Logistics vehicle access management method, device and equipment and storage medium

Abstract

The present application relates to the technical field of logistics, and particularly relates to a logistics vehicle passing management method, device and equipment and a storage medium, comprising collecting vehicle images of a logistics vehicle team, and generating a first training set and a second training set according to the vehicle images; the first training set comprises license plate annotation features; the second training set comprises license plate number character annotation features; a first pre-recognition model is built based on a DBNet model; a second pre-recognition model is built based on an ABINet model; the first training set is input into the first pre-recognition model for training to obtain a license plate recognition model; the second training set is input into the second pre-recognition model for training to obtain a license plate number recognition model; real-time vehicle image data is acquired, and the license plate recognition model and the license plate number recognition model are used for recognition to obtain a license plate recognition result; the license plate number is visually recognized through the construction of the license plate recognition model and the license plate number recognition model, the vehicle identity is identified by the collection and distribution center, and the safety of passing is ensured.

Description

Technical Field

[0001] This invention relates to the field of logistics technology, and in particular to a method, apparatus, equipment and storage medium for managing the passage of logistics vehicles. Background Technology

[0002] Currently, logistics companies are developing on a large scale, and the number of large parcel distribution centers is also increasing. However, due to the large number of logistics vehicles, existing distribution centers are unable to manage the passage of all logistics vehicles, which may lead to the problem of illegal vehicles entering the distribution centers.

[0003] It is evident that existing technologies still need improvement and enhancement. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, the present invention aims to provide a method, device, equipment and storage medium for the passage management of logistics vehicles, which can automatically collect the license plate information of vehicles requesting passage and identify the vehicles requesting passage based on the vehicle information entered by each distribution center, thereby achieving the purpose of logistics vehicle passage management.

[0005] The first aspect of this invention provides a method for managing the passage of logistics vehicles, comprising: acquiring vehicle images of a logistics fleet, and generating a first training set and a second training set based on the vehicle images; the first training set including license plate annotation features; the second training set including license plate number character annotation features; constructing a first pre-recognition model based on a DBNet model; constructing a second pre-recognition model based on an ABINet model; inputting the first training set into the first pre-recognition model for training to obtain a license plate recognition model; inputting the second training set into the second pre-recognition model for training to obtain a license plate number recognition model; acquiring real-time vehicle image data, and performing recognition through the license plate recognition model and the license plate number recognition model to obtain a license plate recognition result.

[0006] Optionally, in a first implementation of the first aspect of the present invention, the step of collecting vehicle images of a logistics fleet and generating a first training set and a second training set based on the vehicle images includes: filtering the vehicle images to retain vehicle images containing license plate features, and generating a vehicle sample set based on the filtered vehicle images; performing a first annotation process on the vehicle sample set to label the license plate features in the images of the vehicle sample set to obtain the first training set; and performing a second annotation process on the vehicle sample set to label the license plate number features in the images of the vehicle sample set to obtain the second training set.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the construction of the first pre-identification model based on the DBNet model includes: constructing an infrastructure based on the DBNet model, the infrastructure including a Backbone network, an FPN network, and a Head network; inputting a MobileNetV3 model into the infrastructure so that the MobileNetV3 model serves as the Backbone network of the infrastructure; the MobileNetV3 model including a depthwise separable convolutional structure, an inverse residual structure, and an SE module; and obtaining the first pre-identification model composed of the DBNet model and the MobileNetV3 model.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the step of inputting the first training set into the first pre-recognition model for training to obtain a license plate recognition model includes: acquiring first preset weight ratio data, allocating weights to the first training set to obtain a license plate training set, a license plate verification set, and a license plate test set; inputting the license plate training set into the first pre-recognition model for training to obtain multiple first training models with different results; inputting the license plate verification set into the multiple first training models for verification to obtain multiple first prediction results; when the first prediction result meets the first preset condition, inputting the license plate test set into the first training model, scoring each first training model, and selecting the first training model with the highest score as the license plate recognition model.

[0009] Optionally, in the fourth implementation of the first aspect of the present invention, the step of inputting the second training set into the second pre-recognition model for training to obtain a license plate number recognition model includes: iteratively training the second pre-recognition model using a learning rate cosine decay strategy; obtaining second preset weight ratio data and weighting the second training set to obtain a license plate number training set, a license plate number verification set, and a license plate number test set; inputting the license plate number training set into the second pre-recognition model for training to obtain multiple second training models with different results; inputting the license plate number verification set into the multiple second training models for verification to obtain multiple second prediction results; when the second prediction result meets the second preset condition, inputting the license plate number test set into the second training model, scoring each second training model, and selecting the second training model with the highest score as the license plate number recognition model.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, the step of acquiring real-time vehicle image data and recognizing it through a license plate recognition model and a license plate number recognition model to obtain a license plate recognition result includes: collecting license plate number information of the logistics fleet and establishing a license plate number information database based on the license plate number information; inputting real-time vehicle image data into the license plate recognition model to obtain a license plate recognition result; inputting the license plate recognition result into the license plate number recognition model to obtain a license plate number recognition result; performing matching processing between the license plate number recognition result and the license plate number information database; and issuing a passage instruction if the license plate number recognition result matches the license plate number information in the license plate number information database.

[0011] Optionally, in the sixth implementation of the first aspect of the present invention, the step of matching the license plate number recognition result with the license plate number information database, and issuing a passage instruction if the license plate number recognition result matches the license plate number information in the license plate number information database, includes: extracting character information from the license plate number recognition result; using the character information as a filtering condition to filter the information in the license plate number information database; if the license plate number information database contains a filtering result, issuing a passage instruction; if the license plate number information database does not contain a filtering result, issuing an illegal passage instruction.

[0012] A second aspect of the present invention provides a traffic management device for logistics vehicles, comprising: a data acquisition module for acquiring vehicle images of a logistics fleet and generating a first training set and a second training set based on the vehicle images; the first training set including license plate annotation features; the second training set including license plate number character annotation features; a first modeling module for building a first pre-recognition model based on a DBNet model; a second modeling module for building a second pre-recognition model based on an ABINet model; a first training module for inputting the first training set into the first pre-recognition model for training to obtain a license plate recognition model; a second training module for inputting the second training set into the second pre-recognition model for training to obtain a license plate number recognition model; and a recognition module for acquiring real-time vehicle image data and performing recognition through the license plate recognition model and the license plate number recognition model to obtain a license plate recognition result.

[0013] Optionally, in a first implementation of the second aspect of the present invention, the acquisition module includes: a filtering unit, used to filter vehicle images, retain vehicle images containing license plate features, and generate a vehicle sample set based on the filtered vehicle images; a first annotation unit, used to perform a first annotation process on the vehicle sample set, and to label the license plate features in the images of the vehicle sample set to obtain a first training set; and a second annotation unit, used to perform a second annotation process on the vehicle sample set, and to label the license plate number features in the images of the vehicle sample set to obtain a second training set.

[0014] Optionally, in a second implementation of the second aspect of the present invention, the first modeling module includes: a construction unit for constructing an infrastructure based on a DBNet model, the infrastructure including a Backbone network, an FPN network, and a Head network; a reconstruction unit for inputting a MobileNetV3 model into the infrastructure so that the MobileNetV3 model serves as the Backbone network of the infrastructure; the MobileNetV3 model including a depthwise separable convolutional structure, an inverse residual structure, and an SE module; and a generation unit for obtaining a first pre-identification model composed of the DBNet model and the MobileNetV3 model.

[0015] Optionally, in a third implementation of the second aspect of the present invention, the first training module includes: a first allocation unit, configured to acquire first preset weight ratio data and allocate weights to the first training set to obtain a license plate training set, a license plate verification set, and a license plate test set; a first iteration unit, configured to input the license plate training set into a first pre-recognition model for training to obtain multiple first training models with different results; a first verification unit, configured to input the license plate verification set into multiple first training models for verification to obtain multiple first prediction results; and a first scoring unit, configured to input the license plate test set into the first training model when the first prediction result meets the first preset condition, and score each first training model, and select the first training model with the highest score as the license plate recognition model.

[0016] Optionally, in a fourth implementation of the second aspect of the present invention, the second training module includes: a setting unit, used to iteratively train the second pre-recognition model using a learning rate cosine decay strategy; a second allocation unit, used to obtain second preset weight ratio data and allocate weights to the second training set to obtain a license plate number training set, a license plate number verification set, and a license plate number test set; a second iteration unit, used to input the license plate number training set into the second pre-recognition model for training to obtain multiple second training models with different results; a second verification unit, used to input the license plate number verification set into the multiple second training models for verification to obtain multiple second prediction results; and a second scoring unit, used to input the license plate number test set into the second training model when the second prediction result meets the second preset condition, and score each second training model, taking the second training model with the highest score as the license plate number recognition model.

[0017] Optionally, in a fifth implementation of the second aspect of the present invention, the recognition module includes: a database building unit, used to collect license plate number information of the logistics fleet and build a license plate number information database based on the license plate number information; a first input unit, used to input real-time vehicle image data into the license plate recognition model to obtain a license plate recognition result; a second input unit, used to input the license plate recognition result into the license plate number recognition model to obtain a license plate number recognition result; and a matching unit, used to match the license plate number recognition result with the license plate number information database, and if the license plate number recognition result matches the license plate number information in the license plate number information database, then a passage instruction is issued.

[0018] Optionally, in a sixth implementation of the second aspect of the present invention, the matching unit includes: an extraction unit for extracting character information from the license plate number recognition result; and a filtering unit for filtering the information in the license plate number information database using the character information as a filtering condition; if the license plate number information database contains a filtering result, a passage instruction is issued; if the license plate number information database does not contain a filtering result, an illegal passage instruction is issued.

[0019] A third aspect of the present invention provides a passage management device for logistics vehicles, the passage management device comprising: a memory and at least one processor, the memory storing instructions; the at least one processor calling the instructions in the memory to cause the passage management device for logistics vehicles to execute the various steps of the passage management method for logistics vehicles described in any of the preceding claims.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the above-described method for managing the passage of logistics vehicles.

[0021] In the technical solution of this invention, visual recognition of license plate numbers is achieved by constructing a license plate recognition model and a license plate number recognition model. Real-time vehicle image data is input into the license plate recognition model, which identifies and calibrates the license plate features in the real-time image. The license plate recognition model also denoises the content of the real-time image, selecting images containing only the license plate area from the real-time image, thus reducing the difficulty of extracting the license plate number. Subsequently, the output of the license plate recognition model is input into the license plate number recognition model, which extracts the characters from the license plate image, thereby accurately obtaining the license plate number of the relevant vehicle. The license plate recognition result facilitates the identification of vehicles by the distribution center, ensuring the safety of vehicle passage and reducing the management difficulty of passing vehicles by the distribution center. Attached Figure Description

[0022] Figure 1 This is a first flowchart of a method for managing the passage of logistics vehicles provided in an embodiment of the present invention;

[0023] Figure 2 This is a second flowchart of a method for managing the passage of logistics vehicles provided in an embodiment of the present invention;

[0024] Figure 3 This is a third flowchart of a method for managing the passage of logistics vehicles provided in an embodiment of the present invention;

[0025] Figure 4 This is a fourth flowchart of a method for managing the passage of logistics vehicles provided in an embodiment of the present invention;

[0026] Figure 5 A fifth flowchart of the logistics vehicle access management method provided in an embodiment of the present invention;

[0027] Figure 6 A schematic diagram of a passage management device for logistics vehicles provided in an embodiment of the present invention;

[0028] Figure 7 Another structural schematic diagram of the passage management device for logistics vehicles provided in an embodiment of the present invention;

[0029] Figure 8 A schematic diagram of the structure of a logistics vehicle access management device provided in an embodiment of the present invention. Detailed Implementation

[0030] This invention provides a method, device, equipment, and storage medium for managing the passage of logistics vehicles. The invention achieves visual recognition of license plate numbers by constructing a license plate recognition model and a license plate number recognition model. Real-time vehicle image data is input into the license plate recognition model, which identifies and calibrates the license plate features in the real-time image. The model also denoises the real-time image, selecting only the license plate area to reduce the difficulty of extracting the license plate number. The output of the license plate recognition model is then input into the license plate number recognition model, which extracts characters from the license plate image, thereby accurately obtaining the license plate number of the relevant vehicle. This license plate recognition result facilitates vehicle identification at the distribution center, ensuring vehicle passage safety and reducing the management difficulty for the distribution center.

[0031] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the logistics vehicle traffic management method in this invention includes:

[0033] 101. Collect vehicle images of the logistics fleet, and generate a first training set and a second training set based on the vehicle images; the first training set includes license plate annotation features; the second training set includes license plate number character annotation features;

[0034] In this embodiment, the acquired vehicle images contain various features, such as vehicle body, railings, pedestrians, and drivers. To obtain accurate annotation information for license plates and license plate numbers, each vehicle image needs to be annotated. The regions in the image containing license plates and license plate numbers are identified and marked, thereby obtaining annotated data containing license plate and license plate number regions. This generates the first and second training sets for subsequent training. For ease of understanding, the following example illustrates this: the vehicle images are input into image annotation software (e.g., but not limited to Labelme, BeaverDam, and Imglab tools). Then, the annotation tools in the image annotation software are used to select the license plate and license plate number characters in the image using closed lines to obtain images containing the license plate and license plate number regions. After annotation is completed, the corresponding annotation data is generated by the image annotation software, which is then used to train the subsequent model.

[0035] It should be noted that vehicle images can be obtained through the gate system of the distribution center. By connecting to each gate system, images of passing vehicles can be obtained. These images include images of the front and rear areas of the vehicle, and both images contain license plate features.

[0036] 102. Build the first pre-identification model based on the DBNet model;

[0037] In this embodiment, DBNet is a segmentation-based text detection algorithm that introduces a differentiable binarization module into the segmentation model, enabling the segmentation model to perform binarization through an adaptive threshold. The DBNet model introduces a sigmoid function with a k-gradient gain factor as the binarization equation to predict the probability map of the text and the probability map of the threshold. The threshold plus DB is used to increase the gradient of the prediction error, thereby improving the text detection effect.

[0038] 103. Build a second pre-identification model based on the ABINet model;

[0039] In this embodiment, the ABINet model is a natural scene recognition model with automatic, bidirectional, and iterative characteristics. It can achieve unsupervised training on unlabeled data. ABINet has strong recognition capabilities on low-quality images and achieves state-of-the-art performance on some benchmarks.

[0040] It should be noted that the ABINet model includes an input terminal, a visual model ABIVisionModel, a text model ABILanguageDecoder, a loss function, and inference. The workflow of the ABINet model is as follows: 1. Obtain an image with license plate numbers through the input terminal, and the label is to convert each character into a one-to-one corresponding 38 numbers, with a maximum of 26 numbers, where 37 is the terminator.

[0041] 2. The input text image is processed by ResNet+Transformer to extract features. The output result is a tensor (8, 512, 8, 32), where 8 represents the batch size, 512 represents the number of feature maps, and 8*32 represents the size of the feature maps.

[0042] 3. Apply Position Attention to the output to obtain logits, which has dimensions of tensor(8,26,38), where 8 represents the batch size and 26 represents the output length of 26. Perform a 38 classification task at each position.

[0043] 4. Correction is performed using the text model ABILanguageDecoder to check whether the output of the visual model is reasonable. Each letter is checked one by one, and the check is iterated n times (default is 3) to improve the model accuracy.

[0044] 5. Combine the results from the visual model and the text model;

[0045] 6. The final output is compensated by using a loss function and inference to obtain the text recognition result.

[0046] 104. Input the first training set into the first pre-recognition model for training to obtain the license plate recognition model;

[0047] In this embodiment, a first training set containing license plate annotation features is input into the first pre-recognition model, which then undergoes iterative training. Based on different output results, the model continuously optimizes itself to obtain the optimal license plate recognition model. The license plate recognition model is used to identify the license plate, thereby extracting the license plate area from the vehicle image containing multiple features. This facilitates the subsequent acquisition of the license plate number and reduces the amount of subsequent data processing.

[0048] 105. Input the second training set into the second pre-recognition model for training to obtain the license plate number recognition model;

[0049] In this embodiment, a second training set containing license plate number annotation features is input into the second pre-recognition model, enabling iterative training of the second pre-recognition model. Based on different output results, the model continuously optimizes itself to obtain the optimal license plate number recognition model. The license plate number recognition model is used to identify the license plate number, extracting the characters from the license plate image to facilitate the verification of the legality of passing vehicles, preventing illegal vehicles from entering the distribution center, and achieving the purpose of logistics vehicle traffic management.

[0050] 106. Acquire real-time vehicle image data, and perform recognition using a license plate recognition model and a license plate number recognition model to obtain the license plate recognition result;

[0051] In this embodiment, after the first and second pre-identification models are trained, real-time images of the distribution center's gate can be acquired. The real-time vehicle image data is first input into the license plate recognition model, which identifies and calibrates the license plate features in the real-time images to obtain an output result containing only license plate features. The license plate recognition model also denoises the real-time images containing clutter, selecting images containing only license plate areas from the real-time images, reducing the difficulty of extracting license plate numbers. Subsequently, the output result of the license plate recognition model is input into the license plate number recognition model, which extracts characters from the license plate images in the output results and selects the license plate numbers from the license plate images, thereby accurately obtaining the license plate numbers of the relevant vehicles. The license plate recognition results facilitate the distribution center's identification of vehicles, preventing illegal vehicles from entering the distribution center and reducing the management difficulty of the distribution center for passing vehicles.

[0052] In this embodiment of the invention, a license plate recognition model and a license plate number recognition model are constructed to achieve visual recognition of license plate numbers. Real-time vehicle image data is input into the license plate recognition model, which identifies and calibrates the license plate features in the real-time image. The license plate recognition model also denoises the content of the real-time image, selecting images containing only the license plate area from the real-time image, thus reducing the difficulty of extracting the license plate number. Subsequently, the output of the license plate recognition model is input into the license plate number recognition model, which extracts the characters from the license plate image, thereby accurately obtaining the license plate number of the relevant vehicle. The license plate recognition result facilitates the identification of vehicles by the distribution center, ensuring the safety of vehicle passage and reducing the management difficulty of passing vehicles by the distribution center.

[0053] Please see Figure 2 The second embodiment of the logistics vehicle traffic management method in this invention includes:

[0054] 201. Filter the vehicle images, retain the vehicle images containing license plate features, and generate a vehicle sample set based on the filtered vehicle images.

[0055] In this embodiment, since vehicle images are easily affected by factors such as shooting angle, lighting, and obstacles, there may be some invalid images in the vehicle images. By filtering, images without license plates, images with incomplete license plates, or images with blurry license plates are removed, and only images with clear license plates are retained as the data of the vehicle sample set. Filtering the vehicle sample set data improves the effect of model training and provides good data support for model training.

[0056] 202. Perform a labeling process on the vehicle sample set, and label the license plate features in the images of the vehicle sample set to obtain the first training set;

[0057] In this embodiment, the Labelme tool is used to annotate the images in the vehicle sample set, the license plate area is selected, and the license plate area is annotated by means of boxes, lines, etc., and the area containing license plate features is selected to obtain the annotated data containing license plate features.

[0058] 203. Perform secondary annotation processing on the vehicle sample set, and label the license plate number features in the images of the vehicle sample set to obtain the second training set;

[0059] In this embodiment, the Labelme tool is used to annotate the images in the vehicle sample set. First, the license plate area is selected and annotated using methods such as boxes and line segments. From this, areas containing license plate features are selected to obtain annotated data containing license plate features. Then, a second annotation is performed based on the annotated data containing license plate features. The small area of ​​the image containing only license plate features is annotated with characters, and the license plate number on the license plate is annotated. From this, areas containing license plate number features are selected to obtain annotated data containing license plate number features.

[0060] Please see Figure 3 The third embodiment of the logistics vehicle traffic management method in this invention includes:

[0061] 301. Construct an infrastructure based on the DBNet model, wherein the infrastructure includes a Backbone network, an FPN network, and a Head network;

[0062] In this embodiment, the Backbone network is responsible for extracting image features; the FPN network is used to enhance features using a feature pyramid structure; and the Head network is used to calculate the probability map of text regions.

[0063] The Backbone network's workflow is as follows: Assuming the input image is [1, 3, 640, 640], it enters the Backbone network. After a series of convolutional calculations and downsampling, it outputs four scale feature maps: original... Figure 1 / 4 Feature map: [1,16,160,160], original Figure 1 / 8 Feature map: [1,24,80,80], original Figure 1 / 16 Feature map: [1,56,40,40] and original Figure 1 / 32 Feature map: [1,480,20,20].

[0064] The workflow of the FPN network is as follows: The FPN network is a convolutional network that efficiently extracts features from various dimensions of an image. The input to the FPN network is the output of the Backbone network. After calculation by the FPN network, the height and width of the output feature map are 1 / 4 of the original image, i.e., [1, 256, 160, 160], as detailed below:

[0065] 1. Calculate the new feature map

[0066] New 1 / 32 feature map: [1,64,20,20];

[0067] New 1 / 16 feature map: [1,64,40,40], original Figure 1 The 1 / 16 feature map is added to the upsampled version of the new 1 / 32 feature map;

[0068] New 1 / 8 feature map: [1,64,80,80], original Figure 1 The / 8 feature map is added to the upsampled map of the new 1 / 16 feature map;

[0069] New 1 / 4 feature map: [1,64,160,160], original Figure 1 Add the / 4 feature map to the upsampled map of the new 1 / 8 feature map.

[0070] 2. Feature map fusion

[0071] New 1 / 32 feature map: [1,64,20,20], convolution + 8x upsampling, [1,64,160,160]

[0072] New 1 / 16 feature map: [1,64,40,40], convolution + 4x upsampling, [1,64,160,160]

[0073] New 1 / 8 feature map: [1,64,80,80], convolution + 2x upsampling, [1,64,160,160]

[0074] New 1 / 4 feature map: [1,64,160,160], convolution, [1,64,160,160]

[0075] The feature maps are merged as follows: [1,256,160,160]. The 1 / 4, 1 / 8, 1 / 16, and 1 / 32 feature maps are merged according to the channel layer.

[0076] The Head network works as follows: it calculates the text region probability map, the text region threshold map, and the text region binary map; the Head network upsamples the FPN features, mapping the FPN features from 1 / 4 of their original size to the original image size, and finally merges the three generated images, outputting [1,3,640,640].

[0077] 302. Input the MobileNetV3 model into the infrastructure so that the MobileNetV3 model serves as the backbone network of the infrastructure; the MobileNetV3 model includes a deep separable convolutional structure, an inverse residual structure, and an SE module;

[0078] 303. Obtain the first pre-identification model composed of the DBNet model and the MobileNetV3 model;

[0079] In this embodiment, using the lightweight backbone network MobileNetV3 model as the backbone network of the DBNet model greatly improves the overall computing speed of the model, making the recognition process more efficient and the recognition results more accurate. The MobileNetV3 model has fewer parameters and is faster. Furthermore, MobileNetV3 combines the depthwise separating convolutional structure of MobileNetV1, the inverse residual structure of MobileNetV2, and the SE module, and uses neural architecture search to search for the network configuration and parameters.

[0080] Please see Figure 4 The third embodiment of the logistics vehicle traffic management method in this invention includes:

[0081] 401. Obtain the first preset weight ratio data, and perform weight allocation on the first training set to obtain the license plate training set, license plate verification set and license plate test set;

[0082] In this embodiment,

[0083] 402. Input the license plate training set into the first pre-recognition model for training to obtain multiple first training models with different results;

[0084] 403. Input the license plate verification set into multiple first training models for verification to obtain multiple first prediction results;

[0085] 404. When the first prediction result meets the first preset condition, the license plate test set is input into the first training model, and each first training model is scored. The first training model with the highest score is taken as the license plate recognition model.

[0086] In this embodiment, a license plate training set, a license plate verification set, and a license plate test set are input into a first pre-recognition model for model training to obtain a trained license plate recognition model for license plate recognition. The first pre-recognition model is iteratively trained using the license plate training set to obtain a trained first training model. Then, license plate verification data is randomly obtained from the license plate verification set and input into the first training model for data prediction to obtain a prediction result. When the prediction result meets the preset optimization conditions, the first training model is tested and scored. The license plate test set is input into the first training model to obtain the training model with the highest score, and the training model with the highest score is used as the license plate recognition model.

[0087] In this embodiment, the proportions of the license plate training set, license plate verification set, and license plate test set are 60%, 30%, and 10%, respectively.

[0088] 405. The second pre-identification model is iteratively trained using a learning rate cosine decay strategy;

[0089] In this embodiment, the learning rate cosine decay strategy controls the update speed of parameters when training the second pre-identification model. When the learning rate is small, it greatly reduces the update speed of parameters. When the learning rate is large, it causes oscillations in the search process, causing the parameters to hover around the optimal value. By introducing the learning rate cosine decay strategy, the learning rate can gradually decay as training progresses, thereby avoiding the problem of overfitting in the second pre-identification model during training and improving the training effect of the second pre-identification model.

[0090] 406. Obtain the second preset weight ratio data, and perform weight allocation on the second training set to obtain the license plate number training set, the license plate number verification set, and the license plate number test set;

[0091] 407. Input the license plate number training set into the second pre-recognition model for training to obtain multiple second training models with different results;

[0092] 408. Input the license plate number verification set into multiple second training models for verification to obtain multiple second prediction results;

[0093] 409. When the second prediction result meets the second preset condition, the license plate number test set is input into the second training model, and each second training model is scored. The second training model with the highest score is used as the license plate number recognition model.

[0094] In this embodiment, a license plate number training set, a license plate number verification set, and a license plate number test set are input into a second pre-recognition model for model training to obtain a trained license plate number recognition model for license plate number recognition. The second pre-recognition model is iteratively trained using the license plate number training set to obtain a trained second training model. Then, license plate number verification data is randomly obtained from the license plate number verification set and input into the second training model for data prediction to obtain a second prediction result. When the second prediction result meets the second preset optimization condition, the second training model is tested and scored. The license plate number test set is input into the second training model to obtain the training model with the highest score, and the training model with the highest score is used as the license plate number recognition model.

[0095] In this embodiment, the proportions of the license plate training set, license plate verification set, and license plate test set are 60%, 30%, and 10%, respectively.

[0096] Please see Figure 5 The fourth embodiment of the logistics vehicle traffic management method in this invention includes:

[0097] 501. Collect the license plate information of the logistics fleet and establish a license plate information database based on the license plate information;

[0098] In this embodiment, a pre-set license plate number information database facilitates the identification of license plate numbers of passing vehicles. The license plate number information database stores the license plate number information of legitimate logistics vehicles in various distribution centers. In addition, the license plate number information database can be updated regularly to ensure the real-time nature of the comparison data.

[0099] 502. Input real-time vehicle image data into the license plate recognition model to obtain the license plate recognition result;

[0100] 503. Input the license plate recognition result into the license plate number recognition model to obtain the license plate number recognition result;

[0101] 504. Match the license plate recognition result with the license plate information database. If the license plate recognition result matches the license plate information in the license plate information database, issue a passage instruction.

[0102] In this embodiment, real-time vehicle image data is first input into a license plate recognition model. The license plate recognition model identifies and calibrates the license plate features in the real-time image, thus obtaining an output result containing only license plate features. The license plate recognition model also denoises the real-time image containing clutter, selecting images containing only license plate areas from the real-time image, reducing the difficulty of extracting license plate numbers. Subsequently, the output result of the license plate recognition model is input into a license plate number recognition model. The license plate number recognition model extracts characters from the license plate image in the output result and selects the license plate number from the license plate image, thereby accurately obtaining the license plate number of the relevant vehicle. Finally, the license plate number recognition result is compared with the license plate number information database. If there is consistent information, a passage instruction is sent to the barrier gate, allowing the vehicle to enter the distribution center. If there is no consistent information, passage through the barrier gate is prohibited, and a warning message is sent to relevant departments or personnel to prevent illegal vehicles from entering the distribution center and reduce the management difficulty of the distribution center for passing vehicles.

[0103] 505. Extract character information from the license plate recognition results;

[0104] 506. Use character information as a filter to filter information in the license plate number database;

[0105] 507. If the license plate number information database contains the filtered results, then issue a passage instruction;

[0106] 508. If no results are found in the license plate database, an illegal passage instruction will be issued;

[0107] In this embodiment, character information is first extracted from the license plate recognition results and displayed in the order of license plate numbers. The character information is then used as a filtering condition to filter all information in the license plate information database. If a filtering result appears, it indicates that the vehicle is a legitimate vehicle and should be allowed to pass. If the filtering result is 0, it indicates that the vehicle is suspected of illegal entry and an illegal passage instruction needs to be sent to relevant departments or personnel to verify the vehicle. This prevents illegal vehicles from entering the distribution center and reduces the difficulty of managing passing vehicles in the distribution center.

[0108] The passage management method for logistics vehicles in this embodiment of the invention has been described above. The passage management device for logistics vehicles in this embodiment of the invention is described below. Please refer to [link / reference]. Figure 6 One embodiment of the logistics vehicle access management device in this invention includes:

[0109] The acquisition module 601 is used to acquire vehicle images of the logistics fleet and generate a first training set and a second training set based on the vehicle images; the first training set includes license plate annotation features; the second training set includes license plate number character annotation features.

[0110] The first modeling module 602 is used to build the first pre-identification model based on the DBNet model;

[0111] The second modeling module 603 is used to build a second pre-identification model based on the ABINet model;

[0112] The first training module 604 is used to input the first training set into the first pre-recognition model for training, so as to obtain the license plate recognition model;

[0113] The second training module 605 is used to input the second training set into the second pre-recognition model for training, so as to obtain the license plate number recognition model.

[0114] The recognition module 606 is used to acquire real-time vehicle image data and perform recognition through a license plate recognition model and a license plate number recognition model to obtain the license plate recognition result.

[0115] In this embodiment, a license plate recognition model and a license plate number recognition model are constructed through a data acquisition module, a first modeling module, a second modeling module, a first training module, and a second training module to achieve visual recognition of license plate numbers. Real-time vehicle image data is input into the license plate recognition model via the recognition module. The license plate recognition model identifies and calibrates the license plate features in the real-time image and denoises the content of the real-time image, selecting images containing only the license plate area to reduce the difficulty of extracting the license plate number. Subsequently, the output of the license plate recognition model is input into the license plate number recognition model, which extracts characters from the license plate image to accurately obtain the license plate number of the relevant vehicle. This license plate recognition result facilitates vehicle identification at the distribution center, ensuring vehicle traffic safety and reducing the management difficulty of vehicles at the distribution center.

[0116] Please see Figure 7 Another embodiment of the logistics vehicle access management device in this invention includes:

[0117] The acquisition module 601 is used to acquire vehicle images of the logistics fleet and generate a first training set and a second training set based on the vehicle images; the first training set includes license plate annotation features; the second training set includes license plate number character annotation features.

[0118] The first modeling module 602 is used to build the first pre-identification model based on the DBNet model;

[0119] The second modeling module 603 is used to build a second pre-identification model based on the ABINet model;

[0120] The first training module 604 is used to input the first training set into the first pre-recognition model for training, so as to obtain the license plate recognition model;

[0121] The second training module 605 is used to input the second training set into the second pre-recognition model for training, so as to obtain the license plate number recognition model.

[0122] The recognition module 606 is used to acquire real-time vehicle image data and perform recognition through a license plate recognition model and a license plate number recognition model to obtain the license plate recognition result.

[0123] In this embodiment, the acquisition module 601 includes: a filtering unit 6011, used to filter vehicle images, retain vehicle images containing license plate features, and generate a vehicle sample set based on the filtered vehicle images; a first annotation unit 6012, used to perform a first annotation process on the vehicle sample set, selecting license plate features in the images of the vehicle sample set to obtain a first training set; and a second annotation unit 6013, used to perform a second annotation process on the vehicle sample set, selecting license plate number features in the images of the vehicle sample set to obtain a second training set.

[0124] In this embodiment, the first modeling module 602 includes: a construction unit 6021, used to construct an infrastructure based on the DBNet model, the infrastructure including a Backbone network, an FPN network, and a Head network; a reconstruction unit 6022, used to input a MobileNetV3 model into the infrastructure so that the MobileNetV3 model serves as the Backbone network of the infrastructure; the MobileNetV3 model includes a depthwise separable convolutional structure, an inverse residual structure, and an SE module; and a generation unit 6023, used to obtain a first pre-identification model composed of the DBNet model and the MobileNetV3 model.

[0125] In this embodiment, the first training module includes 604: a first allocation unit 6041, used to acquire first preset weight ratio data and allocate weights to the first training set to obtain a license plate training set, a license plate verification set, and a license plate test set; a first iteration unit 6042, used to input the license plate training set into the first pre-recognition model for training to obtain multiple first training models with different results; a first verification unit 6043, used to input the license plate verification set into the multiple first training models for verification to obtain multiple first prediction results; and a first scoring unit 6044, used to input the license plate test set into the first training model when the first prediction result meets the first preset condition, and score each first training model, taking the first training model with the highest score as the license plate recognition model.

[0126] In this embodiment, the second training module 605 includes: a setting unit 6051, used to iteratively train the second pre-recognition model using a learning rate cosine decay strategy; a second allocation unit 6052, used to obtain second preset weight ratio data and allocate weights to the second training set to obtain a license plate number training set, a license plate number verification set, and a license plate number test set; a second iteration unit 6053, used to input the license plate number training set into the second pre-recognition model for training to obtain multiple second training models with different results; a second verification unit 6054, used to input the license plate number verification set into multiple second training models for verification to obtain multiple second prediction results; and a second scoring unit 6055, used to input the license plate number test set into the second training model when the second prediction result meets the second preset condition, and score each second training model, selecting the second training model with the highest score as the license plate number recognition model.

[0127] In this embodiment, the recognition module 606 includes: a database building unit 6061, used to collect license plate number information of logistics fleet vehicles and build a license plate number information database based on the license plate number information; a first input unit 6062, used to input real-time vehicle image data into the license plate recognition model to obtain the license plate recognition result; a second input unit 6063, used to input the license plate recognition result into the license plate number recognition model to obtain the license plate number recognition result; and a matching unit 6064, used to match the license plate number recognition result with the license plate number information database. If the license plate number recognition result matches the license plate number information in the license plate number information database, a passage instruction is issued.

[0128] In this embodiment, the matching unit 6064 includes: an extraction unit 60641, used to extract character information from the license plate number recognition result; and a filtering unit 60642, used to filter the information in the license plate number information database using the character information as a filtering condition; if the license plate number information database contains a filtering result, a passage instruction is issued; if the license plate number information database does not contain a filtering result, an illegal passage instruction is issued.

[0129] above Figure 6 and Figure 7 The passage management device for logistics vehicles in this embodiment of the invention is described in detail from the perspective of modular functional entities. The passage management equipment for logistics vehicles in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0130] Figure 8This is a schematic diagram of the structure of a logistics vehicle access management device 700 provided in an embodiment of the present invention. The logistics vehicle access management device 700 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 710 (e.g., one or more processors) and a memory 720, and one or more storage media 730 (e.g., one or more mass storage devices) storing application programs 733 or data 732. The memory 720 and storage media 730 can be temporary or persistent storage. The program stored in the storage media 730 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the logistics vehicle access management device 700. Furthermore, the processor 710 may be configured to communicate with the storage media 730 and execute the series of instruction operations in the storage media 730 on the logistics vehicle access management device 700 to implement the steps of the logistics vehicle access management method provided in the above-described method embodiments.

[0131] The access control device 700 for logistics vehicles may also include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input / output interfaces 760, and / or one or more operating systems 731, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 8 The illustrated access control equipment structure for logistics vehicles does not constitute a limitation on access control equipment based on logistics vehicles. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0132] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a method for managing the passage of logistics vehicles.

[0133] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0134] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0135] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for managing the passage of logistics vehicles, characterized in that, include: Collect vehicle images of the logistics fleet and generate a first training set and a second training set based on the vehicle images; the first training set includes license plate annotation features. The second training set includes license plate number character annotation features; An infrastructure is built based on the DBNet model, comprising a Backbone network, an FPN network, and a Head network. The Backbone network is responsible for extracting image features, the FPN network enhances features using a feature pyramid structure, and the Head network computes text region probability maps. A MobileNetV3 model is input into the infrastructure to serve as the Backbone network. The MobileNetV3 model includes a depthwise separable convolutional structure, an inverse residual structure, and an SE module. A first pre-recognition model composed of the DBNet model and the MobileNetV3 model is obtained. A second pre-recognition model is built based on the ABINet model, which includes an input terminal, a visual model ABIVisionModel, a text model ABILanguageDecoder, a loss function, and Inference. The first training set is input into the first pre-recognition model for training to obtain the license plate recognition model; The second training set is input into the second pre-recognition model for training to obtain the license plate number recognition model; The system acquires real-time vehicle image data, collects license plate information of the logistics fleet, and establishes a license plate information database based on the license plate information. The real-time vehicle image data is input into a license plate recognition model to obtain the license plate recognition result. The license plate recognition result is then input into the license plate number recognition model to obtain the license plate number recognition result. Character information is extracted from the license plate number recognition result and displayed in the same arrangement as the license plate number. This character information is used as a filter to filter the information in the license plate number information database. If the database contains a filtered result, a passage instruction is issued; otherwise, an illegal passage instruction is issued.

2. The method for managing the passage of logistics vehicles according to claim 1, characterized in that, The process of collecting vehicle images from the logistics fleet and generating a first training set and a second training set based on the vehicle images includes: The vehicle images are filtered to retain those containing license plate features, and a vehicle sample set is generated based on the filtered vehicle images. The vehicle sample set is labeled once, and the license plate features in the images of the vehicle sample set are labeled to obtain the first training set; The vehicle sample set is then subjected to secondary annotation processing, in which the license plate number features in the images of the vehicle sample set are labeled to obtain the second training set.

3. The method for managing the passage of logistics vehicles according to claim 1, characterized in that, The step of inputting the first training set into the first pre-recognition model for training to obtain the license plate recognition model includes: Obtain the first preset weight ratio data, and perform weight allocation on the first training set to obtain the license plate training set, license plate verification set and license plate test set; The license plate training set is input into the first pre-recognition model for training, resulting in multiple first training models with different results; The license plate verification set is input into multiple first training models for verification, resulting in multiple first prediction results; When the first prediction result meets the first preset condition, the license plate test set is input into the first training model, and each first training model is scored. The first training model with the highest score is used as the license plate recognition model.

4. The method for managing the passage of logistics vehicles according to claim 1, characterized in that, The step of inputting the second training set into the second pre-recognition model for training to obtain the license plate number recognition model includes: The second pre-identification model is iteratively trained using a learning rate cosine decay strategy. Obtain the second preset weight ratio data, and perform weight allocation on the second training set to obtain the license plate number training set, license plate number verification set, and license plate number test set; The license plate number training set is input into the second pre-recognition model for training, resulting in multiple second training models with different results; The license plate number verification set is input into multiple second training models for verification, resulting in multiple second prediction results; When the second prediction result meets the second preset condition, the license plate number test set is input into the second training model, and each second training model is scored. The second training model with the highest score is used as the license plate number recognition model.

5. A passage management device for logistics vehicles, characterized in that, include: The acquisition module is used to acquire vehicle images of the logistics fleet and generate a first training set and a second training set based on the vehicle images; the first training set includes license plate annotation features; the second training set includes license plate number character annotation features. The first modeling module is used to construct an infrastructure based on the DBNet model. This infrastructure includes a Backbone network, an FPN network, and a Head network. The Backbone network is responsible for extracting image features, the FPN network is used to enhance features using a feature pyramid structure, and the Head network is used to calculate text region probability maps. A MobileNetV3 model is input into the infrastructure to serve as the Backbone network. The MobileNetV3 model includes a depthwise separable convolutional structure, an inverse residual structure, and an SE module. This results in a first pre-recognition model composed of the DBNet model and the MobileNetV3 model. The second modeling module is used to build a second pre-recognition model based on the ABINet model. The ABINet model includes an input terminal, a visual model ABIVisionModel, a text model ABILanguageDecoder, a loss function, and Inference. The first training module is used to input the first training set into the first pre-recognition model for training, so as to obtain the license plate recognition model; The second training module is used to input the second training set into the second pre-recognition model for training, so as to obtain the license plate number recognition model. The recognition module is used to acquire real-time vehicle image data, collect license plate information of logistics fleets, and establish a license plate information database based on the license plate information. It inputs real-time vehicle image data into a license plate recognition model to obtain license plate recognition results, and then inputs the license plate recognition results back into the license plate number recognition model to obtain license plate number recognition results. It extracts character information from the license plate number recognition results and displays the character information in the license plate number arrangement. The character information is used as a filtering condition to filter information in the license plate number information database. If the license plate number information database contains a filtered result, a passage instruction is issued; otherwise, an illegal passage instruction is issued.

6. A passage management device for logistics vehicles, characterized in that, The access management device for the logistics vehicle includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the logistics vehicle access management device to perform the steps of the logistics vehicle access management method as described in any one of claims 1-4.

7. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the access management method for logistics vehicles as described in any one of claims 1-4.

Images