Traffic flow monitoring method and system based on multi-source data fusion
Through the dual-branch fusion model combined with the characteristics of multi-spectral and synthetic aperture radar data, the problems of real-time and data heterogeneity in vehicle flow monitoring are solved, efficient and accurate vehicle detection is achieved, and real-time traffic management is supported.
Patent Information
- Application Number
- CN202411479425.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-23
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing multi-source data fusion technology has problems in real-time, data heterogeneity and feature extraction in vehicle traffic monitoring, resulting in the inability to achieve accurate, efficient and real-time monitoring.
采用双分支融合模型,通过并行的卷积神经网络CNN和注意力机制Transformer提取局部和全局特征,并通过交互方式融合,结合多光谱数据的光谱特征,生成具有优良纹理和光谱信息的融合数据, Used for traffic monitoring.
It realizes more efficient and comprehensive image data fusion, improves the accuracy and robustness of vehicle detection, and can effectively support traffic management in real-time applications.
Smart Images

Figure CN119007137A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of traffic management, and in particular relates to a vehicle flow monitoring method and system based on multi-source data fusion. Background Art
[0002] With the comprehensive development of UAV technology, UAVs play an increasingly important role in the field of traffic management. UAVs equipped with various sensors can realize real-time monitoring of traffic conditions, which is of great significance to traffic management. Therefore, UAV-based aerial image processing technology has become the focus of current research.
[0003] Among the current technologies, multispectral data (MS) provides rich spectral information, but it is difficult to achieve real-time monitoring due to factors such as weather, time and climate; synthetic aperture radar (SAR) can provide monitoring data in all weather and all time periods, and the image structure information is good, but it is a single-band imaging, although the texture information is rich, but lacks multispectral information. Therefore, the monitoring technology based on single data has obvious defects, and the fusion of multi-source data through multi-source sensors carried by drones has become a hot issue that needs to be solved urgently.
[0004] The existing multi-source data fusion technology has the following difficulties that need to be overcome: First, the real-time problem. From the perspective of real-time applications, multi-source data fusion needs to control the calculation time under the premise of calculation complexity and resource consumption to achieve real-time vehicle flow monitoring and achieve the purpose of efficient traffic management; second, the data heterogeneity problem. The image data collected by different sensors or modalities have large differences in characteristics such as resolution, noise level, brightness and contrast, and need to be effectively normalized and fused; finally, the problem of feature extraction and selection. There is a large amount of repeated or redundant information in the image data collected by different sensors. How to extract and select key features is an existing problem.
[0005] Therefore, existing solutions cannot achieve accurate, efficient and real-time traffic flow monitoring. Summary of the invention
[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a vehicle flow monitoring method and system based on multi-source data fusion, which fuses the multispectral data and synthetic aperture radar data collected by UAV, gives full play to the advantages of various technologies, and obtains fused data with excellent texture and spectral information for accurate, efficient and real-time monitoring of vehicle flow.
[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: A first aspect of the present invention provides a vehicle flow monitoring method based on multi-source data fusion.
[0008] A vehicle flow monitoring method based on multi-source data fusion, comprising: Acquire multispectral data and synthetic aperture radar data collected in real time and perform preprocessing; The pre-processed multispectral data and synthetic aperture radar data are input into the dual-branch fusion model to obtain fused data; Carry out traffic flow monitoring based on the fused data; Among them, the dual-branch fusion model extracts local features and global features respectively through parallel convolutional neural network CNN and attention mechanism Transformer, and fuses local features and global features in an interactive manner to generate a fused image with enhanced spatial details. After fusing the spectral features of multi-spectral data again, fused data with excellent texture and spectral information is obtained.
[0009] Furthermore, the multispectral data and synthetic aperture radar data are collected in real time by a multispectral camera and a synthetic aperture radar respectively carried by the UAV.
[0010] Furthermore, the preprocessing includes geometric correction, spatial alignment and data enhancement; The data enhancement is to enhance the edge of the image.
[0011] Furthermore, the image edge enhancement is performed by performing linear interpolation on adjacent pixels according to different weights or interpolating through a cubic convolution function to make the edge of the multispectral image coherent and smooth.
[0012] Furthermore, the dual-branch fusion model includes two branches: a detail enhancement branch and a spectrum preservation branch; The detail enhancement branch is based on a parallel convolutional neural network (CNN) and an attention mechanism (Transformer) to generate a fused image that enhances spatial details. The spectrum holding branch obtains the spectrum characteristics of the multi-spectral data by sampling the multi-spectral data.
[0013] Furthermore, the detail enhancement branch includes parallel convolutional neural network CNN and attention mechanism Transformer and PFM modules; The parallel convolutional neural network CNN and attention mechanism Transformer are used to extract local features and global features; The PFM module, based on the improved PFM method, interactively fuses local features and global features through feature similarity calculation and adaptive threshold strategy in the fusion process to generate a fused image with enhanced spatial details.
[0014] Furthermore, the dual-branch fusion model also includes a residual fusion network and a decoder network; The residual fusion network is used to perform multi-level fusion of fused images and spectral features; The decoder network is used to sample the results of multi-level fusion to the same scale to obtain fused data with excellent texture and spectral information.
[0015] A second aspect of the present invention provides a vehicle flow monitoring system based on multi-source data fusion.
[0016] A vehicle flow monitoring system based on multi-source data fusion, comprising: The acquisition module is configured to: acquire multispectral data and synthetic aperture radar data collected in real time and perform preprocessing; The fusion module is configured to: input the pre-processed multispectral data and synthetic aperture radar data into a dual-branch fusion model to obtain fused data; The monitoring module is configured to: monitor vehicle flow based on the fused data; Among them, the dual-branch fusion model extracts local features and global features respectively through parallel convolutional neural network CNN and attention mechanism Transformer, and fuses local features and global features in an interactive manner to generate a fused image with enhanced spatial details. After fusing the spectral features of multi-spectral data again, fused data with excellent texture and spectral information is obtained. A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a vehicle flow monitoring method based on multi-source data fusion as described in the first aspect of the present invention.
[0017] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in a vehicle flow monitoring method based on multi-source data fusion as described in the first aspect of the present invention are implemented.
[0018] One or more of the above technical solutions have the following beneficial effects: The present invention provides an airborne multi-source data fusion method, which realizes the fusion of airborne synthetic aperture radar data and multispectral data by using a convolutional neural network CNN and an attention mechanism Transformer to extract multi-scale features in parallel and perform parallel fusion. It can effectively combine local and global features, improve detail loss and spectral distortion, and achieve more efficient and comprehensive image data fusion, thus providing a new solution for the multi-sensor system carried by unmanned aerial vehicles in applications such as traffic management and real-time monitoring.
[0019] The present invention provides a dynamic PFM strategy that uses a weight adjustment mechanism to evaluate the importance of features in real time during the fusion process. It not only calculates feature similarity through pixel-level matching, but also combines an adaptive threshold strategy to dynamically select important features for fusion. This method allows the model to flexibly adjust the fusion strategy according to the different feature distributions of the input image, thereby maximizing the retention of key details and contextual information.
[0020] The present invention also provides an improved residual fusion network, in particular, introduces an attention module and a deconvolution visualization module to improve model performance and interpretability. The attention mechanism is integrated in each fusion layer of the residual fusion network, and the weights of important features are automatically enhanced by modeling the saliency of the feature map, while suppressing the influence of irrelevant features, thereby improving the accuracy and robustness of target detection. The attention module is based on the self-attention mechanism, which effectively captures long-distance dependencies, allowing the model to focus more on the target area when processing complex scenes, thereby significantly improving the detection performance of targets such as vehicles.
[0021] The present invention also provides a deconvolution visualization module for visual analysis of the decision-making process of the model. By introducing deconvolution operations at specific layers of the network, high-dimensional feature maps are restored to the input image space, which can intuitively show which features have an important impact on the final detection results. This innovative design not only enhances the interpretability of the model, but also provides strong support for subsequent model optimization and debugging.
[0022] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0024] Figure 1 This is a flow chart of the method of the first embodiment.
[0025] Figure 2 This is a structural diagram of the dual-branch fusion model of the first embodiment.
[0026] Figure 3 This is a residual fusion network structure diagram of the first embodiment. DETAILED DESCRIPTION
[0027] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.
[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0029] Embodiment 1 In one embodiment of the present disclosure, a vehicle flow monitoring method based on multi-source data fusion is provided. Figure 1 As shown, the following steps are included: Step S1: Acquire multispectral data and synthetic aperture radar data collected in real time and perform preprocessing.
[0030] Step S2: input the preprocessed multispectral data and synthetic aperture radar data into the dual-branch fusion model to obtain fused data.
[0031] Step S3: Carry out traffic flow monitoring based on the fused data.
[0032] The following is a detailed description of the implementation process of a vehicle flow monitoring method based on multi-source data fusion in this embodiment.
[0033] The fusion of multispectral images and SAR images has significant advantages in traffic management and planning. By combining the advantages of these two data sources, more comprehensive and accurate information can be obtained to support more effective decision-making and management. The specific implementation method involves multiple steps, including data acquisition, preprocessing, fusion, analysis and application. The following focuses on image fusion technology: The multispectral and SAR data are aligned to the same spatial resolution and coordinate system, and feature information is extracted from the multispectral data and SAR data respectively. The features of SAR and multispectral data are processed in parallel using a dual-branch convolutional neural network (CNN) and Transformer. Through parallel CNN and Transformer models, the feature information of multispectral and SAR data is comprehensively extracted and fused to obtain more comprehensive surface feature data. At the same time, the extracted image data is fused through the residual fusion network.
[0034] This embodiment provides a dual-branch fusion model, which extracts local features and global features respectively through parallel CNN and Transformer, and fuses the local features and global features in an interactive manner to generate a fused image with enhanced spatial details. After fusing the spectral features of multispectral data again, fused data with excellent texture and spectral information is obtained.
[0035] The dual-branch fusion model, such as Figure 2 As shown, it mainly includes two branches: detail enhancement branch and spectrum preservation branch.
[0036] 1. Detail enhancement branch The detail enhancement branch generates a fused image with enhanced spatial details based on parallel CNN and Transformer, and includes parallel CNN and Transformer, and PFM modules.
[0037] (1) Parallel CNN and Transformer, used to extract local features and global features, and interactively fuse them; The CNN branch extracts local features through the detail feature extraction (DFE) module, and the Transformer branch extracts global features through the GCF (Global Context Feature) module. Based on the improved PFM method, the local features and global features are interactively fused through feature similarity calculation and adaptive threshold strategy in the fusion process. Specifically: 1) The CNN branch processes local features of the image through the Detail Feature Extraction (DFE) module. The DFE module deeply mines the details in the image, such as edges, textures, and small objects, through multi-layer convolution and pooling operations, thereby generating high-quality local feature maps. This module can adaptively adjust the size and stride of the convolution kernel to optimize the extraction of features of different scales, ensuring that the model can accurately identify targets even in complex backgrounds.
[0038] The working principle of the DFE module is as follows: Preprocess the input image, such as denoising and grayscale conversion; use multi-layer convolution and pooling operations to extract detailed features of the image; convert the extracted detailed features into a form that can be used for subsequent processing, such as feature vectors, feature maps, etc.; apply specific feature detection algorithms to identify the detailed features of the target of interest.
[0039] 2) The Transformer branch extracts global features of the image through the Global Context Feature (GCF) module. The core idea of the GCF module is to capture the semantics and context of the image by analyzing the global context information in the image, and process it as part of the image features; it can help the model better understand the overall structure and semantic information of the image, thereby improving the accuracy and robustness of image analysis; the GCF module uses the self-attention mechanism to analyze the relationship between all pixels in the image, thereby capturing long-distance dependencies and global semantic information. The design of this module allows the model to not only focus on local details when learning features, but also effectively integrate global background information and enhance its ability to understand the target.
[0040] The working principle of the GCF module is as follows: The attention mechanism is used to combine the global context information with the local features extracted by DFE to extract the global context features; the extracted global context features are converted into a form that can be processed later.
[0041] 3) Due to the asymmetry between local features and global features, a dynamic PFM strategy is introduced in the feature fusion stage, which uses a learnable weight adjustment mechanism to evaluate the importance of features in real time during the fusion process; specifically, the PFM module not only calculates feature similarity through pixel-level matching, but also combines an adaptive threshold strategy to dynamically select important features for fusion, specifically: First, by calculating the similarity between pixels, we determine which features are important in a specific context, usually using similarity metrics such as correlation, Euclidean distance, and cosine similarity. Next, based on the feature similarity calculation results, we dynamically adjust the weight of the features, and dynamically set the threshold based on the image content and feature distribution to select appropriate features under different conditions; this method can better adapt to image changes and improve matching accuracy.
[0042] The weight adjustment process is mainly achieved through several steps: First, the input sequence is linearly transformed to generate query (Q), key (K) and value (V). Then, the attention score is obtained by calculating the dot product of the query and the key. The formula is scores = , in The dimension of the key is used for scaling to prevent the gradient from vanishing. Then, the score is converted into a probability distribution by applying the Softmax function so that the weight of each row sums to 1, thus reflecting the importance of each position. Finally, the value vector is weighted and summed according to the calculated attention weights to get the output This series of steps ensures that the self-attention mechanism can dynamically adjust the weights of different elements and effectively capture the relationship between elements in the sequence, whether short-range or long-range dependencies, thereby improving the performance of the model in complex tasks. Finally, by analyzing the contribution of features, it is decided which features should be retained or merged to reduce the impact of noise on the results.
[0043] The advantage of the dynamic PFM strategy is that it dynamically adjusts feature weights according to different data characteristics and application requirements to better utilize the advantages of different features, improve feature fusion effects, and adapt to changes in different data conditions; this method allows the model to flexibly adjust the fusion strategy according to the different feature distributions of the input image, thereby maximizing the retention of key details and contextual information.
[0044] 2. Spectrum Preserving Branching The spectrum preserving branch obtains the spectrum characteristics of the multi-spectral data by performing triple sampling on the multi-spectral data.
[0045] 3. Residual Fusion Network This embodiment adopts an improved residual fusion network MRFN, especially introduces an attention module and a deconvolution visualization module to perform multi-level fusion of fused images and spectral features to improve model performance and interpretability; Figure 3 As shown in the figure, the residual fusion network consists of a residual module, an attention module, a feature fusion layer and a deconvolution visualization module connected in sequence.
[0046] Specifically, the residual module is composed of three convolution blocks that are sequentially subjected to the ReLU operation and finally input into the attention module. After being screened by the attention mechanism, it is processed by the fusion layer and input into the deconvolution visualization module. The above structure is a multi-scale residual fusion network structure that nests the attention module and the visualization module.
[0047] The attention module is based on the self-attention mechanism, which effectively captures long-distance dependencies, allowing the model to focus more on the target area when processing complex scenes, thereby significantly improving the detection performance of targets such as vehicles. In the implementation process, first, the input sequence is linearly transformed to generate queries (Q), keys (K), and values (V). Then, the attention score is obtained by calculating the dot product of the query and the key. The formula is scores= , in The dimension of the key is used for scaling to prevent the gradient from vanishing. Then, the score is converted into a probability distribution by applying the Softmax function so that the weight of each row sums to 1, thus reflecting the importance of each position. Finally, the value vector is weighted and summed according to the calculated attention weights to get the output In the Transformer architecture, the output is also added to the original input and layer normalization is applied to enhance the stability of the model. This mechanism enables the model to better capture the relationship and long-range dependencies between elements in the sequence, thereby improving performance.
[0048] The feature fusion layer includes feature map concatenation and convolution operations. By modeling the saliency of feature maps, the weights of important features are automatically enhanced while suppressing the influence of irrelevant features, thereby improving the accuracy and robustness of target detection.
[0049] The deconvolution visualization module is used to visualize the decision-making process of the model. The image input to the deconvolution visualization module is processed by two convolution kernels and activated by ReLU. The result feature map is obtained by max pooling. After depooling, ReLU activation and deconvolution, the visualization effect is finally achieved, which helps to understand the feature fusion process and improve the interpretability of the model. By introducing deconvolution operations at specific layers of the network, the high-dimensional feature map is restored to the input image space, which can intuitively show which features have an important impact on the final detection results. This innovative design not only enhances the interpretability of the model, but also provides strong support for subsequent model optimization and debugging.
[0050] 4. Decoder Network The decoder samples the multi-level fusion results in the MRFN network to the same scale to obtain a fused image with enhanced spatial details and achieve full fusion of image features.
[0051] 5. Traffic flow monitoring Use YOLOv5 for vehicle detection. The specific steps are as follows: (1) Filter out diverse vehicle image data from the collected fusion images, including images of different vehicle models, colors, angles, and environments, to form a preliminary data set.
[0052] (2) Use annotation tools to annotate the collected images and generate annotation files that conform to the YOLO format, recording the category and bounding box coordinates of each vehicle.
[0053] (3) Perform data enhancement by expanding the diversity of the data set through techniques such as random cropping, rotation, and color jittering to improve the generalization ability of the model and effectively reduce the overfitting phenomenon of the model in different environments.
[0054] (4) During the environment setup phase, install the necessary dependent libraries, including Python and PyTorch, and clone the YOLOv5 code repository; then, create a custom data configuration file containing the paths and category information of the training and validation sets, and select the appropriate YOLOv5 model for training.
[0055] (5) During the training process, the model is trained using the specified training command and appropriate hyperparameters are set, such as image size, batch size, and number of training rounds.
[0056] (6) After training is completed, the validation set is used to evaluate the performance of the model, calculate the mean average precision (mAP) and loss value, and adjust the model parameters based on the evaluation results.
[0057] (7) Use the trained model for inference to detect traffic flow in new images and optimize the detection results through the non-maximum suppression (NMS) method, thereby achieving real-time and high-precision vehicle detection. It can be deployed on edge computing devices and is suitable for various drone application scenarios, such as traffic monitoring, automatic driving assistance systems, etc., providing strong support for intelligent traffic management.
[0058] Embodiment 2 In one embodiment of the present disclosure, a vehicle flow monitoring system based on multi-source data fusion is provided, comprising: The acquisition module is configured to: acquire multispectral data and synthetic aperture radar data collected in real time and perform preprocessing; The fusion module is configured to: input the pre-processed multispectral data and synthetic aperture radar data into a dual-branch fusion model to obtain fused data; The monitoring module is configured to: monitor vehicle flow based on the fused data; Among them, the dual-branch fusion model extracts local features and global features respectively through parallel CNN and Transformer, and fuses the local features and global features in an interactive manner to generate a fused image with enhanced spatial details. After fusing the spectral features of multispectral data again, fused data with excellent texture and spectral information is obtained.
[0059] Embodiment 3 The purpose of this embodiment is to provide a computer-readable storage medium.
[0060] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in a vehicle flow monitoring method based on multi-source data fusion as described in the first embodiment of the present disclosure.
[0061] Embodiment 4 The purpose of this embodiment is to provide an electronic device.
[0062] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in a vehicle flow monitoring method based on multi-source data fusion as described in the first embodiment of the present disclosure are implemented.
[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A vehicle flow monitoring method based on multi-source data fusion, characterized in that: include: Acquire multispectral data and synthetic aperture radar data collected in real time and perform preprocessing; The pre-processed multispectral data and synthetic aperture radar data are input into the dual-branch fusion model to obtain fused data; Carry out traffic flow monitoring based on the fused data; Among them, the dual-branch fusion model extracts local features and global features respectively through parallel convolutional neural network CNN and attention mechanism Transformer, and fuses local features and global features in an interactive manner to generate a fused image with enhanced spatial details. After fusing the spectral features of multi-spectral data again, fused data with excellent texture and spectral information is obtained.
2. A vehicle flow monitoring method based on multi-source data fusion as claimed in claim 1, characterized in that: The multispectral data and synthetic aperture radar data are collected in real time by a multispectral camera and a synthetic aperture radar respectively carried by the UAV.
3. The vehicle flow monitoring method based on multi-source data fusion as claimed in claim 1, characterized in that: The preprocessing includes geometric correction, spatial alignment and data enhancement; The data enhancement is to enhance the edge of the image.
4. The vehicle flow monitoring method based on multi-source data fusion as claimed in claim 3, characterized in that: The image edge enhancement is performed by performing linear interpolation on adjacent pixel points according to different weights or interpolating through a cubic convolution function to make the edge of the multispectral image coherent and smooth.
5. The vehicle flow monitoring method based on multi-source data fusion as claimed in claim 1, characterized in that: The dual-branch fusion model includes two branches: a detail enhancement branch and a spectrum preservation branch; The detail enhancement branch is based on a parallel convolutional neural network (CNN) and an attention mechanism (Transformer) to generate a fused image that enhances spatial details. The spectrum holding branch obtains the spectrum characteristics of the multi-spectral data by sampling the multi-spectral data.
6. A vehicle flow monitoring method based on multi-source data fusion as claimed in claim 5, characterized in that: The detail enhancement branch includes a parallel convolutional neural network CNN and an attention mechanism Transformer and PFM module; The parallel convolutional neural network CNN and attention mechanism Transformer are used to extract local features and global features; The PFM module, based on the improved PFM method, interactively fuses local features and global features through feature similarity calculation and adaptive threshold strategy in the fusion process to generate a fused image with enhanced spatial details.
7. The vehicle flow monitoring method based on multi-source data fusion as claimed in claim 5, characterized in that: The dual-branch fusion model also includes a residual fusion network and a decoder network; The residual fusion network is used to perform multi-level fusion of fused images and spectral features; The decoder network is used to sample the results of multi-level fusion to the same scale to obtain fused data with excellent texture and spectral information.
8. A vehicle flow monitoring system based on multi-source data fusion, characterized in that: include: The acquisition module is configured to: acquire multispectral data and synthetic aperture radar data collected in real time and perform preprocessing; The fusion module is configured to: input the pre-processed multispectral data and synthetic aperture radar data into a dual-branch fusion model to obtain fused data; The monitoring module is configured to: monitor vehicle flow based on the fused data; Among them, the dual-branch fusion model extracts local features and global features respectively through parallel convolutional neural network CNN and attention mechanism Transformer, and fuses local features and global features in an interactive manner to generate a fused image with enhanced spatial details. After fusing the spectral features of multi-spectral data again, fused data with excellent texture and spectral information is obtained.
9. An electronic device, comprising: a memory for non-transitory storage of computer readable instructions; a processor for executing the computer-readable instructions; Wherein, when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 7 is executed.
10. A storage medium, characterized in that: The computer-readable instructions are non-transitorily stored, wherein when the computer-readable instructions are executed by a computer, the method of any one of claims 1 to 7 is performed.
Citation Information
Patent Citations
Image fusion algorithm
CN114821261A
Space-time mixed video super-resolution method based on deformable attention
CN115861068A
Red tide detection method and system based on CNN-Transform spectrum reconstruction
CN117690025A
Cited By
High-speed traffic flow detection method and system based on deep learning
CN119672964A
