Intelligent driving target detection method and device, electronic equipment and storage medium
Through the ResNet50-based object detection model, combined with the technical means of zero-filling layer and attention model, the problem of difficult to take into account the real-time and accuracy of driving object detection in the existing technology is solved, and efficient detection in complex environments is achieved, ensuring the safety of drivers and passengers.
Patent Information
- Application Number
- CN202311573235.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-23
- Publication Date
- 2025-05-23
AI Technical Summary
The prior art is difficult to take into account both real-time and accuracy in driving target detection, especially when facing complex conditions such as vehicles of different types and sizes, mutual occlusion between vehicles, interference from environmental factors, and different pitch angles of roads.
The object detection model based on ResNet50 is adopted, and the initial model is pre-trained, the convolutional neural network with zero-filling layer is trained, and the feature spectrum is fused through the attention model, and finally the migration training is carried out to improve the detection performance.
It realizes the accuracy and real-time performance of driving target detection in complex environments, enhances the detection capabilities of vision boundaries and distant vehicles, reduces interference from environmental factors, avoids traffic accidents, and ensures the safety of drivers and passengers.
Smart Images
Figure CN120032329A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of automobile and intelligent driving technology, and in particular to an intelligent driving target detection method, device, electronic equipment and storage medium. Background Art
[0002] Frontal vehicle recognition and distance detection constitute one of the core technologies of intelligent driving environment perception. However, in real-world scenarios, there are many challenges, including vehicles of different types and sizes, mutual occlusion between vehicles, interference from environmental factors, and different pitch angles of the road. These factors increase the difficulty of forward vehicle detection to varying degrees. Some existing detections for specific targets face the problem of severe loss of spatial information, which affects the accuracy of detection.
[0003] Therefore, the existing technology has the problem that it is difficult to take into account both the real-time performance and the accuracy when detecting driving targets. Summary of the invention
[0004] The main purpose of the embodiments of the present invention is to propose an intelligent driving target detection method, device, electronic device and storage medium, which achieves accuracy and real-time detection of driving targets.
[0005] One aspect of the present invention provides an intelligent driving target detection method, comprising:
[0006] According to the target detection request, obtain the target detection image of the intelligent driving vehicle;
[0007] Performing target detection on the target detection image using a target detection model based on ResNet50 to obtain a target detection result;
[0008] According to the detection result, executing control processing of the intelligent driving vehicle;
[0009] The training of the ResNet50-based target detection model includes:
[0010] The first data set is used to train the initial model based on ResNet50 to obtain a pre-trained model;
[0011] The pre-trained model is trained using a second data set and a convolutional neural network with a zero-filling layer to obtain first training data;
[0012] Training the first training data through an attention model to obtain an attention spectrum, and fusing the attention with a feature spectrum of the first training data to obtain second training data;
[0013] Classifying the second training data to obtain a classification result and a trained model;
[0014] The trained model is subjected to migration training to obtain the ResNet50-based target detection model.
[0015] According to the intelligent driving target detection method, the first data set is used to train the initial model based on ResNet50 to obtain a pre-trained model, including:
[0016] Build the target task model and use the non-first data set for training to obtain weight parameters, where the target task model is based on ResNet50;
[0017] The weight parameters are added to the target task model to obtain the pre-trained model.
[0018] According to the intelligent driving target detection method, the pre-trained model is trained using a second data set and a convolutional neural network with a zero-filling layer to obtain first training data, including:
[0019] When the convolution neural network is used to perform convolution processing on the images in the second data set, the borders of the images are zero-filled through a zero-filling layer so that the images are convolved at least twice by the convolution kernel.
[0020] According to the intelligent driving target detection method, the first training data is trained by an attention model to obtain an attention spectrum, and the attention is fused with a feature spectrum of the first training data to obtain second training data, including:
[0021] Using the first training data as input of the attention model, calculating an attention spectrum;
[0022] Performing weighted processing on the attention spectrum and the feature spectrum of the first data to obtain a weighted feature spectrum;
[0023] The weighted feature spectrum is superimposed on the feature spectrum of the first data to obtain the second training data, wherein the second training data is an attention feature spectrum.
[0024] According to the intelligent driving target detection method, the second training data is classified to obtain a classification result and a trained model, including:
[0025] A fully connected layer is used to receive the attention feature spectrum, and the fully connected layer includes a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence, wherein the first fully connected layer is used to receive the attention feature spectrum;
[0026] Performing dimension transformation and spatial cascade processing on the attention feature spectrum through the second fully connected layer, and merging the attention feature spectrum after dimension transformation and spatial cascade processing with the received attention feature spectrum;
[0027] The merged attention feature spectrum is classified through the third fully connected layer to obtain the classification result and the trained model.
[0028] According to the intelligent driving target detection method, the trained model is subjected to migration training to obtain the target detection model based on ResNet50, including:
[0029] The pre-trained model based on ResNet50 is subjected to migration training to obtain a target detection model, wherein the model structure used for migration training includes a convolutional layer, a maximum pooling layer, a self-attention layer and a fully connected layer in sequence, wherein the convolutional layer is a convolutional layer of an initial model based on ResNet50, wherein the maximum pooling layer is a 2×2 sliding window, wherein the self-attention layer performs a transition between the convolutional layer and the fully connected layer through a flatten function, wherein the first fully connected layer in the fully connected layer uses ReLU as an activation function, the second fully connected layer uses a batch normalization function as an activation function, and the third fully connected layer uses Softmax as an activation function.
[0030] According to the intelligent driving target detection method, the method further includes:
[0031] The detection result is sent to a decision-making system of the intelligent driving vehicle, and the decision-making system performs vehicle control processing based on the detection result.
[0032] Another aspect of an embodiment of the present invention provides an intelligent driving target detection device, including:
[0033] An identification module and a training module, wherein the identification module includes a first module, a second module and a third module, and the training module includes a fourth module, a fifth module, a sixth module, a seventh module and an eighth module;
[0034] The first module is used to obtain a target detection image of the intelligent driving vehicle according to a target detection request;
[0035] The second module is used to perform target detection on the target detection image using a target detection model based on ResNet50 to obtain a target detection result;
[0036] The third module is used to perform control processing of the intelligent driving vehicle according to the detection result;
[0037] The training module is used to perform the training of the ResNet50 target detection model:
[0038] The fourth module uses the first data set to train the initial model based on ResNet50 to obtain a pre-trained model;
[0039] The fifth module trains the pre-trained model using the second data set and a convolutional neural network with a zero-filling layer to obtain first training data;
[0040] The sixth module trains the first training data through the attention model to obtain an attention spectrum, and fuses the attention with the feature spectrum of the first training data to obtain second training data;
[0041] The seventh module classifies the second training data to obtain a classification result and a trained model;
[0042] The eighth module performs migration training on the trained model to obtain the target detection model based on ResNet50.
[0043] Another aspect of an embodiment of the present invention provides an electronic device, including a processor and a memory;
[0044] The memory is used to store programs;
[0045] The processor executes the program to implement the method described above.
[0046] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method described above.
[0047] The beneficial effects of the present invention are as follows: the forward vehicle distance detection based on vision has an overall simple structure, is easy to implement, and has a relatively low cost; adding a zero-filling layer enhances recognition at the boundary of the field of view and reduces interference from environmental factors; adding a self-attention mechanism enhances detection of vehicles far away from the field of view and vehicles under occlusion. The problem of inaccurate obstacle detection under conditions such as vehicles of different types and sizes, mutual occlusion between vehicles, interference from environmental factors, and different pitch angles of the road is solved, that is, the detection accuracy within the field of view is optimized. This further prevents the occurrence of failure to identify surrounding vehicles, thereby avoiding traffic accidents and protecting the lives and property of drivers and passengers. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:
[0049] Figure 1 Schematic diagram of an intelligent driving target detection system according to an embodiment of the present invention.
[0050] Figure 2 It is a schematic diagram of the intelligent driving target detection process of an embodiment of the present invention.
[0051] Figure 3 It is a schematic diagram of the initial model training process of an embodiment of the present invention.
[0052] Figure 4 Schematic diagram of zero filling layer processing according to an embodiment of the present invention.
[0053] Figure 5 It is a schematic diagram of a training process based on a self-attention mechanism according to an embodiment of the present invention.
[0054] Figure 6 It is a schematic diagram of the data classification process of an embodiment of the present invention.
[0055] Figure 7 It is a schematic diagram of the model structure of the self-attention according to an embodiment of the present invention.
[0056] Figure 8 It is a model structure diagram when migrating a trained model according to an embodiment of the present invention.
[0057] Fig. 9 Schematic diagram of an intelligent driving target detection device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0058] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. In the subsequent description, the use of suffixes such as "module", "component" or "unit" used to represent elements is only for the purpose of facilitating the description of the present invention, and has no specific meaning in itself. Therefore, "module", "component" or "unit" can be used in a mixed manner. "First", "second" and the like are only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the order of the indicated technical features. In this subsequent description, the continuous numbering of the method steps is for the convenience of review and understanding. In combination with the overall technical solution of the present invention and the logical relationship between the various steps, adjusting the implementation order between the steps will not affect the technical effect achieved by the technical solution of the present invention. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.
[0059] Figure 1Schematic diagram of the intelligent driving target detection system of an embodiment of the present invention. It includes an intelligent driving vehicle 100 and a server 200, wherein the intelligent driving vehicle 100 includes a camera 110, a processing device 120 and a decision system 130. The camera 110, the processing device 120 and the decision system 130 in the driving vehicle 100 complete the target detection and vehicle control processing during intelligent driving, including: according to the target detection request, the camera 110 obtains the target detection image of the intelligent driving vehicle; the processing device 120 obtains the target detection image and uses the target detection model based on ResNet50 to perform target detection to obtain the target detection result; according to the detection result, the control processing of the intelligent driving vehicle is performed. The server 200 is used to perform target detection model training, including: using the first data set to train the initial model based on ResNet50 to obtain a pre-trained model; using the second data set and a convolutional neural network with a zero-filling layer to train the pre-trained model to obtain first training data; training the first training data through the attention model to obtain an attention spectrum, fusing the attention with the feature spectrum of the first training data to obtain second training data; classifying the second training data to obtain a classification result and a trained model; performing migration training on the trained model to obtain a target detection model based on ResNet50.
[0060] In some embodiments, the server 200 may also be a device capable of performing a target detection model based on ResNet50, such as a PC (personal computer), a workstation, etc.
[0061] Reference Figure 2 , Figure 2 : is a schematic diagram of the intelligent driving target detection process of an embodiment of the present invention, which includes but is not limited to steps S100 to S800:
[0062] S100 to S300 are the target detection process, including:
[0063] S100, acquiring a target detection image of the intelligent driving vehicle according to a target detection request;
[0064] In some embodiments, the target detection image is an image obtained during vision-based forward vehicle distance detection, which includes images in front of and around the vehicle.
[0065] S200, performing target detection on the target detection image using a target detection model based on ResNet50 to obtain a target detection result;
[0066] S300: Execute control processing of the intelligent driving vehicle according to the detection result.
[0067] In some embodiments, the detection results are sent to a decision system of the intelligent driving vehicle, and the decision system performs vehicle control processing based on the detection results. For example, when an obstacle is detected in front of the vehicle, the vehicle is controlled to stop.
[0068] Steps S400 to S800 are a target detection model training flow for target detection, including:
[0069] S400: Use the first data set to train an initial model based on ResNet50 to obtain a pre-trained model.
[0070] In some embodiments, reference Figure 3 The schematic diagram of the initial model training process shown includes but is not limited to steps S410 to S420:
[0071] S410, constructing a target task model, using a non-first data set for training, and obtaining weight parameters, wherein the target task model is based on ResNet50;
[0072] S420, adding the weight parameters to the target task model to obtain a pre-trained model.
[0073] Exemplarily, deep transfer learning for the network requires selecting a suitable training model before training. An appropriate training model can significantly reduce the cost of the task, thereby accelerating the convergence speed and improving the accuracy of the model. In some embodiments, a model trained on the KITTI dataset is used. Before training, the model parameters are trained on other datasets that are the same as the target task, and then the weight parameters are transferred to the target task model as the initial training parameters. This approach may find the best model for the solution faster than randomly initializing the parameters.
[0074] S500: Use the second data set and a convolutional neural network with a zero-filling layer to train the pre-trained model to obtain first training data.
[0075] In some embodiments, when a convolutional neural network is used to perform convolution processing on an image in the second data set, the border of the image is zero-filled through a zero-filling layer so that the image is convolved at least twice by the convolution kernel.
[0076] For example, reference Figure 4The zero padding layer processing diagram shown in the figure shows the addition of the zero padding layer. Its significance is that when the boundary is not filled with "0" during neural network training, the convolution kernel only performs convolution calculation on the boundary once, which may miss some important information, while the middle part of the image can be convolved by the convolution kernel multiple times, which may reduce the model's reference to the edge of the image to a certain extent. The edge features of the image can reflect some important feature signals in some scenarios. Filling the boundary of the image with "0" is more conducive to the acquisition of boundary information.
[0077] S600, training the first training data through the attention model to obtain an attention spectrum, and fusing the attention with the feature spectrum of the first training data to obtain second training data.
[0078] In some embodiments, reference Figure 5 The training process diagram based on the self-attention mechanism shown includes but is not limited to steps S610 to S630:
[0079] S610, using the first training data as an input of the attention model to calculate an attention spectrum;
[0080] S620, performing weighted processing on the attention spectrum and the feature spectrum of the first data to obtain a weighted feature spectrum;
[0081] S630, superimposing the weighted feature spectrum with the feature spectrum of the first data to obtain second training data, wherein the second training data is an attention feature spectrum.
[0082] S700, classify the second training data to obtain a classification result and a trained model.
[0083] In some embodiments, reference Figure 6 The data classification process diagram shown includes but is not limited to steps S710 to S730:
[0084] S710 uses a fully connected layer to receive the attention feature spectrum, where the fully connected layer includes a first fully connected layer, a second fully connected layer, and a third fully connected layer that are connected in sequence, wherein the first fully connected layer is used to receive the attention feature spectrum;
[0085] S720, performing dimensional transformation and spatial cascade processing on the attention feature spectrum through a second fully connected layer, and merging the attention feature spectrum after the dimensional transformation and spatial cascade processing with the received attention feature spectrum;
[0086] S730, performing classification processing on the merged attention feature spectrum through the third fully connected layer to obtain a classification result and a trained model.
[0087] In some embodiments, reference Figure 7 The model structure diagram of self-attention is shown in the figure. The model structure uses the self-attention mechanism to take the output of the last convolutional layer of the convolutional layer as the input of this attention model, and calculates the corresponding attention spectrum. Then the original feature spectrum is weighted using the attention spectrum. Then the weighted feature spectrum is superimposed on the original feature spectrum. Then the fused attention feature spectrum is passed to the fully connected layer. In the second fully connected layer, the attention feature spectrum after dimensional transformation and spatial cascading is merged with the attention feature spectrum of the first fully connected layer, and then sent to the classification layer to predict the image category.
[0088] S800: Perform migration training on the trained model to obtain a target detection model based on ResNet50.
[0089] In some embodiments, a pre-trained model based on ResNet50 is subjected to transfer training to obtain a target detection model, wherein the model structure used for transfer training includes a convolutional layer, a maximum pooling layer, a self-attention layer and a fully connected layer in sequence, wherein the convolutional layer is a convolutional layer of an initial model based on ResNet50, wherein the maximum pooling layer is a 2×2 sliding window, wherein the self-attention layer performs a transition between the convolutional layer and the fully connected layer through a flatten function, wherein the first fully connected layer in the fully connected layer uses ReLU as an activation function, the second fully connected layer uses a batch normalization function as an activation function, and the third fully connected layer uses Softmax as an activation function.
[0090] In some embodiments, reference Figure 8 The model structure diagram of the migration of the trained model shown in the figure, in which the convolution layer is mainly from the ResNet50 network on the kitti dataset. A maximum pooling layer with a sliding window size of 2×2 is added after the convolution layer. Next is the self-attention layer, and the flatten function method is added after the self-attention layer. Three calculation layers are added after the flatten method. The first layer is a fully connected layer with ReLU as the activation function, the second layer is the Batch Normalization function, and the third layer is a fully connected layer with Softmax as the activation function. It is put into use in the perception decision system test of the self-driving car.
[0091] Fig. 9 Schematic diagram of an intelligent driving target detection device according to an embodiment of the present invention. The device includes a recognition module and a training module, wherein the recognition module includes a first module 910, a second module 920 and a third module 930, wherein the training module includes a fourth module 940, a fifth module 950, a sixth module 960, a seventh module 970 and an eighth module 980.
[0092] The first module is used to obtain a target detection image of the intelligent driving vehicle according to a target detection request; the second module is used to perform target detection on the target detection image using a target detection model based on ResNet50 to obtain a target detection result; the third module is used to perform control processing of the intelligent driving vehicle according to the detection result;
[0093] Among them, the training module is used to perform the training of the ResNet50 target detection model: the fourth module uses the first data set to train the initial model based on ResNet50 to obtain a pre-trained model; the fifth module uses the second data set and a convolutional neural network with a zero-filling layer to train the pre-trained model to obtain the first training data; the sixth module trains the first training data through the attention model to obtain the attention spectrum, and fuses the attention with the feature spectrum of the first training data to obtain the second training data; the seventh module classifies the second training data to obtain the classification result and the trained model; the eighth module performs migration training on the trained model to obtain a target detection model based on ResNet50.
[0094] Exemplarily, with the cooperation of the first module, the second module and the third module in the device, and the training module including the fourth module, the fifth module, the sixth module, the seventh module and the eighth module, the embodiment device can implement any of the aforementioned intelligent driving target detection methods, that is, according to the target detection request, obtaining the target detection image of the intelligent driving vehicle; performing target detection on the target detection image using the target detection model based on ResNet50 to obtain the target detection result; performing control processing of the intelligent driving vehicle according to the detection result; the training of the target detection model based on ResNet50 includes: using the first data set to train the initial model based on ResNet50 to obtain a pre-trained model; using the second data set and a convolutional neural network with a zero-filling layer to train the pre-trained model to obtain first training data; training the first training data through the attention model to obtain an attention spectrum, and fusing the attention with the feature spectrum of the first training data to obtain second training data; classifying the second training data to obtain the classification result and the trained model; performing migration training on the trained model to obtain a target detection model based on ResNet50. The beneficial effects of the present invention are as follows: the forward vehicle distance detection based on vision has an overall simple structure, is easy to implement, and has a relatively low cost; adding a zero-filling layer enhances recognition at the boundary of the field of view and reduces interference from environmental factors; adding a self-attention mechanism enhances detection of vehicles far away from the field of view and vehicles under occlusion. The problem of inaccurate obstacle detection under conditions such as vehicles of different types and sizes, mutual occlusion between vehicles, interference from environmental factors, and different pitch angles of the road is solved, that is, the detection accuracy within the field of view is optimized. This further prevents the occurrence of failure to identify surrounding vehicles, thereby avoiding traffic accidents and protecting the lives and property of drivers and passengers.
[0095] An embodiment of the present invention further provides an electronic device, the electronic device comprising a processor and a memory;
[0096] The memory stores a program;
[0097] The processor executes the program to execute the aforementioned intelligent driving target detection method; the electronic device has the function of carrying and running the software system for intelligent driving target detection provided by an embodiment of the present invention, for example, a personal computer, a minicomputer, a main frame, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or communicates with a charged particle tool or other imaging device, etc.
[0098] An embodiment of the present invention further provides a computer-readable storage medium, which stores a program, and the program is executed by a processor to implement the intelligent driving target detection method as described above.
[0099] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.
[0100] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the aforementioned intelligent driving target detection method.
[0101] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features described may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.
[0102] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.
[0103] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.
[0104] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.
[0105] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0106] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0107] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.
[0108] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. A method for intelligent driving target detection, It is characterized in that include: According to the target detection request, obtain the target detection image of the intelligent driving vehicle; Performing target detection on the target detection image using a target detection model based on ResNet50 to obtain a target detection result; According to the detection result, executing control processing of the intelligent driving vehicle; The training of the ResNet50-based target detection model includes: The first data set is used to train the initial model based on ResNet50 to obtain a pre-trained model; The pre-trained model is trained using a second data set and a convolutional neural network with a zero-filling layer to obtain first training data; Training the first training data through an attention model to obtain an attention spectrum, and fusing the attention with a feature spectrum of the first training data to obtain second training data; Classifying the second training data to obtain a classification result and a trained model; The trained model is subjected to migration training to obtain the ResNet50-based target detection model.
2. The intelligent driving target detection method according to claim 1, It is characterized in that The first data set is used to train the initial model based on ResNet50 to obtain a pre-trained model, including: Build the target task model and use the non-first data set for training to obtain weight parameters, where the target task model is based on ResNet50; The weight parameters are added to the target task model to obtain the pre-trained model.
3. The intelligent driving target detection method according to claim 1, It is characterized in that The pre-training model is trained using the second data set and a convolutional neural network with a zero-filling layer to obtain first training data, including: When the convolution neural network is used to perform convolution processing on the images in the second data set, the borders of the images are zero-filled through a zero-filling layer so that the images are convolved at least twice by the convolution kernel.
4. The intelligent driving target detection method according to claim 1, It is characterized in that The step of training the first training data by using the attention model to obtain an attention spectrum, and fusing the attention with the feature spectrum of the first training data to obtain second training data includes: Using the first training data as input of the attention model, calculating an attention spectrum; Performing weighted processing on the attention spectrum and the feature spectrum of the first data to obtain a weighted feature spectrum; The weighted feature spectrum is superimposed on the feature spectrum of the first data to obtain the second training data, wherein the second training data is an attention feature spectrum.
5. The intelligent driving target detection method according to claim 4, It is characterized in that The classifying the second training data to obtain a classification result and a trained model includes: A fully connected layer is used to receive the attention feature spectrum, and the fully connected layer includes a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence, wherein the first fully connected layer is used to receive the attention feature spectrum; Performing dimension transformation and spatial cascade processing on the attention feature spectrum through the second fully connected layer, and merging the attention feature spectrum after dimension transformation and spatial cascade processing with the received attention feature spectrum; The merged attention feature spectrum is classified through the third fully connected layer to obtain the classification result and the trained model.
6. The intelligent driving target detection method according to claim 5, It is characterized in that The step of performing migration training on the trained model to obtain the ResNet50-based target detection model includes: The pre-trained model based on ResNet50 is subjected to migration training to obtain a target detection model, wherein the model structure used for migration training includes a convolutional layer, a maximum pooling layer, a self-attention layer and a fully connected layer in sequence, wherein the convolutional layer is a convolutional layer of an initial model based on ResNet50, wherein the maximum pooling layer is a 2×2 sliding window, wherein the self-attention layer performs a transition between the convolutional layer and the fully connected layer through a flatten function, wherein the first fully connected layer in the fully connected layer uses ReLU as an activation function, the second fully connected layer uses a batch normalization function as an activation function, and the third fully connected layer uses Softmax as an activation function.
7. The intelligent driving target detection method according to claim 1, It is characterized in that The method further comprises: The detection result is sent to a decision-making system of the intelligent driving vehicle, and the decision-making system performs vehicle control processing based on the detection result.
8. An intelligent driving target detection device, It is characterized in that include: An identification module and a training module, wherein the identification module includes a first module, a second module and a third module, and the training module includes a fourth module, a fifth module, a sixth module, a seventh module and an eighth module; The first module is used to obtain a target detection image of the intelligent driving vehicle according to a target detection request; The second module is used to perform target detection on the target detection image using a target detection model based on ResNet50 to obtain a target detection result; The third module is used to perform control processing of the intelligent driving vehicle according to the detection result; The training module is used to perform the training of the ResNet50 target detection model: The fourth module uses the first data set to train the initial model based on ResNet50 to obtain a pre-trained model; The fifth module trains the pre-trained model using the second data set and a convolutional neural network with a zero-filling layer to obtain first training data; The sixth module trains the first training data through the attention model to obtain an attention spectrum, and fuses the attention with the feature spectrum of the first training data to obtain second training data; The seventh module classifies the second training data to obtain a classification result and a trained model; The eighth module performs migration training on the trained model to obtain the target detection model based on ResNet50.
9. An electronic device, It is characterized in that including a processor and a memory; The memory is used to store programs; The processor executes the program to implement the intelligent driving target detection method according to any one of claims 1 to 7.
10. A computer-readable storage medium, It is characterized in that The storage medium stores a program, and the program is executed by a processor to implement the intelligent driving target detection method according to any one of claims 1 to 7.