Deep learning-based multi-mode security and protection equipment early warning method and system

Through deep learning technology, combining long-term memory neural networks and generation adversarial networks, multimodal data is processed to generate warning instructions, which solves the shortcomings of traditional security equipment in obtaining and processing multimodal data and improves safety warnings at work sites Ability and intelligence level.

CN119091384APending Publication Date: 2024-12-06SHANDONG XIANXING ZHIYE INFORMATION TECHNOLOGY CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411341561.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-25
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Traditional security equipment is difficult to obtain and process multimodal data, resulting in the inability to detect potential safety hazards in a timely manner in complex and dynamic work sites, and the reliability and stability in harsh environments are insufficient, which cannot meet the security needs of modern industrial.

Method used

Using a multimodal security device early warning method based on deep learning, video, audio and environmental data are obtained through high-definition cameras, preprocessing, and data prediction and judgment are used to predict and judge data by using long and short-term memory neural network model and generation adversarial network to generate warning instructions to improve Warning accuracy and reliability.

Benefits of technology

It realizes comprehensive acquisition and efficient processing of multimodal data on the work site, improves the timeliness and accuracy of early warnings, reduces safety risks, adapts to dynamic changes in different environments and conditions, and enhances the intelligence level of security equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091384A_ABST
    Figure CN119091384A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of deep learning, in particular to a multi-mode security and protection equipment early warning police method and system based on deep learning. The method comprises the following steps: acquiring field work multi-modal data; carrying out preprocessing on the obtained field work multi-modal data; a neural network model based on long and short term memory is constructed, and the neural network model based on long and short term memory is utilized to perform result prediction on the preprocessed field work multi-modal data; inputting the prediction result into a generative adversarial network, and obtaining a discrimination result based on the generative adversarial network; and a warning instruction is generated after the judgment result is uploaded. According to the invention, the field video is collected, the image quality is improved by using the image enhancement technology, and the field construction can be effectively monitored in real time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a multimodal security equipment early warning method and system based on deep learning. Background Art

[0002] In today's society, industrial production and various work sites continue to expand in scale and complexity. Whether it is a factory workshop, construction site or other workplace, safety is always a critical issue.

[0003] Most traditional security equipment uses a single monitoring method. For example, a simple video surveillance system can only provide visual images, but cannot obtain multiple information such as audio and environmental data. This makes it easy to miss key safety hazards in actual applications. For example, in some noisy working environments, video surveillance alone may not be able to detect subtle sound changes caused by abnormal operation of equipment; and in some cases, scenes that appear normal visually may pose huge safety risks due to changes in environmental factors. Without comprehensive analysis of multimodal data, these hidden dangers are difficult to detect in a timely manner.

[0004] Moreover, traditional security equipment is relatively weak in data processing capabilities. Faced with a large amount of multimodal data from different channels, traditional methods are often difficult to efficiently integrate and analyze. It is impossible to deeply understand and accurately judge complex on-site situations, which greatly reduces the timeliness and accuracy of early warnings.

[0005] Furthermore, with the continuous advancement of technology, the dynamics and uncertainty of the work site are also increasing. The existing security warning systems are difficult to meet actual needs in terms of intelligence. They usually cannot make adaptive adjustments based on real-time changing site conditions, nor can they accurately predict potential security risks. For example, it is impossible to dynamically adjust the warning strategy based on factors such as work characteristics, personnel activities, and equipment operating status in different time periods.

[0006] In addition, in some special working scenarios, such as high temperature, high pressure, high dust and other harsh environments, the reliability and stability of traditional security equipment also face severe challenges. These devices may be affected by environmental factors and fail or misreport, causing great trouble to security management.

[0007] In summary, in order to better ensure the safety of the work site, there is an urgent need for a multimodal security equipment early warning method and system based on deep learning to achieve comprehensive acquisition, efficient processing and accurate early warning of multimodal data of on-site work. Summary of the invention In order to solve the above-mentioned problems, the present invention provides a multimodal security equipment early warning method and system based on deep learning.

[0008] In the first aspect, the present invention provides a multi-modal security equipment early warning method based on deep learning, which adopts the following technical solution: A multimodal security equipment early warning method based on deep learning, comprising: Acquire multimodal data for field work; Preprocess the acquired multimodal data of field work; Construct a neural network model based on long short-term memory, and use it to predict the results of preprocessed multimodal data of field work; The prediction results are input into the generative adversarial network, and the discrimination results are obtained based on the generative adversarial network; After uploading the judgment result, a warning instruction is generated.

[0009] Furthermore, the obtaining of multimodal data of on-site work includes obtaining audio, video and environmental data of the work site through a high-definition camera, wherein key frames are periodically extracted from the video stream as image data.

[0010] Furthermore, the acquired multimodal data of field work is preprocessed, including cropping, scaling and normalizing the collected videos and images, wherein image enhancement technology is used to improve the quality of image data; and noise reduction and filtering technology are used to process audio data, which is converted into a format suitable for LSTM network input after preprocessing.

[0011] Furthermore, the construction of the neural network model based on long short-term memory includes constructing an input layer, an LSTM layer and an output layer of the neural network, wherein the number of LSTM layers is determined to be 13 and the number of units in each LSTM layer is set to 50, and a fully connected layer based on a sigmoid activation function is set in the output layer to output probability values.

[0012] Furthermore, the method uses a long short-term memory-based neural network model to predict the results of preprocessed field work multimodal data, including inputting the preprocessed multimodal data into the LSTM network, calculating the output result through forward propagation, and calculating the error between the predicted result and the true label according to the loss function, and updating the weight and bias parameters of the LSTM network by back-propagating the error through an optimization algorithm until the loss function converges.

[0013] Furthermore, the prediction results are input into a generative adversarial network, and a discrimination result is obtained based on the generative adversarial network, including constructing a generator and a discriminator, inputting the prediction results output by the neural network model through the generator, and generating corresponding samples, and judging the possibility that the input samples are real samples through the discriminator to obtain the output results.

[0014] Furthermore, generating a warning instruction after uploading the judgment result includes receiving the judgment result of the generative adversarial network, generating a warning instruction, and storing and managing the data, wherein the warning instruction is output to workers and managers in the form of sound alarms, flashing lights, and display screen prompts.

[0015] In the second aspect, a multimodal security equipment early warning system based on deep learning includes: The data acquisition module is configured to acquire multimodal data of field work; A preprocessing module is configured to preprocess the acquired field work multimodal data; The model prediction module is configured to construct a neural network model based on long short-term memory, and use the neural network model based on long short-term memory to predict the results of the pre-processed field work multimodal data; The discrimination module is configured to input the prediction result into a generative adversarial network and obtain a discrimination result based on the generative adversarial network; The warning module is configured to generate a warning instruction after uploading the judgment result.

[0016] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device for the multimodal security equipment early warning method based on deep learning.

[0017] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the multimodal security equipment early warning method based on deep learning.

[0018] In summary, the present invention has the following beneficial technical effects: The present invention can effectively monitor on-site construction in real time by collecting on-site videos and using image enhancement technology to improve image quality; the neural network model based on long short-term memory can effectively process time series data and fully extract features to improve prediction accuracy. At the same time, the network model is used to effectively judge the standardization of workers' actions and provide a more reliable basis for warning instructions. Warning instructions are output in various forms such as sound alarms, flashing lights, and display screen prompts to ensure that workers and managers receive warning information in a timely manner. At the same time, data is stored and managed to provide data support and analysis basis for safety management.

[0019] The solution uses a variety of technologies to achieve real-time monitoring and early warning of on-site work, effectively improve the early warning capabilities of security equipment, reduce safety risks, and improve safety management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 It is a schematic diagram of a multimodal security equipment early warning method based on deep learning according to Example 1 of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0021] Example 1 Reference Figure 1 , a multi-modal security equipment early warning based on deep learning in this embodiment S1. Acquire multimodal data of on-site work, including obtaining audio, video and environmental data of the work site through high-definition cameras, wherein key frames are regularly extracted from the video stream as image data.

[0022] Among them, video acquisition: use high-definition cameras installed at key locations on the work site to cover the workers' work area. The camera should have good image quality and a wide-angle field of view to clearly capture the workers' movements and working environment. The video acquisition frame rate should be set according to actual needs to ensure that the worker's movements can be accurately recorded.

[0023] Image acquisition: Key frames are periodically extracted from the video stream as image data, or supplemented with a separate image acquisition device. The image should have sufficient resolution to clearly show the details of the worker's operation.

[0024] Sound collection: Install microphones at the work site to collect sound signals from workers at work. The microphones should have good sensitivity and noise immunity, and be able to accurately record the workers' operating sounds and other sounds in the work environment.

[0025] Work environment data collection: Sensors can be used to collect data such as temperature, humidity, and light intensity in the work environment to help determine whether workers' operations are affected by environmental factors.

[0026] S2. Preprocess the acquired multimodal data from field work, including cropping, scaling and normalizing the collected videos and images. Image enhancement technology is used to improve the quality of image data. Noise reduction and filtering technology is used to process audio data.

[0027] Among them, image enhancement technology is used to improve image quality Contrast enhancement adjusts the brightness difference between different areas in the image, making dark areas darker and bright areas brighter, thereby enhancing the contrast of the image. Linear contrast stretching expands the pixel value range of an image through a linear transformation function. The pixel value range in the original image is stretched from [a, b] to [c, d], expressed as: , where the pixel value of a pixel in the original image is x, and the pixel value of the pixel after linear contrast stretching is y.

[0028] Histogram equalization adjusts the image's histogram to make the image's pixel values ​​more evenly distributed, thereby improving contrast. The specific approach is to calculate the image's histogram and then remap the pixel values ​​based on the histogram.

[0029] Laplace sharpening is used to perform convolution operations on the image using the Laplace operator to highlight the edge information in the image. The Laplace operator is a second-order derivative operator that can detect grayscale mutations in an image.

[0030] Audio noise reduction and filtering Using spectral subtraction, assume that the noise is additive, that is, the sum of the clean audio signal and the noise signal is the noisy audio signal. By estimating the power spectrum of the noise, subtract the noise power spectrum from the power spectrum of the noisy audio signal to obtain the power spectrum estimate of the clean audio signal. Then, the denoised audio signal is obtained by inverse Fourier transform.

[0031] The audio signal is frequency-selected through the filter to remove unnecessary frequency components and retain useful frequency components.

[0032] S3. Construct a neural network model based on long short-term memory. Among them, determine the network structure Input layer: Split into consecutive image frames in time series, each image frame is represented as a vector. Sound data is converted into a spectrogram or other feature representation and also input as a vector. The dimension of the input vector is determined by the preprocessing method of the image and sound data and the specific feature extraction method. LSTM layers: Determine the number of LSTM layers. Generally, you can start with one layer and gradually increase the number of layers based on performance. Set the number of units in each LSTM layer. The number of units determines the memory and expression capabilities of the LSTM network. You can determine the appropriate number of units through experiments, usually between dozens and hundreds.

[0033] Output layer: The structure of the output layer is determined according to the task requirements. If it is a binary classification problem to determine whether the worker's operation is standard, a fully connected layer with a sigmoid activation function can be used to output a probability value between 0 and 1, indicating the probability of non-standard operation.

[0034] (3) Tuning based on stochastic gradient descent (SGD) or Adam. The Adam algorithm usually performs well in most cases, with adaptive learning rate and good convergence.

[0035] Set the parameters of the optimization algorithm, including learning rate, momentum, etc. The learning rate can start from a small value, such as 0.001, and then adjust it according to the training situation.

[0036] S4. Input the training data into the LSTM network and calculate the output result through forward propagation; calculate the error between the predicted result and the true label according to the loss function; use the optimization algorithm to backpropagate the error and update the weight and bias parameters of the network; repeat the above steps until the loss function converges or the preset number of training rounds is reached.

[0037] in, 1. Forward propagation, the input training data is time series data, which can be a series of image frames for video data and a processed spectral feature sequence for sound data. The input sequence length is T, and the input feature vector dimension of each time step is n. The entire input can be expressed as The matrix , where x t Represents the input vector at the tth time step.

[0038] 2. The LSTM unit mainly consists of three gates (forget gate, input gate, output gate) and a cell state.

[0039] Forget gate ft: decides which information to discard from the cell state. It does this by t and the hidden state ht at the previous moment -1 As input, a value between 0 and 1 is obtained through the sigmoid activation function, which indicates the degree of forgetting the cell state at the previous moment. The formula is: , where W f is the weight matrix of the forget gate, b f is the bias term, and σ is the sigmoid function. Input Gate : Determines what new information will be stored in the cell state. t and the hidden state at the previous moment As input, two values ​​are obtained after a sigmoid activation function and a tanh activation function. The formula is: , where W i and W Care the weight matrices of the input gate and the new cell state, respectively, and b i and b C is the bias term.

[0040] Cell status update: ,in Represents element-by-element multiplication. Cell state C t is the cell state C at the previous moment t - 1 and fused with new information.

[0041] Output gate o t :Determines what information will be output. t and the hidden state h at the previous moment t-1 As input, a value is obtained through the sigmoid activation function. The formula is: .

[0042] Hide status updates: , hidden state h t It is calculated based on the output gate and the cell state, and it will serve as the input of the next time step and the output of the network.

[0043] Forward propagation process For the first time step t = 1, initialize the cell state C0 and hidden state h0 (usually initialized to all zero vectors).

[0044] For each time step t, calculate the forget gate ft, the input gate , new cell state , cell status , output gate ot and hidden state ht.

[0045] The hidden state h at the last time step T As the output of the LSTM network.

[0046] Loss function selection Cross entropy loss function, assuming the true label is The prediction result is , then the cross entropy loss function is: ,in Represents the category index.

[0047] Comparison of predicted results with true labels The output of the LSTM network The prediction result is obtained through a fully connected layer and an activation function (such as sigmoid or softmax) .

[0048] Calculate the value of the loss function based on the true label and the predicted result.

[0049] Use an optimization algorithm (such as Adam) to perform backpropagation to update the weights and bias parameters of the network.

[0050] Calculate the gradient of the loss function for each parameter, and then update the parameter according to the direction of gradient descent.

[0051] First calculate the gradient: for the parameters , whose gradient is , where $L$ is the loss function.

[0052] Calculate the first-order moment estimate (momentum): ,in is the momentum parameter, .

[0053] Compute the second moment estimate (RMSProp): ,in is the RMSProp parameter, .

[0054] Perform bias correction on the first-order and second-order moment estimates: .

[0055] Update parameters: ,in is the learning rate, is a small constant used to prevent the denominator from being zero.

[0056] Repeat the training process, and keep repeating the steps of forward propagation, error calculation, and back propagation until the loss function converges or the preset number of training rounds is reached. During the training process, the performance of the model can be regularly evaluated on the validation set to adjust the hyperparameters in time and prevent overfitting.

[0057] S5. Input the prediction results into the generative adversarial network, and obtain the discrimination results based on the generative adversarial network, including constructing a generator and a discriminator, inputting the prediction results output by the neural network model through the generator, and generating corresponding samples, and judging the possibility that the input samples are real samples through the discriminator to obtain the output results.

[0058] Among them, GAN consists of a generator and a discriminator. The goal of the generator is to generate realistic samples to deceive the discriminator, while the task of the discriminator is to distinguish between real samples and samples generated by the generator.

[0059] Generator Design The generator consists of multiple layers of fully connected layers or convolutional neural network layers. The input is the prediction result of LSTM, and the output is the generated sample, which should be similar to the real sample in form so as to deceive the discriminator. The loss function of the generator is usually designed based on the feedback of the discriminator. Common loss functions include least squares generative adversarial network (LSGAN) loss, Wasserstein generative adversarial network (WGAN) loss, etc. The goal is to make the samples generated by the generator as close to the real samples as possible, so that the discriminator is difficult to distinguish.

[0060] Discriminator design The discriminator has a different structure from the generator. The input is a real sample or a sample generated by the generator, and the output is a probability value, which indicates the possibility that the input sample is a real sample. For the discriminator, the loss function is designed based on its ability to distinguish between real samples and generated samples. In the standard GAN, the loss function of the discriminator is to maximize the probability that the real sample is judged as real, while minimizing the probability that the generated sample is judged as real.

[0061] Use the mini-batch stochastic gradient descent algorithm to update the parameters of the generator and discriminator. For each mini-batch of data, calculate the loss function of the generator and discriminator, and update the parameters according to the gradient of the loss function. Adjust parameters such as learning rate, batch size, network structure, etc. to obtain better training results.

[0062] After the training is completed, the prediction results of LSTM are input into the generator to generate corresponding samples. Then the generated samples and real samples are input into the discriminator together, and the discriminator outputs a probability value, indicating the possibility that the input sample is a real sample. According to the output of the discriminator, it can be judged whether the prediction results of LSTM are reasonable or standard.

[0063] S6. Collect the sound, video, image and working environment data of on-site work through security equipment.

[0064] Background processing layer: receives the judgment results of GAN, generates warning instructions, and stores and manages data. Warning output layer: outputs warning instructions to workers and managers in the form of sound alarms, flashing lights, display screen prompts, etc.

[0065] When the GAN discrimination result is input to the background, the result is first parsed. The discrimination result is usually a probability value, indicating the possibility that the input sample is a real sample (standard operation).

[0066] According to specific application requirements, a threshold can be set to determine whether the operation is standardized. If the probability value is greater than 0.5, the operation is considered standardized; otherwise, the operation is considered unstandardized.

[0067] The discrimination results and related input data (such as video, image, sound, etc.) are stored in the database for subsequent analysis and query. The data is marked with timestamps to track the time sequence of operations.

[0068] According to the judgment results and preset rules, corresponding warning instructions are generated. The rules can be customized according to actual conditions.

[0069] When the operation is not in compliance with regulations, an audible alarm will sound immediately and a specific description of the irregular behavior will be displayed on the display.

[0070] If irregular operations occur repeatedly, a notification will be sent to the management staff so that further measures can be taken.

[0071] Different levels of warnings are set according to the severity of irregular operations, such as yellow warnings for minor irregularities and red warnings for serious irregularities.

[0072] The warning instructions adopt a standardized format to facilitate system processing and output. The warning instructions are represented in JSON format, including the warning type, level, description and other information.

[0073] Among them, a loudspeaker is used to send out sound alarms of different frequencies and volumes to attract the attention of workers. Different sound modes can be set according to the warning level, such as a rapid beep to indicate serious irregularities.

[0074] Install indicator lights at the work site to indicate the warning level by flashing lights of different colors. For example, flashing red lights indicate serious irregularities, and flashing yellow lights indicate minor irregularities.

[0075] Set up a display screen at the work site to display warning information and descriptions of irregular behaviors. The display screen can be updated in real time so that workers can understand their own operation status in time.

[0076] If it is necessary to notify the management personnel, the notification can be sent via email, SMS or special management software. The notification content should include the time, location, description and other information of the irregular operation so that the management personnel can take appropriate measures.

[0077] Establish a feedback mechanism to allow workers and managers to provide feedback on warning instructions. For example, workers can confirm receipt of warnings and indicate whether corrective actions have been taken. Managers can evaluate the accuracy and effectiveness of warnings and make suggestions for improvement.

[0078] Based on feedback information and new data, the deep learning model and warning rules are continuously optimized. The model can be retrained regularly to improve its accuracy and generalization ability. At the same time, the warning rules are adjusted according to the actual situation to make them more reasonable and effective.

[0079] Example 2 This embodiment provides a multi-modal security equipment early warning system based on deep learning, including: The data acquisition module is configured to acquire multimodal data of field work; A preprocessing module is configured to preprocess the acquired field work multimodal data; The model prediction module is configured to construct a neural network model based on long short-term memory, and use the neural network model based on long short-term memory to predict the results of the pre-processed field work multimodal data; The discrimination module is configured to input the prediction result into a generative adversarial network and obtain a discrimination result based on the generative adversarial network; The warning module is configured to generate a warning instruction after uploading the judgment result.

[0080] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for a multimodal security equipment early warning method based on deep learning.

[0081] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor to implement a multimodal security equipment early warning method based on deep learning.

[0082] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-modal security equipment early warning method based on deep learning, characterized in that: include: Acquire multimodal data for field work; Preprocess the acquired multimodal data of field work; Construct a neural network model based on long short-term memory, and use it to predict the results of preprocessed multimodal data of field work; The prediction results are input into the generative adversarial network, and the discrimination results are obtained based on the generative adversarial network; After uploading the judgment result, a warning instruction is generated.

2. According to the multi-modal security equipment early warning method based on deep learning in claim 1, it is characterized in that: The obtaining of multimodal data of on-site work includes obtaining audio, video and environmental data of the work site through a high-definition camera, wherein key frames are periodically extracted from the video stream as image data.

3. The multi-modal security equipment early warning method based on deep learning according to claim 2 is characterized in that: The preprocessing of the acquired multimodal data of field work includes cropping, scaling and normalizing the collected videos and images, wherein image enhancement technology is used to improve the quality of image data; and noise reduction and filtering technology are used to process audio data, which is converted into a format suitable for LSTM network input after preprocessing.

4. The multi-modal security equipment early warning method based on deep learning according to claim 3 is characterized in that: The method for constructing a neural network model based on long short-term memory includes constructing an input layer, an LSTM layer and an output layer of the neural network, wherein the number of LSTM layers is determined to be 13 and the number of units in each LSTM layer is set to 50, and a fully connected layer based on a sigmoid activation function is set in the output layer to output probability values.

5. The multi-modal security equipment early warning method based on deep learning according to claim 4 is characterized in that: The method uses a long short-term memory-based neural network model to predict the results of preprocessed field work multimodal data, including inputting the preprocessed multimodal data into the LSTM network, calculating the output result through forward propagation, and calculating the error between the predicted result and the true label according to the loss function, and updating the weight and bias parameters of the LSTM network through back-propagation of the error through an optimization algorithm until the loss function converges.

6. The multi-modal security equipment early warning method based on deep learning according to claim 5 is characterized in that: The prediction results are input into a generative adversarial network, and a discrimination result is obtained based on the generative adversarial network, including constructing a generator and a discriminator, inputting the prediction results output by the neural network model through the generator, and generating corresponding samples, and judging the possibility that the input samples are real samples through the discriminator to obtain the output results.

7. The multi-modal security equipment early warning method based on deep learning according to claim 6 is characterized in that: The method of generating a warning instruction after uploading the discrimination result includes receiving the discrimination result of the generative adversarial network, generating a warning instruction, and storing and managing the data, wherein the warning instruction is output to workers and managers in the form of sound alarms, flashing lights, and display screen prompts.

8. A multi-modal security equipment early warning system based on deep learning, characterized in that: include: The data acquisition module is configured to acquire multimodal data of field work; A preprocessing module is configured to preprocess the acquired field work multimodal data; The model prediction module is configured to construct a neural network model based on long short-term memory, and use the neural network model based on long short-term memory to predict the results of the pre-processed field work multimodal data; The discrimination module is configured to input the prediction result into a generative adversarial network and obtain a discrimination result based on the generative adversarial network; The warning module is configured to generate a warning instruction after uploading the judgment result.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the method according to claim 1 .

10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method as claimed in claim 1 .

Citation Information

Patent Citations

  • Bridge static displacement prediction technology based on deep learning LSTM network

    CN112784336A

  • Multi-mode lane changing trajectory prediction and safety evaluation method in strong interaction scene

    CN118538022A

  • Data Augmentation Method Based On Generative Adversarial Networks In Tool Condition Monitoring

    US20210197335A1