Passenger head injury prediction method and device based on sparse weight neural network, and storage medium

Abstract

The invention provides a passenger head injury prediction method and device based on a sparse weight neural network and a storage medium, and relates to the technical field of deep learning. Performing sparsification on the head injury neural network prediction model according to the importance to obtain a sparsification model; the sparse model is trained again, and the trained sparse model is adopted to predict the acceleration of the head of the passenger; predicting the head injury of the passenger according to the head acceleration of the passenger; the complexity of the model in time and space is reduced, the prediction speed is improved, and the generalization ability of the model can be improved to a certain extent.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a method, device and storage medium for predicting occupant head injury based on a sparse weighted neural network. Background Art

[0002] The probability of occupants suffering head injuries in traffic accidents is relatively high, and the fatality and injury rate of occupant head injuries is relatively high, so occupant head injuries deserve attention. In the field of passive safety, the protection performance of vehicles for occupants is a concern of various car companies and consumers. As an assessment item, the head damage value is also one of the main evaluation indicators reflecting the passive safety performance of the vehicle. In the prior art, the head injury value of automobile collision is mainly obtained through real vehicle collision tests or simulation tests. The main problems are: real vehicle collision tests will cost a lot of money and time, and the accuracy of simulation test results is generally strongly related to the benchmarking accuracy of the simulation model.

[0003] Based on the above problems, the present invention is proposed. Summary of the invention

[0004] In response to the above problems, the present invention proposes a method, device and storage medium for predicting occupant head injuries based on a sparse weight neural network. The weight matrix is ​​sparse to reduce the complexity of the model in time and space, improve the prediction speed, and at the same time, the generalization ability of the model can be improved to a certain extent.

[0005] In a first aspect, the present invention provides a method for predicting occupant head injury based on a sparse weighted neural network, comprising: Training a neural network prediction model for head injury, and extracting a weight matrix between neurons in the front and rear layers from the trained neural network prediction model for head injury; the neural network prediction model for head injury is used to input collision test data of the vehicle and output acceleration of the occupant's head; Evaluate the importance of weights according to collision test data; and sparsify the head injury neural network prediction model according to the importance to obtain a sparse model; Retraining the sparse model, and using the trained sparse model to predict the occupant's head acceleration; The occupant's head injury is obtained according to the occupant's head acceleration.

[0006] In a second aspect, the present invention provides an electronic device, comprising: at least one processor, and a memory communicatively coupled to at least one of the processors; The memory stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors so that at least one of the processors can execute the occupant head injury prediction method based on a sparse weight neural network described in any embodiment.

[0007] In a third aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions can be executed by one or more processors to implement the occupant head injury prediction method based on a sparse weight neural network as described in any embodiment.

[0008] Compared with the prior art, the present invention has the following technical effects: According to the known collision test data, the present invention uses a sparse weighted neural network-based occupant head injury prediction model to finally obtain the occupant head acceleration, and then obtain the cumulative 3ms synthetic acceleration and HIC value that characterize the head injury. The input of the head injury prediction neural network of the present invention can be extended to more different forms of input, and the output of the present invention can be extended to more different outputs. The method of the present patent can be extended to different collision forms.

[0009] When the present invention performs sparse weighting, the importance of the weighting is evaluated according to the collision test data; thus, the model can be sparsely constructed by considering the impact of the collision test data on head injuries, and the weights with small impact can be pruned to obtain a more accurate sparse model.

[0010] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above description and other purposes, features and advantages of the present invention more obvious and easy to understand, the preferred embodiments are specifically cited and described in detail as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 A flowchart of a method for predicting occupant head injury based on a sparse weighted neural network provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of training a neural network prediction model for head injury provided by an embodiment of the present invention; Figure 3 is a curve chart of vehicle collision pulse data provided by this embodiment; Figure 4A structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0013] In the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be connected, detachably connected, or integrated; it can be mechanically connected or electrically connected; it can be directly connected or indirectly connected through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0014] Example 1 The present invention proposes a method for predicting occupant head injuries based on a sparse weighted neural network, which is suitable for obtaining occupant head injuries directly based on known collision test data without the need for actual testing, thereby saving test costs and improving the operating efficiency of the model. The method can be executed by an electronic device.

[0015] See also Figure 1 , the method specifically comprises the following steps: S110, training a neural network prediction model for head injury, and extracting a weight matrix between neurons in the front and back layers from the trained neural network prediction model for head injury.

[0016] The head injury neural network prediction model includes multiple convolutional layers in series. The head injury neural network prediction model is used to output the occupant's head acceleration (n-2 dimensions) based on the input vehicle collision test data (n dimensions).

[0017] When training the neural network prediction model for head injury, first, a training sample is obtained, and the training sample includes a collision test data sample of the vehicle and an occupant head acceleration label. The collision test data sample includes a collision speed, a vehicle collision pulse data, and a seat belt force limit value. Among them, the collision speed is the preset speed when the vehicle contacts the obstacle, and the seat belt force limit values ​​are all known. The vehicle collision pulse data is a sequence of changes in vehicle acceleration over time, presented as a pulse waveform. The collision speed, vehicle collision pulse data, and seat belt force limit value are spliced ​​to obtain a collision test data sample. Obviously, in this sample, the vehicle collision pulse data has the largest number because it contains acceleration values ​​at multiple time points, while the other data are all one value. The occupant head acceleration label includes the head acceleration value at the same time point as the vehicle collision pulse data, which can be obtained by testing the head acceleration in advance through experiments. It can be seen that the occupant head acceleration is a time series with the same dimension as the vehicle collision pulse data.

[0018] When collecting training samples, data with the same collision form and the same dummy type are grouped together. For each group of data, damaged data and obviously unreasonable data are removed, and the remaining data are normalized to obtain samples for training.

[0019] Figure 2 Schematic diagram of the training of the neural network prediction model for head injury provided by the embodiment of the present invention; Figure 2 Detailed description of the training process of the neural network prediction model for head injury.

[0020] A set of input samples of the model is denoted as , is an n-dimensional vector. For example, for a full-width head-on collision test, the input can be the vehicle collision pulse data, recorded as , collision speed , seat belt force limiter , all inputs are represented as a concatenated vector, denoted as .

[0021] The output of the head injury neural network prediction model is the occupant's head acceleration. Its sampling frequency is consistent with the vehicle collision pulse data, so its data points are consistent with the vehicle collision pulse data points. The output of the head injury neural network prediction model is recorded as , which is an n-2 dimensional vector. Where n-2 is the number of vehicle collision pulse data points / occupant head acceleration data points, and k is the total number of neuron layers in the head injury neural network prediction model. Figure 2 , each neuron receives input from the neurons in the previous layer and outputs to the neurons in the next layer. The most important thing in neurons is the size of the connection weight. For the convenience of description, the weight value between the two layers is recorded as , which means Layer The neuron to the next layer The weight value of the neuron. , . Figure 2 Shows the weight value from the first neuron in layer 1 to the first neuron in layer 2 , the weight value from the first neuron in the first layer to the second neuron in the second layer .

[0022] when hour, . Layer to The weight matrix of the layer is ,when hour, .

[0023] In the head injury neural network prediction model, the relationship between the upper and lower layers (or front and back layers) of neurons is: ; in, For the The input value of the layer to the jth neuron, For the -1 layer input value to the nth neuron ， For the -The weight value from the nth neuron in the first layer to the jth neuron in the next layer. , , but when hour, , to match the dimension of the output occupant head acceleration data.

[0024] Then, the neural network prediction model for head injury is trained based on the training samples. The specific operations performed on the input data by each layer in the neural network prediction model for head injury are completed through the weight of the layer, and the weight is essentially a string of numbers. The training process is the process of finding the most appropriate weight value for the neural network prediction model for head injury. The loss function is used to measure the gap between the predicted value output by the model and the label, and the value of the loss function is used as a feedback signal to update the weight. The loss function in the present invention adopts the mean square error (MSE): ; in, is the true value of the occupant's head acceleration at the i-th time point, It is the predicted value of the i-th time point output by the last layer (k layer) of the head injury neural network prediction model.

[0025] The weight matrix of the neural network prediction model for head injury is recorded as , which is A collection of: .

[0026] The obtained training samples are divided into two parts, one for the training of the head injury neural network prediction model and the other for the subsequent sparse model training. First, the weight matrix of the head injury prediction neural network is Random initialization to get the initial weight matrix , after the first iteration, the weight matrix is ​​updated to , if the model accuracy meets the requirements at this time, stop training, if the accuracy does not meet the requirements, continue iterative training. Similarly, after the second iteration, the weight matrix is ​​updated to , after the pth iteration, the weight matrix is ​​updated to , at this time, the neural network prediction model of head injury meets the accuracy requirements. Among them, the number of iterations p is determined according to the required accuracy. When the accuracy requirement is higher, the number of iterations is generally more.

[0027] When the accuracy of the head injury prediction neural network reaches the required a%, the weight matrix is , Among them, when hour, ,at this time ;when hour, ,at this time -2.

[0028] l Each neuron pair in the layer l The weight matrix of each neuron in the +1 layer is recorded as . ,at this time , The weight matrix of each neuron from the second-to-last layer to each neuron in the last layer is: ,at this time , .

[0029] S120. Evaluate the importance of the weights based on the collision test data.

[0030] S120 specifically includes the following three steps: S121. Select target collision test data from the collision test data.

[0031] The collision test data includes collision speed, vehicle collision pulse data and seat belt force limit value. Vehicle collision pulse data is the most abundant data and has an important impact on the prediction results; and when the pulse value / acceleration is different, the degree of impact on head injury is different. Therefore, this embodiment performs a hierarchical weighted importance assessment on the pulse waveform according to its characteristics and its impact on head injury. A negative acceleration time series whose absolute value is less than the acceleration threshold is selected from the collision test data. Figure 3 is a curve diagram of the vehicle collision pulse data provided by this embodiment, or a vehicle collision pulse waveform diagram. The horizontal axis is time, and the vertical axis is the acceleration at the B-pillar of the vehicle. In the process of performing a frontal collision test of a vehicle, when the vehicle begins to contact an obstacle, the vehicle is subjected to resistance, the acceleration value increases and the direction is negative; then the negative direction reaches an extreme value and then returns to zero. Since the smaller part of the negative pulse value / acceleration value (absolute value) (i.e. Figure 3 The two ends of the middle curve) will have a certain impact on head injury, but it is less than the impact of the larger part of the negative pulse value / acceleration value (absolute value) on head injury; moreover, the slopes of the two ends of the curve are large, and the acceleration changes rapidly. In order to effectively highlight the impact of each acceleration stage on head injury and improve the accuracy of prediction, this embodiment divides the smaller part of the negative pulse value / acceleration value (absolute value) into layers and determines the weighted values ​​of different layers. Figure 3 , define the acceleration threshold of the smaller part of the pulse value / acceleration (absolute value) as ,in is the maximum value in the collision pulse, is a percentage value. For example, when =30, which means that the smaller part of the pulse value / acceleration (absolute value) is less than 30% of the maximum value.

[0032] S122. Determine, according to the duration of the target collision test data, a weighted value of the weight of the first layer of neurons of the head injury neural network prediction model.

[0033] Specifically, the negative acceleration time series is divided into a plurality of subsequences according to a set acceleration interval; and the weighted value of the weight of the first layer of neurons input to each subsequence is calculated according to the duration ratio of the subsequences.

[0034] The parts with smaller pulse value / acceleration (absolute value) are averaged and layered, and the interval of the layers is ,in For example, when =10, that is, the smaller part of the pulse value / acceleration (absolute value) is set to the maximum value. The number of layers is Each layer obtained by division is a subsequence, and it can be seen that a subsequence includes two pulse curves on the left and right sides of the curve.

[0035] Next, determine the weight of each subsequence. For the first layer with smaller pulse values, see Figure 3 , whose weighted value is: ; in is the time corresponding to the acceleration point passing through 0 in the negative direction in the collision pulse, is the time corresponding to the acceleration point passing through 0 in the positive direction in the collision pulse, is the time when the absolute value of acceleration at the beginning of the first layer is the maximum, The time corresponding to the maximum absolute value of acceleration at the end of the first layer. The weighted values ​​of other layers are calculated in the same way as the first layer. When the duration of a layer accounts for a large proportion, its impact on head injury is also greater. By using this feature, the weighted value of the subsequence with a large duration is also set to a larger value, which can highlight its impact on the prediction results and improve the accuracy of the prediction.

[0036] S123. According to the weighted value, weight the weights of the neurons in the first layer to obtain the importance of each neuron in the first layer.

[0037] Define the weight matrix The importance of , there exists the following formula: ; in, , , is the weighted value of the smaller part of the pulse value (i.e. the target collision test data), =1,2... .in addition, Only the first layer of the head injury neural network prediction model, that is, the input layer, is weighted to affect other layers of the model through forward propagation. The weighted value is calculated as follows: ; , ; in, After p iterations, Layer Neuron pairs The importance of the weight matrix of each neuron in the +1 layer (it is actually a matrix), n is the After the above analysis, the collision test data will be spliced ​​into column vectors and input into the head injury neural network prediction model. Each data of the collision test data is input into the neurons of the input layer. For the target collision test data, the weighted value of the weight matrix input into the first layer of neurons is Figure 3 The layered calculation method shown in the figure. , the weighted value of the weight matrix input to the first layer of neurons is 0, because these data indicate that no collision has occurred or the collision has ended, and no longer affects the head injury. For data that have not been stratified and have a large proportion of negative pulse values / acceleration values ​​(absolute values), the weighted value is set to 1. The collision speed and seat belt force limit also have an important impact on head injuries, so the weighted value of the weight matrix input to the first layer of neurons is set to 1.

[0038] S130. Sparsely calculate the head injury neural network prediction model according to the importance to obtain a sparse model.

[0039] S130 includes the following two steps: S131, iterating the head injury neural network prediction model for multiple times until the set sparsity rate is met to obtain a sparse model; in each iteration, determining the sparsification matrix of the first layer weight matrix according to the size relationship between the importance matrix and the importance threshold.

[0040] The importance matrix obtained from S120 The head injury neural network prediction model is sparsed. For the convenience of description and distinction, the sparse head injury neural network prediction model is called a sparse model.

[0041] Specifically, the sparseness method is: make Among them, b is the variable of the number of iterations, and s is the total number of iterations after reaching the set sparsity rate. is the sparse matrix after the bth iteration. When b=p, = , which is obtained from S120 Starting from the pth iteration, the training samples are continued to be used to train the head injury neural network prediction model. When the weights between neurons are iterated for p+1 and subsequent times, each iteration needs to calculate , and Each element in the matrix is ​​related to the importance threshold at the bth iteration Perform size comparison.

[0042] The set sparsity rate is q%. The sparsity rate q% means that the importance of q% of the neuron weight matrices in the head injury neural network prediction model is set to 0. The head injury neural network prediction model is sparsely iterated, with a total of sp iterations. The sparsity rate of each iteration is The method for determining is: the importance of each iteration Arrange all the elements in descending order, and remove the less important ones. The elements of the sparse matrix are set to 0. The elements of the sparse matrix are parameter 0 or parameter 1.

[0043] S132. Sparse the first-layer weight matrix according to the sparsification matrix.

[0044] Specifically, at each iteration, the Hadamard product is calculated between the sparsified matrix and the first-layer weight matrix to obtain a new weight matrix for the first layer: ; For the bth iteration, the weight matrix The weight matrix after sparseness, that is, sparse weight, is the Hadamard product.

[0045] After sparsification, the weights of some neurons in the first layer are set to 0 and remain unchanged in subsequent iterations or training, that is, they no longer participate in forward propagation and the neurons are essentially pruned away.

[0046] S140: retrain the sparse model, and use the trained sparse model to predict the occupant's head acceleration.

[0047] The same initial weight matrix as in the first iteration of the head injury neural network prediction model is used Train the sparse model to obtain the final sparse weight matrix , that is, the available sparse model is determined. At this time, the accuracy of the sparse model is generally not less than a%.

[0048] The collision test data to be predicted is input into the sparse model, and the final weight matrix is ​​used The neural network calculation is performed to obtain the occupant head acceleration output by the sparse model.

[0049] A three-dimensional spatial coordinate system is established in advance with the occupant's head center and the origin. The occupant's head acceleration output by the sparse model includes acceleration in the x, y, and z directions. , and .

[0050] S150. Predicting occupant head injury according to occupant head acceleration.

[0051] The acceleration of the occupant's head in three directions in the three-dimensional coordinate system of space , and Perform synthesis to obtain the cumulative 3ms synthetic acceleration value ; ; Then, the occupant's head injury index HIC is calculated based on the cumulative 3ms synthetic acceleration value and the head collision duration.

[0052] ; in, , are any two moments in the process from the initial contact of the occupant's head with the obstacle to the final departure from the obstacle in the collision test, and Less than .

[0053] Compared with the prior art, the present invention has the following technical effects: 1. Based on the known collision test data, the present invention uses a sparse weighted neural network-based occupant head injury prediction model to finally obtain the occupant head acceleration, and then obtain the cumulative 3ms synthetic acceleration and HIC value that characterize the head injury. The input of the head injury prediction neural network of the present invention can be extended to more different forms of input, and the output of this patent can be extended to more different outputs. The method of this patent can be extended to different collision forms.

[0054] 2. When the present invention performs sparse weighting, the importance of the weighting is evaluated based on the collision test data; in this way, the model can be sparsely constructed by considering the impact of the collision test data on head injuries, and the weights with small impact can be pruned to obtain a more accurate sparse model.

[0055] 3. Based on the known collision test data, the present invention uses a sparse weighted neural network-based occupant head injury prediction model to finally obtain the three-dimensional acceleration of the occupant's head, and then obtain the cumulative 3ms synthetic acceleration and HIC value that characterize the head injury. The input of the head injury prediction neural network of the present invention can be extended to more different forms of input, and the output of this patent can be extended to more different outputs. The method of this patent can be extended to different collision forms.

[0056] Example 2 like Figure 4 As shown, an embodiment of the present application provides an electronic device, including: at least one processor; and a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions executable by at least one of the processors, and the instructions are executed by at least one of the processors to enable at least one of the processors to perform the above method. At least one processor in the electronic device can perform the above method, and thus has at least the same advantages as the above method.

[0057] Optionally, the electronic device also includes interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the electronic device, including instructions stored in or on a memory to display graphical information of a GUI (Graphical User Interface) on an external input / output device (such as a display device coupled to an interface). In other embodiments, if necessary, multiple processors can be used together with multiple memories, and / or multiple buses can be used together with multiple memories. Similarly, multiple electronic devices can be connected (for example, as a server array, a group of blade servers, or a multi-processor system), and each device provides some necessary operations. Figure 4 A processor 301 is taken as an example.

[0058] The memory 302, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as the program instructions / modules corresponding to the occupant head injury prediction method based on sparse weight neural network on the intelligent networked car in the embodiment of the present application. The processor 301 executes various functional applications and data processing of the electronic device by running the software programs, instructions and modules stored in the memory 302, that is, realizes the above-mentioned occupant head injury prediction method based on sparse weight neural network.

[0059] The memory 302 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 302 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some examples, the memory 302 may further include a memory remotely arranged relative to the processor 301, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0060] The electronic device may further include: an input device 303 and an output device 304. The processor 301, the memory 302, the input device 303 and the output device 304 may be connected via a bus or other means, and the figure takes the connection via a bus as an example.

[0061] The input device 303 can receive input digital or character information, and the output device 304 can include a display device, an auxiliary lighting device (e.g., LED), and a tactile feedback device (e.g., a vibration motor), etc. The display device can include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some embodiments, the display device can be a touch screen.

[0062] Example 3 This embodiment provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to enable the computer to execute the above method. The computer instructions on the computer-readable storage medium are used to enable the computer to execute the above method, and thus have at least the same advantages as the above method.

[0063] The computer-readable storage medium in this application may adopt any combination of one or more computer-readable media. The medium may be a computer-readable signal medium or a computer-readable storage medium. The medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of the medium (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, the medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device, or device.

[0064] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0065] The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF (Radio Frequency), etc., or any suitable combination of the above.

[0066] Computer program code for performing the operation of the present application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect through the Internet).

[0067] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for predicting occupant head injury based on sparse weighted neural network, characterized in that: include: Training a neural network prediction model for head injury, and extracting a weight matrix between neurons in the front and back layers from the trained neural network prediction model for head injury; The head injury neural network prediction model is used to input the collision test data of the vehicle and output the acceleration of the occupant's head; Evaluate the importance of weights according to collision test data; and sparsify the head injury neural network prediction model according to the importance to obtain a sparse model; Retraining the sparse model, and using the trained sparse model to predict the occupant's head acceleration; The occupant's head injury is obtained according to the occupant's head acceleration.

2. The method according to claim 1, characterized in that Training of a neural network prediction model for head injury, including: Acquire training samples, wherein the training samples include collision test data samples of the vehicle and occupant head acceleration labels; A neural network prediction model for head injury is trained according to the training samples.

3. The method according to claim 1, characterized in that The importance of evaluating weights based on crash test data includes: selecting target collision test data from the collision test data; Determining a weighted value of a weight of a first layer of neurons of the head injury neural network prediction model according to the duration of the target collision test data; The weights of the neurons in the first layer are weighted according to the weighted values ​​to obtain the importance of each neuron in the first layer.

4. The method according to claim 3, characterized in that The collision test data includes collision speed, vehicle collision pulse data and seat belt force limit value; the vehicle collision pulse data is a sequence of changes in vehicle acceleration over time; Selecting target collision test data from the collision test data comprises: Selecting a negative acceleration time series whose absolute value is less than an acceleration threshold from the collision test data; Determining the weighted value of the weight of the first layer of neurons in the head injury neural network prediction model according to the duration of the target collision test data includes: According to a set acceleration interval, the negative acceleration time series is divided into a plurality of subsequences; According to the duration ratio of the subsequences, the weighted value of the weight of the first layer of neurons input to each subsequence is calculated.

5. The method according to claim 1, characterized in that The head injury neural network prediction model is sparsely constructed according to the importance to obtain a sparse model, including: Iterating the head injury neural network prediction model for multiple times until a set sparsification rate is met to obtain a sparsified model; In each iteration, the sparsification matrix of the first layer weight matrix is ​​determined according to the size relationship between the importance matrix and the importance threshold; the elements in the sparsification matrix are parameter 0 or parameter 1; and the first layer weight matrix is ​​sparsified according to the sparsification matrix.

6. The method according to claim 5, characterized in that The first layer weight matrix is ​​sparsely processed according to the sparse matrix, comprising: The Hadamard product is calculated between the sparsified matrix and the weight matrix of the first layer to obtain a new weight matrix of the first layer.

7. The method according to claim 1, characterized in that Retraining the sparsified model includes: The sparse model is retrained using the initial weights used when the head injury neural network prediction model is trained.

8. The method according to claim 1, characterized in that Predicting the occupant's head injury according to the occupant's head acceleration includes: The accelerations of the occupant's head in three directions in the three-dimensional coordinate system of space are synthesized to obtain a cumulative 3ms synthetic acceleration value; The occupant's head injury index HIC is calculated according to the accumulated 3 ms synthetic acceleration value and the head collision duration.

9. An electronic device, characterized in that: include: at least one processor, and a memory communicatively coupled to at least one of the processors; The memory stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors to enable at least one of the processors to execute the occupant head injury prediction method based on a sparse weight neural network as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, which can be executed by one or more processors to implement the occupant head injury prediction method based on a sparse weight neural network as described in any one of claims 1-8.

Images