Personnel behavior identification method and system based on big data
By combining bilateral attention mechanism with convolutional neural networks, extracting and fusing target features in personnel behavior data, the problem of difficulty in capturing long-term dependencies and complex interaction patterns in behavior data in the existing technology is solved, and a higher level of Behavior identification accuracy and efficiency.
Patent Information
- Application Number
- CN202411582907.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-07
- Publication Date
- 2025-02-14
AI Technical Summary
Existing personnel behavior recognition technology has limitations in processing spatiotemporal information, and it is difficult to fully capture long-term dependencies and complex interaction patterns in behavioral data.
Combining the bilateral attention mechanism with the convolutional neural network, extracting target features through the bilateral attention mechanism, and using the convolutional neural network to fusion, to build a behavior recognition model.
It significantly improves the accuracy and efficiency of behavior recognition, and enhances the adaptability and robustness of the model to complex scenarios.
Smart Images
Figure CN119445670A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of personnel behavior recognition, and in particular, to a personnel behavior recognition method and system based on big data. Background Art
[0002] In today's society, with the rapid development of big data and artificial intelligence technology, human behavior recognition technology has become a hot topic in research and application. This technology has broad application prospects in many fields such as security monitoring, health monitoring, intelligent transportation, and human-computer interaction. Traditional human behavior recognition methods mainly rely on manual feature extraction and simple classifiers. These methods often require a lot of expertise and time to design features, and have limited adaptability and generalization capabilities for complex scenarios. With the emergence of deep learning technology, especially the successful application of convolutional neural networks or CNNs in the field of image recognition, people have begun to try to apply them to human behavior recognition tasks. However, pure convolutional neural networks have certain limitations in processing spatiotemporal information, and it is difficult to fully capture the long-term temporal dependencies and complex interaction patterns in behavioral data. In order to solve this problem, researchers have proposed a method to combine attention mechanisms with convolutional neural networks, hoping to enhance the model's ability to capture key information through attention mechanisms, thereby improving the accuracy of behavior recognition. As a special attention mechanism, the bilateral attention mechanism can simultaneously consider the interaction between two information sources, which helps the model better understand and analyze human behavior. Nevertheless, how to effectively integrate the bilateral attention mechanism with convolutional neural networks, and how to process and utilize large-scale behavioral data, are still challenges in current research. Summary of the invention
[0003] The present invention provides a method and system for personnel behavior recognition based on big data, which improves the accuracy and efficiency of personnel behavior recognition by integrating a bilateral attention mechanism and a convolutional neural network.
[0004] The technical solution adopted by the present invention is:
[0005] A method for identifying human behavior based on big data includes the following specific steps:
[0006] S1, collect original image data of personnel behavior;
[0007] S2, preprocessing the original image data to obtain target image data;
[0008] S3, extracting features from the target image data based on a bilateral attention mechanism to obtain a first target feature, and fusing the first target feature based on a convolutional neural network to obtain a second target feature;
[0009] S4. constructing a behavior recognition model based on the second target feature;
[0010] S5. Perform personnel behavior recognition based on the behavior recognition model to obtain the target personnel behavior.
[0011] Furthermore, the step S2 specifically includes:
[0012] S21, performing data cleaning processing on the original image data to fill missing values in the original image data and remove abnormal values in the original image data to obtain first image data;
[0013] S22, performing Gaussian filtering on the first image data to remove noise in the original image data to obtain second image data;
[0014] S23: Normalize the second image data to obtain target image data.
[0015] Furthermore, the Gaussian filtering process specifically includes:
[0016] 1) Calculating the weight corresponding to each pixel in the first image data based on a Gaussian function, the Gaussian function calculation formula is as follows:
[0017]
[0018] In formula (1), G(m,n) is the value of the Gaussian function at the point (m,n); σ is the standard deviation, which controls the width of the Gaussian distribution; the constant 2πσ 2 is a normalization factor that ensures that the integral of the Gaussian function on the entire plane is 1; the exponential term Describes the shape of a Gaussian distribution, where m and n are coordinate variables;
[0019] 2) For each pixel point in the first image data, the pixel values in its neighborhood are convolved with the Gaussian kernel to obtain the weighted average of the pixel values in the neighborhood of each pixel point after filtering, where the convolution formula is as follows:
[0020]
[0021] In formula (2), I filtered (i, j) represents the pixel value of the pixel point (i, j) in the first image data after filtering; I(i+a, j+b) represents the pixel value of the pixel point (i+a, j+b) in the first image data; G(a, b) represents the element value of the Gaussian kernel at the position (a, b); c represents the size of the Gaussian kernel neighborhood, that is, the radius of the kernel;
[0022] 3) Substituting the weighted average of the pixel values in the neighborhood of each pixel for the original value of each pixel to obtain a new value of each pixel to obtain the second image data;
[0023] Furthermore, the normalization process specifically includes:
[0024] The minimum-maximum normalization method is used to map the second image data to the [0,1] interval to obtain the target image data. The calculation formula is as follows:
[0025]
[0026] In formula (3), x norm represents the normalized value; x represents the value of the second image data; x min represents the minimum value in the second image data; x max Indicates the maximum value in the second image data.
[0027] Furthermore, the step S3 specifically includes:
[0028] 1) Extract information source P and information source Q based on the target image data and calculate element p in information source P α and element q in the information source Q β The attention weight between is calculated as follows:
[0029]
[0030] In formula (4), w αβ Represents element p in information source P α and element q in the information source Q β The attention weight between g(p α ,q β ) is a measure of p α and q β The function of the correlation between them; k and l represent the index variables used for iteration;
[0031] 2) Calculate the sum of the products of each element in the information source P and its corresponding attention weight, and update it to obtain the information source P', the calculation formula of which is as follows:
[0032]
[0033] In formula (5), P' represents the updated representation of the information source P after the bilateral attention mechanism. The update is to replace each element p in the information source P with α Multiply it by its corresponding attention weight w αβ And do this by summing all these products;
[0034] 3) Calculate the sum of the products of each element in the information source Q and its corresponding attention weight, and update it to obtain the information source Q', the calculation formula of which is as follows:
[0035]
[0036] In formula (6), Q' represents the updated representation of the information source Q after the bilateral attention mechanism; the update is to replace each element q in the information source Q with β Multiply it by its corresponding attention weight w αβ And do this by summing all these products;
[0037] 4) The updated information source P' and information source Q' are used as the first target feature extracted from the target image data;
[0038] 5) The first target feature extracted by the bilateral attention mechanism is taken as input and sequentially fused through the two-dimensional convolution layer or Conv2d, batch normalization or BN, ReLU and maximum pooling layer or Max pooling in the convolutional neural network to obtain the second target feature.
[0039] Another technical solution adopted by the present invention is:
[0040] A personnel behavior recognition system based on big data, comprising:
[0041] Data collection module, used to collect raw image data of personnel behavior;
[0042] A data processing module, used for preprocessing the original image data to obtain target image data;
[0043] A model building module, used for extracting features from the target image data based on a bilateral attention mechanism to obtain a first target feature, fusing the first target feature based on a convolutional neural network to obtain a second target feature, and building a behavior recognition model based on the second target feature;
[0044] The recognition module is used to perform personnel behavior recognition based on the behavior recognition model to obtain the target personnel behavior.
[0045] Compared with the prior art, the present invention has the following beneficial effects:
[0046] By combining the bilateral attention mechanism with the convolutional neural network, the present invention can more effectively extract and fuse the target features in the personnel behavior data, thereby significantly improving the accuracy of behavior recognition; the bilateral attention mechanism can not only enhance the model's ability to capture the relationship between behavioral features, but also help the model focus on the most important information for behavior recognition, thereby enhancing the model's adaptability and robustness to complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments:
[0048] Figure 1 It is a flow chart of a method for identifying personnel behavior based on big data of the present invention;
[0049] Figure 2 It is a structural schematic diagram of a personnel behavior recognition system based on big data of the present invention. DETAILED DESCRIPTION
[0050] Example 1
[0051] like Figure 1 As shown, a method for identifying human behavior based on big data includes the following specific steps:
[0052] S1, collect original image data of personnel behavior;
[0053] Specifically, raw image data is collected from open source websites. The raw image data includes but is not limited to PAMAP2 (Physical Activity Monitoring dataset), OPPORTUNITY and MHELTH (Mobile Health datasets). These datasets are multi-position wearable human behavior recognition data, which are used to provide a large number of real samples for the construction and training of subsequent models and improve the accuracy of the network;
[0054] S2, preprocessing the original image data to obtain target image data, specifically including:
[0055] S21, performing data cleaning processing on the original image data to fill missing values in the original image data and remove abnormal values in the original image data to obtain first image data;
[0056] Specifically, first check the integrity of the data, and then perform data cleaning:
[0057] 1) Identify missing values and deal with them by deleting samples containing missing values, filling them with the mean or median, etc. according to the specific situation;
[0058] 2) Identify outliers through statistical methods such as mean and standard deviation or visual methods such as box plots; choose to delete or modify outliers;
[0059] S22, performing Gaussian filtering on the first image data to remove noise in the original image data to obtain second image data;
[0060] Specifically, 1) the weight corresponding to each pixel in the first image data is calculated based on a Gaussian function, and the Gaussian function calculation formula is as follows:
[0061]
[0062] In formula (1), G(m,n) is the value of the Gaussian function at the point (m,n); σ is the standard deviation, which controls the width of the Gaussian distribution; the constant 2πσ 2 is a normalization factor that ensures that the integral of the Gaussian function on the entire plane is 1; the exponential term Describes the shape of a Gaussian distribution, where m and n are coordinate variables;
[0063] 2) For each pixel point in the first image data, the pixel values in its neighborhood are convolved with the Gaussian kernel to obtain the weighted average of the pixel values in the neighborhood of each pixel point after filtering, where the convolution formula is as follows:
[0064]
[0065] In formula (2), I filtered (i, j) represents the pixel value of the pixel point (i, j) in the first image data after filtering; I(i+a, j+b) represents the pixel value of the pixel point (i+a, j+b) in the first image data; G(a, b) represents the element value of the Gaussian kernel at the position (a, b); c represents the size of the Gaussian kernel neighborhood, that is, the radius of the kernel;
[0066] 3) Substituting the weighted average of the pixel values in the neighborhood of each pixel for the original value of each pixel to obtain a new value of each pixel to obtain the second image data;
[0067] S23, performing normalization processing on the second image data to obtain target image data;
[0068] Specifically, the minimum-maximum normalization method is used to map the second image data to the [0,1] interval to obtain the target image data, and the calculation formula is as follows:
[0069]
[0070] In formula (3), x norm represents the normalized value; x represents the value of the second image data; x min represents the minimum value in the second image data; x max represents the maximum value in the second image data;
[0071] S3, extracting features from the target image data based on a bilateral attention mechanism to obtain a first target feature, and fusing the first target feature based on a convolutional neural network to obtain a second target feature;
[0072] Specifically, 1) extract information source P and information source Q based on the target image data and calculate the element p in the information source P. α and element q in the information source Q βThe attention weight between is calculated as follows:
[0073]
[0074] In formula (4), w αβ Represents element p in information source P α and element q in the information source Q β The attention weight between g(p α ,q β ) is a measure of p α and q β The function of the correlation between them; k and l represent the index variables used for iteration;
[0075] 2) Calculate the sum of the products of each element in the information source P and its corresponding attention weight, and update it to obtain the information source P', the calculation formula of which is as follows:
[0076]
[0077] In formula (5), P' represents the updated representation of the information source P after the bilateral attention mechanism. The update is to replace each element p in the information source P with α Multiply it by its corresponding attention weight w αβ And do this by summing all these products;
[0078] 3) Calculate the sum of the products of each element in the information source Q and its corresponding attention weight, and update it to obtain the information source Q', the calculation formula of which is as follows:
[0079]
[0080] In formula (6), Q' represents the updated representation of the information source Q after the bilateral attention mechanism; the update is to replace each element q in the information source Q with β Multiply it by its corresponding attention weight w αβ And do this by summing all these products;
[0081] 4) The updated information source P' and information source Q' are used as the first target feature extracted from the target image data;
[0082] 5) The first target feature extracted by the bilateral attention mechanism is used as input and sequentially fused through the two-dimensional convolution layer or Conv2d, batch normalization or BN, ReLU and maximum pooling layer or Max pooling in the convolutional neural network to obtain the second target feature;
[0083] S4. constructing a behavior recognition model based on the second target feature;
[0084] S5. Perform personnel behavior recognition based on the behavior recognition model to obtain the target personnel behavior.
[0085] The behavior recognition model constructed in the method of the present invention can simultaneously focus on the relationship between two information sources, namely, information source P and information source Q, so as to better extract and fuse useful information; after the feature extraction is completed, the convolutional neural network part is used to fuse the extracted features, that is, the feature fusion step integrates the features extracted by the bilateral attention mechanism, so that the model can effectively capture the complex patterns in the data.
[0086] Example 2
[0087] like Figure 2 As shown, a personnel behavior recognition system based on big data includes:
[0088] Data collection module, used to collect raw image data of personnel behavior;
[0089] A data processing module, used for preprocessing the original image data to obtain target image data;
[0090] A model building module, used to build a behavior recognition model based on the target image data;
[0091] The recognition module is used to recognize human behavior based on the constructed behavior recognition model.
[0092] The embodiments described above are merely descriptions of preferred implementation modes of the present invention and are not intended to limit the scope of the present invention. Without departing from the principles and essence of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the scope of protection determined by the claims of the present invention.
Claims
1. A method for identifying human behavior based on big data, characterized in that: The specific steps include: S1, collect original image data of personnel behavior; S2, preprocessing the original image data to obtain target image data; S3, extracting features from the target image data based on a bilateral attention mechanism to obtain a first target feature, and fusing the first target feature based on a convolutional neural network to obtain a second target feature; S4. constructing a behavior recognition model based on the second target feature; S5. Perform personnel behavior recognition based on the behavior recognition model to obtain the target personnel behavior.
2. According to the method for identifying human behavior based on big data in claim 1, it is characterized in that: The step S2 specifically includes: S21, performing data cleaning processing on the original image data, filling missing values in the original image data and removing abnormal values in the original image data to obtain first image data; S22, performing Gaussian filtering on the first image data to remove noise in the original image data to obtain second image data; S23: Normalize the second image data to obtain target image data.
3. According to the method for identifying human behavior based on big data in claim 2, it is characterized in that: The Gaussian filtering process specifically includes: 1) Calculating the weight corresponding to each pixel in the first image data based on a Gaussian function, the Gaussian function calculation formula is as follows: In formula (1), G(m,n) is the value of the Gaussian function at the point (m,n); σ is the standard deviation, which controls the width of the Gaussian distribution; the constant 2πσ 2 is a normalization factor that ensures that the integral of the Gaussian function on the entire plane is 1; the exponential term Describes the shape of a Gaussian distribution, where m and n are coordinate variables; 2) For each pixel point in the first image data, the pixel values in its neighborhood are convolved with the Gaussian kernel to obtain the weighted average of the pixel values in the neighborhood of each pixel point after filtering, where the convolution formula is as follows: In formula (2), I filtered (i, j) represents the pixel value of the pixel point (i, j) in the first image data after filtering; I(i+a, j+b) represents the pixel value of the pixel point (i+a, j+b) in the first image data; G(a, b) represents the element value of the Gaussian kernel at the position (a, b); c represents the size of the Gaussian kernel neighborhood, that is, the radius of the kernel; 3) Substituting the weighted average of the pixel values in the neighborhood of each pixel for the original value of each pixel to obtain a new value of each pixel to form the second image data.
4. According to the method for identifying human behavior based on big data in claim 2, it is characterized in that: The normalization process specifically includes: The minimum-maximum normalization method is used to map the second image data to the [0,1] interval to obtain the target image data. The calculation formula is as follows: In formula (3), x norm represents the normalized value; x represents the value of the second image data; x min represents the minimum value in the second image data; x max Indicates the maximum value in the second image data.
5. According to the method of personnel behavior recognition based on big data in claim 1, it is characterized in that: The step S3 specifically includes: 1) Extract information source P and information source Q based on the target image data and calculate element p in information source P α and element q in the information source Q β The attention weight between is calculated as follows: In formula (4), w αβ Represents element p in information source P α and element q in the information source Q β The attention weight between g(p α ,q β ) is a measure of p α and q β A function of the correlation between ; k and l represent index variables used for iteration; 2) Calculate the sum of the products of each element in the information source P and its corresponding attention weight, and update it to obtain the information source P', the calculation formula of which is as follows: In formula (5), P' represents the updated representation of the information source P after the bilateral attention mechanism. The update is to replace each element p in the information source P with α Multiply it by its corresponding attention weight w αβ And do this by summing all these products; 3) Calculate the sum of the products of each element in the information source Q and its corresponding attention weight, and update it to obtain the information source Q', the calculation formula of which is as follows: In formula (6), Q' represents the updated representation of the information source Q after the bilateral attention mechanism; the update is to replace each element q in the information source Q with β Multiply it by its corresponding attention weight w αβ And do this by summing all these products; 4) The updated information source P' and information source Q' are used as the first target feature extracted from the target image data; 5) The first target feature extracted by the bilateral attention mechanism is taken as input and sequentially fused through the two-dimensional convolution layer or Conv2d, batch normalization or BN, ReLU and maximum pooling layer or Max pooling in the convolutional neural network to obtain the second target feature.
6. A personnel behavior recognition system based on big data, the system is used to implement the method described in any one of claims 1 to 5, characterized in that: include: A data collection module (101), used for collecting raw image data of personnel behavior; A data processing module (102), used for preprocessing the original image data to obtain target image data; A model building module (103) is used to extract features from the target image data based on a bilateral attention mechanism to obtain a first target feature, fuse the first target feature based on a convolutional neural network to obtain a second target feature, and build a behavior recognition model based on the second target feature; The identification module (104) is used to identify the behavior of a person based on the behavior identification model to obtain the behavior of a target person.