Pollution monitoring method based on behavior pattern recognition

Abstract

The invention discloses a pollution monitoring method based on behavior pattern recognition, which is used for monitoring whether a real-time environment is polluted or not, and comprises the following steps: obtaining real-time environment image data and reference environment image data, and determining a query vector of the real-time environment image data; calling reference environment element data in the environment element reference library, and performing feature extraction to obtain a reference environment feature vector; acquiring real-time environment element data, and performing feature extraction to obtain a real-time environment element data feature map; distributing the weights of the environment elements according to the query vector, obtaining a reference environment feature vector and a real-time environment feature vector, and calculating a transfer matrix; and inputting the transfer matrix into an environment classifier to obtain a classification result, and sending a pollution warning signal according to the classification result, so that pollution information can be provided for maintenance personnel in time, the pollution level can be effectively prevented from rising in time, the calculation complexity is reduced, and the recognition rate and accuracy of real-time environment pollution are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pollution monitoring, and particularly relates to a pollution monitoring method based on behavior pattern recognition. BACKGROUND

[0002] In recent years, with the improvement of social productivity, a large amount of pollution is generated in the manufacturing process of many industries, for example, the main pollution source types of construction sites, vehicle repair and catering industries, and the generation of pollution not only threatens the living environment of human beings, but also may cause great safety hazards and reduce production efficiency. Therefore, the monitoring of industrial pollution, the timely discovery of pollution problems, the adoption of corresponding measures to reduce or eliminate pollution measures, the protection of the integrity and stability of the ecological system, pollution monitoring plays a very key role in environmental protection, public health, sustainable development and policy making, and strengthening pollution monitoring has important significance for maintaining ecological balance, protecting people's health and promoting economic and social development. The traditional pollution monitoring method generally relies on manual on-site detection, cannot timely discover the occurrence of pollution sources, wastes many human resources, and the elements of pollution judgment are relatively single, and the remote cannot determine the corresponding repair manpower and repair time according to the pollution level of the pollution source. SUMMARY

[0003] The technical problem solved by the present application is that the existing detection technology cannot timely discover the occurrence of pollution sources, the elements of pollution judgment are relatively single, and the remote cannot judge the pollution level of the pollution source.

[0004] To solve the above technical problems, the present application provides the following technical scheme: a pollution monitoring method based on behavior pattern recognition, comprising the following steps

[0005] Step S100, acquiring real-time environment image data of a to-be-detected site and reference environment image data of the to-be-detected site, and determining a query vector of the real-time environment image data according to the reference environment image data;

[0006] Step S200, calling reference environment element data corresponding to the reference environment image data in an environment element reference library, performing feature extraction on the reference environment element data, and obtaining a reference environment feature vector;

[0007] Step S300, acquiring real-time environment element data of a real-time environment, performing feature extraction on the real-time environment element data, and obtaining a real-time environment element data feature map;

[0008] Step S400, assigning a weight of an environment element according to the query vector, obtaining the reference environment feature vector and the real-time environment feature vector, and calculating a transition matrix of the reference environment feature vector relative to the real-time environment feature vector;

[0009] Step S500, input the transfer matrix into an environment classifier to obtain a classification result, and send a pollution warning signal according to the classification result.

[0010] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, the step S100 specifically comprises:

[0011] Step S101, call the monitoring video data of the real-time environment of the to-be-detected site, intercept the image frame data of the monitoring video data according to a unit time, and obtain the real-time environment image data;

[0012] Step S102, call the reference environment image data of the to-be-detected site in the reference image library data;

[0013] Step S103, determine the query vector according to the detection task of the reference environment image data.

[0014] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, the step S200 specifically comprises:

[0015] Step S201, match the reference environment element data corresponding to the environment element reference library according to the reference environment image data corresponding to the real-time environment image data, wherein the reference environment element data comprises reference human behavior element data, reference temperature element data, reference humidity element data and reference odor element data;

[0016] Step S202, input the reference environment element data into an encoder for encoding, wherein the encoder has three layers, namely an input layer, a hidden layer and an output layer, the reference environment element data is taken as an input of an autoencoder, the output layer outputs a reference environment element data feature map of the reference environment element data, and the reference environment element data feature map comprises a reference human behavior element feature map, a reference temperature element feature map, a reference humidity element feature map and a reference odor element feature map;

[0017] Step S203, convert the reference environment element data feature map from a three-dimensional array to a one-dimensional array, convert the one-dimensional array into a feature vector, and obtain a reference environment feature vector.

[0018] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, the step S300 specifically comprises:

[0019] Step S301, according to the real-time environment element data corresponding to the real-time environment image data, the corresponding real-time environment element data is detected by an environment element detection instrument, the real-time environment element data detection instrument includes a camera, a temperature sensor, a humidity sensor, an odor sensor, the real-time environment element data includes real-time human behavior elements, real-time temperature elements, real-time humidity elements and real-time odor elements;

[0020] The human behavior of the site to be detected is monitored by the camera;

[0021] The temperature of the site to be detected is monitored by the temperature sensor;

[0022] The humidity of the site to be detected is monitored by the humidity sensor;

[0023] The odor concentration of the site to be detected is monitored by the odor sensor;

[0024] Step S302, the real-time environment element data is subjected to data segmentation and data enhancement processing to obtain first real-time environment element data after preliminary processing, the abnormal values in the first real-time environment element data are deleted and the missing values are supplemented to obtain second real-time environment element data;

[0025] Step S303, the real-time environment element data is input into an encoder for encoding, the encoder has three layers, namely an input layer, a hidden layer and an output layer, the real-time environment element data is taken as the input of the autoencoder, the output layer outputs the real-time environment element data feature map of the real-time environment element data, and the real-time environment element data feature map includes a human behavior element feature map, a temperature element feature map, a humidity element feature map and an odor element feature map;

[0026] Step S304, the real-time environment element data feature map is converted from a three-dimensional array to a one-dimensional array, and the one-dimensional array is converted into a feature vector to obtain a real-time environment feature vector.

[0027] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, step S400 specifically includes:

[0028] Step S401, according to the query vector of the real-time environment image data obtained from the reference environment image data, the similarity of the query vector with the reference human behavior element data, the reference temperature element data, the reference humidity element data and the reference odor element data is calculated, the similarity is subjected to normalization processing to obtain a weight matrix;

[0029] Step S402, the reference human behavior element feature vector, the reference temperature element feature vector, the reference humidity element feature vector and the reference smell element feature vector are weighted and summed through the weight matrix to obtain a reference environment feature vector;

[0030] Step S403, the real-time human behavior element feature vector, the real-time temperature element feature vector, the real-time humidity element feature vector and the real-time smell element feature vector are weighted and summed through the weight matrix to obtain a real-time environment feature vector;

[0031] Step S404, a transition matrix of the reference environment feature vector relative to the real-time environment feature vector is calculated, and the mathematical expression is as follows:

[0032]

[0033] Wherein, V a represents the reference environment feature vector, V b represents the real-time environment feature vector, M represents the transition matrix, represents the matrix multiplication.

[0034] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, wherein: step S500 specifically comprises:

[0035] Step S501, the transition matrix is unfolded as a row vector to obtain a classification feature vector;

[0036] Step S502, the classification feature vector is input into the full connection layer of the environment classifier, and the full connection layer performs full connection coding on the classification feature vector to obtain a coded classification feature vector;

[0037] Step S503, the coded classification feature vector is input into the Softmax classification function of the environment classifier, and the classification function outputs the classification result;

[0038] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, wherein: the mathematical expression of the Softmax classification function is as follows:

[0039]

[0040] Wherein, y i represents a probability value, x i represents the i-th row vector of the classification feature vector, k represents the k-th row vector of the classification feature vector, N represents that the classification feature vector has N row vectors, and e represents exponential operation.

[0041] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, the classification result has three categories, including no pollution, first-class pollution and second-class pollution, the output of the Softmax classification function is the probability distribution of different real-time environment states, and different classification results are obtained.

[0042] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, the sum of the row elements of the transition matrix is 1, and the number of row elements is equal to the number of row elements of the reference environment feature vector.

[0043] As a preferred scheme of the power equipment state data monitoring method based on artificial intelligence, the dimension of the hidden layer of the encoder is smaller than the dimension of the input layer, and the dimension reduction operation is performed from the input layer to the hidden layer, so as to describe the reference environment element data and the real-time environment element data with smaller dimensions.

[0044] The beneficial effects of the present application are as follows: by calling the monitoring video data of the real-time environment of the to-be-detected site, intercepting the real-time environment image data, inputting the environment elements of the real-time environment image data and the reference environment image data into the encoder to obtain the reference environment feature vector and the real-time environment feature vector, and calculating the transition matrix of the reference environment feature vector and the real-time environment feature vector, the change trend of the real-time environment state to the reference environment state can be obtained, the pollution classification result direction of the real-time environment can be predicted according to the change trend, the pollution message can be provided to the maintenance personnel in time, and the pollution level can be prevented from rising in time and effectively.

[0045] The real-time environment image data is intercepted, the reference environment image data corresponding to the real-time environment of the to-be-detected site in the reference image library data is called, each real-time environment of the to-be-detected site has its corresponding reference environment, and the real-time environment element data only needs to be input into the encoder with the corresponding reference environment element data to obtain the feature vector, calculate the corresponding transition matrix, input the transition matrix into the full connection layer of the environment classifier, and then input the environment classifier to obtain the final classification result, so as to reduce the complexity of calculation and improve the accuracy of real-time environment pollution identification.

[0046] By calling the reference environment element data corresponding to the reference environment image data in the environment element reference library, the reference environment element data corresponding to each reference environment image data is different, different query vectors are determined according to different detection tasks, the weights of the environment elements are allocated according to the query vectors, the pollution is considered from multiple elements, the judgment result is more comprehensive, the weights of the reference environment element data can be adjusted according to the differences of the real-time environment, the judgment result is more targeted and more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 A basic flowchart of a pollution monitoring method based on behavior pattern recognition is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments.

[0049] Embodiments, with reference to Figure 1 For an embodiment of the present application, a pollution monitoring method based on behavior pattern recognition is provided, comprising:

[0050] Step S100, acquiring real-time environmental image data of a to-be-detected site and reference environmental image data of the to-be-detected site, determining a query vector of the real-time environmental image data according to the reference environmental image data;

[0051] Step S200, calling reference environmental element data corresponding to the reference environmental image data in an environmental element reference library, performing feature extraction on the reference environmental element data to obtain a reference environmental feature vector;

[0052] Step S300, acquiring real-time environmental element data of the real-time environment, performing feature extraction on the real-time environmental element data to obtain a real-time environmental element data feature map;

[0053] Step S400, assigning weights to environmental elements according to the query vector, obtaining the reference environmental feature vector and the real-time environmental feature vector, and calculating a transition matrix of the reference environmental feature vector relative to the real-time environmental feature vector;

[0054] Step S500, inputting the transition matrix into an environmental classifier to obtain a classification result, and sending a pollution warning signal according to the classification result.

[0055] The step S100 specifically comprises:

[0056] Step S101, calling monitoring video data of a real-time environment of a to-be-detected site, and according to unit time, intercepting image frame data of the monitoring video data to obtain the real-time environmental image data;

[0057] Step S102, calling the reference environmental image data of the to-be-detected site in the reference image library data;

[0058] Step S103, determining the query vector according to a detection task of the reference environmental image data.

[0059] In this embodiment, the to-be-detected site includes places to be detected for pollution, such as major pollution source construction sites, vehicle repair and catering areas, and identifies pollution such as bare soil, material uncovered and soil on construction site roads. A monitoring camera is installed at the to-be-detected site, and image frame data of monitoring video data is intercepted per unit time, which is 5 minutes. Real-time environmental image data of the to-be-detected environment is obtained, and an image library stores corresponding reference environmental image data of each to-be-detected site, such as no bare soil, material covering and no soil on construction site roads. According to different detection tasks, different query vectors are determined. For example, when the detection object is a garbage can, the detection task is whether the garbage in the garbage can is full. The query vector is that people pour garbage to increase the garbage in the garbage can until it is full.

[0060] Step S200 specifically includes:

[0061] Step S201, according to the reference environmental image data corresponding to the real-time environmental image data, match the reference environmental element data corresponding to the environmental element reference library, the reference environmental element data includes reference human behavior element data, reference temperature element data, reference humidity element data and reference odor element data;

[0062] Step S202, input the reference environmental element data into the encoder for encoding, the encoder has three layers, respectively, the input layer, the hidden layer and the output layer, the reference environmental element data as the input of the autoencoder, the output layer outputs the reference environmental element data feature map of the reference environmental element data, the reference environmental element data feature map includes reference human behavior element feature map, reference temperature element feature map, reference humidity element feature map and reference odor element feature map;

[0063] Step S203, convert the reference environmental element data feature map from three-dimensional array to one-dimensional array, and convert the one-dimensional array into a feature vector to obtain a reference environmental feature vector.

[0064] In this embodiment, according to the reference environmental image data corresponding to the real-time environmental image data, the reference environmental element data corresponding to the environmental element reference library is matched. For example, when the detection object is a garbage can, the reference environmental image data is no garbage overflow, no odor stimulation, and the garbage humidity cannot be too large to prevent the garbage from overflowing as a fluid and polluting the road and affecting pedestrians. According to the detection task of the real-time environmental image data, the reference environmental element data includes reference human behavior element data, reference humidity element data and reference odor element data. The reference environmental element data is input into the encoder for encoding to obtain the reference environmental element data feature map of the reference environmental image data. The reference environmental feature map is flattened, which can convert the three-dimensional array into a one-dimensional array. The one-dimensional array is transposed to obtain the reference environmental feature vector. Feature extraction is performed on each environmental element, which can consider pollution from multiple environmental elements, making pollution judgment more targeted.

[0065] Step S300 specifically comprises:

[0066] Step S301, according to the real-time environment element data corresponding to the real-time environment image data, the corresponding real-time environment element data is detected by an environment element detection instrument, the real-time environment element data detection instrument includes a camera, a temperature sensor, a humidity sensor, an odor sensor, the real-time environment element data includes real-time human behavior elements, real-time temperature elements, real-time humidity elements and real-time odor elements;

[0067] The human behavior of the site to be detected is monitored by the camera;

[0068] The temperature of the site to be detected is monitored by the temperature sensor;

[0069] The humidity of the site to be detected is monitored by the humidity sensor;

[0070] The odor concentration of the site to be detected is monitored by the odor sensor;

[0071] Step S302, the real-time environment element data is subjected to data segmentation and data enhancement processing to obtain first real-time environment element data after preliminary processing, the abnormal values in the first real-time environment element data are deleted and the missing values are supplemented to obtain second real-time environment element data;

[0072] Step S303, the real-time environment element data is input into an encoder for encoding, the encoder has three layers, respectively, an input layer, a hidden layer and an output layer, the real-time environment element data is taken as the input of the autoencoder, the output layer outputs the real-time environment element data feature map of the real-time environment element data, the real-time environment element data feature map includes human behavior element feature map, temperature element feature map, humidity element feature map and odor element feature map;

[0073] Step S304, the real-time environment element data feature map is converted from a three-dimensional array to a one-dimensional array, the one-dimensional array is converted into a feature vector to obtain a real-time environment feature vector.

[0074] In this embodiment, according to the real-time environment element data corresponding to the real-time environment image data, for example, when the detected object is a garbage can, the real-time environment element data including real-time human behavior element data, real-time humidity element data and real-time odor element data is determined according to the reference environment element data corresponding to the real-time environment image data, the real-time environment element data obtained in a unit time is subjected to data segmentation and data enhancement processing to obtain first real-time environment element data after preliminary processing, and the abnormal values in the first real-time environment element data are deleted and the missing values are supplemented to obtain second real-time environment element data, which can improve the data quality, expand the data set size, improve the model generalization ability and performance, input the processed real-time environment image data into the encoder for coding to obtain the real-time environment feature map of the real-time environment image data, the obtained feature representation is more pure and robust, which can better cope with the uncertainty in the real-time environment image data, and the real-time environment feature map is flattened, which can convert the three-dimensional array into a one-dimensional array, and the one-dimensional array is transposed to obtain the reference environment feature vector.

[0075] Step S400 specifically includes:

[0076] Step S401, obtaining a query vector of the real-time environment image data according to the reference environment image data, calculating the similarity of the query vector with the reference human behavior element data, the reference temperature element data, the reference humidity element data and the reference odor element data, and performing normalization processing on the similarity to obtain a weight matrix;

[0077] Step S402, performing weighted summation on the reference human behavior element feature vector, the reference temperature element feature vector, the reference humidity element feature vector and the reference odor element feature vector through the weight matrix to obtain a reference environment feature vector;

[0078] Step S403, performing weighted summation on the real-time human behavior element feature vector, the real-time temperature element feature vector, the real-time humidity element feature vector and the real-time odor element feature vector through the weight matrix to obtain a real-time environment feature vector;

[0079] Step S404, calculating the transition matrix of the reference environment feature vector relative to the real-time environment feature vector, and the mathematical expression is as follows:

[0080]

[0081] Wherein, V a represents the reference environment feature vector, V b represents the real-time environment feature vector, and M represents the transition matrix, represents the matrix multiplication.

[0082] In this embodiment, the query vector of the real-time environment image data obtained according to the reference environment image data is, for example, when the detected object is a garbage can, the main task of detection is whether the garbage in the garbage can is full, and the secondary task is that the smell is not irritating, and the humidity of the garbage cannot be too large. Then the main query vector is that the person puts garbage in the garbage can, which increases the garbage in the garbage can until it is full. The secondary query vector is the smell of the garbage and the water content of the garbage. For the detection task of the real-time environment image data, the reference environment element data includes reference human behavior element data, reference humidity element data and reference odor element data. According to the similarity calculation of the query vector and the reference environment element data, a weight matrix is obtained. The weight matrix is used to weight and sum the reference human behavior element feature vector, the reference temperature element feature vector, the reference humidity element feature vector and the reference odor element feature vector to obtain the reference environment feature vector. The weight matrix is used to weight and sum the real-time human behavior element feature vector, the real-time temperature element feature vector, the real-time humidity element feature vector and the real-time odor element feature vector to obtain the reference environment feature vector. The weight matrix can distribute the weight size of the reference environment element data, so that important reference environment element data can become the main judgment element in the pollution judgment, so that the judgment is more targeted. The transfer matrix of the reference environment feature vector relative to the real-time environment feature vector is calculated, and the transfer probability of the real-time environment image data to the reference environment image data is obtained. The greater the value in the transfer matrix, the greater the possibility of the real-time environment image data changing into the reference environment image data, and the greater the probability of the real-time environment of the detection site being uncontaminated. If it is 1, it represents that the detection site is uncontaminated, and if it is 0, it represents that the detection site is contaminated.

[0083] Step S500 specifically includes:

[0084] Step S501, the transfer matrix is unfolded according to the row vector to obtain a classification feature vector;

[0085] Step S502, inputting the classification feature vector into the full connection layer of the environment classifier, the full connection layer performing full connection coding on the classification feature vector to obtain an encoded classification feature vector;

[0086] Step S503, inputting the encoded classification feature vector into the Softmax classification function of the environment classifier, and the classification function outputs the classification result;

[0087] In the embodiment, the transition matrix is expanded as a row vector to obtain a classification feature vector. Each row of the transition matrix represents a transition probability distribution from a starting state to all other possible states. Expanding the row vectors into feature vectors can extract the probability distribution as a feature. The classification feature vector can reflect the internal relationship and transition rule between different states. Inputting the classification feature vector into the fully connected layer of the environment classifier and combining the Softmax classification function can obtain a classification result, which can make the classification result more accurate and comprehensive. The classification result has three categories, including no pollution, first-level pollution and second-level pollution.

[0088] The mathematical expression of the Softmax classification function is as follows:

[0089]

[0090] where y i represents a probability value, x i represents the i-th row vector of the classification feature vector, k represents the k-th row vector of the classification feature vector, N represents the total number of row vectors of the classification feature vector, and e represents an exponential operation.

[0091] In the embodiment, the row vectors of the transition matrix are input into the fully connected layer of the environment classifier to obtain an encoded classification feature vector. The i-th row vector of the classification function represents the probability distribution of a specific environment state transforming into a reference environment, and the N row vectors represent the respective probability distributions of N environment states transforming into the reference environment.

[0092] The classification result has three categories, including no pollution, first-level pollution and second-level pollution. The output of the Softmax classification function is the probability distribution of different real-time environment states, and different classification results are obtained.

[0093] In the embodiment, the classification result includes no pollution, first-level pollution and second-level pollution, which can give maintenance personnel clear pollution warning signals.

[0094] When the pollution warning signal is no pollution, the maintenance personnel do not need to perform maintenance tasks.

[0095] When the pollution warning signal is first-level pollution and second-level pollution, the maintenance personnel perform maintenance tasks according to the specific situation.

[0096] The sum of the row elements of the transition matrix is 1, and the number of row elements is equal to the number of row elements of the reference environment feature vector.

[0097] In the embodiment, the sum of the row elements of the transition matrix is 1, and the row elements represent the probability of each specific environment state transforming into a reference environment state. The number of row elements represents the number of specific environment states.

[0098] The dimension of the hidden layer of the encoder is less than the dimension of the input layer, and a dimension reduction operation is performed from the input layer to the hidden layer, for describing the reference environment element data and the real-time environment element data in a smaller dimension.

[0099] In this embodiment, the dimension of the hidden layer of the encoder is less than the dimension of the input layer, the encoder can compress the data amount of the input image, convert high-dimensional image data into low-dimensional feature representation, retain key feature information, reduce the amount of calculation in the subsequent processing process, and improve the processing efficiency.

[0100] By calling the monitoring video data of the real-time environment of the to-be-detected site, the real-time environment image data is intercepted, the real-time environment image data and the environment element of the reference environment image data are input into the encoder to obtain the reference environment feature vector and the real-time environment feature vector, and the transition matrix of the reference environment feature vector and the real-time environment feature vector is calculated, so that the change trend of the real-time environment state to the reference environment state can be obtained, the pollution classification result direction of the real-time environment can be predicted according to the change trend, the pollution message can be provided to the maintenance personnel in time, and the increase of the pollution level can be prevented in time and effectively.

[0101] The real-time environment image data is intercepted, and the reference environment image data corresponding to the real-time environment of the to-be-detected site in the reference image library data is called. Each real-time environment of the to-be-detected site has its corresponding reference environment, and only the corresponding reference environment element data needs to be input into the encoder with the real-time environment element data to obtain the feature vector, the corresponding transition matrix is calculated, the transition matrix is input into the full connection layer of the environment classifier, and then the environment classifier is input to obtain the final classification result, which can reduce the complexity of calculation and improve the accuracy of real-time environment pollution identification.

[0102] By calling the reference environment element data corresponding to the reference environment image data in the environment element reference library, the reference environment element data corresponding to each reference environment image data is different, and different query vectors are determined according to different detection tasks, and the weights of the environment elements are allocated according to the query vectors, which not only considers pollution from multiple elements to make the judgment result more comprehensive, but also adjusts the weights of the reference environment element data according to the differences of the real-time environment, so that the judgment result is more targeted and more accurate.

[0103] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (or computer- readable storage media) having computer-usable program code embodied in the medium. The medium can be any available medium or combination thereof that is accessible by a general purpose or special purpose computer. By way of example, such computer-usable storage media can include a volatile memory, such as a random access memory (RAM), a non-volatile memory, such as a read-only memory (ROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a floppy diskette, a compact disk, a tape, a flash memory, etc. The computer-usable program code can include any suitable set of instructions directly readable by a computer or any suitable set of instructions executable by a computer after being processed by an interpreter. The computer-usable program code can also be stored in a computer- readable storage medium that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the computer- readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement the flowchart Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or multiple blocks.

[0104] It should be noted that the above-mentioned embodiments are only used to illustrate but not to limit the technical solutions of the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalent replaced without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A pollution monitoring method based on behavior pattern recognition, characterized in that, The method comprises the following steps: Step S100, acquiring real-time environment image data of a to-be-detected site and reference environment image data of the to-be-detected site, determining a query vector of the real-time environment image data according to the reference environment image data; Step S200, calling reference environment element data corresponding to the reference environment image data in an environment element reference library, performing feature extraction on the reference environment element data to obtain a reference environment feature vector; Step S300, acquiring real-time environment element data of the real-time environment, performing feature extraction on the real-time environment element data to obtain a real-time environment feature vector; Step S400, assigning a weight of an environment element according to the query vector, obtaining the reference environment feature vector and the real-time environment feature vector, and calculating a transition matrix of the reference environment feature vector relative to the real-time environment feature vector; Step S500, inputting the transition matrix into an environment classifier to obtain a classification result, and sending a pollution warning signal according to the classification result.

2. A pollution monitoring method based on behavior pattern recognition as claimed in claim 1, wherein: The step S100 specifically comprises: Step S101, calling monitoring video data of a real-time environment of a to-be-detected site, and obtaining image frame data of the real-time environment according to unit time; Step S102, calling the reference environment image data of the to-be-detected site in a reference image library; Step S103, determining the query vector according to a detection task of the reference environment image data.

3. The pollution monitoring method based on behavior pattern recognition according to claim 2, wherein: The step S200 specifically comprises: Step S201, matching reference environment element data corresponding to the reference environment image data in an environment element reference library according to the reference environment image data corresponding to the real-time environment image data, wherein the reference environment element data comprises reference human behavior element data, reference temperature element data, reference humidity element data, and reference odor element data; Step S202, inputting the reference environment element data into an encoder for encoding, wherein the encoder comprises three layers, namely an input layer, a hidden layer, and an output layer, the reference environment element data is taken as an input of an autoencoder, the output layer outputs a reference environment element data feature map of the reference environment element data, and the reference environment element data feature map comprises a reference human behavior element feature map, a reference temperature element feature map, a reference humidity element feature map, and a reference odor element feature map; Step S203, converting the reference environment element data feature map from a three-dimensional array to a one-dimensional array, converting the one-dimensional array into a feature vector, and obtaining a reference environment feature vector.

4. The pollution monitoring method based on behavior pattern recognition according to claim 3, wherein: The step S300 specifically comprises: Step S301, detecting corresponding real-time environmental element data according to the real-time environmental element data corresponding to the real-time environmental image data through an environmental element detection instrument, the real-time environmental element data detection instrument including a camera, a temperature sensor, a humidity sensor, and an odor sensor, and the real-time environmental element data including real-time human behavior element data, real-time temperature element data, real-time humidity element data, and real-time odor element data; monitoring human behavior in the detection site through the camera; monitoring the temperature of the detection site through the temperature sensor; monitoring the humidity of the detection site through the humidity sensor; monitoring the odor concentration of the detection site through the odor sensor; Step S302, performing data segmentation and data enhancement processing on the real-time environmental element data to obtain first real-time environmental element data after preliminary processing, deleting abnormal values and supplementing missing values in the first real-time environmental element data to obtain second real-time environmental element data; Step S303, inputting the real-time environmental element data into an encoder for encoding, the encoder having three layers, namely an input layer, a hidden layer, and an output layer, the real-time environmental element data serving as the input of the autoencoder, the output layer outputting a real-time environmental element data feature map of the real-time environmental element data, and the real-time environmental element data feature map including a human behavior element feature map, a temperature element feature map, a humidity element feature map, and an odor element feature map; Step S304, converting the real-time environmental element data feature map from a three-dimensional array to a one-dimensional array, and converting the one-dimensional array into a feature vector to obtain a real-time environmental feature vector.

5. The pollution monitoring method based on behavior pattern recognition according to claim 4, characterized in that: Step S400 specifically includes: Step S401, obtaining a query vector of the real-time environmental image data according to the reference environmental image data, calculating the similarity of the query vector with reference human behavior element data, reference temperature element data, reference humidity element data, and reference odor element data, and performing normalization processing on the similarity to obtain a weight matrix; Step S402, performing weighted summation on the reference human behavior element feature vector, the reference temperature element feature vector, the reference humidity element feature vector, and the reference odor element feature vector through the weight matrix to obtain a reference environmental feature vector; Step S403, performing weighted summation on the real-time human behavior element feature vector, the real-time temperature element feature vector, the real-time humidity element feature vector, and the real-time odor element feature vector through the weight matrix to obtain a real-time environmental feature vector; Step S404, calculating the transition matrix of the reference environmental feature vector relative to the real-time environmental feature vector, and the mathematical expression being as follows: wherein V a represents the reference environment feature vector, V b represents the real-time environment feature vector, M represents the transition matrix, represents the matrix multiplication.

6. A pollution monitoring method based on behavior pattern recognition as claimed in claim 5, wherein: Step S500 specifically includes: Step S501, expanding the transition matrix as a row vector to obtain a classification feature vector; Step S502, inputting the classification feature vector into the fully connected layer of the environmental classifier, performing fully connected coding on the classification feature vector through the fully connected layer to obtain a coded classification feature vector; Step S503, input the encoded classification feature vector into a Softmax classification function of the environment classifier, and output the classification result.

7. A pollution monitoring method based on behavior pattern recognition as claimed in claim 6, wherein: The mathematical expression of the Softmax classification function is as follows: wherein y i represents a probability value, x i represents an i-th row vector of the classification feature vector, k represents a k-th row vector of the classification feature vector, N represents that the classification feature vector has N row vectors in total, and e represents an exponential operation.

8. A pollution monitoring method based on behavior pattern recognition as claimed in claim 7, wherein: The classification result has three categories, including no pollution, first-level pollution and second-level pollution, and the output of the Softmax classification function is the probability distribution of different real-time environment states, and different classification results are obtained.

9. A pollution monitoring method based on behavior pattern recognition as claimed in claim 8, wherein: The sum of the row elements of the transition matrix is 1, and the number of row elements is equal to the number of row elements of the reference environment feature vector.

10. A pollution monitoring method based on behavior pattern recognition as claimed in claim 9, wherein: The dimension of the hidden layer of the encoder is smaller than that of the input layer, and the dimension reduction operation is performed from the input layer to the hidden layer, so as to describe the reference environment element data and the real-time environment element data in a smaller dimension.

Images