Verification method and device of vehicle-mounted chip convolutional neural network

Abstract

The invention provides a verification method and device for a convolutional neural network of a vehicle-mounted chip. The method comprises the following steps: acquiring environment data to be detected of a target vehicle; respectively generating a first convolution result and a second convolution result from the to-be-detected environment data through a parallel computing path of a vehicle-mounted chip convolutional neural network; and based on the to-be-detected environment data and a convolution kernel of the vehicle-mounted chip convolutional neural network, calculating checksum data, and based on the first convolution result, the second convolution result and the checksum data, determining a verification result of the to-be-detected environment data passing through the vehicle-mounted chip convolutional neural network, and reporting the verification result to an automobile master control. Therefore, computing resources are reduced, and power consumption and hardware resource consumption are saved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a method and device for verifying a convolutional neural network on a vehicle-mounted chip. Background Art

[0002] The development of autonomous driving technology presents a series of standardized tasks, the most important of which are environmental perception, decision-making, and planning. As a vital tool for human safety, automotive chipsets place high demands on safety, requiring timely detection and correction of all potential errors. When correction is impossible, an error signal should be immediately returned and an emergency plan should be activated.

[0003] In the on-board chip safety system of autonomous driving, the Triple Modular Redundancy (TMR) method is often used. This method is a reliability method that achieves fault tolerance through redundant design and majority voting mechanism. Its core goal is to ensure that the system can still operate normally when some modules fail by actively correcting errors rather than simply detecting errors.

[0004] However, the traditional triple-module redundancy method requires three times the resources to implement the function, which greatly increases hardware resource consumption and power consumption, resulting in a waste of resources. Summary of the Invention

[0005] The present invention provides a verification method and device for a convolutional neural network of an on-board chip, which is used to solve the problem that the triple-module redundancy method in the prior art requires three times the resources to implement the function, greatly increases hardware resource consumption, increases power consumption, and causes waste of resources. It achieves a reduction in computing resources and saves power consumption and hardware resource consumption.

[0006] The present invention provides a method for verifying a convolutional neural network of an on-board chip, comprising the following steps: Obtain the target vehicle's environmental data to be detected; Passing the detected environmental data through a parallel computing path of a vehicle-mounted chip convolutional neural network to generate a first convolution result and a second convolution result, respectively; and calculating checksum data based on the detected environmental data and the convolution kernel of the vehicle-mounted chip convolutional neural network; Based on the first convolution result, the second convolution result and the checksum data, determine the verification result of the environmental data to be detected after passing the on-board chip convolutional neural network, and report the verification result to the vehicle main control.

[0007] According to a verification method for a vehicle-mounted chip convolutional neural network provided by the present invention, based on the environmental data to be detected and the convolution kernel of the vehicle-mounted chip convolutional neural network, verification and data are calculated, including: Based on the size of the convolution kernel of the on-board chip convolutional neural network, the to-be-detected environmental data is slidingly divided into multiple groups of sub-region data, and the data in the multiple groups of sub-region data are summed respectively to obtain a sub-region sum matrix consisting of the sum of the multiple groups of sub-region data; The sub-region and the matrix are convolved with the convolution kernel and then summed to obtain checksum data.

[0008] According to a verification method for a vehicle-mounted chip convolutional neural network provided by the present invention, based on the first convolution result, the second convolution result and the verification sum data, determining the verification result of the environmental data to be detected after the vehicle-mounted chip convolutional neural network, and reporting the verification result to the vehicle main control, including: respectively summing the data in the first convolution result and the second convolution result to obtain first convolution sum data and second convolution sum data; The first convolution sum data, the second convolution sum data and the checksum data are compared, and based on the comparison result, the verification result of the environmental data to be detected after passing the convolutional neural network of the on-board chip is determined, and the verification result is reported to the vehicle main control.

[0009] According to a verification method for a vehicle-mounted chip convolutional neural network provided by the present invention, based on a comparison result, determining a verification result of the environmental data to be detected after the vehicle-mounted chip convolutional neural network, and reporting the verification result to the vehicle main control, the method includes: When at least two of the first convolution sum data, the second convolution sum data, and the checksum data are equal, the two equal data are used as the accurate result of the on-board chip convolutional neural network, and the accurate result is reported to the vehicle main control; When the first convolution sum data, the second convolution sum data, and the checksum data are not equal, an error signal is reported to the vehicle main control, and an emergency state is entered.

[0010] The present invention also provides a vehicle-mounted chip convolutional neural network verification device, comprising the following modules: An acquisition module is used to obtain the environmental data to be detected of the target vehicle; a computing module, configured to pass the detected environmental data through a parallel computing path of a convolutional neural network of an on-board chip to generate a first convolution result and a second convolution result, respectively; and, based on the detected environmental data and a convolution kernel of the convolutional neural network of the on-board chip, calculate checksum data; A determination and reporting module is used to determine the verification result of the environmental data to be detected after the convolutional neural network of the on-board chip based on the first convolution result, the second convolution result and the checksum data, and report the verification result to the vehicle main control.

[0011] According to the present invention, a verification device for a vehicle-mounted chip convolutional neural network is provided, which calculates verification data based on the environmental data to be detected and the convolution kernel of the vehicle-mounted chip convolutional neural network, including: Based on the size of the convolution kernel of the on-board chip convolutional neural network, the to-be-detected environmental data is slidingly divided into multiple groups of sub-region data, and the data in the multiple groups of sub-region data are summed respectively to obtain a sub-region sum matrix consisting of the sum of the multiple groups of sub-region data; The sub-region and the matrix are convolved with the convolution kernel and then summed to obtain checksum data.

[0012] According to a verification device for a vehicle-mounted chip convolutional neural network provided by the present invention, based on the first convolution result, the second convolution result, and the checksum data, determining the verification result of the environmental data to be detected after the vehicle-mounted chip convolutional neural network, and reporting the verification result to the vehicle main control, including: respectively summing the data in the first convolution result and the second convolution result to obtain first convolution sum data and second convolution sum data; The first convolution sum data, the second convolution sum data and the checksum data are compared, and based on the comparison result, the verification result of the environmental data to be detected after passing the convolutional neural network of the on-board chip is determined, and the verification result is reported to the vehicle main control.

[0013] According to a verification device for a vehicle-mounted chip convolutional neural network provided by the present invention, based on a comparison result, determining a verification result of the environmental data to be detected after the vehicle-mounted chip convolutional neural network, and reporting the verification result to the vehicle main control, the device includes: When at least two of the first convolution sum data, the second convolution sum data, and the checksum data are equal, the two equal data are used as the accurate result of the on-board chip convolutional neural network, and the accurate result is reported to the vehicle main control; When the first convolution sum data, the second convolution sum data, and the checksum data are not equal, an error signal is reported to the vehicle main control, and an emergency state is entered.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements a verification method for the on-board chip convolutional neural network as described in any one of the above-mentioned methods.

[0015] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements a verification method for the on-board chip convolutional neural network as described in any of the above-mentioned methods.

[0016] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements a verification method for the on-board chip convolutional neural network as described in any of the above-mentioned methods.

[0017] The verification method and device for the on-board chip convolutional neural network provided by the present invention calculate the checksum data based on the environmental data to be detected and the convolution kernel of the on-board chip convolutional neural network, and then determine the verification result of the on-board chip convolutional neural network of the environmental data to be detected based on the two convolution results generated by passing the on-board chip convolutional neural network through the environmental data to be detected and the checksum data, thereby reducing computing resources and saving power consumption and hardware resource consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 It is a flow chart of the on-board chip convolutional neural network verification method provided by the present invention.

[0020] Figure 2 It is a schematic diagram of the arbitration process provided by the present invention.

[0021] Figure 3 This is a schematic diagram of the broadcast range provided by the present invention, using a 5×5 feature map and a 3×3 convolution as an example.

[0022] Figure 4 This is a schematic diagram of the application process of convolutional neural networks in autonomous driving provided by related technologies.

[0023] Figure 5 It is a schematic diagram of a traditional triple-module redundancy method provided by related technology.

[0024] Figure 6 It is a schematic diagram of the verification and validation method provided by the present invention.

[0025] Figure 7 It is a structural schematic diagram of the vehicle-mounted chip convolutional neural network verification device provided by the present invention.

[0026] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0027] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0028] Figure 1 This is a flow chart of the on-board chip convolutional neural network verification method provided by the present invention, such as Figure 1 As shown, the method includes the following steps: Step 100: Obtain the target vehicle's environmental data to be detected.

[0029] Step 101: Pass the environmental data to be detected through the parallel computing path of the on-board chip convolutional neural network to generate a first convolution result and a second convolution result respectively; and, based on the environmental data to be detected and the convolution kernel of the on-board chip convolutional neural network, calculate the checksum data.

[0030] Step 102: Based on the first convolution result, the second convolution result, and the checksum data, determine the verification result of the on-board chip convolutional neural network on the environmental data to be detected, and report the verification result to the vehicle main control.

[0031] Specifically, the target vehicle is any vehicle equipped with an intelligent driving onboard chip. Onboard chips place high demands on safety. They require timely detection and correction of any potential errors. If correction is not possible, an error signal should be immediately returned and an emergency plan activated.

[0032] Convolutional Neural Networks (CNN) are one of the important methods of environmental perception in autonomous driving. In an embodiment of the present invention, it is necessary to verify the data generated by the CNN used by the on-board chip in environmental perception to prevent errors.

[0033] First, the target vehicle's environmental data to be detected can be obtained through on-board sensors (such as cameras, lidar, etc.). This type of data usually contains information such as roads, pedestrians, obstacles, etc., and can be used for feature extraction, pattern recognition and other detection through convolutional neural networks.

[0034] After the environmental data to be detected is input into the onboard chip's convolutional neural network, a designed parallel computation path is used to generate the first and second convolution results. Simultaneously, based on the same environmental data and convolutional neural network kernel parameters, a checksum is independently calculated for data verification, ensuring the reliability of the computation process.

[0035] After obtaining the first and second convolution results and the checksum data, the credibility of the first and second convolution results can be determined by comparing them with the checksum data. For example, if the two convolution results match the checksum data, it indicates that the convolutional neural network processed the results correctly; if there is a deviation, recalculation or an exception reporting mechanism may be triggered. The final verification results can be fed back to the vehicle's main control system in real time, providing a basis for autonomous driving decision-making.

[0036] The verification method for the on-board chip convolutional neural network provided by the present invention calculates checksum data based on the environmental data to be detected and the convolution kernel of the on-board chip convolutional neural network, and then determines the verification result of the on-board chip convolutional neural network of the environmental data to be detected based on the two convolution results generated by passing the on-board chip convolutional neural network through the environmental data to be detected and the checksum data, thereby reducing computing resources and saving power consumption and hardware resource consumption.

[0037] According to a verification method for a vehicle-mounted chip convolutional neural network provided by the present invention, verification and data are calculated based on the environmental data to be detected and the convolution kernel of the vehicle-mounted chip convolutional neural network, including: Based on the size of the convolution kernel of the on-board chip convolutional neural network, the environmental data to be detected is slidingly divided into multiple groups of sub-region data, and the data in the multiple groups of sub-region data are summed up respectively to obtain a sub-region sum matrix consisting of the sum of the multiple groups of sub-region data; The sub-region and matrix are convolved with the convolution kernel and then summed to obtain the checksum data.

[0038] Specifically, in the embodiment of the present invention, during the calculation of the checksum data, the environmental data to be detected (e.g., captured image data) can first be divided into multiple groups of sub-region data that match the convolution kernel size (e.g., 3×3) of the convolutional neural network in the vehicle chip using a sliding window method. For example, if the image data is 5×5 and the convolution kernel size is 3×3, the image data will be divided into nine 3×3 local regions.

[0039] After obtaining multiple sets of sub-region data, all data elements in each set are summed to obtain a matrix consisting of the sums of the sub-regions (referred to as the sub-region sum matrix in this embodiment of the present invention). Here, only the sum of the regional data is retained, rather than the complex features, to simplify the calculation and adapt to the real-time requirements of the onboard chip.

[0040] Then, the above sub-regions and matrices are convolved element-by-element with the convolution kernel and the sum is calculated to obtain the final checksum data.

[0041] The method provided by the embodiment of the present invention obtains checksum data through staged sub-region processing and convolution kernel convolution operation, thereby realizing rapid verification of the output results of the neural convolution network, meeting the requirements of real-time and reliability in vehicle-mounted scenarios.

[0042] According to a verification method for a vehicle-mounted chip convolutional neural network provided by the present invention, based on a first convolution result, a second convolution result, and checksum data, a verification result of the environmental data to be detected after the vehicle-mounted chip convolutional neural network is determined, and the verification result is reported to the vehicle main control, including: respectively summing the data in the first convolution result and the second convolution result to obtain first convolution sum data and second convolution sum data; The first convolution sum data, the second convolution sum data and the checksum data are compared. Based on the comparison result, the verification result of the environmental data to be detected after passing through the on-board chip convolutional neural network is determined, and the verification result is reported to the vehicle main control.

[0043] Specifically, in the process of determining the verification results of the environmental data to be detected after passing through the on-board chip convolutional neural network, first, the first convolution result and the second convolution result generated by the on-board chip convolutional neural network are globally summed to obtain the first convolution sum data and the second convolution sum data.

[0044] The summation operation compresses the two convolution results into a scalar value, which can simplify the comparison complexity.

[0045] The first convolution sum data and the second convolution sum data are compared with the checksum data. In an embodiment of the present invention, the value of the checksum data can represent the expected accurate calculation result. Therefore, based on the comparison results of the first convolution sum data, the second convolution sum data and the checksum data, it can be determined whether the first convolution result and the second convolution result have passed the verification. If the first convolution result and the second convolution result have passed the verification, it indicates that there is no error in the convolution neural network processing of the on-board chip; if there is a significant difference, an abnormality reporting or recalculation mechanism can be triggered.

[0046] According to a verification method for a vehicle-mounted chip convolutional neural network provided by the present invention, based on the comparison result, the verification result of the environmental data to be detected after the vehicle-mounted chip convolutional neural network is determined, and the verification result is reported to the vehicle main control, including: If at least two of the first convolution sum data, the second convolution sum data, and the checksum data are equal, the two equal data are regarded as the accurate results of the on-board chip convolutional neural network, and the accurate results are reported to the vehicle main control; When the first convolution sum data, the second convolution sum data, and the checksum data are not equal, an error signal is reported to the vehicle main control and an emergency state is entered.

[0047] Specifically, in this embodiment of the present invention, if at least two of the first convolution sum data, the second convolution sum data, and the checksum data are equal (for example, the first convolution sum data and the checksum data are equal; or the second convolution sum data and the checksum data are equal; or the first convolution sum data and the second convolution sum data are equal), these two equal values ​​can be used as the accurate output of the convolutional neural network. This design, based on redundant checksum logic, eliminates random errors that may be introduced by single-path calculations through the majority consensus principle.

[0048] For example, if the first convolution sum data and the checksum data are both 150, and the second convolution sum data is 148, 150 can be determined as the correct result and reported to the vehicle's main control system for subsequent decision-making (such as path planning).

[0049] If all three values ​​are unequal (for example, the first convolution sum is 150, the second convolution sum is 148, and the checksum is 152), this indicates a serious error in the convolutional neural network calculation (such as a hardware failure or data transmission interference). In this case, the vehicle's main control system can report the error and immediately initiate an emergency state (such as disabling the autonomous driving function). This mechanism meets the functional safety requirements of automotive chips and prevents dangerous behaviors caused by erroneous results.

[0050] The following further illustrates the on-board chip convolutional neural network verification method provided by the present invention through examples in specific application scenarios.

[0051] The development of autonomous driving technology presents a series of standardized tasks, the most important of which are environmental perception, decision-making, and planning. Environmental perception is one of the key areas of autonomous driving development. At the same time, as a vital tool for human safety, automotive chipsets place high demands on safety. They require timely detection and correction of all potential errors. When correction is impossible, an error signal should be immediately returned and emergency response plans activated.

[0052] CNN is one of the most important methods for environmental perception. YOLO is a classic convolutional neural network. Taking the YOLO neural network as an example, the most important computing resource of the YOLO network is convolution calculation.

[0053] In view of the characteristics of convolution operations, this embodiment reduces hardware overhead by improving the calculation process. Triple modular redundancy technology actually only requires one set of correct data. The other two sets are used to verify whether the data is correct. Considering that errors themselves are rare events (which is also the theoretical basis of triple modular redundancy technology), the triple operation can be optimized to two sets of data alternatives and a checksum is introduced for arbitration. Figure 2 It is a schematic diagram of the arbitration process provided by the present invention.

[0054] Figure 3 This is a schematic diagram of the broadcast range provided by the present invention using a 5×5 feature map and a 3×3 convolution as an example. Figure 3 As shown, the checksum is the sum of a group of data, not the specific content of each group. Therefore, the calculation strategy can be optimized by pre-summing the broadcast range of a convolution kernel on the feature map before performing the convolution operation. This can greatly optimize the number of multiplications, reducing hardware overhead and system power consumption.

[0055] A feature map is a core concept in convolutional neural networks (CNNs). It represents the output of each convolutional layer in the network and can be viewed as a stack of multiple two-dimensional images. In a CNN, after the original image is processed by the convolutional layers, multiple feature maps are generated. Each feature map represents a characteristic response to the input image, such as edges, textures, and other information. The number and size of feature maps processed from the original image are typically large, so the convolution operation on the feature maps is the main computational overhead in the CNN network and the largest consumption of hardware resources after triple-module redundancy. Furthermore, the size of the convolution kernel in a typical CNN network is relatively fixed (mostly 3×3), so optimizing the computational path can significantly improve performance.

[0056] Figure 4 This is a schematic diagram of the application process of convolutional neural networks in autonomous driving provided by related technologies, such as Figure 4 As shown, since autonomous driving involves personal safety, if data errors are caused by environmental factors (such as vibration, static electricity, electromagnetic interference, etc.), it may cause serious consequences. Therefore, the safety requirements for automotive chips are very high.

[0057] Figure 5 This is a schematic diagram of the traditional triple-module redundancy method provided by the related technology, such as Figure 5As shown, the triple modular redundancy method independently calculates three sets of data and compares them, and votes on the calculation results. If two are the same and different from the third, it is considered that the third data is wrong. If all three sets of data are different, an error signal is reported and an emergency state is entered.

[0058] Depend on Figure 5 It can be seen that triple modular redundancy consumes a lot of resources. Ultimately, only one set of correct data is needed. The two independent sets of data are only calculated to ensure the accuracy of the final data. Therefore, the test data does not actually need to obtain all the characteristics of the final result; it only needs to verify the correctness of the final result.

[0059] Figure 6 It is a schematic diagram of the verification and validation method provided by the present invention, such as Figure 6 As shown, in this embodiment, the characteristics of convolution calculation are used to optimize the calculation process, and the checksum (check_sum) of the final result is calculated. By comparing the checksums of the other two data, it can be verified whether the data is calculated correctly.

[0060] By optimizing the calculation process to obtain the checksum for verification, the computing resources can be saved by almost half (the main computational effort in convolution calculation is multiplication. The checksum in this embodiment uses very few multiplications, so the computational effort is far less than that of a single convolution calculation). The improved security mechanism has an ISO standard diagnostic coverage rate of over 99% (slightly fluctuating depending on different network structures), meeting the security verification standards.

[0061] The on-board chip convolutional neural network verification device provided by the present invention is described below. The on-board chip convolutional neural network verification device described below and the on-board chip convolutional neural network verification method described above can be referenced to each other.

[0062] Figure 7 This is a schematic diagram of the structure of the vehicle-mounted chip convolutional neural network verification device provided by the present invention. Figure 7 As shown, the device includes the following modules: An acquisition module 700 is used to acquire the target vehicle's environmental data to be detected; A computing module 710 is configured to pass the detected environmental data through the parallel computing path of the onboard chip convolutional neural network to generate a first convolution result and a second convolution result, respectively; and to calculate checksum data based on the detected environmental data and the convolution kernel of the onboard chip convolutional neural network; The determination and reporting module 720 is used to determine the verification result of the environmental data to be detected after the on-board chip convolutional neural network based on the first convolution result, the second convolution result and the checksum data, and report the verification result to the vehicle main control.

[0063] According to the present invention, a verification device for a vehicle-mounted chip convolutional neural network is provided, which calculates verification and data based on the environmental data to be detected and the convolution kernel of the vehicle-mounted chip convolutional neural network, including: Based on the size of the convolution kernel of the on-board chip convolutional neural network, the environmental data to be detected is slidingly divided into multiple groups of sub-region data, and the data in the multiple groups of sub-region data are summed up respectively to obtain a sub-region sum matrix consisting of the sum of the multiple groups of sub-region data; The sub-region and matrix are convolved with the convolution kernel and then summed to obtain the checksum data.

[0064] According to the present invention, a verification device for a vehicle-mounted chip convolutional neural network is provided. Based on a first convolution result, a second convolution result, and checksum data, the verification result of the environmental data to be detected after the vehicle-mounted chip convolutional neural network is determined, and the verification result is reported to the vehicle main control, including: respectively summing the data in the first convolution result and the second convolution result to obtain first convolution sum data and second convolution sum data; The first convolution sum data, the second convolution sum data and the checksum data are compared. Based on the comparison result, the verification result of the environmental data to be detected after passing through the on-board chip convolutional neural network is determined, and the verification result is reported to the vehicle main control.

[0065] According to the present invention, a verification device for a vehicle-mounted chip convolutional neural network is provided. Based on the comparison result, the verification result of the environmental data to be detected after the vehicle-mounted chip convolutional neural network is determined, and the verification result is reported to the vehicle main control, including: If at least two of the first convolution sum data, the second convolution sum data, and the checksum data are equal, the two equal data are regarded as the accurate results of the on-board chip convolutional neural network, and the accurate results are reported to the vehicle main control; When the first convolution sum data, the second convolution sum data, and the checksum data are not equal, an error signal is reported to the vehicle main control and an emergency state is entered.

[0066] Figure 8 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the vehicle chip convolutional neural network verification method, which includes: Obtain the target vehicle's environmental data to be detected; Passing the detected environmental data through the parallel computing path of the on-board chip convolutional neural network to generate a first convolution result and a second convolution result, respectively; and calculating checksum data based on the detected environmental data and the convolution kernel of the on-board chip convolutional neural network; Based on the first convolution result, the second convolution result and the checksum data, the verification result of the environmental data to be detected after passing through the on-board chip convolutional neural network is determined, and the verification result is reported to the vehicle main control.

[0067] Furthermore, the logic instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0068] On the other hand, the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the vehicle-mounted chip convolutional neural network verification method provided by the above methods, which includes: Obtain the target vehicle's environmental data to be detected; Passing the detected environmental data through the parallel computing path of the on-board chip convolutional neural network to generate a first convolution result and a second convolution result, respectively; and calculating checksum data based on the detected environmental data and the convolution kernel of the on-board chip convolutional neural network; Based on the first convolution result, the second convolution result and the checksum data, the verification result of the environmental data to be detected after passing through the on-board chip convolutional neural network is determined, and the verification result is reported to the vehicle main control.

[0069] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for verifying a convolutional neural network for an on-board chip provided by the above methods is implemented. The method includes: Obtain the target vehicle's environmental data to be detected; Passing the detected environmental data through the parallel computing path of the on-board chip convolutional neural network to generate a first convolution result and a second convolution result, respectively; and calculating checksum data based on the detected environmental data and the convolution kernel of the on-board chip convolutional neural network; Based on the first convolution result, the second convolution result and the checksum data, the verification result of the environmental data to be detected after passing through the on-board chip convolutional neural network is determined, and the verification result is reported to the vehicle main control.

[0070] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0071] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for verifying a convolutional neural network on a vehicle chip, characterized in that: include: Obtain the target vehicle's environmental data to be detected; Passing the detected environmental data through a parallel computing path of a vehicle-mounted chip convolutional neural network to generate a first convolution result and a second convolution result, respectively; and calculating checksum data based on the detected environmental data and the convolution kernel of the vehicle-mounted chip convolutional neural network; Based on the first convolution result, the second convolution result and the checksum data, determine the verification result of the environmental data to be detected after passing the on-board chip convolutional neural network, and report the verification result to the vehicle main control.

2. The verification method for the on-board chip convolutional neural network according to claim 1, characterized in that: Calculating checksum data based on the to-be-detected environment data and the convolution kernel of the on-board chip convolutional neural network includes: Based on the size of the convolution kernel of the on-board chip convolutional neural network, the to-be-detected environmental data is slidingly divided into multiple groups of sub-region data, and the data in the multiple groups of sub-region data are summed respectively to obtain a sub-region sum matrix consisting of the sum of the multiple groups of sub-region data; The sub-region and the matrix are convolved with the convolution kernel and then summed to obtain checksum data.

3. The verification method for the on-board chip convolutional neural network according to claim 1 or 2, characterized in that: Determining a verification result of the on-board chip convolutional neural network on the environmental data to be detected based on the first convolution result, the second convolution result, and the checksum data, and reporting the verification result to the vehicle main control, including: respectively summing the data in the first convolution result and the second convolution result to obtain first convolution sum data and second convolution sum data; The first convolution sum data, the second convolution sum data and the checksum data are compared, and based on the comparison result, the verification result of the environmental data to be detected after passing the convolutional neural network of the on-board chip is determined, and the verification result is reported to the vehicle main control.

4. The verification method for the on-board chip convolutional neural network according to claim 3 is characterized in that: Based on the comparison result, determining the verification result of the environmental data to be detected after the on-board chip convolutional neural network, and reporting the verification result to the vehicle main control, including: If at least two of the first convolution sum data, the second convolution sum data, and the checksum data are identical, the two identical data are used as the accurate result of the on-board chip convolutional neural network, and the accurate result is reported to the vehicle main control; When the first convolution sum data, the second convolution sum data, and the checksum data are not equal, an error signal is reported to the vehicle main control, and an emergency state is entered.

5. A vehicle-mounted chip convolutional neural network verification device, characterized in that: include: An acquisition module is used to obtain the environmental data to be detected of the target vehicle; A computing module, configured to pass the detected environmental data through a parallel computing path of a vehicle-mounted chip convolutional neural network to generate a first convolution result and a second convolution result, respectively; and, based on the detected environmental data and the convolution kernel of the vehicle-mounted chip convolutional neural network, calculate checksum data; A determination and reporting module is used to determine the verification result of the environmental data to be detected after the convolutional neural network of the on-board chip based on the first convolution result, the second convolution result and the checksum data, and report the verification result to the vehicle main control.

6. The verification device for the on-board chip convolutional neural network according to claim 5, characterized in that: Calculating checksum data based on the to-be-detected environment data and the convolution kernel of the on-board chip convolutional neural network includes: Based on the size of the convolution kernel of the on-board chip convolutional neural network, the to-be-detected environmental data is slidingly divided into multiple groups of sub-region data, and the data in the multiple groups of sub-region data are summed respectively to obtain a sub-region sum matrix consisting of the sum of the multiple groups of sub-region data; The sub-region and the matrix are convolved with the convolution kernel and then summed to obtain checksum data.

7. The verification device for the on-board chip convolutional neural network according to claim 5 or 6, characterized in that: Determining a verification result of the on-board chip convolutional neural network on the environmental data to be detected based on the first convolution result, the second convolution result, and the checksum data, and reporting the verification result to the vehicle main control, including: respectively summing the data in the first convolution result and the second convolution result to obtain first convolution sum data and second convolution sum data; The first convolution sum data, the second convolution sum data and the checksum data are compared, and based on the comparison result, the verification result of the environmental data to be detected after passing the convolutional neural network of the on-board chip is determined, and the verification result is reported to the vehicle main control.

8. The verification device for the on-board chip convolutional neural network according to claim 7, characterized in that: Based on the comparison result, determining the verification result of the environmental data to be detected after the on-board chip convolutional neural network, and reporting the verification result to the vehicle main control, including: If at least two of the first convolution sum data, the second convolution sum data, and the checksum data are identical, the two identical data are used as the accurate result of the on-board chip convolutional neural network, and the accurate result is reported to the vehicle main control; When the first convolution sum data, the second convolution sum data, and the checksum data are not equal, an error signal is reported to the vehicle main control, and an emergency state is entered.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, it implements the verification method of the on-board chip convolutional neural network as described in any one of claims 1 to 4.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the verification method of the vehicle-mounted chip convolutional neural network as described in any one of claims 1 to 4 is implemented.

Images