CN113984946B - Crayfish freshness detection method based on gas phase electronic nose and machine learning - Google Patents

Abstract

The invention discloses a crawfish freshness detection method based on a gas-phase electronic nose and machine learning, which comprises the steps of placing a crawfish sample in a beaker, sealing the sample by using a double-layer preservative film, and standing for headspace; preheating an ultra-fast gas phase electronic nose instrument, and deeply inserting a sample injection needle into a beaker for sampling to obtain a chromatogram map; normalizing the maximum value and the minimum value of the chromatogram peak height; preprocessing the baseline data of the peak height, and eliminating the label noise of the chromatogram by using belief learning; performing feature extraction on the chromatogram by using a sequence model to obtain the trend features of the chromatogram with different freshness and odor changes; extracting the content characteristics of the volatile compounds corresponding to each retention time through a multilayer perceptron according to the chromatogram trend characteristics, and splicing the chromatogram trend characteristics and the content characteristics of the volatile compounds; performing feature classification by using the spliced features of the feedforward neural network; the method can accurately obtain the odor information of the crayfishes with different freshness, and realizes the accurate classification of the freshness of the crayfishes.

Description

A crayfish freshness detection method based on gas-phase electronic nose and machine learning

technical field

The invention relates to the technical field of crayfish freshness detection, in particular to a crayfish freshness detection method based on gas-phase electronic nose and machine learning.

Background technique

Crayfish, also known as Procambarus clarkii, is one of the important freshwater economic aquatic products in my country. Crayfish are deeply loved by consumers because of their tender meat, delicious taste and rich nutrition. In recent years, my country's crayfish industry has developed rapidly, and the aquaculture area and output have increased rapidly. In 2020, the total output of crayfish aquaculture in my country will reach 2.3937 million tons. However, due to the complex breeding environment of crayfish, many microorganisms are usually carried on the surface and inside of the crayfish, which leads to a decrease in the freshness of the crayfish to varying degrees during the fresh storage, transportation and processing, and even corruption and deterioration due to death, resulting in small crayfish. Potential safety risks of processed lobster products.

The electronic nose is a device that comprehensively imitates the biological olfactory system, and classifies and identifies samples by identifying volatile compounds in the sample. The electronic nose does not require any sample pretreatment, nor does it require solvents. It has a wide range of applications, short detection time, and high sensitivity, and can give more comprehensive and objective results. The types of electronic noses can be divided into sensor electronic noses, mass spectrometry electronic noses and ultra-fast gas phase electronic noses. Sensor-type electronic noses are most commonly used in the fields of food, medicine, and traditional Chinese medicine, but sensor-type electronic noses also have limitations such as time-consuming, complicated sensors, and large external influences. Heracles II ultra-fast gas-phase electronic nose is a new type of odor analysis instrument, equipped with two chromatographic columns with different polarities, the chromatographic peaks obtained from the gas phase replace the sensor signals in the traditional sensor-type electronic nose, and more compound signals are obtained. It can accurately separate volatile compounds with different polarities, has the advantages of high sensitivity, short detection time, and wide application range. It has played an important role in the classification and identification of milk, liquor, mutton, and fruits. However, the difference in the smell of live crayfish for different periods of time and different freshness is small, and it is difficult to accurately judge the freshness of crayfish through simple data dimensionality reduction methods in the data processing stage, such as principal component analysis.

Contents of the invention

The purpose of this section is to outline some aspects of embodiments of the invention and briefly describe some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and titles of this application, to avoid obscuring the purpose of this section, the abstract and titles, and such simplifications or omissions should not be used to limit the scope of the invention.

In view of the above existing problems, the present invention is proposed.

Therefore, the present invention provides a method for detecting the freshness of crayfish based on gas-phase electronic nose and machine learning, which can accurately obtain the odor information of crayfish with different freshness, and accurately judge the freshness of crayfish.

In order to solve the above technical problems, the present invention provides the following technical solutions: including, placing the crayfish sample in the beaker, sealing it with double-layer plastic wrap, and leaving it in the headspace; preheating the ultra-fast gas-phase electronic nose instrument, and inserting the sampling needle into the beaker Perform sampling to obtain chromatograms; normalize and preprocess the maximum and minimum peak heights of the chromatograms; preprocess the baseline data of peak heights, and use the confidence learning strategy to remove the label noise of crayfish samples; use the sequence The model extracts the features of the chromatograms to obtain the chromatogram trend features of different freshness odor changes; according to the chromatogram trend features and extracts the volatile compound content features corresponding to each retention time through a multi-layer perceptron, and splices the chromatogram Graph trend features and volatile compound content features; use feedforward neural network spliced features for feature classification.

As a preferred solution of the crayfish freshness detection method based on gas-phase electronic nose and machine learning according to the present invention, wherein: the normalized preprocessing includes,

Among them, h _scale is the peak height of the chromatogram after normalization, h is the peak height of the chromatogram, h _min is the minimum value of the peak height of the chromatogram, and h _max is the maximum value of the peak height of the chromatogram.

对于峰高h的取值范围R＝{h|0＜h＜+∞}，对于任意给定的正常数s存在一个划分S＝{S₁，S₂，...，S_r}，满足：As a preferred solution of the crayfish freshness detection method based on gas-phase electronic nose and machine learning in the present invention, wherein: preprocessing the baseline data of the peak height includes calculating the peak height empirical distribution

For the value range R={h|0<h<+∞} of the peak height h, there is a division S={S ₁ , S ₂ ,...,S _r } for any given normal constant s, satisfying :

S _i ={h|(i-1)×s≤h≤i×s, sup(R)≤r×s}, i=1, 2,...r;

计算估计的基线值

Define the event A _i ={h|h∈S _i } whose peak height h falls in different data segment intervals, then the probability of occurrence of this event

Calculate the estimated baseline value

为第i个划分的经验分布，

为第i-1个划分的经验分布。Among them, S _r is the rth data segment divided; m is the serial number of the interval corresponding to the event A _i with the highest probability of occurrence, S _m is the division corresponding to the event with the highest probability of occurrence, and n is the total number of peak heights,

is the empirical distribution for the i-th partition,

Empirical distribution for the i-1th partition.

包括，将色谱图峰高h₁，h₂，...，h_n视为独立同分布的实随机变量，累积分布函数为F(k)，得到峰高经验分布

As a preferred solution of the crayfish freshness detection method based on gas-phase electronic nose and machine learning according to the present invention, wherein: the peak height empirical distribution

Including, the chromatogram peak heights h ₁ , h ₂ ,..., h _n are regarded as independent and identically distributed real random variables, the cumulative distribution function is F(k), and the empirical distribution of peak heights is obtained

为{h_i|h_i≤k}的指示函数。in,

is an indicator function of {h _i |h _i ≤ k}.

真实标签定义为y^*，样本总数为N，类别数量为M；将N个样本平均分为a份，取其中一份作为测试集，剩余a-1份作为训练集，计算测试集样本的估计概率p＝{p_j，j＝0，1，...，M}，重复a次，得到所有样本的折外预测；计算每个标定类别j下的平均概率t_j，并将其作为置信度阈值：As a preferred solution of the crayfish freshness detection method based on gas-phase electronic nose and machine learning according to the present invention, wherein: removing the label noise of the chromatogram includes defining the initial label and the label of days that may have errors as

The real label is defined as y ^* , the total number of samples is N, and the number of categories is M; the N samples are evenly divided into a parts, one of which is taken as the test set, and the remaining a-1 parts are used as the training set, and the estimation of the test set samples is calculated Probability p={p _j , j=0, 1,...,M}, repeat a times to get the out-of-the-box prediction of all samples; calculate the average probability t _j under each calibration category j, and take it as the confidence Degree Threshold:

Calculate count matrix

Calibration count matrix:

和真实标签y^*的联合分布

Estimating the initial label

and the joint distribution of the true label y ^*

均非对角单元，选取

个样本进行过滤，并按照最大间隔

排序，过滤每一类别的

个最大间距样本；For count matrix

are non-diagonal units, choose

samples and filter according to the maximum interval

Sort, filter for each category

maximum distance samples;

为初始标记

约个数；l表示满足

均标签；

为计数矩阵

的标定值。Among them, the probability that sample x belongs to the jth category

mark as initial

Approximate number; l means satisfied

average label;

is the count matrix

calibration value.

As a preferred solution of the crayfish freshness detection method based on gas-phase electronic nose and machine learning in the present invention, it is characterized in that: the trend feature of the chromatogram includes that the sequence model is preliminarily obtained through multiple convolutions. Get the rough trend feature X, and then extract the trend feature SLSTM(X) of X based on the LSTM network, that is, the chromatogram trend feature:

Among them, LSTM ₁ and LSTM ₂ are LSTM networks.

As a preferred solution of the crayfish freshness detection method based on gas-phase electronic nose and machine learning in the present invention, it also includes that the trend feature X is a sequence with a length of 65, and each position t contains the corresponding time The 64 numerical features x _t of the segment.

As a preferred solution of the crayfish freshness detection method based on gas-phase electronic nose and machine learning according to the present invention, wherein: the volatile compound content characteristics include,

layer _i (X)=ReLU(XW _i )

Among them, layer _i is the i-th layer network; W _i is the parameter of the i-th layer; x is the design matrix of the position feature; layer _o is the o-th layer network.

Beneficial effects of the present invention: the present invention utilizes an ultra-fast gas-phase electronic nose to acquire live crayfish of different freshness and the smell changes of crayfish at different times after death, so that the odor information of crayfish with different freshness can be obtained more intuitively and accurately; Use confidence learning to remove label noise and improve prediction accuracy; at the same time, use LSTM and MLP to extract the trend characteristics and relative content characteristics of volatile compounds from ultra-fast gas-phase electronic nose chromatography data, and use the feedforward neural network to realize the comparison after splicing the extracted features. The classification of crayfish freshness has good stability and accuracy.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without any creative effort. in:

Fig. 1 is the chromatogram of the crayfish freshness detection method based on gas-phase electronic nose and machine learning described in the first embodiment of the present invention;

Fig. 2 is a schematic diagram of peak height data PCA of the crayfish freshness detection method based on gas-phase electronic nose and machine learning described in the second embodiment of the present invention;

Fig. 3 is a confusion matrix of the crayfish freshness detection method based on gas-phase electronic nose and machine learning according to the second embodiment of the present invention.

detailed description

In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Example. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative efforts shall fall within the protection scope of the present invention.

In the following description, a lot of specific details are set forth in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described here, and those skilled in the art can do it without departing from the meaning of the present invention. By analogy, the present invention is therefore not limited to the specific examples disclosed below.

Second, "one embodiment" or "an embodiment" referred to herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. "In one embodiment" appearing in different places in this specification does not all refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

The present invention is described in detail in conjunction with schematic diagrams. When describing the embodiments of the present invention in detail, for the convenience of explanation, the cross-sectional view showing the device structure will not be partially enlarged according to the general scale, and the schematic diagram is only an example, which should not limit the present invention. scope of protection. In addition, the three-dimensional space dimensions of length, width and depth should be included in actual production.

At the same time, in the description of the present invention, it should be noted that the orientation or positional relationship indicated by "upper, lower, inner and outer" in the terms is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention. The invention and the simplified description do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and thus should not be construed as limiting the present invention. In addition, the terms "first, second or third" are used for descriptive purposes only, and should not be construed as indicating or implying relative importance.

Unless otherwise specified and limited in the present invention, the term "installation, connection, connection" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integrated connection; it can also be a mechanical connection, an electrical connection or a direct connection. A connection can also be an indirect connection through an intermediary, or it can be an internal communication between two elements. Those of ordinary skill in the art can understand the specific meanings of the above terms in the present invention in specific situations.

Example 1

Referring to Fig. 1, it is the first embodiment of the present invention, which provides a crayfish freshness detection method based on gas-phase electronic nose and machine learning, including:

S1: Place the crayfish sample in a beaker, seal it with double-layer plastic wrap, and let it stand in the headspace.

S2: Preheat the ultra-fast gas-phase electronic nose instrument, insert the sampling needle into the beaker for sampling, and obtain the chromatogram.

S3: Perform normalization preprocessing on the maximum and minimum peak heights of the chromatogram.

Normalized preprocessing:

Among them, h _scale is the peak height of the chromatogram after normalization, h is the peak height of the chromatogram, h _min is the minimum value of the peak height of the chromatogram, and h _max is the maximum value of the peak height of the chromatogram.

S4: Preprocess the baseline data of peak height, and use the confidence learning strategy to remove the label noise of crayfish samples.

(1) Preprocess the baseline data of peak height

①Calculation of peak height empirical distribution

The chromatogram peak heights h ₁ , h ₂ ,..., h _n are regarded as independent and identically distributed real random variables, the cumulative distribution function is F(k), and the peak height empirical distribution is obtained

为{h_i|h_i≤k}的指示函数。in,

is an indicator function of {h _i |h _i ≤ k}.

②For the value range R={h|0<h<+∞} of the peak height h, there is a partition S={S ₁ , S ₂ ,..., S _r } for any given constant s, satisfy:

S _i ={h|(i-1)×s≤h≤i×s, sup(R)≤r×s}, i=1, 2,...r;

Among them, S _r is the divided rth data segment.

计算估计的基线值

③ Define the event A _i ={h|h∈S _i } whose peak height h falls in different data intervals, then the probability of occurrence of this event

Calculate the estimated baseline value

为第i个划分的经验分布，

为第i-1个划分的经验分布。Among them, m is the number of the interval corresponding to the event A _i with the highest probability of occurrence, S _m is the division corresponding to the event with the highest probability of occurrence, n is the total number of peak heights,

is the empirical distribution for the i-th partition,

Empirical distribution for the i-1th partition.

(2) Eliminate label noise of crayfish samples including,

真实标签定义为y^*，样本总数为N，类别数量为M。①Define the initial label and the label of days with possible errors as

The true label is defined as y ^* , the total number of samples is N, and the number of categories is M.

② Divide the N samples into a parts on average, take one part as the test set, and the remaining a-1 part as the training set, and calculate the estimated probability p={p _j , j=0,1,... , M}, repeated a times to get the out-of-the-box prediction of all samples;

Among them, the probability that sample x belongs to the jth category

③Calculate the average probability t _j under each calibration category j, and use it as the confidence threshold:

为初始标记

的个数。in,

mark as initial

the number of .

④ Calculate the count matrix

的标签。Among them, l means satisfying

Tag of.

⑤ Calibration count matrix:

为计数矩阵

的标定值。in,

is the count matrix

calibration value.

和真实标签y^*的联合分布

⑥ Estimate initial label

and the joint distribution of the true label y ^*

均非对角单元，选取

个样本进行过滤，并按照最大间隔

排序，过滤每一类别的

个最大间距样本；⑦For count matrix

are non-diagonal units, choose

samples and filter according to the maximum interval

Sort, filter for each category

maximum distance samples;

Preferably, crayfish freshness labels may be misjudged due to differences in origin, transportation time, and individual differences. Artificial labels can only be used as a priori estimates of freshness. In this embodiment, wrong artificial labels are eliminated by using a confidence learning strategy. , that is, samples with obvious errors in the results, which improves the prediction accuracy.

S5: Use the sequence model to extract the features of the chromatogram, and obtain the trend characteristics of the chromatogram with different freshness odor changes.

The sequence model initially obtains a rough trend feature X through multiple convolutions (the trend feature X is a sequence with a length of 65, and each position t contains 64 numerical features x _t of the corresponding time period), and then extracts X based on the LSTM network The trend feature SLSTM(X), that is, the chromatogram trend feature:

Among them, LSTM ₁ and LSTM ₂ are LSTM networks.

It should be noted that LSTM is a recurrent neural network that can learn long-term dependencies and extract deep trend features of odor information.

The LSTM network processes the sequence sequentially in chronological order, feeds the feature x _t of each position into the input gate and the forget gate respectively, and obtains the control vectors it and _{f t ,} the calculation formula is as follows:

i _t = σ(W _ii x _t +b _ii +W _hi h _t-1 +b _hi )

f _t ＝σ(W _if x _t +b _if +W _hf h _t-1 +b _hf )

Each LSTM network contains a memory vector c, which is passed between different locations; the LSTM network uses the control vector f _t obtained by the forget gate to determine the information to be forgotten:

f _t ＝σ(W _if x _t +b _if +W _hf h _t-1 +b _hf )

Get new memories after discarding useless information

The LSTM network contains a hidden layer feature h for introducing sequence information; in the input gate, the LSTM network uses the hidden layer feature h _t-1 of the previous moment to recorrect the deep representation g _t extracted by the input gate, so as to capture the features of different positions interaction;

g _t ＝hanh(W _ig x _t +b _ig +W _hg h _t-1 +b _ng )

Among them, g _t contains the odor quantity information at time t and the overall information (including trend and quantity) before time t;

The input gate determines the information that the _LSTM network needs to remember through the vector it

To sum up, at time t, the updated memory vector c _t of the LSTM network.

After updating the memory vector, the LSTM network also needs to obtain h _t that incorporates information at this moment through the output gate; similar to the input gate, the LSTM network uses the hidden layer features of the previous moment to assist in calculating the output gate control vector σ _t :

o _t ＝σ(W _io x _t +b _io +W _ho h _t-1 +b _ho )

Finally, using the output gate control vector to decide what information needs to be kept in the hidden layer vector, _ht is updated:

h _t ＝o _t ⊙tanh(c _t )

S6: According to the trend feature of the chromatogram and extract the content feature of the volatile compound corresponding to each retention time through the multi-layer perceptron, and stitch the trend feature of the chromatogram and the feature of the content of the volatile compound.

This embodiment uses a multi-layer perceptron (MLP) to extract the volatile compound content characteristics corresponding to each retention time:

layer _i (X)=ReLU(XW _i )

Among them, layer _i is the i-th layer network; W _i is the parameter of the i-th layer; x is the design matrix of the position feature; layer _o is the o-th layer network.

S7: Classify the features by using the spliced features of the feed-forward neural network.

Example 2

In order to verify and illustrate the technical effects adopted in this method, this embodiment selects principal component analysis, LDA, RF, SVM algorithms and adopts this method to carry out comparative tests, and compares the test results by means of scientific demonstration to verify the advantages of this method. real effect.

For fresh and live crayfish samples, the samples collected after harvesting were recorded as day 0, placed in a refrigerator at 4°C, and stored for 1, 2, 3, 4, and 5 days. The dead shrimp samples were stored at 4°C for 6h and 12h, and at room temperature (25°C) for 3h and 24h, and the group stored at 25°C for 24h was the spoilage group.

Among them, the dead shrimp samples are processed as follows:

(1) Select crayfish of similar size, remove dead shrimp and shrimp with residual limbs; 5 crayfish per bag, suffocate the crayfish to death under the condition of vacuum degree of 0.1MPa.

(2) Sample 20 crayfish from the refrigerator at 4°C every day, and place them at room temperature for 1 hour to allow the crayfish to return to room temperature; put each crayfish into a 500mL beaker and seal it with double-layer plastic wrap. Stand in the headspace for 30min.

(3) The instrument was preheated for 30 minutes, and the sampling needle was 5 cm deep into the beaker to sample 5000 μL. The ultra-fast gas-phase electronic nose Heracles II was used to detect crayfish with different storage days. The instrument parameters were set as shown in Table 1.

Table 1: Analysis parameters.

serial number parameter condition 1 Injection volume 5000μL 2 Inlet temperature 200℃ 3 Injection duration 45s 4 Trap initial temperature 40℃ 5 Trap split rate 10mL/min 6 capture duration 50s 7 Trap final temperature 240°C 8 The initial temperature of the column temperature 50℃ 9 Column temperature programming method 1°C/s to 80°C, 2°C/s to 250°C for 60s 10 collection time 177s 11 Detector temperature 260°C

Principal component analysis was used to reduce the dimensionality of the chromatographic peak data, and the principal component analysis diagram (Fig. 2) was obtained, in which 1-5 were live crayfish stored at 4°C for 1-5 days, and 6 and 7 were dead crayfish stored at 4°C. Stored at 25°C for 3 hours and 24 hours, 8 and 9 are dead crayfish stored at 4°C for 6 hours and 12 hours respectively; PCA visualization analysis results show that the contribution rate of PC1 is 21.8%, the contribution rate of PC2 is 12.6%, and the total contribution rate is 34.4% . There is overlap between the sample points of each classification, and the different freshness of crayfish cannot be distinguished; therefore, the freshness of crayfish cannot be distinguished by using principal component analysis to process the sample odor information.

Based on embodiment 1, the specific parameters of this method are set as follows:

(1) Preprocessing the chromatographic data obtained from the ultrafast gas-phase electronic nose Heracles II.

(2) Use the sample to calculate the empirical distribution of peak height, obtain the maximum probability interval of the peak height of the first 666 records, and use the interval mean to estimate the baseline value.

A total of 35,407 pieces of peak height data were obtained in the ultra-fast gas-phase electronic nose analysis time of 177s, and the first 666 pieces of data were selected; according to the maximum value and minimum value of the first 666 pieces of data, the step length was 10 as a unit and divided into 67 data segments. Count the number of the 666 pieces of data falling in different data segments, find the data segment with the most data, and take the average value of the data in this segment as the baseline value.

真实标签定义为y*，样本总数为380，共有9类，分为活虾(4℃存放1天、2天、3天、4天、5天)、死虾(4℃贮藏6h、12h、25℃贮藏3h、24h)，其中活虾样本每类60只，死虾样本每类20只。(3) Define the initial label and the label of days with possible errors as

The true label is defined as y*, the total number of samples is 380, and there are 9 categories in total, divided into live shrimp (stored at 4°C for 1 day, 2 days, 3 days, 4 days, 5 days), dead shrimp (stored at 4°C for 6h, 12h, Store at 25°C for 3h, 24h), with 60 live shrimp samples per type and 20 dead shrimp samples per type.

即样本x属于第j个类别的概率。Further, divide the 380 samples into 5 parts on average for each category, take one of them as the test set, and the remaining 4 as the training set, and calculate the estimated probability p={p _j , j=0,1,. .., 9}, repeated 5 times to get the out-of-fold predictions of all samples, where

That is, the probability that sample x belongs to the jth category.

个样本进行过滤，并按照最大间隔

排序，过滤每一类别的22个最大间距样本；得到的22个过滤样本中，根据真实标签剔除错判活、死虾样本共10个，其余12个样本将原始标签修改为真实标签。select

samples and filter according to the maximum interval

Sorting and filtering the 22 samples with the largest distance in each category; among the obtained 22 filtered samples, a total of 10 falsely judged live and dead shrimp samples were eliminated according to the real label, and the original label was modified to the real label for the remaining 12 samples.

(4) After splicing the 128 features obtained by inputting the multi-layer perceptron and the 64 features obtained by the sequence model, 9 classifications are obtained by using the feedforward neural network classification.

Using this method, LDA, RF, and SVM algorithms respectively, 300 samples of live shrimp (stored at 4°C for 1 day, 2 days, 3 days, 4 days, and 5 days) and dead shrimp (stored at 4°C for 6h, 12h, and 25°C Store 3h, 24h) 80 samples to model; Each group randomly selects 80% of the crayfish samples as a training set, and 20% of the samples as a verification set; calculate the accuracy of the model prediction according to the confusion matrix (Fig. 3);

Table 2: Crayfish freshness prediction model results.

It can be seen from Table 2 that the prediction accuracy rates of the five modelings using this method are 92.00%, 93.33%, 95.95%, 97.30%, and 97.30%, respectively, and the overall prediction result can reach 95.16%; the prediction accuracy rate of this method is much higher than that using The accuracy rates of LDA, RF, and SVM modeling (79.16%, 75.99%, 71.50%) indicate that this method has good stability and accuracy.

It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention without limitation, although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be carried out Modifications or equivalent replacements without departing from the spirit and scope of the technical solution of the present invention shall be covered by the claims of the present invention.

Claims (7)

1. A crayfish freshness detection method based on a gas phase electronic nose and machine learning is characterized by comprising the following steps: comprises the steps of (a) preparing a substrate,

putting a crayfish sample into a beaker, sealing the sample with a double-layer preservative film, and standing for headspace;

preheating an ultra-fast gas phase electronic nose instrument, and inserting a sample injection needle into a beaker for sampling to obtain a chromatogram;

carrying out normalization pretreatment on the maximum value and the minimum value of the peak height of the chromatogram;

preprocessing the baseline data of the peak height, and eliminating the label noise of the crayfish sample by using a belief learning strategy;

performing feature extraction on the chromatogram by using a sequence model to obtain the trend features of the chromatogram with different freshness and odor changes;

extracting the volatile compound content characteristics corresponding to each retention time through a multilayer perceptron according to the chromatogram trend characteristics, and splicing the chromatogram trend characteristics and the volatile compound content characteristics;

performing feature classification by using the spliced features of the feedforward neural network;

the sequence model preliminarily obtains rough trend characteristics X through multiple convolution, and then extracts trend characteristics SLSTM (X) of X based on an LSTM network, namely the chromatogram trend characteristics:

wherein, LSTM ₁ 、LSTM ₂ Is an LSTM network.

2. The crayfish freshness detection method based on the gas-phase electronic nose and the machine learning as claimed in claim 1, characterized in that: the normalization pre-processing includes the steps of,

wherein h is _scale Is the peak height of the chromatogram after normalization, h is the peak height of the chromatogram, h _min Minimum value of the peak height of the chromatogram, h _max The maximum value of the chromatogram peak height.

3. The crayfish freshness detection method based on the gas-phase electronic nose and the machine learning as claimed in claim 2, characterized in that: pre-processing the baseline data for the peak heights includes,

calculating a peak height empirical distribution

For the value range R = { h |0 of peak height h<h<+ ∞, there is one division S = { S } for any given normal number S ₁ ,S ₂ ,…,S _r And satisfies:

S _i ＝{h|(i-1)×s≤j≤i×s,sup(R)≤r×s}，i＝1,2,…r；

defining event A with peak height h falling in different data segment intervals _i ＝{h|h∈S _i The probability of occurrence of the event

Calculating an estimated baseline value

Wherein S is _r Is the r-th divided data segment; m is the event A with the maximum occurrence probability _i Number of corresponding section, S _m N is the total peak height number of the corresponding division of the event with the maximum occurrence probability,

for the empirical distribution of the ith partition,

is the empirical distribution of the i-1 th partition.

4. The crayfish freshness detection method based on the gas-phase electronic nose and the machine learning as claimed in claim 3, characterized in that: empirical distribution of peak heights

Comprises the steps of (a) preparing a mixture of a plurality of raw materials,

the peak height h of the chromatogram map ₁ ,h ₂ ,…,h _n The real random variables which are regarded as independent and same distribution are accumulated to be distributed as a function F (k), and the peak height is obtainedEmpirical distribution

Wherein,

is { h _i |h _i K ≦ k).

5. The crayfish freshness detection method based on gas phase electronic nose and machine learning of claim 4, wherein: the label noise for eliminating the chromatogram includes,

defining the initial annotation, the number of days of possible error label as

The real tag is defined as y ^* The total number of samples is N, and the number of categories is M;

averagely dividing N samples into a parts, taking one part as a test set, taking the rest a-1 parts as a training set, and calculating the estimated probability p = { p } of the test set samples _j J =0,1, \8230;, M }, repeating a times to obtain the folded prediction of all samples;

calculating the average probability t under each calibration category j _j And as a confidence threshold:

calculating a count matrix

Calibrating a counting matrix:

estimating initial tags

And a genuine label y ^* Joint distribution of

For a counting matrix

Off diagonal cell of (1), selecting

Filtering the samples at a maximum interval

Sorting, filtering of each category

A maximum-spaced sample;

wherein the probability that a sample x belongs to the jth class

Is an initial mark

The number of (2); l represents the satisfaction of

The label of (2);

is a counting matrix

To the calibration value of (c).

6. The crayfish freshness detection method based on the gas-phase electronic nose and the machine learning as claimed in claim 5, characterized in that: also comprises the following steps of (1) preparing,

the trend feature X is a sequence of length 65, each position t containing 64 numerical features X for a corresponding time segment _t 。

7. The crayfish freshness detection method based on the gas-phase electronic nose and the machine learning as claimed in claim 6, characterized in that: the volatile compound content characteristics include,

layer _i (X)＝ReLU(XW _i )

wherein, layer _i Is the i-layer network; w _i Is a parameter of the ith layer; x is a design matrix of the position characteristics; layer _o Is a layer o network.

Images