th sample process time in forecast model process of establishing; for the mean value of sample predicted values process time all in forecast model process of establishing; N _correctlyfor hickory nut processing quality differentiates correct number of samples.

In the present embodiment, adopt the forecast model of partial least squares regression (PLSR) and BP neural network hickory nut process time respectively, wherein, the PLSR regression model set up based on Electronic Nose data in checksum set is shown below:

T (process time)=-428.132-870.818 × S1-4.532 × S2+89.975 × S3+180.936 × S4+898.908 × S5+248.104 × S6+34.335 × S7-402.819 × S8+394.985 × S9-112.526 × S10 wherein S1-S10 represents the eigenwert of each sensor in electric nasus system.

In the present embodiment, based on hickory nut forecast model process time of PLSR performance as shown in Figure 3.Can carry out the discriminating of hickory nut processing quality according to predicting the outcome of process time, the hickory nut processing quality identification result wherein predicted the outcome based on PLSR is as shown in table 2 and table 3.

The checksum set hickory nut processing quality identification result of table 2PLSR

The test set hickory nut attribute classification result of table 3PLSR

In the present embodiment, based on BP neural network forecast model as shown in Figure 4, this model is three-layer neural network, and ground floor is input layer, and neuron number is 10 (10 Electronic Nose data feature values that input is chosen); The second layer is hidden layer, the neuron number of hidden layer directly affects the performance of institute's Modling model, through experimental verification, has higher estimated performance when the neuron number of hidden layer is 14, the computer resource simultaneously consumed is also relatively less, reaches optimum cost performance; Third layer is output layer, and output layer exports predicting the outcome of storage time, and neuron number is 1.Based on the forecast model of BP neural network and performance thereof as shown in Figure 5.According to the process time of BP neural network prediction, carry out the discriminating of processing quality, its identification result is as shown in table 4 and table 5.

The checksum set hickory nut attribute classification result of table 4BP neural network

The test set hickory nut attribute classification result of table 5BP neural network

Hickory nut forecast model process time of contrast two kinds of method establishment can find, based on the checksum set of the forecast model of BP neural network and the determination coefficients R of test set ²all large than the determination coefficient of the checksum set of the forecast model based on PLSR and test set, and root-mean-square error RMSE is less, select regression model based on BP neural network as the forecast model of hickory nut process time so final.In addition, according to the discriminating of the hickory nut processing quality carried out that predicts the outcome of hickory nut process time, also there is larger difference in its result, the discriminating accuracy wherein predicted the outcome based on PLSR in checksum set and test set is respectively 62.4% and 56%, and be respectively 98.4% and 96% based on the discriminating accuracy of BP neural network prediction result in checksum set and test set, illustrate that using BP neural network to carry out hickory nut predicts process time, prediction effect is best.

According to the method in summary of the invention, the operation such as detection by electronic nose, characteristics extraction is carried out to hickory nut sample to be measured, then the eigenwert extracted is imported the BP neural network prediction model set up, the accurate machining time of final acquisition unknown sample, carrying out processing quality discriminating according to predicting the outcome in addition, final processing quality can be obtained.

The present embodiment sufficient proof present invention achieves the fast prediction to hickory nut process time, and predictablity rate is higher, and have very important significance at hickory nut manufacture field, therefore the method should be widely promoted.

Claims (4)

1., based on the method for Electronic Nose fast prediction hickory nut process time, it is characterized in that, specifically comprise the steps:

(1) hickory nut detects sample processing procedure: choose fresh hickory nut, boils and remove astringent taste in 3-4 hour in boiling water; Take out well-done hickory nut, at 35 DEG C, the climatic chamber inner drying 3.5-4.5 hour of 30% relative humidity; Dried hickory nut is placed in hot-air oven and is divided into some batches and toasts, each batch of baking time difference, thus obtain process time different hickory nut and detect sample;

(2) detection by electronic nose process: detect in sample by obtaining process time different hickory nut in step (1), the hickory nut of Identical Processing has 3 groups of different detection samples at least, detect sample according to the ratio of 1/50ml headspace gas by each group and put into tasteless closed container respectively, the volume of described closed container is not less than 500ml; Leave standstill 45-75 minute at room temperature, the smell that hickory nut is distributed reaches capacity; Before each detection by electronic nose starts, use dried and clean air to clean electric nasus system, arranging cleaning flow velocity is 500ml/min-700ml/min, and scavenging period is 60-80 second; After having cleaned, extract the headspace gas in closed container, arranging the flow velocity extracting headspace gas is 150ml/min-300ml/min, and detection time is 70-90 second; The hickory nut of detection record sensor array to different process time detects the response signal of sample, and then obtains the response curve that the hickory nut of sensor array to different process time detects sample;

(3) characteristics extraction process: the eigenwert on the sensor array response curve adopting stationary value method extraction step (2) to obtain, described stationary value method is specially: the response signal extracting the stabilization sub stage on sensor array response curve;

(4) forecast model process of establishing: using the eigenwert that extracts in step (3) as independent variable, the process time of the detection sample that process time is known is as dependent variable, set up forecast model process time, select wherein to determine that coefficient is large and root-mean-square error is little forecast model is as final forecast model process time;

(5) process time forecasting process: the eigenwert being obtained the detection sample response curve of the unknown process time by step (2) and step (3), by forecast model process time set up in the eigenwert steps for importing (4) obtained, thus obtain the process time of the hickory nut of the unknown process time.

2. the method based on Electronic Nose fast prediction hickory nut process time according to claim 1, it is characterized in that, in described step (1), the baking temperature of hot-air oven is 150 DEG C, each batch of baking time all must not more than 35 minutes, and each batch of baking time difference must not be less than 1 minute.

3. the method based on Electronic Nose fast prediction hickory nut process time according to claim 1, is characterized in that, the determination coefficient in described step (4) and the computing formula of root-mean-square error as follows:

$R^2=\frac{{\left(\underset{i=1}{\overset{N}{\Σ}}\right(X_i-\stackrel{\‾}{X}\left)\right(Y_i-\stackrel{\‾}{Y}\left)\right)}^2}{\underset{i=1}{\overset{N}{\Σ}}{\left(X_i-\stackrel{\‾}{X}\right)}^2\underset{i=1}{\overset{N}{\Σ}}{(Y_i-\stackrel{\‾}{Y})}^2}$

$RMSE=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{(X_i-{\stackrel{\‾}{Y}}_i)}^2}$

Wherein, R ²for determining coefficient; RMSE is root-mean-square error; N is process time of using in forecast model process of establishing of known number of samples; X _ifor the actual value of i-th sample process time in forecast model process of establishing; for the mean value of the actual value of all samples process time in forecast model process of establishing; Y _ifor the predicted value of i-th sample process time in forecast model process of establishing; for the mean value of sample predicted values process time all in forecast model process of establishing.

4. the method based on Electronic Nose fast prediction hickory nut process time according to claim 1, is characterized in that, described step (4) forecast model method for building up middle process time is data statistical analysis method or neural network algorithm.