Human body behavior recognition method and system based on DualBiNet and storage medium

Abstract

The invention discloses a DualBiNet-based human body behavior recognition method and system and a storage medium, relates to the technical field of deep learning, and aims to solve the problems that an existing human body behavior recognition method utilizing channel state information is easily influenced by noise and irrelevant features, the calculation complexity is high or the resource consumption is large, and information leakage often exists. Tiny changes in human behaviors are difficult to accurately capture. Comprising the following steps: S1, collecting a human body behavior CSI data set, and preprocessing CSI data; s2, a DualBiNet model is constructed, wherein the DualBiNet model comprises a BiTCAN module and a BiLSTM module, and the BiTCAN module and the BiLSTM module are arranged in a bidirectional mode; s3, inputting the human behavior CSI data into a DualBiNet model, performing feature extraction through a BiTCAN module, capturing a long-term dependency relationship of the extracted features through BiLSTM, weighting the output of the BiLSTM module by adopting an attention module, and splicing the output of the BiLSTM module and the output of the attention module by adopting residual connection; and finally, the features subjected to multi-layer processing are integrated through a full connection layer, and human behaviors are classified.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a human behavior recognition method, system and storage medium based on DualBiNet. Background Art

[0002] When WiFi signals come into contact with the human body during propagation, reflection and scattering will occur, causing changes in signal propagation. These changes can actually reflect the dynamic behavior of objects and humans in the environment. In recent years, using WiFi signals for environmental perception and human behavior recognition has gradually become a research hotspot that has attracted much attention.

[0003] In the field of WiFi signal-based perception technology, there are two commonly used parameters, namely, Received Signal Strength Indicator (RSSI) and Channel State Information (CSI). RSSI has been widely used in early research because it is easy to obtain and simple to process; and with the advancement of technology and the demand for high-precision perception, CSI, which contains richer information, has received more and more attention.

[0004] CSI is extremely sensitive to environmental changes and can capture subtle physiological signals such as breathing and heartbeat. However, this feature also makes CSI data susceptible to noise, resulting in outliers and irrelevant features.

[0005] Traditional convolutional neural networks use channel state information to perform human behavior recognition methods. However, they are affected by noise and irrelevant features, have high computational complexity or high resource consumption, and often have information leakage problems. They are not suitable for processing time series data, that is, the model may introduce future information in the convolution operation, resulting in the model being non-causal. Ultimately, it is difficult to accurately capture subtle changes in human behavior, affecting the accuracy and robustness of recognition. Summary of the invention

[0006] The technical problems to be solved by the present invention are:

[0007] Existing human behavior recognition methods that use channel state information are easily affected by noise and irrelevant features, have high computational complexity or large resource consumption, and often have information leakage problems, making it difficult to accurately capture subtle changes in human behavior.

[0008] The present invention adopts the following technical solutions to solve the above technical problems:

[0009] The present invention provides a human behavior recognition method based on DualBiNet, comprising the following steps:

[0010] S1, collect human behavior CSI data set and preprocess the CSI data;

[0011] S2. Construct a DualBiNet model, wherein the DualBiNet model includes a BiTCAN module and a BiLSTM module with a bidirectional setting, wherein the BiTCAN module adopts 1D-CBAM, wherein the 1D-CBAM adjusts two-dimensional convolution and two-dimensional pooling to one-dimensional convolution and one-dimensional pooling to obtain the 1D-CBAM, and introduces the 1D-CBAM into TCN to replace the standard pooling layer, and replaces the convolution operation of TCN with bidirectional convolution to simultaneously capture the forward and reverse information of the input data;

[0012] S3. Input the human behavior CSI data into the DualBiNet model, extract features through the BiTCAN module, capture long-term dependencies on the extracted features through BiLSTM, weight the output of the BiLSTM module using the attention module, and concatenate the output of the BiLSTM module with the output of the attention module using residual connection; finally, integrate the features processed by multiple layers through the fully connected layer to classify the human behavior.

[0013] Furthermore, the preprocessing of the CSI data described in S1 includes filtering the data using Hampel filtering and Butterworth filtering.

[0014] Furthermore, the attention module in S3 includes two linear layers, a tanh activation function and a softmax.

[0015] Furthermore, the DualBiNet model described in S3 adopts NLLLoss as the loss function.

[0016] Furthermore, in S3, an early stopping method is used in the process of training the DualBiNet model, and the performance indicators on the validation set are monitored in real time to determine when to stop the training.

[0017] The present invention provides a human behavior recognition system based on DualBiNet, which has a program module corresponding to the steps of the method described in any one of the above technical solutions, and executes the steps in the above human behavior recognition method based on DualBiNet during operation.

[0018] The present invention provides a computer-readable storage medium, which stores a computer program. The computer program is configured to implement the steps in the DualBiNet-based human behavior recognition method described in any one of the above technical solutions when called by a processor.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The present invention proposes a composite classification model based on a bidirectional temporal convolutional attention network BiTCAN and a long short-term memory network Bi-LSTM, and adopts a one-dimensional convolutional attention module 1D-CBAM, and combines the attention mechanism and the residual network to achieve a deep fusion of the attention mechanism and the convolutional neural network, significantly improving the performance of the model in processing one-dimensional data, and only adding a small amount of computing resource overhead, so that the model can more accurately capture subtle changes in human behavior, thereby greatly improving the accuracy of human behavior recognition. The model of the present invention is carried out on two public data sets, UT-HAR and NTU-HAR, and can achieve recognition accuracy rates of 99.03% and 98.47% respectively, indicating that the model of the present invention shows excellent recognition performance on different data sets, and provides new ideas and methods for research in related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flow chart of a human behavior recognition method based on DualBiNet in an embodiment of the present invention;

[0022] Figure 2 This is a data preprocessing effect diagram in an embodiment of the present invention;

[0023] Figure 3 Schematic diagram of the DualBiNet model structure in an embodiment of the present invention;

[0024] Figure 4 Schematic diagram of the internal structure of BiTCAN in an embodiment of the present invention;

[0025] Figure 5 Schematic diagram of the structure of a 1D-CBAM module in an embodiment of the present invention;

[0026] Figure 6 Schematic diagram of bidirectional convolution in an embodiment of the present invention;

[0027] Figure 7 Schematic diagram of the structure of the attention module in an embodiment of the present invention;

[0028] Figure 8 is a confusion matrix diagram in an embodiment of the present invention;

[0029] Fig. 9 This is a comparison chart of the accuracy of DualBiNet in an embodiment of the present invention and the existing CNN;

[0030] Fig.10 This is a comparison chart of recognition effects of different algorithms in an embodiment of the present invention;

[0031] Fig.11This is a comparison chart of the effects of the DualBiNet model in an embodiment of the present invention on two data sets. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the scheme of the present invention, exemplary implementations or embodiments of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described implementations or embodiments are only implementations or embodiments of a part of the present invention, not all of them. Based on the implementations or embodiments of the present invention, all other implementations or embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0033] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0034] The present invention provides a human behavior recognition method based on DualBiNet, such as Figure 1 As shown, the following steps are included:

[0035] S1, collect human behavior CSI data set and preprocess the CSI data;

[0036] S2, build Figure 3 The DualBiNet model shown in the figure includes a BiTCAN module and a BiLSTM module in a bidirectional setting, the BiTCAN module adopts 1D-CBAM, the 1D-CBAM adjusts the convolution, pooling, splicing and weight calculation operations originally applicable to two-dimensional data to obtain the 1D-CBAM, the 1D-CBAM is introduced into the TCN to replace the standard pooling layer, and the convolution operation of the TCN is replaced by a bidirectional convolution to simultaneously capture the forward and reverse information of the input data;

[0037] S3. Input the human behavior CSI data into the DualBiNet model, extract features through the BiTCAN module, capture long-term dependencies on the extracted features through BiLSTM, weight the output of the BiLSTM module using the attention module, and concatenate the output of the BiLSTM module with the output of the attention module using residual connection; finally, integrate the features processed by multiple layers through the fully connected layer to classify the human behavior.

[0038] In this embodiment, the BiTCAN module is a bidirectional temporal convolutional attention network (Bidirectional Temporal Convolutional Attention Network). First, the traditional convolutional block attention module (Convolutional Block Attention Module, CBAM) is improved and converted into a one-dimensional convolutional attention module (1D-Convolutional Block Attention Module, 1D-CBAM), so that it can be correctly applied to the processing of one-dimensional signals. The 1D-CBAM module is introduced into TCN (Temporal Convolutional Network, Time Convolutional Network), replacing the standard pooling layer, realizing the deep fusion of the attention mechanism and the convolutional neural network, significantly improving the performance of the model when processing one-dimensional data, and only increasing a small amount of computing resource overhead. The improved TCN model is called the Temporal Convolutional Attention Network (TCAN).

[0039] Based on the construction of the TCAN model, this embodiment further optimizes the convolution operation and replaces it with a bidirectional convolution to simultaneously capture the forward and reverse information of the input data, significantly enhancing the model's ability to perceive time series features, and obtaining the following: Figure 4 The bidirectional convolutional attention network BiTCAN shown in Figure 1 not only improves the efficiency of feature extraction, but also provides more powerful support for processing complex one-dimensional signals.

[0040] TCN, or Temporal Convolutional Network, is a deep learning model for processing sequence data. TCN mainly captures the temporal information in the time series through a series of one-dimensional convolutional layers, and introduces causal convolution, dilated convolution, and residual connection on the basis of one-dimensional CNN. It is often used in tasks such as time series prediction and sequence classification.

[0041] TCN uses causal convolution, which ensures that the convolution operation does not introduce future information by properly padding the input, thereby enhancing the model's ability to analyze time series data. The introduction of dilated convolution allows the model to capture longer dependencies in time series without increasing parameters, improving the ability to recognize complex sequence patterns. In addition, TCN also includes residual connections, which add the input before convolution and the output after convolution, allowing the model to effectively transfer information between different layers, alleviate gradient vanishing and retain important features, allowing the model to learn complex sequence features more stably.

[0042] 1D-CBAM adjusts the convolution, pooling, splicing and weight calculation operations originally applicable to two-dimensional data to adapt them to one-dimensional structures, thus achieving the following results: Figure 5 1D-CBAM shown in Figure 1. This module introduces channel attention and spatial attention mechanisms, which enables the model to adaptively focus on important parts of the input features, significantly improving the recognition ability of convolutional neural networks on key features and important positions.

[0043] This attention mechanism not only enhances the overall interpretability of the model, but also makes the features and positions it focuses on more transparent. By giving higher weights to important features, 1D-CBAM can effectively reduce the impact of noise and irrelevant features, and improve the robustness of the model in noisy and abnormal data environments. In addition, 1D-CBAM mainly relies on global pooling and simple convolution operations, so it has a low computational overhead and does not significantly increase the computational complexity or resource consumption of the model, providing a good balance for practical applications.

[0044] In traditional TCN, the main function of the pooling layer is to extract important features and reduce the feature size. This embodiment uses 1D-CBAM to replace the traditional pooling layer. By introducing the attention mechanism, it realizes the adaptive selection of important features, rather than relying solely on simple maximum pooling or average pooling, thereby significantly improving the feature extraction capability. At the same time, due to the small original size of the input data, 1D-CBAM not only maintains the stability of the feature size, but also reduces information loss, avoiding the feature map from being reduced too early in the extraction process to the extent that information cannot be effectively obtained. Therefore, the convolutional attention network (TCAN) after using 1D-CBAM to replace the pooling layer has shown significant advantages in human behavior recognition tasks compared to traditional TCN.

[0045] Since only forward or reverse information is processed, traditional convolution may face difficulties in understanding the global context, especially in the case of long sequences, where the dependency between the previous and next information is difficult to fully capture. At the same time, unidirectional convolution is highly sensitive to local noise or data loss, which may cause the model to perform poorly under unstable input conditions. The core idea of ​​bidirectional convolution is to flip the input data at the same time during the regular convolution process, and then perform a reverse convolution. This means that when processing the input sequence, the network not only considers its forward information, but also captures the reverse information. Finally, the results of the two convolutions are concatenated to form the output of the bidirectional convolution, such as Figure 6 shown.

[0046] The dual information processing mechanism of this embodiment enables the network to extract forward and reverse features at the same time, thereby achieving a more comprehensive and rich feature representation; by integrating information from different directions, the network can focus on key features more effectively and improve the generalization ability of the model. In addition, since the bidirectional convolution utilizes more information sources, the model has a stronger resistance to the influence of local noise or missing data, thereby enhancing the robustness of the model.

[0047] BiLSTM (Bidirectional Long Short-Term Memory) is an improved recurrent neural network (RNN) used to process sequence data, such as natural language processing and time series analysis. BiLSTM combines the advantages of traditional LSTM and enhances the model's contextual understanding ability through a bidirectional structure.

[0048] This embodiment combines BiTCAN with BiLSTM to effectively improve the performance of sequence data processing. BiTCAN can quickly identify key patterns and trends in input sequences, and is suitable for processing tasks with significant local properties such as time series and text data. This structure not only speeds up feature extraction, but also provides a more refined feature representation for subsequent sequence modeling. BiLSTM, through its unique bidirectional structure, effectively integrates context information, improves the model's ability to understand context, and thus enhances the accuracy of prediction of complex data patterns. This architecture gives full play to the advantages of both. First, BiTCAN is used to extract local features efficiently to generate simplified and expressive feature representations, and then the extracted features are analyzed in depth by BiLSTM to capture long-term dependencies. The combination of BiTCAN and BiLSTM realizes the efficient fusion of feature extraction and context modeling, which not only improves overall efficiency, but also reduces the amount of data required to be processed by BiLSTM, so that it can focus more on capturing complex time dependencies. This strategy not only improves the performance of the model, but also provides a new perspective for the application of deep learning in the field of sequence data analysis. The model of this embodiment shows a low computational complexity when processing data, proving its superior balance between speed and effect. This innovation not only provides a new solution for sequence data analysis, but also points out the direction for future related research and broadens the boundaries of deep learning applications.

[0049] Since the attention module involves weight calculation, the matrix processed by the module may lose some useful information. In order to solve this problem, this embodiment adopts the strategy of residual connection, and the output of the bidirectional long short-term memory network (BiLSTM) is sorted and spliced ​​with the output of the attention module. The residual connection introduces shortcut connections between layers, so that the gradient can be directly propagated in the network, effectively alleviating the problems of gradient disappearance and gradient explosion. It not only enhances the training stability of the model, but also provides a direct gradient propagation path, so that the input can bypass the attention module and pass it directly to the output. It significantly reduces the negative impact that the attention module may bring, further improves the expression ability and overall performance of the model, and enables the network to better capture and utilize information, thereby achieving more outstanding performance in behavior recognition tasks.

[0050] Preferably, the preprocessing of the CSI data in S1 includes filtering the data using Hampel filtering and Butterworth filtering.

[0051] This embodiment uses Hampel filtering to detect and remove outliers in data. Hampel filtering compares each data point in a signal window with the median of the window. If the difference between a data point and the median exceeds a set threshold, the point is considered an outlier and is replaced by the median. Figure 2 The figure shows the waveform changes before and after the Hampel filter is applied. The outliers in the signal are effectively removed, and there are no large jumps in the data. It shows a smoother feature, which significantly improves the reliability and availability of the data.

[0052] Usually, the frequency of human activities is in the low-frequency range, while the noise and other components contained in the acquired CSI data are in the high-frequency range. By applying the Butterworth low-pass filter, the high-frequency components can be effectively suppressed, thereby retaining the low-frequency components and significantly improving the quality and reliability of the data. The filtering process of Butterworth filtering is:

[0053]

[0054] Among them, is the order of the filter, c is the cut-off frequency.

[0055] like Figure 2 As shown in the figure, after Butterworth low-pass filtering, it can be clearly seen that the noise in the signal has been effectively removed. The two-step filtering process improves the smoothness and stability of the signal, laying a solid foundation for subsequent analysis.

[0056] Preferably, the attention module comprises two linear layers, a tanh activation function and a softmax; its structure is as follows Figure 7 As shown. For the output of BiLSTM The output h at each time step t All of them will go through a linear transformation and Tanh activation function to get the query vector t :

[0057] Q t =tanh(W q h t )#(19)

[0058] in, is the weight matrix of the query layer.

[0059] Next, another linear layer maps the query to a score t :

[0060] S t =W s Q t #(20)

[0061] in, is the weight matrix of the scoring layer.

[0062] Then, the scores for all time steps are t Apply the softmax function to obtain the attention weights t :

[0063]

[0064] Finally, the context vector C is calculated by multiplying the attention weights with the output of the BiLSTM:

[0065]

[0066] The attention mechanism provides a rich information basis for subsequent processing, which not only optimizes the feature extraction process, but also greatly improves the adaptability and robustness of the model in dynamic scenarios, and gives stronger analytical capabilities for complex behavior recognition tasks.

[0067] Preferably, the DualBiNet model adopts NLLLoss as the loss function.

[0068] In the final stage of the model, this embodiment integrates the features processed by multiple layers through the fully connected layer and inputs them into the LogSoftmax layer for normalization. The LogSoftmax layer is usually used in combination with the negative log-likelihood loss function (NLLLLoss) to have higher stability. In addition, since the gradient of the logarithmic function is relatively smooth, the combination of LogSoftmax and NLLLoss helps the effective propagation of the gradient in the network, thereby significantly improving the training efficiency, which not only optimizes the learning process of the model, but also promotes the close connection between feature extraction and classification results, providing a more accurate and reliable output for the final behavior recognition.

[0069] Preferably, in S3, an early stopping method is used in the process of training the DualBiNet model, and when to stop training is determined by real-time monitoring of performance indicators on the validation set.

[0070] The human behavior recognition method (algorithm) based on DualBiNet proposed in the present invention is the underlying technical core of the present invention, and various products can be derived based on the algorithm.

[0071] Based on the method proposed in the present invention, a human behavior recognition system based on DualBiNet is developed using a programming language. The system has program modules corresponding to the steps of the above technical solution, and executes the steps in the above human behavior recognition method based on DualBiNet during operation.

[0072] The computer program of the developed system (software) is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the above-mentioned human behavior recognition method based on DualBiNet when called by a processor, that is, the present invention is materialized on a carrier to become a computer program product.

[0073] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0074] The computer programs (also referred to as programs, software, software applications, or codes) of the present invention include machine instructions for programmable processors, and these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or device (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0075] Example 1

[0076] This embodiment first uses the public dataset UT-HAR for systematic testing, which includes seven actions: lying down, falling, picking up, running, sitting down, standing up and walking. The CSI size is 1x 250x 90, and the dataset is divided into training set, validation set and test set in a ratio of 8:1:1. During the training process of the DualBiNet model, an early stopping mechanism is introduced to determine when to stop training by real-time monitoring of performance indicators on the validation set. This can effectively avoid unnecessary training processes, not only save computing resources, but also significantly improve the generalization ability of the model and prevent overfitting. Some parameter settings are shown in Table 1.

[0077] Table 1

[0078]

[0079] After adopting the early stopping method, the model automatically stopped training at about 100 epochs, and the classification accuracy reached 99.03%. Figure 8 The following is the confusion matrix obtained after testing, which clearly shows the performance of the DualBiNet model in various behavior recognition tasks. This result not only shows that the model has excellent classification ability, but also further verifies the effectiveness of the proposed method, providing strong support for behavior recognition in practical applications.

[0080] It can be seen that the DualBiNet model has a strong recognition ability for seven categories of actions, among which the recognition accuracy of falling, walking, picking up, running, and standing up has reached 100%. This fully demonstrates that the model has shown excellent performance on the UT-HAR dataset.

[0081] To further verify the improvement of BiTCAN over conventional CNN, this example replaces the BiTCAN module in the model with the traditional CNN and conducts a comparative experiment. Fig. 9 As shown, the performance difference between the two in the action recognition task is demonstrated.

[0082] Conventional CNN no longer uses the CBAM module, retains the pooling layer for further feature extraction, and reduces the size of the convolution kernel to avoid losing effective information. The results show that the performance of the model has been significantly improved after using BiTCAN.

[0083] In order to more intuitively demonstrate the superiority of the model, this embodiment also compares it with several commonly used classification algorithms including Multi-Layer Perceptron (MLP), Recurrent Neural Network (RNN), Bidirectional Long Short-Term Memory Network (BiLSTM) and Convolutional Neural Network (CNN). The results are as follows: Fig.10 shown.

[0084] In addition, this embodiment adjusts the input and output size and the number of channels of the convolutional layer of the model parameters to adapt to the CSI information of the NTU-HAR dataset, and trains and tests on the dataset again. The test results are compared with those of UT-HAR. Fig.11 As shown. It can be seen that the DualBiNet model shows an accuracy of more than 98% on both datasets. It is worth noting that on the NTU-HAR dataset, the relatively simple BiLSTM can also achieve similar results. This is mainly due to the fact that the dataset is relatively easy to identify. It contains six actions: running, walking, falling, boxing, hugging arms, and cleaning the floor. There is a more significant distinction between these behaviors, so the shape of the CSI signal also shows a more obvious difference. The NTU-HAR dataset is equipped with a different system during the acquisition process, and each antenna has 114 subcarriers, which provides more representative feature information. In addition, the scale of this dataset is much larger than UT-HAR, which significantly increases the amount of data available for training.

[0085] Driven by many factors, for the NTU-HAR dataset, using BiLSTM alone is sufficient to achieve highly accurate recognition. Although DualBiNet can mine deeper features, its recognition ability does not show a particularly obvious advantage.

[0086] Nevertheless, considering the actual application scenarios, it is impossible to ensure that the collected data always has the same high quality and high discrimination as NTU-HAR. Therefore, when facing relatively difficult recognition tasks, the DualBiNet model shows significant advantages. In addition, the model can achieve excellent recognition results on two datasets with obvious differences, proving its excellent generalization ability and strong robustness.

[0087] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A human behavior recognition method based on DualBiNet, characterized in that: The steps include: S1, collect human behavior CSI data set and preprocess the CSI data; S2. Construct a DualBiNet model, wherein the DualBiNet model includes a BiTCAN module and a BiLSTM module with a bidirectional setting, wherein the BiTCAN module adopts 1D-CBAM, wherein the 1D-CBAM adjusts two-dimensional convolution and two-dimensional pooling to one-dimensional convolution and one-dimensional pooling to obtain the 1D-CBAM, and introduces the 1D-CBAM into TCN to replace the standard pooling layer, and replaces the convolution operation of TCN with bidirectional convolution to simultaneously capture the forward and reverse information of the input data; S3, inputting human behavior CSI data into the DualBiNet model, extracting features through the BiTCAN module, capturing long-term dependencies of the extracted features through the BiLSTM, using the attention module to weight the output of the BiLSTM module, and using the residual connection to splice the output of the BiLSTM module with the output of the attention module; Finally, the features processed by multiple layers are integrated through the fully connected layer to classify human behavior.

2. The human behavior recognition method based on DualBiNet according to claim 1, characterized in that: The CSI data is preprocessed as described in S1, including filtering the data using Hampel filtering and Butterworth filtering.

3. The human behavior recognition method based on DualBiNet according to claim 2, characterized in that: The attention module described in S3 contains two linear layers, a tanh activation function and a softmax.

4. The human behavior recognition method based on DualBiNet according to claim 3 is characterized in that: The DualBiNet model described in S3 uses NLLLoss as the loss function.

5. The human behavior recognition method based on DualBiNet according to claim 4, characterized in that: In S3, the early stopping method is used in the process of training the DualBiNet model, and the performance indicators on the validation set are monitored in real time to determine when to stop training.

6. A human behavior recognition system based on DualBiNet, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 5, and executes the steps of the human behavior recognition method based on DualBiNet when running.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps in the DualBiNet-based human behavior recognition method according to any one of claims 1 to 5 when called by a processor.

Images