Methods for predicting the relapse risk of drug addicts

Abstract

The present invention relates to the technical field of drug control monitoring, and discloses a method for predicting the relapse risk of drug addicts, including: conducting a drug visual stimulation test on drug addicts and non-drug addicts, collecting image data of the tested persons, and obtaining the drug craving value of the tested persons, obtaining the motion feature data and expression feature data of the tested persons based on the image data, training a first prediction model based on training samples formed by expression feature data and corresponding craving values ​​and person identity attributes, and training a second prediction model based on training samples formed by motion feature data and corresponding craving values ​​and person identity attributes, using the two models to predict drug addicts respectively, and predicting the relapse risk by combining the prediction results of the two prediction models and the model weights. The present invention can achieve high-efficiency and accurate prediction of the relapse risk of drug addicts.

Description

Technical Field

[0001] The present invention relates to the technical field of drug control monitoring, and more specifically, to a method for predicting the relapse risk of drug addicts. Background Art

[0002] Currently, there is no effective way to predict whether drug addicts will relapse. The only way is to conduct reexaminations through regular manual interviews and random sampling monitoring, and predict the possibility of relapse based on the subjective judgment of the staff. This greatly wastes manpower and material resources and is very inefficient. In addition, due to the subjective concealment of drug addicts and the different professional levels of reexamination personnel, the prediction accuracy is low, the prediction results cannot be quantified, and the prediction timeliness is poor.

[0003] In view of this, how to overcome the defects of existing relapse prediction methods and solve the problem of difficulty in predicting the relapse risk of drug addicts is a problem to be solved in this technical field. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting the relapse risk of drug addicts, so as to achieve efficient and accurate prediction of the relapse risk of drug addicts.

[0005] To achieve the above object, the present invention proposes a method for predicting the relapse risk of drug addicts, comprising:

[0006] Through conducting drug visual stimulation tests on drug addicts, people in the process of drug rehabilitation, people who have successfully rehabilitated drugs, people who have relapsed after drug rehabilitation, and non-drug addicts, collecting image data of the tested persons during the test, and obtaining the drug craving values ​​of the tested persons through drug-related questionnaire tests;

[0007] Obtaining motion feature data and expression feature data of the person being measured based on the image data;

[0008] The drug craving value of the tested person and the identity attribute of the tested person are used as first label data of the corresponding expression feature data, the expression feature data and the corresponding first expression feature data are used as first training samples, part of the first training samples are used as a first training set, and another part of the first training samples are used as a first test set;

[0009] Using the drug craving value of the tested person and the identity attribute of the tested person as the second label data of the corresponding action feature data, using the action feature data and the corresponding second label data as second training samples, using part of the second training samples as a second training set, and using another part of the second training samples as a second test set;

[0010] Using the first training set to train a first prediction model, the trained first prediction model can output corresponding identity attributes and desire level prediction values ​​according to the input expression feature data;

[0011] The second prediction model is trained using the second training set, so that the trained second prediction model can output corresponding identity attributes and desire level prediction values ​​according to the input action feature data;

[0012] Using the first test set and the second test set to evaluate the accuracy of the prediction results of the first prediction model and the second prediction model respectively, and determining the weights of the prediction results output by the first prediction model and the second prediction model according to the accuracy of the prediction results of the two models;

[0013] Acquire the action feature data and expression feature data of the drug rehabilitation personnel to be evaluated when undergoing a drug visual stimulation test, input the expression feature data of the drug rehabilitation personnel to be evaluated into the first prediction model to obtain a first prediction result, and input the action feature data of the drug rehabilitation personnel to be evaluated into the second prediction model to obtain a second prediction result;

[0014] Based on the first prediction result, the second prediction result, and the weights of the prediction results output by the first prediction model and the second prediction model, a comprehensive relapse risk prediction result of the drug rehabilitation person to be evaluated is obtained.

[0015] Optionally, the weight value of the prediction result output by the model with higher prediction accuracy among the first prediction model and the second prediction model is greater.

[0016] Optionally, it also includes:

[0017] Values ​​are assigned to different identity attributes, where the scores of drug addicts, drug addicts who relapse after drug rehabilitation, people in the drug rehabilitation process, people who have successfully quit drugs, and non-drug addicts decrease in order.

[0018] Optionally, the first prediction result includes a first identity attribute and a first craving degree prediction value corresponding to the drug-free person to be evaluated, and the second prediction result includes a second identity attribute and a second craving degree prediction value corresponding to the drug-free person to be evaluated;

[0019] The method of obtaining a comprehensive relapse risk prediction result of the drug rehabilitation person to be evaluated based on the first prediction result, the second prediction result, and the weight of the prediction results output by the first prediction model and the second prediction model includes:

[0020] determining identity attribute scores corresponding to the first identity attribute and the second identity attribute,

[0021] The relapse probability of the drug-free person to be evaluated is calculated using the following formula:

[0022]

[0023] In the formula, is the relapse probability value of the drug rehabilitation personnel to be evaluated. The larger the value, the higher the relapse risk. is the first eagerness prediction value output by the first prediction model, is a second desire level prediction value output by the second prediction model, is the identity attribute score corresponding to the first identity attribute output by the first prediction model, is the identity attribute score corresponding to the second identity attribute output by the second prediction model, is the weight value of the prediction result of the first prediction model, The weight value of the prediction result of the second prediction model.

[0024] Optionally, obtaining the drug craving value of the tested person through a drug-related questionnaire test includes:

[0025] While conducting the drug visual stimulation test or after completing the drug visual stimulation test, conducting a drug-related questionnaire test on the testee, obtaining the testee's answer score, and using the answer score as the testee's drug desire degree value;

[0026] The drug-related questionnaire includes a plurality of questions on quantifiable behavioral characteristics or psychological characteristics related to the degree of drug craving, and a higher score indicates a higher degree of drug craving.

[0027] Optionally, when conducting drug-related question-and-answer tests on drug addicts, people in the process of drug rehabilitation, people who have successfully quit drugs, and people who relapse after drug rehabilitation, a polygraph may be used to assist in obtaining the degree of drug craving of the person being tested.

[0028] Optionally, obtaining facial expression feature data of the person being measured based on the image data includes:

[0029] The facial image data of the person under test is obtained from the image data, and facial expression features are extracted from the facial image data using a facial expression recognition algorithm to obtain expression feature data.

[0030] Optionally, obtaining motion feature data of the measured person based on the image data includes:

[0031] The limb image data of the measured person is obtained from the image data, and the action features of the limb image data are extracted by a motion recognition algorithm to obtain the action feature data.

[0032] Optionally, the first prediction model adopts a convolutional neural network model, and the second prediction model adopts a recurrent neural network model.

[0033] Optionally, after completing the relapse risk prediction for the drug rehabilitation personnel to be assessed, it also includes:

[0034] Adding the facial expression feature data, action feature data and prediction results of the drug rehabilitation personnel to be evaluated into the first training set and the second training set;

[0035] Using the updated first training set and the second training set to complete the self-learning of the first prediction model and the second prediction model;

[0036] After predicting the relapse risk of a specific number of drug rehabilitation personnel, the accuracy of the prediction results of the first prediction model and the second prediction model is re-evaluated, and the weights of the model prediction results are adjusted based on the prediction accuracy of the two models.

[0037] The beneficial effects of the present invention are:

[0038] The present invention performs drug visual stimulation tests on drug addicts, people in the process of drug rehabilitation, people who have successfully detoxified, people who relapse after drug rehabilitation, and non-drug addicts, collects image data of the tested people during the test, obtains the drug craving degree value of the tested people, then obtains action feature data and expression feature data of the tested people based on the image data, trains a first prediction model based on training samples formed by the expression feature data and the corresponding craving degree value and the identity attributes of the people, and trains a second prediction model based on training samples formed by the action feature data and the corresponding craving degree value and the identity attributes of the people, then collects the expression feature data and action feature data of the drug rehabilitation people when they are undergoing the drug visual stimulation test, and uses the trained first prediction model and the second prediction model to train the first prediction model. The prediction models predict the drug craving degree and identity attributes of drug addicts respectively, and finally the weights corresponding to the prediction results of the two prediction models are combined to comprehensively predict the relapse risk of drug addicts. This method can collect real facial expressions and movement feature data of people with different drug abuse levels and non-drug addicts with different identities through drug visual stimulation tests, thereby improving the accuracy of sample data. The two models are used to make predictions respectively, and the weights of the prediction accuracy of the two models are comprehensively considered to improve the accuracy of the prediction results. The same drug visual stimulation test method is used for drug addicts to ensure the consistency of personnel test conditions and model input data, further improving the accuracy of the prediction. At the same time, this method is more timely and efficient than the existing manual method of judging relapse risk.

[0039] Furthermore, the method of the present invention is based on the desire level values ​​and identity attributes in the prediction results of the two models. The identity attribute values ​​of people with different degrees of association with drugs are determined by assigning values ​​to the identity attributes. The relapse risk probability is comprehensively calculated by combining the desire level values ​​and identity attribute values ​​predicted by the two models and the weight values ​​of the two models, so as to obtain a more scientific and accurate relapse risk prediction result.

[0040] The system of the present invention has other characteristics and advantages, which will be apparent from the drawings incorporated herein and the following detailed description, or will be described in detail in the drawings incorporated herein and the following detailed description, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, in which like reference numerals generally represent like components.

[0042] Figure 1 A step diagram of a method for predicting the relapse risk of drug addicts according to the present invention is shown. DETAILED DESCRIPTION

[0043] The present invention will be described in more detail below with reference to the accompanying drawings. Although preferred embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0044] like Figure 1 As shown, this embodiment provides a method for predicting the relapse risk of drug addicts, including:

[0045] S1: Conduct drug visual stimulation tests on drug addicts, people in the process of drug rehabilitation, people who have successfully rehabilitated drugs, people who relapse after drug rehabilitation, and non-drug addicts, collect image data of the tested persons during the test, and obtain the drug craving values ​​of the tested persons through drug-related questionnaire tests;

[0046] In this step, video capture can be used to capture images of the above-mentioned people with different degrees of association with drugs during the drug visual stimulation test.

[0047] In this step, obtaining the drug craving value of the tested person includes:

[0048] While conducting the drug visual stimulation test or after completing the drug visual stimulation test, conducting a drug-related questionnaire test on the testee, obtaining the testee's answer score, and using the answer score as the testee's drug desire degree value;

[0049] The drug-related questionnaire includes a plurality of questions on quantifiable behavioral characteristics or psychological characteristics related to the degree of drug craving, and a higher score indicates a higher degree of drug craving.

[0050] Preferably, when conducting drug-related question-and-answer tests on drug addicts, people in the process of drug rehabilitation, people who have successfully quit drugs, and people who relapse after drug rehabilitation, a polygraph is used to assist in obtaining the degree of drug craving of the tested person, so as to ensure the accuracy of the answers of former drug users, thereby improving the authenticity and accuracy of the degree of drug craving of the relevant persons.

[0051] S2: obtaining motion feature data and expression feature data of the person being measured based on the image data;

[0052] In this step, the facial expression feature data of the person being measured is obtained based on the image data, including:

[0053] The facial image data of the person under test is obtained from the image data, and facial expression features are extracted from the facial image data using a facial expression recognition algorithm to obtain expression feature data.

[0054] In this step, obtaining the motion feature data of the person being measured based on the image data includes:

[0055] The limb image data of the measured person is obtained from the image data, and the action features of the limb image data are extracted by a motion recognition algorithm to obtain the action feature data.

[0056] Among them, the facial expression recognition algorithm can adopt the existing mature facial expression recognition algorithms, such as the Holistic method based on global information, which operates on the entire face image to extract features, including algorithms such as principal component analysis (PCA) and linear discriminant analysis (LDA). These methods reduce the dimension of the face image by finding a set of projection vectors, and then send the low-dimensional features to machine learning classifiers such as SVM for facial expression feature recognition. For example, the PCA method finds a set of feature vectors (called "feature faces"), each of which captures or describes a change or characteristic between faces, so that each face can be represented as a linear combination of these feature faces.

[0057] The action recognition algorithm can adopt the existing mature human action feature extraction algorithm, such as the human action recognition algorithm based on the human joint angle: This method mainly involves the acquisition of 3D joint point coordinates of the human body, feature engineering, model integration and model training. First, the 3D coordinates of the key points of the human body are extracted using the Openpose body25 pre-trained model combined with the depth map. Then, the feature vector is formed by calculating the cosine value of the angle between the 24 limb vectors. Next, the variance-based Filter method and the random forest-based RFE method are used for feature selection.

[0058] S3: using the drug desire value of the tested person and the identity attribute of the tested person as the first label data of the corresponding expression feature data, using the expression feature data and the corresponding first expression feature data as the first training samples, using part of the first training samples as the first training set, and using another part of the first training samples as the first test set;

[0059] S4: using the drug craving value of the tested person and the identity attribute of the tested person as the second label data of the corresponding action feature data, using the action feature data and the corresponding second label data as second training samples, using part of the second training samples as a second training set, and using another part of the second training samples as a second test set;

[0060] S5: training a first prediction model using the first training set, so that the trained first prediction model can output corresponding identity attributes and predicted values ​​of desire according to the input expression feature data;

[0061] In this embodiment, the first prediction model adopts a convolutional neural network (CNN) model, which uses a convolutional neural network (CNN) to perform feature recognition on expression feature image data through a convolutional layer, a pooling layer, and a fully connected layer, and inputs these features into a classifier to classify identity attributes and desire level values.

[0062] S6: training the second prediction model using the second training set, so that the trained second prediction model can output corresponding identity attributes and desire level prediction values ​​according to the input action feature data;

[0063] In this embodiment, the second prediction model adopts a recurrent neural network model (RNN), which includes an input layer, a hidden layer, and an output layer. The recurrent neural network introduces a loop in the hidden layer so that the network can remember previous information and use it in the processing of the current input. This structure is particularly suitable for processing data with forward and backward dependencies, such as time series data. This embodiment preferably adopts an improved RNN model, such as a long short-term memory network (LSTM), inputs the image frame data of the action feature into the LSTM network, then recognizes the action feature, and outputs the corresponding identity attribute and the predicted value of the desire degree based on the recognized action feature.

[0064] S7: using the first test set and the second test set to evaluate the accuracy of the prediction results of the first prediction model and the second prediction model respectively, and determining the weights of the prediction results output by the first prediction model and the second prediction model according to the accuracy of the prediction results of the two models;

[0065] Among them, the model with higher prediction accuracy in the first prediction model and the second prediction model has a greater weight value of the prediction result output by the model.

[0066] S8: obtaining the action feature data and expression feature data of the drug-free person to be evaluated when undergoing a drug visual stimulation test, inputting the expression feature data of the drug-free person to be evaluated into the first prediction model to obtain a first prediction result, and inputting the action feature data of the drug-free person to be evaluated into the second prediction model to obtain a second prediction result;

[0067] In this step, the first prediction result includes a first identity attribute and a first desire level prediction value corresponding to the drug rehabilitation person to be evaluated, and the second prediction result includes a second identity attribute and a second desire level prediction value corresponding to the drug rehabilitation person to be evaluated.

[0068] The method also includes: assigning values ​​to different identity attributes in advance, wherein the identity attribute scores of drug addicts, drug addicts who relapse after drug rehabilitation, people in the process of drug rehabilitation, people who have successfully rehabilitated drugs, and non-drug addicts decrease in sequence.

[0069] S9: Based on the first prediction result, the second prediction result, and the weights of the prediction results output by the first prediction model and the second prediction model, a comprehensive relapse risk prediction result of the drug rehabilitation person to be evaluated is obtained.

[0070] Step S9 specifically includes:

[0071] determining identity attribute scores corresponding to the first identity attribute and the second identity attribute,

[0072] The relapse probability of the drug-free person to be evaluated is calculated using the following formula:

[0073]

[0074] In the formula, is the relapse probability value of the drug rehabilitation personnel to be evaluated, is the first eagerness prediction value output by the first prediction model, is a second desire level prediction value output by the second prediction model, is the identity attribute score corresponding to the first identity attribute output by the first prediction model, is the identity attribute score corresponding to the second identity attribute output by the second prediction model, is the weight value of the prediction result of the first prediction model, The weight value of the prediction result of the second prediction model.

[0075] The relapse probability value of the drug rehabilitation personnel to be evaluated is calculated , assess the relapse risk of drug addicts who have quit drug abuse, A larger value indicates a higher risk of relapse.

[0076] In this embodiment, preferably, after the relapse risk prediction of the drug rehabilitation personnel to be assessed is completed, the following steps are further included:

[0077] Adding the facial expression feature data, action feature data and prediction results of the drug rehabilitation personnel to be evaluated into the first training set and the second training set;

[0078] Using the updated first training set and the second training set to complete the self-learning of the first prediction model and the second prediction model;

[0079] After predicting the relapse risk of a specific number of drug rehabilitation personnel, the accuracy of the prediction results of the first prediction model and the second prediction model is re-evaluated, and the weights of the model prediction results are adjusted based on the prediction accuracy of the two models.

[0080] Through the relapse risk assessment of many drug rehabilitation personnel in the subsequent model application process, the samples of the training set are continuously supplemented and improved, and the weight values ​​of the prediction results of the two models are continuously adjusted based on the prediction accuracy of the two models, gradually improving the accuracy of the comprehensive prediction results of multiple models.

[0081] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method for predicting the relapse risk of drug addicts, characterized in that: include: Through conducting drug visual stimulation tests on drug addicts, people in the process of drug rehabilitation, people who have successfully detoxified, people who have relapsed after drug rehabilitation, and non-drug addicts, collecting image data of the tested persons during the test, and obtaining the drug craving values ​​of the tested persons through drug-related questionnaire tests; Obtaining motion feature data and expression feature data of the person being measured based on the image data; The drug craving value of the tested person and the identity attribute of the tested person are used as the first label data of the corresponding expression feature data, the expression feature data and the corresponding first label data are used as the first training samples, part of the first training samples are used as the first training set, and another part of the first training samples are used as the first test set; Using the drug craving value of the tested person and the identity attribute of the tested person as the second label data of the corresponding action feature data, using the action feature data and the corresponding second label data as second training samples, using part of the second training samples as a second training set, and using another part of the second training samples as a second test set; Using the first training set to train a first prediction model, the trained first prediction model can output corresponding identity attributes and desire level prediction values ​​according to the input expression feature data; The second prediction model is trained using the second training set, so that the trained second prediction model can output corresponding identity attributes and predicted values ​​of the desire level according to the input action feature data; Using the first test set and the second test set to evaluate the accuracy of the prediction results of the first prediction model and the second prediction model respectively, and determining the weights of the prediction results output by the first prediction model and the second prediction model according to the accuracy of the prediction results of the two models; Obtaining motion feature data and facial feature data of the drug rehabilitation personnel to be evaluated when undergoing a drug visual stimulation test, inputting the facial feature data of the drug rehabilitation personnel to be evaluated into the first prediction model to obtain a first prediction result, and inputting the motion feature data of the drug rehabilitation personnel to be evaluated into the second prediction model to obtain a second prediction result; Obtaining a comprehensive relapse risk prediction result of the detoxified person to be evaluated based on the first prediction result, the second prediction result, and the weights of the prediction results output by the first prediction model and the second prediction model; The model with higher prediction accuracy in the first prediction model and the second prediction model has a greater weight value of the prediction result output; Also includes: Assign values ​​to different identity attributes, where the identity attribute scores of drug addicts, drug relapsers, drug rehabilitation personnel, drug rehabilitation personnel, drug rehabilitation personnel, and non-drug addicts decrease in order; The first prediction result includes a first identity attribute and a first craving degree prediction value corresponding to the drug-free person to be evaluated, and the second prediction result includes a second identity attribute and a second craving degree prediction value corresponding to the drug-free person to be evaluated; The method of obtaining a comprehensive relapse risk prediction result of the drug rehabilitation person to be evaluated based on the first prediction result, the second prediction result, and the weight of the prediction results output by the first prediction model and the second prediction model includes: determining identity attribute scores corresponding to the first identity attribute and the second identity attribute, The relapse probability of the drug-free person to be evaluated is calculated using the following formula: ； In the formula, is the relapse probability value of the drug rehabilitation personnel to be evaluated. The larger the value, the higher the relapse risk. is the first eagerness prediction value output by the first prediction model, is a second desire level prediction value output by the second prediction model, is the identity attribute score corresponding to the first identity attribute output by the first prediction model, is the identity attribute score corresponding to the second identity attribute output by the second prediction model, is the weight value of the prediction result of the first prediction model, The weight value of the prediction result of the second prediction model.

2. The method for predicting the relapse risk of drug addicts according to claim 1, characterized in that: The method of obtaining the drug craving value of the tested person through a drug-related questionnaire test includes: While conducting the drug visual stimulation test or after completing the drug visual stimulation test, conducting a drug-related questionnaire test on the testee, obtaining the testee's answer score, and using the answer score as the testee's drug desire degree value; The drug-related questionnaire includes a plurality of questions on quantifiable behavioral characteristics or psychological characteristics related to the degree of drug craving, and a higher score indicates a higher degree of drug craving.

3. The method for predicting the relapse risk of drug addicts according to claim 2, characterized in that: When conducting drug-related question-and-answer tests on drug addicts, people in the process of drug rehabilitation, people who have successfully quit drugs, and people who relapse after drug rehabilitation, a polygraph is used to assist in obtaining the degree of drug desire of the test person.

4. The method for predicting the relapse risk of drug addicts according to claim 1, characterized in that: Obtaining facial expression feature data of the person being measured based on the image data includes: The facial image data of the person under test is obtained from the image data, and facial expression features are extracted from the facial image data using a facial expression recognition algorithm to obtain expression feature data.

5. The method for predicting the relapse risk of drug addicts according to claim 1, characterized in that: Obtaining motion feature data of the person being measured based on the image data includes: The limb image data of the measured person is obtained from the image data, and the action features of the limb image data are extracted by a motion recognition algorithm to obtain the action feature data.

6. The method for predicting the relapse risk of drug addicts according to claim 1, characterized in that: The first prediction model adopts a convolutional neural network model, and the second prediction model adopts a recurrent neural network model.

7. The method for predicting the relapse risk of drug addicts according to claim 6, characterized in that: After the relapse risk prediction is completed for the drug rehabilitation personnel to be assessed, it also includes: Adding the facial expression feature data, action feature data and prediction results of the drug rehabilitation personnel to be evaluated into the first training set and the second training set; Using the updated first training set and the second training set to complete the self-learning of the first prediction model and the second prediction model; After predicting the relapse risk of drug addicts, the accuracy of the prediction results of the first prediction model and the second prediction model is re-evaluated, and the weights of the model prediction results are adjusted based on the prediction accuracy of the two models.

Images