Method, cloud platform and system for rapid screening of drug addicts based on skin spectrum

Abstract

The embodiment of the present invention relates to the field of biometric identification technology, and discloses a method, cloud platform and system for rapid screening of drug addicts based on skin spectra. The method includes: using a skin spectrum information collector to collect spectrum sequence data of the non-dominant palm of the screened person; preprocessing the spectrum sequence data; based on the gender of the screened person, inputting the preprocessed spectrum sequence data into different drug abuse recognition models for identification to obtain drug abuse screening results; and feeding back the drug abuse screening results to a mobile terminal for display. The scheme for rapid screening of drug addicts based on skin spectra provided by the embodiment of the present invention has the characteristics of in-situ non-destructiveness compared to traditional sampling and detection methods (blood tests, urine tests, hair tests, etc.), and has the characteristics of rapidity and less restrictions compared to non-invasive screening methods (breath tests, iris / pupil tests). In terms of rapid on-site screening, the method of the present invention is more effective, simple, reliable, and has a higher accuracy rate.

Description

Technical Field

[0001] The present invention relates to the field of biometric identification technology, and in particular to a method, a cloud platform and a system for rapid screening of drug addicts based on skin spectra. Background Art

[0002] Optical non-invasive diagnosis is one of the most promising application areas of human skin tissue spectrum. A large number of experiments have verified that light, as a carrier of energy, carries a large amount of physiological and pathological information of the human body after passing through human skin tissue. Based on this, combined with data analysis algorithms, the physiological and pathological states of tissues can be identified.

[0003] Drug testing includes on-site testing, laboratory testing and laboratory retesting. Laboratory testing cannot meet the requirements of on-site testing in drug control work due to problems such as long testing cycle, expensive equipment and high testing cost. At present, the technologies that can meet the requirements of rapid on-site testing mainly include enzyme-linked immunosorbent assay (ELISA), colloidal gold method, time-resolved fluorescence immunochromatography, quantum dot fluorescence immunochromatography, colloidal carbon method, etc. However, these methods require the collection of biological samples and require the high cooperation of the screened person. Common drug samples include urine, blood, saliva, hair, etc. Urine and saliva testing are limited by time and space and are easily contaminated; hair testing has a time window and is not suitable for bald people; although blood testing has high accuracy, it requires samples to be collected from the corresponding institutions. Taking the above biological samples for testing has high requirements for operators and test environment, and there are problems such as cumbersome pre-treatment procedures, long testing time and high cost in the testing process. At the same time, due to object privacy, cheating and non-cooperation, the efficiency of testing work is seriously affected.

[0004] At present, the rapid detection technologies that are different from traditional ones include breath analysis, iris detection, eye / pupil screening, etc. Breath analysis can only better reflect drug use behavior in a short period of time, and there are problems such as complex and time-consuming processes for gas collection-enrichment, pretreatment, and drug component detection; iris detection and eye / pupil screening both require eye image information, but if the person being tested frequently blinks / turns his eyes / closes his eyes, the image quality will be affected. At the same time, it still takes a long time (more than 10 minutes) for iris detection to produce results; eye / pupil screening can only play a role in identifying drug users who use opioid drugs. Summary of the invention

[0005] In view of the technical defects raised by the prior art, the purpose of the embodiments of the present invention is to provide a method, cloud platform and system for quickly screening drug addicts based on skin spectra.

[0006] To achieve the above objectives, in a first aspect, an embodiment of the present invention provides a method for quickly screening drug addicts based on skin spectra, comprising:

[0007] Acquire spectral sequence data and the gender of the screened person; the spectral sequence data is obtained by collecting diffuse reflectance spectral information of the non-dominant palm of the screened person using a skin spectral information collector;

[0008] Preprocessing the spectral sequence data;

[0009] Based on the gender of the screened person, the pre-processed spectral sequence data is input into different drug abuse recognition models for recognition to obtain drug abuse screening results;

[0010] The drug abuse screening result is fed back to the mobile terminal for display; the drug abuse screening result includes the probability of whether the screened person is taking drugs and the probability of the type of drugs taken.

[0011] As a preferred implementation of the present application, the method further includes constructing and training a drug abuse recognition model, specifically:

[0012] A skin spectrum information collector is used to collect spectrum information of the non-dominant palm of a sample person, and diffuse reflectance spectrum data of four points are obtained in sequence; the sample person includes drug addicts and non-drug addicts;

[0013] Collect the gender and types of drugs used by all sample persons;

[0014] The diffuse reflectance spectrum data of each sample person is labeled with gender, drug use, and drug type to obtain label data;

[0015] The label data is used to construct and train a drug abuse recognition model for different genders or different drugs.

[0016] As a specific implementation of the present application, based on the gender of the screened person, the pre-processed spectral sequence data is input into different drug abuse recognition models for recognition to obtain drug abuse screening results, specifically:

[0017] According to the gender of the screened person, different first-type drug abuse identification models are selected to identify the preprocessed spectral sequence data to obtain the probability of whether the screened person is a drug addict; the first-type drug abuse identification model is constructed according to different genders;

[0018] If the person being screened is a drug addict, different second-type drug abuse recognition models are selected to perform identification in the order of marijuana, heroin, methamphetamine, hemp, K powder and others to obtain the probability of the person being screened taking different drugs, and arrange the types of drugs whose probability exceeds a preset value (for example, 50%) in reverse order of probability value; the second-type drug abuse recognition model is constructed based on different drugs.

[0019] As a specific implementation of the present application, for different genders, the label data is used to construct and train a drug abuse recognition model, specifically:

[0020] Normalizing the label data;

[0021] For different genders, a male spectral dataset and a female spectral dataset are established based on the normalized label data; the number of negative samples in the male and female spectral datasets is much larger than the number of positive samples; among them, drug users are positive samples, and non-drug users are negative samples;

[0022] According to the number of positive samples in the male and female spectral datasets, use the sample() function of Pandas to randomly sample negative sample data, and set the parameter frac to 1.2;

[0023] Merge the positive sample data with the negative sample data randomly sampled by the sample() function, and randomly divide the data set into a training set and a test set in a ratio of 7:3;

[0024] A deep learning model is constructed using python, and the training set and test set are used to train and verify the drug abuse recognition model; the drug abuse recognition model is used to screen drug addicts and obtain the probability that the screened person is a drug addict.

[0025] As a specific implementation of the present application, a deep learning model is constructed using Python, and the training set and test set are used to train and verify the drug abuse recognition model, specifically:

[0026] Select SVM algorithm and Xgboost algorithm to establish drug abuse classification models for male and female genders respectively;

[0027] Extracting features from the diffuse reflectance spectrum data in the training set to obtain spectrum sequence features;

[0028] Using the spectral sequence features as input of the classification model, and searching for the best hyperparameter combination of the classification model through a network search method;

[0029] The test set is input into a classification model with an optimal hyperparameter combination to obtain a final drug abuse identification type; the drug abuse identification model includes an SVM drug abuse identification model and an Xgboost drug abuse identification model, the SVM drug abuse identification model is used for drug abuse screening for males, and the Xgboost drug abuse identification model is used for drug abuse screening for females.

[0030] As a specific implementation of the present application, the optimal hyperparameter combination of the classification model is found through network search, specifically:

[0031] The SVM / Xgboost algorithm uses the grid search function GridSearchCV for hyperparameter tuning, and obtains the best parameters found by GridSearchCV through the best_params_ attribute.

[0032] Furthermore, as a preferred implementation of the present application, the method further includes:

[0033] The output results of the drug abuse recognition model were comprehensively evaluated using accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score and area under the ROC curve.

[0034] In the second aspect, an embodiment of the present application also provides a cloud platform for rapid screening of drug addicts based on skin spectra, comprising a processor, an input device, an output device and a memory, wherein the processor, input device, output device and memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the method of the first aspect above.

[0035] In a third aspect, an embodiment of the present invention further provides a system for rapid screening of drug addicts based on skin spectra, comprising a skin spectrum information collector, a cloud platform and a mobile terminal that communicate with each other; wherein the cloud platform is as described in the second aspect above.

[0036] In a fourth aspect, an embodiment of the present invention provides another system for rapid screening of drug addicts based on skin spectrum, including a skin spectrum information collector, a cloud platform and a mobile terminal that communicate with each other; the skin spectrum information collector is used to emit visible light to illuminate the palm skin of the screened person and collect spectrum sequence data;

[0037] The cloud platform is used to preprocess, extract features and identify models of the spectral sequence data to obtain drug screening results;

[0038] The mobile terminal is used to display the drug abuse screening result.

[0039] Furthermore, as a preferred implementation of the present application, the mobile terminal is also used to provide the functions of rapid screening for clinical inspections and mandatory inspection for serious suspects.

[0040] The skin spectrum-based rapid screening solution for drug addicts provided by the embodiment of the present invention has the characteristics of being in situ non-destructive compared to traditional sampling and testing methods (blood tests, urine tests, hair tests, etc.), and is rapid and less restricted compared to non-invasive screening methods (breath tests, iris / pupil tests). In terms of on-site rapid screening, the method of the present invention is more effective, simple, reliable, and has a higher accuracy rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the specific implementation of the present invention or the technical solution in the prior art, the drawings required for use in the specific implementation or the description of the prior art are briefly introduced below.

[0042] Figure 1 is a flow chart of a method for rapid screening of drug addicts based on skin spectrum provided by an embodiment of the present invention;

[0043] Figure 2 It is a flowchart for building a drug abuse identification model;

[0044] Figure 3 It is the ROC curve and AUC value of the SVM model used by the AI ​​recognition model to determine whether a man is addicted to drugs;

[0045] Figure 4 It is the ROC curve and AUC value of the Xgboost model used by the AI ​​recognition model to determine whether a woman is addicted to drugs;

[0046] Figure 5 It is the ROC curve and AUC value of the SVM model used by the AI ​​recognition model to distinguish heroin addiction;

[0047] Figure 6 It is the cloud platform structure diagram;

[0048] Figure 7 This is a schematic diagram of the drug addiction rapid screening process;

[0049] Figure 8 It is the display diagram of the screening results fed back to the front end. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0052] The skin is the largest organ in the human body and an important effector organ of the internal organs, spirit and surrounding environment. Diseases caused by various pathogenic factors can cause changes in skin tissue, shape and function. Therefore, it is like a "mirror" of human health.

[0053] The inventive concept of the present invention is to collect diffuse reflectance spectral information of the non-dominant palm skin of the screened person, perform normalization preprocessing on it, build a deep learning model, train and verify the deep learning model to obtain a classification model for screening drug addicts, and use the classification model to perform rapid screening of drug addicts.

[0054] Please refer to Figure 1 , is a method for rapid screening of drug addicts based on skin spectrum provided by the first embodiment of the present invention, comprising the following steps:

[0055] S1, build and train a drug abuse recognition model.

[0056] In specific implementation, step S1 includes:

[0057] (1) Use a skin spectrum information collector to collect spectrum information from the non-dominant palm of the sample person, and obtain diffuse reflectance spectrum data at four points in sequence.

[0058] Specifically, the portable skin trace element monitor BC408H was used to collect spectral information on the non-dominant palms of 537 drug addicts and 1835 non-drug addicts, and the diffuse reflectance spectral data of four points on the palms were obtained in turn.

[0059] (2) Collect the gender and drug types of all sample persons

[0060] The sample persons refer to the above-mentioned drug users and non-drug users. The gender information and types of drugs used by all sample persons are collected, including marijuana, heroin, methamphetamine, marijuana, K powder and others.

[0061] (3) Generate label data

[0062] The diffuse reflectance spectrum data of each sample person is annotated with gender, drug abuse, and drug type to obtain label data.

[0063] (4) Use labeled data to build and train drug use recognition models for different genders or different drugs

[0064] 1. Construct drug abuse identification models for different genders

[0065] For specific implementation, please refer to Figure 2 , the specific process is as follows:

[0066] After normalizing the spectral data of each sample, two spectral data sets, male and female, were formed according to gender.

[0067] Considering that the number of negative samples is much larger than the number of positive samples, the sample() function of Pandas is used to randomly sample the negative sample data according to the number of positive samples in the male and female data, and frac is set to 1.2, that is, the number of negative samples is 1.2 times that of positive samples, in order to better handle unbalanced data and ensure that the model will not be overly biased towards categories with a larger number, thereby improving the overall classification performance.

[0068] The positive sample data and the negative sample data generated by random sampling of the sample function are merged, and the data set is randomly divided into a training set and a test set in a ratio of 0.7:0.3.

[0069] A deep learning model was constructed using Python, and the SVM and Xgboost algorithms were selected to establish a classification model for drug abuse in males and females, respectively. The classification effect of the model was comprehensively evaluated by sensitivity, specificity, accuracy, F1 score, and the area under the ROC curve (roc_auc). In this embodiment, the data set was randomly divided 5 times, and the average of the 5 experimental results was taken as the evaluation index.

[0070] It should be noted that the parameters are tuned by grid search during model training. The SVM algorithm tuning parameters and parameter value ranges are: 'kernel': ['rbf', 'poly'], 'C': [0.001, 0.01, 0.1, 1, 10, 100], 'gamma': [0.001, 0.01, 0.1, 1, 10, 100]. The Xgboost algorithm tuning parameters and parameter value ranges are: 'max_depth': range(3,10,2), 'min_child_weight': range(1,6,2), 'gamma': [i / 10.0for i in range(0,5)], 'subsample': [i / 10.0 for i in range(6,10)], 'colsample_bytree': [i / 10.0 for i in range(6,10)], 'reg_alpha': [0.85, 0.9, 0.95, 1,1.05, 1.1]. At the same time, the cross-validation fold is set to 5, that is, the grid parameter tuning uses cross-validation technology to train and evaluate different model parameter combinations. Each model parameter combination will be cross-validated and fitted on the training data set 5 times to find the best parameter combination. Once the best combination is found, it will run the fit again (without cross-validation) on all the training data passed to the fit to build a new model with the best parameter settings.

[0071] The SVM algorithm uses the grid search function GridSearchCV for hyperparameter tuning. The best parameters found by GridSearchCV are obtained through best_params_attribute. The best parameters for males are {'kernel': 'poly', 'gamma': 0.1, 'C': 100}; the best parameters for females are {'kernel': 'poly', 'gamma': 10, 'C': 10}.

[0072] The Xgboost algorithm uses GridSearchCV for hyperparameter tuning. The best parameters found by GridSearchCV are obtained through best_params_ attribute. The best parameters for men are {'learning_rate':0.1, 'max_depth':3,'min_child_weight':5, 'gamma':=0.4, 'subsample':0.7, 'colsample_bytree':0.6,'reg_alpha':0.87}; the best parameters for women are {'learning_rate':0.19, 'max_depth':7, 'min_child_weight':5, 'gamma':=0.03, 'subsample':0.7, 'colsample_bytree':0.6, 'reg_alpha':0.95}.

[0073] Based on the optimal parameters, four prediction models of SVM, Xgboost male and female were constructed respectively, and the test set data was input into the prediction model to verify the effectiveness of the present invention in screening drug addicts based on skin diffuse reflectance spectrum information, and the accuracy, sensitivity, specificity, F1 value and roc_auc of the two algorithms were compared, as shown in the following table:

[0074] Algorithm Model gender Accuracy Sensitivity Specificity F1 roc_auc SVM male 0.9435 0.9290 0.9496 0.944 0.9777 female 0.9706 0.9412 0.9733 0.97 0.9748 Xgboost male 0.8877 0.8851 0.8889 0.89 0.925 female 0.9854 0.9646 0.9878 0.986 0.9960

[0075] Table 1: Comparison of recognition effects of two algorithm models (average value of 5 tests)

[0076] As can be seen from the table, the two algorithms have better female recognition effects than male, and the results of the two algorithms are very close, the SVM algorithm is slightly better than Xgboost; the SVM male recognition effect is better than Xgboost, so the present invention finally selects the SVM algorithm model to identify whether men take drugs, and the Xgboost algorithm model to identify whether women take drugs. The method proposed in the present invention achieves the best accuracy, sensitivity, specificity, F1 value and roc_auc.

[0077] 2. Build a drug abuse recognition model for different drugs

[0078] Since drug addicts may take multiple drugs at the same time, it is necessary to build recognition models for different types of drugs, including marijuana, heroin, methamphetamine, marijuana, K powder and others. In this embodiment, the data collected and labeled with the types of drugs used are used to train and build a recognition model using SVM and Xgboost. Now take heroin as an example:

[0079] The spectral data of each sample is normalized to form a data set. Considering that the number of negative samples is much larger than the positive samples, the sample() function of Pandas is used to randomly sample the negative sample data according to the number of positive samples labeled "heroin". The frac is set to 1.2, that is, the number of negative samples is 1.2 times that of positive samples, in order to better handle unbalanced data and ensure that the model will not be overly biased towards categories with a larger number, thereby improving the overall classification performance.

[0080] The positive sample data and the negative sample data generated by random sampling of the sample function are merged, and the data set is randomly divided into a training set and a test set in a ratio of 0.7:0.3.

[0081] Python was used to build a deep learning model. SVM and Xgboost were selected to establish a classification model for heroin use. The classification effect of the model was comprehensively evaluated by sensitivity, specificity, accuracy, F1 score, and the area under the ROC curve (roc_auc). The data set was randomly divided five times, and the average of the five experimental results was taken as the evaluation index.

[0082] During model training, parameter tuning is performed through grid search. The SVM algorithm tuning parameters and parameter value ranges are: 'kernel':['rbf','poly'], 'C': [0.001, 0.01, 0.1, 1, 10, 100], 'gamma':[0.001, 0.01, 0.1, 1, 10, 100]. The Xgboost algorithm tuning parameters and parameter value ranges are: 'max_depth': range(3,10,2), 'min_child_weight': range(1,6,2), 'gamma': [i / 10.0 for i inrange(0,5)], 'subsample': [i / 10.0 for i in range(6,10)], 'colsample_bytree':[i / 10.0 for i in range(6,10)], 'reg_alpha': [0.85, 0.9, 0.95, 1, 1.05, 1.1]. At the same time, the cross-validation fold is set to 5, that is, the grid parameter tuning uses cross-validation technology to train and evaluate different model parameter combinations. Each model parameter combination will be cross-validated and fitted 5 times on the training data set to find the best parameter combination. Once the best combination is found, it will run the fit again (without cross-validation) on all the training data passed to the fit to build a new model with the best parameter settings.

[0083] The SVM algorithm uses the grid search function GridSearchCV for hyperparameter tuning, and obtains the best parameters found by GridSearchCV through best_params_attribute, {'kernel': 'poly', 'gamma': 1, 'C':100};

[0084] The Xgboost algorithm uses GridSearchCV for hyperparameter tuning. The best parameters found by GridSearchCV are obtained through best_params_ attribute, which are {'learning_rate':0.12, 'max_depth':7,'min_child_weight':1, 'gamma':=0.09, 'subsample':0.8, 'colsample_bytree':0.6,'reg_alpha':0.85}.

[0085] Based on the optimal parameters, SVM and Xgboost prediction models were constructed respectively, and the test set data was input into the prediction model to verify the effectiveness of the present invention in distinguishing heroin addiction based on skin diffuse reflectance spectrum information, and the accuracy, sensitivity, specificity, F1 value and roc_auc of the two algorithms were compared, as shown in the following table:

[0086] Algorithm Model Accuracy Sensitivity Specificity F1 roc_auc SVM 0.8713 0.9634 0.8623 0.872 0.9518 Xgboost 0.8421 0.9375 0.8329 0.852 0.9390

[0087] Table 2: Comparison of recognition effects of two algorithm models (average value of 5 tests)

[0088] As can be seen from the table, the results of the two algorithms are very close, and the SVM algorithm is slightly higher than Xgboost. The present invention finally selects the SVM algorithm model to identify heroin addiction. The method proposed in the present invention achieves the best accuracy, sensitivity, specificity, F1 value and roc_auc.

[0089] It should be noted that machine algorithms based on big data have the characteristic of better generalization performance as the amount of data increases. Therefore, as the amount of data increases, the types of drugs that can be identified by the above-mentioned trained and verified recognition model will also increase.

[0090] S2, obtain the spectral sequence data and the gender of the screened person.

[0091] In the specific implementation, a portable skin trace element monitor BC408H is used to collect spectral information of the non-dominant palm of the screened person to obtain spectral sequence data.

[0092] S3, preprocessing the spectral sequence data.

[0093] In specific implementation, the spectral sequence data is normalized.

[0094] S4, based on the gender of the screened person, the preprocessed spectral sequence data is input into different drug abuse recognition models for recognition to obtain drug abuse screening results.

[0095] As mentioned above, this embodiment finally determines that the SVM drug abuse recognition model is used for drug abuse screening for men, and the Xgboost drug abuse recognition model is used for drug abuse screening for women. The gender of the screened person and the above spectral sequence data are input into different models for prediction, and some of the probability results obtained are as follows:

[0096] Test ID gender True Label Prediction probability (%) Predicted Label 1058 male Non-drug addiction 5.21 Non-drug addiction 1179 female Non-drug addiction 2.01 Non-drug addiction 304 female Drug addiction 98.04 Drug addiction 518 male Non-drug addiction 33.33 Non-drug addiction 222 male Drug addiction 72.55 Drug addiction 398 male Non-drug addiction 13.33 Non-drug addiction 375 male Non-drug addiction 16.08 Non-drug addiction 1191 male Drug addiction 99.64 Drug addiction 627 male Non-drug addiction 2.18 Non-drug addiction 190 male Drug addiction 69.15 Drug addiction 1286 male Drug addiction 80.96 Drug addiction 1259 female Non-drug addiction 10.04 Non-drug addiction 1390 female Non-drug addiction 6.01 Non-drug addiction 292 male Drug addiction 79.85 Drug addiction 343 female Drug addiction 83.68 Drug addiction 140 male Non-drug addiction 36.36 Non-drug addiction 47 male Non-drug addiction 21.04 Non-drug addiction 1232 female Non-drug addiction 0.01 Non-drug addiction 216 male Drug addiction 79.93 Drug addiction 492 female Non-drug addiction 20.57 Non-drug addiction 77 male Non-drug addiction 30.15 Non-drug addiction

[0097] Table 3: Predicted probabilities by gender

[0098] The recognition effects of both algorithms are relatively good (roc_auc is higher than 0.90), which shows that it is feasible to establish a deep learning model based on skin spectrum information to predict drug addiction, and the recognition effect is good. Figure 3 and Figure 4 . Figure 3 In the table, the header refers to the drug addiction SVM algorithm model-male ROC curve, the horizontal axis 1-Specificity refers to the false positive rate, and the vertical axis Sensitivity refers to the true positive rate; Figure 4 In the table, the header refers to the drug addiction Xgboost algorithm model-female ROC curve, the horizontal axis 1-Specificity refers to the false positive rate, and the vertical axis Sensitivity refers to the true positive rate.

[0099] In addition, as mentioned above, taking heroin as an example, this embodiment constructs SVM and Xgboost models, inputs spectral sequence data into the model for prediction, and some probability results are as follows:

[0100] Test ID gender True Label Prediction probability (%) Predicted Label 260 female Non-heroin addiction 16.75 Non-heroin addiction 1025 female Non-heroin addiction 0.68 Non-heroin addiction 1326 male Heroin addiction 96.01 Heroin addiction 1180 male Non-heroin addiction 24.15 Non-heroin addiction 1451 male Non-heroin addiction 32.51 Non-heroin addiction 1396 male Heroin addiction 72.55 Heroin addiction 122 female Non-heroin addiction 11.56 Non-heroin addiction 1070 male Non-heroin addiction 34.75 Non-heroin addiction 908 female Non-heroin addiction 5.32 Non-heroin addiction 1429 male Non-heroin addiction 25.91 Non-heroin addiction 1014 female Non-heroin addiction 0.44 Non-heroin addiction 146 male Non-heroin addiction 2.9 Non-heroin addiction 137 female Heroin addiction 90.81 Heroin addiction 996 female Non-heroin addiction 39.44 Non-heroin addiction 1513 female Non-heroin addiction 31.82 Non-heroin addiction

[0101] Table 4: Predicted probabilities for heroin by different models

[0102] The recognition effects of both algorithms are relatively good (roc_auc is higher than 0.90), which shows that it is feasible to establish a deep learning model based on skin spectrum information to predict heroin addiction, and the recognition effect is good. Figure 5 . Figure 5 In the table, the header refers to the ROC curve of the heroin SVM algorithm model, the horizontal axis 1-Specificity refers to the false positive rate, and the vertical axis Sensitivity refers to the true positive rate

[0103] S5, feeding back the drug screening result to the mobile terminal for display.

[0104] It should be noted that the drug screening results include the probability of whether the screened person is taking drugs and the probability of the type of drugs taken.

[0105] The method for rapid screening of drug addicts based on skin spectroscopy provided by the embodiment of the present invention has the characteristics of being in situ non-destructive compared to traditional sampling and detection methods (blood test, urine test, hair test, etc.), and has the characteristics of being rapid and less restricted compared to non-invasive screening methods (breath test, iris / pupil test). In terms of rapid on-site screening, the method of the present invention is more effective, simple, reliable, and has a higher accuracy rate.

[0106] Based on the same inventive concept, an embodiment of the present invention also provides a system for rapid screening of drug addicts based on skin spectrum, including a skin spectrum information collector, a cloud platform and a mobile terminal that communicate with each other.

[0107] Among them, Figure 6 As shown, the cloud platform may include: one or more processors 101, one or more input devices 102, one or more output devices 103 and a memory 104, wherein the processors 101, the input devices 102, the output devices 103 and the memory 104 are interconnected via a bus 105. The memory 104 is used to store a computer program, wherein the computer program includes program instructions, and the processor 101 is configured to call the program instructions to execute the method of the above method embodiment.

[0108] It should be understood that in the embodiment of the present invention, the processor 101 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0109] The input device 102 may include a keyboard, etc., and the output device 103 may include a display (LCD, etc.), a speaker, etc.

[0110] The memory 104 may include a read-only memory and a random access memory, and provide instructions and data to the processor 101. A portion of the memory 104 may also include a non-volatile random access memory. For example, the memory 104 may also store information about the device type.

[0111] In a specific implementation, the processor 101, input device 102, and output device 103 described in the embodiment of the present invention can execute the implementation method described in the embodiment of the method for rapid screening of drug addicts based on skin spectrum provided by the embodiment of the present invention, which will not be repeated here.

[0112] Optionally, an embodiment of the present invention further provides another system for rapid screening of drug addicts based on skin spectrum, comprising a skin spectrum information collector, a cloud platform and a mobile terminal;

[0113] A skin spectrum information collector, which is used to emit visible light to illuminate the palm skin of the screened person and collect diffuse reflectance spectrum information, and send the diffuse reflectance spectrum data to the cloud platform;

[0114] The cloud platform is used to obtain the probability that the screened person is a drug addict based on the diffuse reflectance spectrum information of the screened person's palm skin through preprocessing, feature extraction and prediction model, and feed it back to the user terminal for display.

[0115] It should be noted that, regarding the specific working processes of each part of the system, please refer to the aforementioned method embodiments, which will not be repeated here.

[0116] When the above system is actually applied, the detection process is as follows Figure 7 As shown, including:

[0117] 1. Before testing, the mobile terminal is connected to the spectrometer via Bluetooth;

[0118] 2. After selecting gender according to the APP prompts, use the spectrometer to detect 4 points on the subject's non-dominant palm in turn;

[0119] 3. The mobile terminal sends the gender and spectral data to the cloud. The cloud normalizes the spectral data and inputs it into the recognition model to obtain the predicted probability value and feed it back to the mobile terminal.

[0120] 4. If the test result is "negative", the "next person" can be tested; if the test result is positive, any identifiable information such as the name / ID number / mobile phone number of the person being tested will be supplemented, and the suspect will be handled in accordance with relevant regulations;

[0121] 5. The current test is finished, and you can repeat steps 2 to 4 to continue testing.

[0122] That is, first select the corresponding algorithm according to gender (the SVM drug recognition model is selected for men, and the Xgboost drug recognition model is selected for women) to obtain the probability of whether the subject is a drug addict. If it is a drug addict, then identify it according to the drug type model, and identify it in the order of marijuana, heroin, methamphetamine, marijuana, K powder and others. Finally, the types with a probability of more than 50 are arranged in reverse order according to the probability value. Finally, the two result values ​​are fed back to the front end for display, such as Figure 8 shown.

[0123] In actual application, the present invention adopts a portable IoT spectrometer + mobile terminal APP + cloud computing to realize on-site rapid screening of drug addicts. The hardware equipment is compact and easy to carry. The spectrometer and the mobile terminal are connected via Bluetooth, without the need for power cord or data cable. The APP operation is simple and does not require professional personnel. Data analysis and prediction models are realized through cloud computing without the need for a PC or laptop computer. The entire detection process takes less than 30 seconds to achieve instant detection and results.

[0124] In addition, at the application level, the above system has APP modules for rapid screening and mandatory testing of serious suspects, and the background has management functions, including query and printing reports.

[0125] Among them, the description of the clinical rapid screening is as follows:

[0126] The spot check rapid screening is used for rapid screening in border inspection, traffic inspection, surprise inspection of entertainment venues and other scenarios. There is no need to enter user information in advance, and the test can be carried out directly, which improves the detection efficiency. The test result has only one indicator, which is used to identify whether the person being tested is suspected of drug use. The process of spot check rapid screening is as follows:

[0127] 1. Connecting devices

[0128] (1) Turn on the switch of BC408H detector;

[0129] (2) Turn on the Bluetooth function of the mobile terminal PDA;

[0130] (3) Open the detection APP on the mobile terminal PDA;

[0131] (4) Click the [Go to Configuration] button in the [Configure Device] section of the APP main interface;

[0132] (5) Select a device with the status "Connectable" in the device list and click to connect.

[0133] 2. Calibration equipment

[0134] (1) Use the calibration box that comes with the device;

[0135] (2) Use a lens cloth to clean the detection probe of the equipment and the white plate of the calibration box;

[0136] (3) Place the white board of the calibration box on the detection probe, click [White Calibration], and wait for the prompt [White Calibration Successful] to appear, and the white calibration is completed;

[0137] (4) Place the blackboard surface of the calibration box onto the detection probe, click [Black Calibration], and wait for the prompt [Black Calibration Successful] to appear. The black calibration is completed.

[0138] 3. Testing

[0139] (1) Make sure the device is ready (connection calibration is completed);

[0140] (2) Click [Quick Screening] on the main interface of the Similar APP;

[0141] (3) Select the gender of the subject;

[0142] (4) Test the four points in turn according to the four positions of the palm shown in the APP: Press the detection probe of the instrument vertically onto the palm skin of the subject, keep the detection probe and skin in close contact to prevent light leakage, press the detection button on the instrument, and wait until the red background of the current point number lights up, and the current point detection is completed; after all four points are tested, the results will be automatically displayed and the detection process is completed.

[0143] (IV) Check the quick screening results

[0144] After the 4-point rapid screening test is completed, the rapid screening results will be automatically displayed. Negative means no suspicion, suspected means weak positive, and positive means suspicion. If the result is [Negative], click [Next] to continue testing other people. If the result is [Positive], the system will automatically pop up a prompt box to fill in the remarks. After filling in, click [OK] to save, or click [Cancel] to not fill in the remarks, and then click [Next] to continue testing other people.

[0145] The following are the instructions for mandatory inspection of serious suspects:

[0146] The mandatory test for serious suspects is used in scenarios where complete information such as ID cards need to be registered. When testing, complete user information including ID card numbers must be entered in advance. The test result has only one indicator, which is used to identify whether the person being tested is suspected of drug use. The process of mandatory test for serious suspects is as follows:

[0147] 1. Enter user information

[0148] Before conducting mandatory testing on serious suspects, you first need to configure the equipment (see the rapid screening for clinical testing for the operating procedures) to ensure that the equipment is ready.

[0149] Click [Repeated Suspects Must Be Inspected] on the main interface of the APP and enter the user information.

[0150] After entering the ID number, if the ID card is already in the system, other information will be automatically filled in; if the ID card has been drug screened before, the last result will be automatically displayed.

[0151] If the user information has not been entered, click [Add]. After completing the information entry, the user information will be displayed, and click [Detect] to start screening.

[0152] 2. Testing

[0153] According to the 4 positions of the palm shown in the APP, test the 4 points in turn: press the detection probe of the instrument vertically onto the palm skin of the subject, keep the detection probe and skin in close contact without light leakage, press the detection button on the instrument, and wait until the red background of the current point number lights up, and the current point detection is completed; after all 4 points are tested, the results will be automatically displayed and the detection process is completed.

[0154] (III) View test results

[0155] After the 4-point test is completed, the test results will be automatically displayed. Negative means no suspicion, suspected means weak positive, and positive means suspicion. If the result is [Positive], the system will automatically pop up a prompt box to fill in the remarks. After filling in, click [OK] to save, or click [Cancel] to not fill in the remarks, and then click [Return] to continue entering the next user information.

[0156] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for rapid screening of drug addicts based on skin spectrum, characterized in that: include: Obtain spectral sequence data and the gender of the person being screened; The spectral sequence data is obtained by collecting diffuse reflectance spectral information of the non-dominant palm of the screened person using a skin spectral information collector; Preprocessing the spectral sequence data; Based on the gender of the screened person, the pre-processed spectral sequence data is input into different drug abuse recognition models for recognition to obtain drug abuse screening results; Feeding back the drug abuse screening result to the mobile terminal for display; the drug abuse screening result includes the probability of whether the screened person is taking drugs and the probability of the type of drugs taken; Among them, the drug screening results are obtained, specifically: According to the gender of the screened person, different first-type drug abuse identification models are selected to identify the preprocessed spectral sequence data to obtain the probability of whether the screened person is a drug addict; the first-type drug abuse identification model is constructed according to different genders; If the screened person is a drug addict, different second-type drug abuse recognition models are selected for identification to obtain the probability of the screened person taking different drugs, and the types of drugs whose probabilities exceed preset values ​​are arranged in reverse order according to the probability values; the second-type drug abuse recognition models are constructed according to different drugs.

2. The method for rapid screening of drug addicts based on skin spectrum according to claim 1, characterized in that: The method also includes constructing and training a drug abuse recognition model, specifically: A skin spectrum information collector is used to collect spectrum information of the non-dominant palm of a sample person, and diffuse reflectance spectrum data of four points are obtained in sequence; the sample person includes drug addicts and non-drug addicts; Collect the gender and types of drugs used by all sample persons; The diffuse reflectance spectrum data of each sample person is labeled with gender, drug use, and drug type to obtain label data; The label data is used to construct and train a drug abuse recognition model for different genders or different drugs.

3. The method for rapid screening of drug addicts based on skin spectrum according to claim 2, characterized in that: For different genders, the label data is used to construct and train a drug abuse recognition model, specifically: Normalizing the label data; For different genders, a male spectral dataset and a female spectral dataset are established based on the normalized label data; the number of negative samples in the male and female spectral datasets is much larger than the number of positive samples; among them, drug users are positive samples, and non-drug users are negative samples; According to the number of positive samples in the male and female spectral datasets, use the sample() function of Pandas to randomly sample negative sample data, and set the parameter frac to 1.2; Merge the positive sample data with the negative sample data randomly sampled by the sample() function, and randomly divide the data set into a training set and a test set in a ratio of 7:3; A deep learning model is constructed using python, and the training set and test set are used to train and verify the drug abuse recognition model; the drug abuse recognition model is used to screen drug addicts and obtain the probability that the screened person is a drug addict.

4. The method for rapid screening of drug addicts based on skin spectrum according to claim 3, characterized in that: A deep learning model was constructed using Python, and the training set and test set were used to train and verify the drug abuse recognition model, specifically: Select SVM algorithm and Xgboost algorithm to establish drug abuse classification models for male and female genders respectively; Extracting features from the diffuse reflectance spectrum data in the training set to obtain spectrum sequence features; Using the spectral sequence features as input of the classification model, and searching for the best hyperparameter combination of the classification model through a network search method; The test set is input into a classification model with an optimal hyperparameter combination to obtain a final drug abuse identification type; the drug abuse identification model includes an SVM drug abuse identification model and an Xgboost drug abuse identification model, the SVM drug abuse identification model is used for drug abuse screening for males, and the Xgboost drug abuse identification model is used for drug abuse screening for females.

5. The method for rapid screening of drug addicts based on skin spectrum according to claim 4, characterized in that: The method further comprises: The output results of the drug abuse recognition model were comprehensively evaluated using accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score and area under the ROC curve.

6. A cloud platform for rapid screening of drug addicts based on skin spectrum, characterized in that: The method comprises a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method according to any one of claims 1 to 5.

7. A system for rapid screening of drug addicts based on skin spectrum, comprising a skin spectrum information collector, a cloud platform and a mobile terminal that communicate with each other, characterized in that: The cloud platform is as described in claim 6.

8. A system for rapid screening of drug addicts based on skin spectrum, comprising a skin spectrum information collector, a cloud platform and a mobile terminal, characterized in that: The skin spectrum information collector is used to emit visible light to illuminate the palm skin of the screened person and collect spectrum sequence data; The cloud platform is used to execute the method according to any one of claims 1 to 5 to obtain drug screening results; The mobile terminal is used to display the drug abuse screening result.

9. The system for rapid screening of drug addicts based on skin spectrum as claimed in claim 8, characterized in that: The mobile terminal is also used to provide the functions of rapid screening for on-site inspections and mandatory inspection for serious suspects.

Images