Engineering machinery identification method, system and equipment and storage medium
By simultaneously acquiring visible light and infrared images and performing deep learning registration and fusion, combined with the CBAM attention mechanism and Wasserstein distance evaluation, the problem of poor recognition of small targets at high altitudes was solved, and efficient and accurate recognition in complex scenes was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-17
- Publication Date
- 2026-03-13
AI Technical Summary
Existing technologies struggle to capture detailed features in high-altitude, overhead scenarios, limiting the effectiveness of small target recognition. Attention mechanisms are insufficient for modeling small targets, and the IoU metric is sensitive to positional deviations, resulting in insufficient accuracy in small target detection.
The model simultaneously acquires visible light and infrared images, fuses them using a deep learning registration model, and combines an improved YOLOv8 model with a CBAM attention mechanism for fine recognition. Wasserstein distance is also introduced as an evaluation metric to optimize the model's feature extraction and localization accuracy.
It significantly improves the recognition rate of small targets in complex scenarios, enhances all-weather operation capability and target discrimination accuracy, solves the accuracy bottleneck problem of small target detection, and achieves synergistic optimization of speed and accuracy.
Smart Images
Figure CN121661381A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of image processing technology, and in particular relates to a method, system, device and storage medium for recognizing engineering machinery. Background Technology
[0002] In complex and dynamic operational scenarios such as engineering construction, mining, and agricultural production, real-time monitoring and accurate identification of construction machinery are crucial for ensuring construction safety, improving operational efficiency, and achieving intelligent management. In recent years, with the rapid development of computer vision and artificial intelligence technologies, mid-to-high-level cameras have gradually become indispensable sensing tools in these fields due to their unique deployment advantages.
[0003] In the process of identifying engineering machinery using mid-to-high-level cameras, existing technologies include detection methods based on single-light images and evaluation methods based on the Interchange of Units (IoU). Single-light image-based detection methods utilize images captured by a single visible light camera, employing deep learning models to learn features from the raw pixels to achieve end-to-end target detection. IoU-based evaluation methods are metrics used to assess the performance of target detection algorithms, measuring the degree of overlap between the model-predicted target bounding box and the manually labeled ground truth bounding box. In engineering machinery identification, researchers and engineers use IoU to quantitatively assess the accuracy of the detection model.
[0004] The existing technology has the following technical problems: 1. Insufficient recognition of small targets. Existing detection models are limited by resolution and viewing angle in high-altitude, top-down scenarios, making it difficult to capture detailed features and thus limiting recognition performance.
[0005] 2. Attention mechanisms are insufficient for modeling small targets. While existing methods can improve feature extraction in complex scenes, they are insufficient in mining and fusing detailed features in high-altitude, small target scenarios, resulting in low sensitivity to small targets.
[0006] 3. The IoU metric is highly sensitive to positional deviations. In small target detection, even slight bounding box offsets can cause a significant drop in the IoU value, becoming a major bottleneck affecting detection accuracy. Summary of the Invention
[0007] This invention provides a method, system, device, and storage medium for identifying engineering machinery, aiming to solve the technical problems existing in the prior art, such as difficulty in capturing detailed features, limited recognition effect, insufficient mining and fusion of detailed features leading to low sensitivity to small targets, and sensitivity to positional deviations.
[0008] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A method for identifying engineering machinery, comprising: Acquire the original image, and preprocess the original image to obtain a preprocessed image; Preliminary detection is performed on the preprocessed image to obtain preliminary detection results; Multiple preprocessed images acquired synchronously and preprocessed under the same field of view are registered and fused to obtain a registered and fused image. A refined recognition model is constructed, and based on the preliminary detection results, the refined recognition model is used to perform target recognition on the registered and fused image to obtain refined recognition results.
[0009] Furthermore, the aforementioned original images include visible light images and infrared images, which are acquired simultaneously by mid-to-high-angle cameras.
[0010] Furthermore, the preprocessing of the original image described above specifically includes: The original image is subjected to image enhancement processing to obtain an enhanced image; The enhanced image is then subjected to noise reduction processing to obtain a denoised image; The denoised image is then subjected to scale normalization to obtain a preprocessed image.
[0011] Furthermore, the preliminary test results obtained above are as follows: A detection model is constructed, and the detection model is trained using the previously collected mid-to-high-angle video frame images to obtain a trained detection model. The preprocessed image is initially detected using the trained inspection model to obtain preliminary detection results.
[0012] Furthermore, the registered and fused image obtained above is specifically as follows: Acquire visible light and infrared images simultaneously acquired within the same field of view; A deep learning registration model is constructed and the modal mapping relationship between visible light images and infrared images is automatically learned using the deep learning registration model; The learned deep learning registration model is used to process visible light images and infrared images under the same field of view to obtain a registered and fused image.
[0013] Furthermore, the detailed identification results obtained above are as follows: An improved YOLOv8 model incorporating the CBAM attention mechanism is adopted as the fine recognition model; Based on the preliminary detection results, the input registered and fused images are subjected to global average pooling and global max pooling respectively using a fine recognition model, resulting in channel statistical vectors and channel saliency vectors. The channel statistical vector and the channel saliency vector are input into a shared fully connected layer to learn the nonlinear relationship between channels, thereby obtaining the channel statistical nonlinear vector and the channel saliency nonlinear vector. The channel statistical nonlinear vector and the channel saliency nonlinear vector are added together, and the channel attention weight vector is generated by passing the Sigmoid activation function. The channel attention weight vector is multiplied channel by channel with the input registration and fusion image to obtain the channel attention image; The channel attention image is subjected to max pooling and average pooling in the channel dimension to obtain a spatial saliency map and a spatial context map, respectively. The spatial saliency map and the spatial context map are then concatenated in the channel dimension to obtain a dual-channel feature map. The dual-channel feature map is input into a convolutional layer for convolution processing to obtain a single-channel spatial attention feature map, which is then processed using the Sigmoid activation function to obtain spatial attention weights. The spatial attention weights are multiplied channel by channel by channel of the input channel attention image to obtain the spatial attention image; The spatial attention image is input into the neck network of the fine recognition model for multi-scale feature fusion. In the detection head, the feature maps at different scales are decoded to predict the bounding box, category, and confidence of the target, thus obtaining the fine recognition result.
[0014] Furthermore, the above also includes: Wasserstein distance is introduced as an evaluation metric for the refined recognition model, specifically: The bounding box of each engineering machinery target is modeled as a probability distribution in two-dimensional space; where the bounding box includes the predicted box and the ground truth box. The Wasserstein distance to the engineering machinery target is calculated using the following formula:
[0015] Among them, B t B represents the true bounding box. p Indicates the prediction box. T represents the Euclidean distance between the center points of the ground truth bounding boxes and the center points of the predicted bounding boxes. r Σ represents the trace of a matrix. t Σ represents the covariance matrix of the true bounding boxes. p This represents the covariance matrix of the predicted bounding boxes; The Wasserstein distance is normalized to obtain the normalized Wasserstein distance; The localization loss component in the loss function of the fine-grained recognition model is replaced with the normalized Wasserstein distance, as shown in the following equation:
[0016] Where, λ cls , λ conf , λ loc L represents the weighting coefficients, respectively. cls L represents the classification loss. conf This represents the confidence loss.
[0017] Secondly, in order to solve the above-mentioned technical problems, the present invention also provides an engineering machinery identification system, comprising: The preprocessing module is used to acquire the original image, preprocess the original image, and obtain a preprocessed image. The preliminary detection module is used to perform preliminary detection on the preprocessed image and obtain preliminary detection results; The registration and fusion module is used to register and fuse multiple preprocessed images that are simultaneously acquired and preprocessed under the same field of view to obtain a registered and fused image. The fine recognition module is used to construct a fine recognition model. Based on the preliminary detection results, the fine recognition model is used to perform target recognition on the registered and fused image to obtain fine recognition results.
[0018] Thirdly, in order to solve the above-mentioned technical problems, the present invention also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the engineering machinery identification method of the present application.
[0019] Fourthly, in order to solve the above-mentioned technical problems, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the engineering machinery identification method of the present application.
[0020] Compared with the prior art, the present invention has the following advantages: 1. This invention simultaneously acquires visible light and infrared images, and then registers and fuses them, effectively combining the rich texture details of visible light images with the obvious thermal radiation characteristics of infrared images. Especially in scenarios where traditional vision is difficult to identify, such as at night, in low light, when obscured by smoke or dust, or when equipment is operating silently, the infrared modality can provide crucial heat source information, significantly improving the system's all-weather operation capability and target identification accuracy.
[0021] 2. In the first stage of this invention, a lightweight YOLOv8n model is used for rapid initial detection on visible light images, achieving efficient screening of a large range of targets and meeting real-time requirements. In the second stage, based on the initial detection results, an improved YOLOv8 model is used on multimodal fusion images for refined identification, focusing on candidate regions and improving classification accuracy and localization accuracy. This architecture achieves synergistic optimization of speed and accuracy, avoiding the performance bottleneck of a single model in complex scenarios.
[0022] 3. This invention achieves feature enhancement in both channel and spatial dimensions by introducing the CBAM attention mechanism into the YOLOv8 model: the channel attention module extracts channel statistical features through global average pooling and max pooling, and learns nonlinear relationships through a shared fully connected layer to strengthen key feature channels; the spatial attention module compresses the channel dimension to generate a spatial weight map, focusing on the target area; this mechanism enables the model to more sensitively capture the weak features of small targets, effectively suppress background noise interference, and significantly improve the recognition rate of engineering machinery under long-distance and low-resolution conditions.
[0023] 4. To further improve the accuracy of the model in detecting small targets, this invention introduces Wasserstein distance as a new evaluation metric. Compared to the traditional IoU metric, Wasserstein distance is more suitable for evaluating the similarity of bounding boxes of small targets, reducing detection errors caused by the small size of the targets.
[0024] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description
[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0026] Figure 1 A flowchart illustrating a method for identifying engineering machinery according to an embodiment of the present invention is shown; Figure 2 A schematic diagram illustrating the feature extraction process of the refined recognition model according to an embodiment of the present invention is shown. Figure 3 A schematic diagram of the structure of an engineering machinery identification system according to an embodiment of the present invention is shown; Figure 4 A schematic diagram of an electronic device structure according to an embodiment of the present invention is shown. Detailed Implementation
[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0028] Figure 1 A schematic flowchart of an engineering machinery identification method according to an embodiment of the present invention is shown, as follows: Figure 1 As shown, an embodiment of the present invention provides a method for identifying engineering machinery, comprising: Acquire the original image, and preprocess the original image to obtain a preprocessed image; Preliminary detection is performed on the preprocessed image to obtain preliminary detection results; Multiple preprocessed images acquired synchronously and preprocessed under the same field of view are registered and fused to obtain a registered and fused image. A refined recognition model is constructed, and based on the preliminary detection results, the refined recognition model is used to perform target recognition on the registered and fused image to obtain refined recognition results.
[0029] Optionally, the original image includes a visible light image and an infrared image, which are acquired simultaneously by a mid-to-high-angle camera.
[0030] Optionally, preprocessing the original image specifically includes: The original image is subjected to image enhancement processing to obtain an enhanced image; The enhanced image is then subjected to noise reduction processing to obtain a denoised image; The denoised image is then subjected to scale normalization to obtain a preprocessed image.
[0031] In this embodiment, since the construction machinery appears in farmland, construction sites, and mines, the background environment is relatively complex. Therefore, image preprocessing is required to enhance the contrast between the construction machinery and the background. During image enhancement, brightness and contrast are adjusted to enhance image details. Noise reduction is performed by using filters (such as Gaussian filters) to reduce noise interference in the image. Scale normalization is used to adjust the image size to meet the input requirements of the subsequent YOLOv8 detection model.
[0032] Optionally, the preliminary test results are as follows: A detection model is constructed, and the detection model is trained using the previously collected mid-to-high-angle video frame images to obtain a trained detection model. The preprocessed image is initially detected using the trained inspection model to obtain preliminary detection results.
[0033] In this embodiment, the detection model selected is the high-precision YOLOv8n model. The video frame images at mid-to-high altitudes have been manually and precisely labeled with bounding boxes and category information for various types of construction machinery (such as excavators, loaders, dump trucks, bulldozers, etc.). By training on this specific dataset, the model can learn the unique visual features of the target from a high-altitude perspective, significantly improving its detection accuracy and robustness in practical application scenarios.
[0034] Optionally, the registered and fused image is obtained as follows: Acquire visible light and infrared images simultaneously acquired within the same field of view; A deep learning registration model is constructed and the modal mapping relationship between visible light images and infrared images is automatically learned using the deep learning registration model; The learned deep learning registration model is used to process visible light images and infrared images under the same field of view to obtain a registered and fused image.
[0035] In this embodiment, a deep learning registration model is used to directly output the transformation matrix or aligned infrared image required for registration, eliminating the need for manually designed feature extraction algorithms. This achieves automated, high-precision registration and deep fusion of infrared and visible light images.
[0036] In this embodiment, before registration and fusion, it is first determined whether the single-light scene meets the alarm conditions. If it does, image fusion is not required; if registration and fusion are required, it is determined whether the image resolution meets the standard. If it does not meet the standard, resampling technology is used to synchronize the two images. This step is a basic preprocessing step to ensure the subsequent registration and fusion effect.
[0037] Optionally, the detailed recognition results are as follows: An improved YOLOv8 model incorporating the CBAM attention mechanism is adopted as the fine recognition model; Based on the preliminary detection results, the input registered and fused images are subjected to global average pooling and global max pooling respectively using a fine recognition model, resulting in channel statistical vectors and channel saliency vectors. The channel statistical vector and the channel saliency vector are input into a shared fully connected layer to learn the nonlinear relationship between channels, thereby obtaining the channel statistical nonlinear vector and the channel saliency nonlinear vector. The channel statistical nonlinear vector and the channel saliency nonlinear vector are added together, and the channel attention weight vector is generated by passing the Sigmoid activation function. The channel attention weight vector is multiplied channel by channel with the input registration and fusion image to obtain the channel attention image; The channel attention image is subjected to max pooling and average pooling in the channel dimension to obtain a spatial saliency map and a spatial context map, respectively. The spatial saliency map and the spatial context map are then concatenated in the channel dimension to obtain a dual-channel feature map. The dual-channel feature map is input into a convolutional layer for convolution processing to obtain a single-channel spatial attention feature map, which is then processed using the Sigmoid activation function to obtain spatial attention weights. The spatial attention weights are multiplied channel by channel by channel of the input channel attention image to obtain the spatial attention image; The spatial attention image is input into the neck network of the fine recognition model for multi-scale feature fusion. In the detection head, the feature maps at different scales are decoded to predict the bounding box, category, and confidence of the target, thus obtaining the fine recognition result.
[0038] In this embodiment, as Figure 2 As shown, in the channel attention branch, global max pooling and global average pooling are performed on the input feature layer, and the outputs are fed into a shared fully connected layer to learn the non-linear relationship between channels. The two output vectors are then added together and the channel attention weight vector M is generated by passing the sigmoid activation function. C In the spatial attention branch, the feature maps are subjected to max pooling and average pooling along the channel dimension, compressing the channel dimension to 1 while preserving spatial information. The pooled features are concatenated and passed through a convolutional layer to extract features, while simultaneously reducing the channel dimension to 1. Finally, a sigmoid activation is applied to obtain the final spatial attention weights.
[0039] Optional, also includes: Wasserstein distance is introduced as an evaluation metric for the refined recognition model, specifically: The bounding box of each engineering machinery target is modeled as a probability distribution in two-dimensional space; where the bounding box includes the predicted box and the ground truth box. The Wasserstein distance to the engineering machinery target is calculated using the following formula:
[0040] Among them, B t B represents the true bounding box. pIndicates the prediction box. T represents the Euclidean distance between the center points of the ground truth bounding boxes and the center points of the predicted bounding boxes. r Σ represents the trace of a matrix. t Σ represents the covariance matrix of the true bounding boxes. p This represents the covariance matrix of the predicted bounding boxes; The Wasserstein distance is normalized to obtain the normalized Wasserstein distance; The localization loss component in the loss function of the fine-grained recognition model is replaced with the normalized Wasserstein distance, as shown in the following equation:
[0041] Where, λ cls , λ conf , λ loc L represents the weighting coefficients, respectively. cls L represents the classification loss. conf This represents the confidence loss.
[0042] In this embodiment, to incorporate Wasserstein distance into the loss function of the YOLOv8 model, it is necessary to combine the existing loss function structure and add a component based on Wasserstein distance. The YOLOv8 loss function typically consists of three parts: classification loss, localization loss, and confidence loss. We replace the localization loss component with a metric based on Wasserstein distance. Classification loss L cls The confidence loss L is used to measure the difference between the predicted class and the true class (typically using cross-entropy loss). conf Used to measure the difference between the confidence level of including the target in the predicted bounding box and the actual situation. Weighting coefficient λ cls , λ conf , λ loc Used to balance the different parts of the loss function.
[0043] Based on and Figure 1 Using the same principle as the method shown, this embodiment of the invention also provides an engineering machinery identification system, such as... Figure 3 As shown, it includes: The preprocessing module is used to acquire the original image, preprocess the original image, and obtain a preprocessed image. The preliminary detection module is used to perform preliminary detection on the preprocessed image and obtain preliminary detection results; The registration and fusion module is used to register and fuse multiple preprocessed images that are simultaneously acquired and preprocessed under the same field of view to obtain a registered and fused image. The fine recognition module is used to construct a fine recognition model. Based on the preliminary detection results, the fine recognition model is used to perform target recognition on the registered and fused image to obtain fine recognition results.
[0044] The metric evaluation module is used to introduce Wasserstein distance as an evaluation metric for the refined recognition model.
[0045] The construction machinery identification system of this invention can execute the construction machinery identification method provided in this invention. The implementation principle is similar. The actions performed by each module and unit in the construction machinery identification system of each embodiment of this invention correspond to the steps in the construction machinery identification method of each embodiment of this invention. For detailed functional descriptions of each module of the construction machinery identification system, please refer to the descriptions in the corresponding construction machinery identification methods shown above. They will not be repeated here.
[0046] The aforementioned construction machinery identification system may be a computer program (including program code) running on a computer device, such as an application software; the application software may be used to execute the corresponding steps in the method provided in the embodiments of the present invention.
[0047] The modules described in the embodiments of the present invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.
[0048] Based on the same principles as the methods shown in the embodiments of the present invention, the embodiments of the present invention also provide an electronic device, which may include, but is not limited to: a processor and a memory; the memory for storing computer programs; and the processor for executing the methods shown in any embodiment of the present invention by invoking the computer programs.
[0049] In one alternative embodiment, an electronic device is provided, such as Figure 4 As shown, Figure 4 The illustrated electronic device includes a processor and a memory. The processor and memory are connected, for example, via a bus. Optionally, the electronic device may also include a transceiver, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver is not limited to one unit, and the structure of this electronic device does not constitute a limitation on the embodiments of the present invention.
[0050] The memory stores application code (computer program) that executes the present invention, and its execution is controlled by a processor. The processor executes the application code stored in the memory to implement the content shown in the foregoing method embodiments.
[0051] Among these, electronic devices can also be terminal devices. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.
[0052] This invention provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.
[0053] According to another aspect of the present invention, a computer program product or computer program is also provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various embodiments described above.
[0054] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.
[0055] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.
Claims
1. A method for identifying engineering machinery, characterized in that, The method includes: Acquire the original image, and preprocess the original image to obtain a preprocessed image; Preliminary detection is performed on the preprocessed image to obtain preliminary detection results; Multiple preprocessed images acquired synchronously and preprocessed under the same field of view are registered and fused to obtain a registered and fused image. A refined recognition model is constructed, and based on the preliminary detection results, the refined recognition model is used to perform target recognition on the registered and fused image to obtain refined recognition results.
2. The method for identifying engineering machinery according to claim 1, characterized in that, The original images include visible light images and infrared images, which are acquired synchronously by mid-to-high-angle cameras.
3. The method for identifying engineering machinery according to claim 1, characterized in that, The preprocessing of the original image specifically includes: The original image is subjected to image enhancement processing to obtain an enhanced image; The enhanced image is then subjected to noise reduction processing to obtain a denoised image; The denoised image is then subjected to scale normalization to obtain a preprocessed image.
4. The method for identifying engineering machinery according to claim 1, characterized in that, The preliminary test results are as follows: A detection model is constructed, and the detection model is trained using the previously collected mid-to-high-angle video frame images to obtain a trained detection model. The preprocessed image is initially detected using the trained inspection model to obtain preliminary detection results.
5. The method for identifying engineering machinery according to claim 2, characterized in that, The resulting registered and fused image is as follows: Acquire visible light and infrared images simultaneously acquired within the same field of view; A deep learning registration model is constructed and the modal mapping relationship between visible light images and infrared images is automatically learned using the deep learning registration model; The learned deep learning registration model is used to process visible light images and infrared images under the same field of view to obtain a registered and fused image.
6. The method for identifying engineering machinery according to claim 1, characterized in that, The detailed recognition results are as follows: An improved YOLOv8 model incorporating the CBAM attention mechanism is adopted as the fine recognition model; Based on the preliminary detection results, the input registered and fused images are subjected to global average pooling and global max pooling respectively using a fine recognition model, resulting in channel statistical vectors and channel saliency vectors. The channel statistical vector and the channel saliency vector are input into a shared fully connected layer to learn the nonlinear relationship between channels, thereby obtaining the channel statistical nonlinear vector and the channel saliency nonlinear vector. The channel statistical nonlinear vector and the channel saliency nonlinear vector are added together, and the channel attention weight vector is generated by passing the Sigmoid activation function. The channel attention weight vector is multiplied channel by channel with the input registration and fusion image to obtain the channel attention image; The channel attention image is subjected to max pooling and average pooling in the channel dimension to obtain a spatial saliency map and a spatial context map, respectively. The spatial saliency map and the spatial context map are then concatenated in the channel dimension to obtain a dual-channel feature map. The dual-channel feature map is input into a convolutional layer for convolution processing to obtain a single-channel spatial attention feature map, which is then processed using the Sigmoid activation function to obtain spatial attention weights. The spatial attention weights are multiplied channel by channel by channel of the input channel attention image to obtain the spatial attention image; The spatial attention image is input into the neck network of the fine recognition model for multi-scale feature fusion. In the detection head, the feature maps at different scales are decoded to predict the bounding box, category, and confidence of the target, thus obtaining the fine recognition result.
7. The method for identifying engineering machinery according to claim 1, characterized in that, Also includes: Wasserstein distance is introduced as an evaluation metric for the refined recognition model, specifically: The bounding box of each engineering machinery target is modeled as a probability distribution in two-dimensional space; where the bounding box includes the predicted box and the ground truth box. The Wasserstein distance to the engineering machinery target is calculated using the following formula: Among them, B t B represents the true bounding box. p Indicates the prediction box. T represents the Euclidean distance between the center points of the ground truth bounding boxes and the center points of the predicted bounding boxes. r Σ represents the trace of a matrix. t Σ represents the covariance matrix of the true bounding boxes. p This represents the covariance matrix of the predicted bounding boxes; The Wasserstein distance is normalized to obtain the normalized Wasserstein distance; The localization loss component in the loss function of the fine-grained recognition model is replaced with the normalized Wasserstein distance, as shown in the following equation: Where, λ cls , λ conf , λ loc L represents the weighting coefficients, respectively. cls L represents the classification loss. conf This represents the confidence loss.
8. An engineering machinery identification system, characterized in that, include: The preprocessing module is used to acquire the original image, preprocess the original image, and obtain a preprocessed image. The preliminary detection module is used to perform preliminary detection on the preprocessed image and obtain preliminary detection results; The registration and fusion module is used to register and fuse multiple preprocessed images that are simultaneously acquired and preprocessed under the same field of view to obtain a registered and fused image. The fine recognition module is used to construct a fine recognition model. Based on the preliminary detection results, the fine recognition model is used to perform target recognition on the registered and fused image to obtain fine recognition results.
9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method of any one of claims 1-7.
10. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, implements the method of any one of claims 1-7.