CN114580511B - Sulfur-smoked dried ginger identification method based on image brightness information and voting mechanism - Google Patents

Abstract

The present invention discloses a sulfur-smoked dried ginger identification method based on image brightness information and a voting mechanism, which comprises the following steps: (1) sample preparation and image data acquisition; (2) image feature extraction; (3) using support vector machine (SVM), BP neural network (BPNN) and random forest algorithm (RF) to establish sulfur-smoked dried ginger identification models; (4) according to the results of the three models, establishing a set of recognition models based on a voting mechanism. The present invention adopts a sulfur-smoked dried ginger identification method based on image brightness information and a voting mechanism for the first time, which can accurately predict the sulfur-smoking degree of dried ginger, and has the advantages of rapid and non-destructive, high recognition accuracy and strong stability.

Description

A sulfur-smoked dried ginger identification method based on image brightness information and voting mechanism

Technical Field

The invention belongs to the technical field of medicinal material detection, and specifically relates to a sulfur-smoked dried ginger identification method based on image brightness information and a voting mechanism.

Background technique

Dried ginger (Zingiber officinale Roscoe) is the dried rhizome of the perennial herb ginger of the ginger family. It has the effects of warming the middle and dispelling cold, restoring yang and unblocking meridians, and warming the lungs and eliminating phlegm. The market demand for dried ginger is huge. A large number of people around the world, especially in Asia, Africa and Europe, have the habit of using dried ginger (including fresh ginger) on a daily basis. Sulfur fumigation of ginger can play a role in anti-corrosion, anti-mildew, and anti-insect effects, and is conducive to drying and color enhancement. Although sulfur fumigation has played a certain positive role in the processing and storage of Chinese medicinal materials, modern research has shown that the chemical properties of Chinese medicinal materials have changed after sulfur fumigation, and even affected the nature, taste and efficacy of the medicinal materials. Heavy metals such as sulfide, arsenic and mercury will also remain. Long-term and large-scale use of such Chinese medicines will cause damage to the human body.

The existing detection technologies for sulfur fumigated medicinal materials include ultra-high performance liquid chromatography, near-infrared spectroscopy, etc. However, the existing methods require expensive equipment to collect information. Therefore, it is not convenient to use the existing methods to identify sulfur-free dried ginger and sulfur-containing dried ginger. With the development of machine learning algorithms, image recognition technology is becoming more and more mature. The image-based sulfur-fumigated dried ginger identification method has the advantages of being fast, pollution-free, and non-destructive to materials. Therefore, how to quickly identify sulfur-fumigated dried ginger through image information is a key problem that needs to be solved at present.

Summary of the invention

Purpose of the invention: To solve the above technical problems, the present invention provides a sulfur-fumigated dried ginger identification method based on image brightness information and voting mechanism. Compared with the traditional identification method, this method can accurately predict the sulfur fumigation degree of dried ginger, and has the advantages of low cost, rapid and non-destructive, high recognition accuracy and strong stability.

Technical solution: To achieve the above purpose, the technical solution provided by the present invention is as follows: a sulfur-smoked dried ginger identification method based on image brightness information and voting mechanism, comprising the following steps:

(1) Sample collection: Select dried ginger with different sulfur contents;

(2) Sample classification: Use sulfur dioxide detection reagent to mark the sulfur fumigation degree of the sample;

(3) Image data collection: Use image acquisition equipment to collect image information of dried ginger, and mark the sulfur fumigation degree (high sulfur, low sulfur, no sulfur) of each image;

(4) LBP feature extraction: Let the RGB value of the pixel (i,j) in the i-th row and j-th column of the image be (R _ij ,G _ij ,B _ij ), and convert the RGB color information of the image into three-channel information in the Lab color space. Let the Lab value of the converted Lab pixel (i,j) be (L _ij ,a _ij ,b _ij ), then the brightness value of the pixel (i,j) is L _ij . Let the brightness values of the eight pixels around the pixel be L ₀ ,L ₁ ,...,L ₇ , respectively. Let L _p be the brightness value of the adjacent p-th pixel, and the brightness relationship value Lr _ij of the pixel can be obtained using formula (1) and formula (2). The brightness relationship value L _ij is an integer between [1,10]. First calculate Lr _ij of each pixel in the entire L channel, and then count the ratio of the number of occurrences of each integer to the total number of brightness relationship values, and then form a ten-dimensional vector. This vector is denoted as f _L1 . Calculate the feature vector values of channel a and channel b in the same way and record them as f _a1 and f _b1 . Concatenate f _L1 , f _a1 and f _b1 to form vector f ₁ . Downsample the original image and repeat the feature extraction process of step (4) for the downsampled image to obtain vector f ₂ . Repeat the downsampling again to obtain vector f ₃ . Concatenate vectors f ₁ , f ₂ and f ₃ to obtain the final feature f of the image.

(5) Model building: The dried ginger image samples were divided into a training set and a validation set. The features of the dried ginger training set samples with different sulfur fumigation degrees were learned using support vector machines, BP neural networks, and random forest algorithms. The LibSVM software package, RF software package, and BPNN software package were used for model training to build three different prediction models for the sulfur fumigation degree of dried ginger.

(6) Establish a voting mechanism: Assume that the weight of the SVM algorithm is w _{SVM_M} , the weight of the BPNN algorithm is w _{BPNN_M} , and the weight of the RF algorithm is w _{RF_M} . Calculate v _i according to formulas (3), (4), (5), and (6). Calculate the maximum value of v ₁ , v ₂ , and v _3. If v ₁ is the largest, then this sample is sulfur-free dried ginger; if v ₂ is the largest, then this sample is low-sulfur dried ginger; if v ₃ is the largest, then this sample is high-sulfur dried ginger. For new test samples, after their feature values are extracted, the prediction model can determine the type of prediction results.

_vi = _{ci_SVM} ×w _{SVM_M} + _{ci_BPNN} ×w _{BPNN_M} + _{ci_RF} ×w _{RF_M} (3)

Compared with the prior art, the beneficial effects of the present invention are:

1. Compared with traditional empirical identification, such as high performance liquid chromatography and Fourier transform near infrared spectroscopy, the present invention has the advantages of being fast, non-destructive, convenient and low-cost.

2. The present invention provides a new identification method for the sulfur fumigation degree of dried ginger, provides a scientific basis for evaluating the sulfur fumigation degree of dried ginger on the market, and has broad application prospects.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1: The sulfur dioxide detection kit of Beijing Zhiyunda Technology Co., Ltd. used in Example 1 of the present invention;

Figure 2: Photos of sulfur-free, low-sulfur and high-sulfur dried ginger collected by a Huawei P30 ELEAL00 mobile phone in Example 1 of the present invention;

Figure 3: Photos of sulfur-free, low-sulfur and high-sulfur dried ginger collected by an Apple XRA2108 mobile phone in Example 1 of the present invention;

FIG4 : Lab three-channel image of Example 1 of the present invention: (a) L channel (b) a channel (c) b channel;

FIG5 : Relationship diagrams of Lab channels in Example 1 of the present invention: (a) L channel brightness relationship diagram (b) a channel value relationship diagram (c) b channel value relationship diagram;

FIG6 : Lab three-channel histogram of Example 1 of the present invention: (a) L channel histogram (b) a channel value histogram (c) b channel value histogram;

FIG. 7 shows the prediction accuracy of the identification method of Example 1 of the present invention under different proportions of training sets on different mobile phones;

FIG8 is a flow chart of a method for identifying sulfur-smoked dried ginger based on image brightness information and a voting mechanism provided by the present invention.

Detailed ways

The present invention is described below by specific examples to make the technical solution of the present invention easier to understand and grasp, but the present invention is not limited thereto. The experimental methods described in the following examples are conventional methods unless otherwise specified; the reagents and materials described are all commercially available unless otherwise specified.

Instruments used in the present invention: Huawei P30 ELEAL00 and Apple XRA2108, sulfur dioxide detection kit of Beijing Zhiyunda Technology Co., Ltd.

Example 1

A sulfur-smoked dried ginger identification method based on image brightness information and voting mechanism, comprising the following steps:

1. Collection of medicinal material samples: 123 batches of dried ginger with different sulfur contents were purchased from the market, and different batches of dried ginger were measured using a sulfur dioxide rapid test kit. According to the different degrees of sulfur fumigation of different batches of dried ginger, they can be divided into three categories, namely low-sulfur dried ginger, high-sulfur dried ginger, and sulfur-free dried ginger. Among them, there were 27 batches of low-sulfur dried ginger, 66 batches of high-sulfur dried ginger, and 30 batches of sulfur-free dried ginger. The standard for dividing the three categories is that the sulfur dioxide content of low-sulfur dried ginger is 50-150 mg/kg, the sulfur dioxide content of high-sulfur dried ginger is greater than 150 mg/kg, and the sulfur dioxide content of sulfur-free dried ginger is 0 mg/kg. Figure 1 shows the colorimetric card results obtained for dried ginger with different sulfur dioxide contents. The final classification is shown in Table 1, where NS means no sulfur, LS means low sulfur, and HS means high sulfur.

Table 1

2. Collect image data of samples: High-definition images of dried ginger from different origins were collected by image acquisition equipment, and the sulfur fumigation degree information of each image was marked. To ensure the integrity of the morphological details of dried ginger, the image data was collected using Huawei P30 ELEAL00 mobile phone and Apple XRA2108 mobile phone under the same conditions. Two images of the front and back were collected for each piece of dried ginger. The number of images of low-sulfur dried ginger, high-sulfur dried ginger and sulfur-free dried ginger were 54, 132 and 60 respectively. Figures 2(a), (b) and (c) are the image data of sulfur-free, low-sulfur and high-sulfur dried ginger taken by Huawei P30 ELEAL00 mobile phone. Figures 3(a), (b) and (c) are the image data of sulfur-free, low-sulfur and high-sulfur dried ginger taken by Apple XRA2108 mobile phone.

3. LBP feature extraction: Let the RGB value of the pixel (i,j) in the i-th row and j-th column of the image be (R _ij ,G _ij ,B _ij ), and convert the RGB color information of the image into three-channel information in the Lab color space. Let the Lab value of the converted Lab pixel (i,j) be (L _ij ,a _ij ,b _ij ), then the brightness value of the pixel (i,j) is L _ij . Let the brightness values of the 8 pixels around the pixel be L ₀ ,L ₁ ,...,L ₇ . Let L _p be the brightness value of the adjacent p-th pixel, and the brightness relationship value Lr _ij of the pixel can be obtained by using formula (1) and formula (2). The brightness relationship value L _ij is an integer in [1,10]. First calculate Lr _ij of each pixel in the entire L channel, and then count the ratio of the number of occurrences of each integer to the total number of brightness relationship values, and then form a ten-dimensional vector. This vector is denoted as f _L1 . Then calculate the feature vector values of the a channel and the b channel in the same way and record them as f _a1 and f _b1 . Concatenate f _L1 , f _a1 and f _b1 to form vector f _1. Downsample the original image, repeat the feature extraction process of step (4) for the downsampled image, and obtain vector f _2. Repeat the downsampling again to obtain vector f _3. Concatenate vectors f ₁ , f ₂ and f ₃ to obtain the final image feature f. Figures 4(a), 4(b) and 4(c) respectively show the Lab space L channel, a channel and b channel images. Figures 5(a), 5(b) and 5(c) respectively show the L channel brightness relationship diagram, the a channel value relationship diagram and the b channel value relationship diagram. Figures 6(a), 6(b) and 6(c) respectively show the L channel histogram, the a channel histogram and the b channel histogram.

4. Establish a voting mechanism: Assume that the weight of the SVM algorithm is w _{SVM_M} , the weight of the BPNN algorithm is w _{BPNN_M} , and the weight of the RF algorithm is w _{RF_M} . Calculate _vi according to formulas (3), (4), (5), and (6). Calculate the maximum value of v ₁ , v ₂ , and v _3. If v ₁ is the largest, then this sample is sulfur-free dried ginger; if v ₂ is the largest, then this sample is low-sulfur dried ginger; if v ₃ is the largest, then this sample is high-sulfur dried ginger. For new test samples, after their feature values are extracted, the prediction model can determine the type of prediction results.

_vi = _{ci_SVM} ×w _{SVM_M} + _{ci_BPNN} ×w _{BPNN_M} + _{ci_RF} ×w _{RF_M} (3)

5. Establish a sulfur content identification model: To avoid experimental errors caused by random sampling, the train-test process of each prediction is repeated 200 times. Each time, the representative eigenvalue is used as the eigenvector, and the sample type is learned using the support vector machine. The model training is performed using the LibSVM software package, BPNN software package, and RF software package. The weight of each algorithm in the voting mechanism is set according to the accuracy of each algorithm. The accuracy obtained by the different methods shown in step (5) is greater than that of the BPNN algorithm and greater than that of the RF algorithm. Therefore, we set the three weights to w _{SVM_M} = 0.4, w _{BPNN_M} = 0.35, and w _{RF_M} = 0.25, respectively, and finally build a prediction model to predict the sulfur content of dried ginger.

Table 2 Average prediction accuracy of images taken by Huawei P30ELEAL00 and Apple XRA2108

Correct rate% ACC _SVM ACC _BPNN ACC _RF Huawei P30 80.8939 71.5648 79.9622 iPhone XR 92.3797 87.1865 88.0789

6. Model training accuracy and stability: The size of the training set will affect the accuracy of the final prediction. The proportion of the training set is obtained incrementally from [0.1-0.9], increasing by 0.1 each time. Repeat the training 200 times using the train-test program. The experimental results are shown in Tables 3 and 4. Among the prediction accuracies of 9 different training set proportions, the accuracy based on the voting mechanism is the highest, and the accuracy of the images taken by Apple XR reaches 93.44%. The accuracy of the images taken by Huawei P30 is 82.98%. In summary, the ideas proposed in this patent have high accuracy and high practicality.

Table 3 Accuracy of images taken by Apple XRA2108 mobile phone

Table 4 Accuracy of images taken by Huawei P30 ELE AL00 mobile phone

The above detailed description is a specific description of one feasible embodiment of the present invention. The embodiment is not intended to limit the patent scope of the present invention. Any equivalent implementation or modification that does not deviate from the present invention should be included in the scope of the technical solution of the present invention.

Claims (6)

1. The sulfur-smoked dried ginger identification method based on the image brightness information and the voting mechanism is characterized by comprising the following steps of:

(1) Sample collection: selecting dried ginger with different sulfur dioxide contents;

(2) Sample classification: marking the sulphitation degree of the sample by using a sulfur dioxide detection reagent;

(3) Acquisition of image data: respectively using image acquisition equipment to acquire image information of the dried ginger, and marking the sulfur fumigation degree of each image according to high sulfur, low sulfur and no sulfur;

(4) LBP feature extraction: setting RGB values of the ith row and the jth column of the image as (R _ij,G_ij,B_ij), and converting RGB color information of the image into three-channel information of a Lab color space; setting the Lab value of the converted Lab pixel point (i, j) as (L _ij,a_ij,b_ij), and setting the brightness value of the pixel point (i, j) as L _ij; the brightness values of 8 adjacent pixel points around the pixel point are respectively L ₀,L₁,...,L₇; let L _p be the luminance value of the p-th pixel point, and use formula (1) and formula (2) to obtain the luminance relation value Lr _ij of the pixel point; the luminance relation value L _ij is an integer of [1,10 ]; calculating Lr _ij of each pixel of the whole L channel, and counting the ratio of the occurrence times of each integer to the total number of brightness relation values to form a ten-dimensional vector; this vector is denoted as f _L1, and the characteristic vector values for the a-channel and the b-channel are calculated in the same manner as f _a1 and f _b1; splicing f _L1,f_a1 and f _b1 to form a vector f ₁, downsampling the original image, and repeating the feature extraction process of the step (4) on the downsampled image to obtain a vector f ₂; then, repeating downsampling to obtain a vector f ₃, and splicing the vectors f ₁,f₂ and f ₃ to obtain the final image characteristic f;

(5) And (3) establishing a model: dividing the dried ginger image sample into a training set and a verification set, and learning the characteristics of the dried ginger training set samples with different sulfuration degrees by using a support vector machine, a BP neural network and a random forest algorithm; model training is carried out by adopting LibSVM software packages, RF software packages and BPNN software packages, and 3 different dry Jiang Liuxun degree prediction models are constructed;

(6) Establishing a voting mechanism: let the weight occupied by the SVM algorithm be w _{SVM_M}, the weight occupied by the BPNN algorithm be w _{BPNN_M}, and the weight occupied by the RF algorithm be w _{RF_M}; v _i is calculated according to formulas (3), (4), (5), (6); calculating the maximum value of v ₁,v₂,v₃, and if v ₁ is the maximum value, then the sample is the sulfur-free dried ginger; if v ₂ is the maximum, then this sample is low sulfur dried ginger; if v ₃ is the maximum, then this sample is high sulfur dried ginger; for the new test sample, after the characteristic value is extracted, the prediction model can judge the type of the prediction result;

v_i＝c_{i_SVM}×w_{SVM_M}+c_{i_BPNN}×w_{BPNN_M}+c_{i_RF}×w_{RF_M} (3)

2. The method for identifying sulfur-smoked dried ginger based on brightness information and voting mechanism according to claim 1, wherein the dried ginger with different sulfur contents in the step (1) is respectively low sulfur dried ginger, high sulfur dried ginger and sulfur-free dried ginger; the criteria for three categories of division are: the sulfur dioxide content of the low sulfur dried ginger is 50-150mg/kg, the sulfur dioxide content of the high sulfur dried ginger is more than 150mg/kg, and the sulfur dioxide content of the sulfur-free dried ginger is 0mg/kg.

3. The method for identifying sulfur-smoked dried ginger based on image brightness information and voting mechanism according to claim 1, wherein the reagent for marking the degree of fumigation of the dried Jiang Liu in the step (2) is a sulfur dioxide rapid detection kit of Beijing Zhiyun reaches science and technology Co., ltd, which detects the sulfur dioxide content range of the dried ginger.

4. The method for authenticating a sulfur-smoked dried ginger having an image based on luminance information and voting scheme as claimed in claim 1, wherein in step (3), in order to ensure the integrity of the morphological detail information of the dried ginger, the data acquisition of the image is respectively carried out by using a Hua P30 ELE AL00 mobile phone and an apple XRA2108 mobile phone under the same condition.

5. The method for identifying the sulfur-smoked dried ginger based on the image brightness information and the voting mechanism according to claim 1, wherein the prediction model in the step (5) is used for collecting pictures of the dried ginger with different sulfur contents, establishing a dried ginger image database, calculating the brightness value relation information among pixels of the dried ginger image, extracting Lab three-channel characteristic information of the dried ginger image, modeling through a machine learning algorithm and the voting mechanism, and finally realizing the prediction of the degree of dry Jiang Liu fumigation.

6. The method for identifying sulfur-smoked dried ginger based on image brightness information and voting mechanism according to claim 1, wherein in step (5), three weights of SVM algorithm, BPNN algorithm and RF algorithm are respectively set to w _{SVM_M}＝0.4,w_{BPNN_M}＝0.35,w_{RF_M} =0.25.