Industrial park group carbon emission detection method, system, equipment and medium

Abstract

The invention discloses an industrial park group carbon emission detection method, system and device and a medium, and the method comprises the steps: inputting a preprocessed image into an object recognition model based on a convolutional neural network and a feature pyramid network, extracting and fusing multi-scale features, obtaining a recognition result containing the category and the position of the recognized object and the corresponding multi-scale object features; calculating global correlation weights among pixel points in the feature map through a self-attention mechanism, and carrying out weighted optimization on features to obtain an optimization recognition result containing the finally confirmed object category and number; the individual carbon emission of each object is obtained by calling a carbon emission calculation model matched with the object type and combining the operation state parameters corresponding to the object type; and accumulating and summarizing the carbon emissions of all individuals to obtain the total carbon emission of the group, and outputting a result in a visual form. According to the method, the automation level, the space coverage capability and the data accuracy of carbon emission detection are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of carbon emission detection technology, and in particular to a method, system, equipment and medium for detecting carbon emissions from industrial parks. Background Technology

[0002] Traditional carbon emission quantification primarily relies on enterprise-level accounting methods, which involve bottom-up calculations and aggregations based on activity-level data (such as fossil fuel consumption) and emission factors. CEMS (Carbon Emission Monitoring System) achieves near real-time monitoring of point source emissions by directly measuring the concentration and flow rate of greenhouse gases such as CO2 in flue gas, offering high data reliability. Internet of Things (IoT) sensor networks, through the widespread deployment of low-cost, distributed environmental sensors (monitoring CO2, CH4 concentrations, etc.) within the park, combined with meteorological data and reverse modeling techniques, construct a real-time monitoring network covering the entire park, capable of retrieving regional emission fluxes and identifying anomalous hotspots. Space-based remote sensing technologies (such as the high-resolution greenhouse gas monitoring satellite GHGSat and UAV-borne spectrometers) provide a macroscopic perspective, enabling rapid screening of regional emission hotspots and tracking of leaks, achieving non-contact, large-scale monitoring.

[0003] Currently, image recognition technology based on carbon emission detection has made significant progress in the field of object detection and classification, but it still faces multiple challenges when applied to the detection of carbon emissions from groups in industrial parks: significant differences in carbon emission characteristics of different objects, strong interference from complex backgrounds, and insufficient adaptability of carbon emission calculation models to scenarios. These issues make it difficult for existing technologies to efficiently and accurately achieve real-time detection and quantification of carbon emissions from groups in industrial parks, thus restricting the level of refinement in carbon emission management in industrial parks. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method, system, equipment, and medium for detecting group carbon emissions in industrial parks, which solves the problems of low detection efficiency and poor accuracy in existing group carbon emission detection methods.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for detecting carbon emissions from industrial parks as a whole, comprising: Obtain the original image of the target scene; The preprocessed image is input into an object recognition model based on convolutional neural networks and feature pyramid networks. Multi-scale features are extracted and fused to obtain recognition results that include the category, location and corresponding multi-scale object features of the recognized objects. Based on the multi-scale object features in the recognition results, the global association weights between each pixel in the feature map are calculated through a self-attention mechanism, and the features are weighted and optimized to obtain an optimized recognition result that includes the finally confirmed object category and number. Based on the object categories and quantities in the optimized identification results, the individual carbon emissions of each object are obtained by calling the carbon emission calculation model that matches the object type and combining the running status parameters corresponding to the object category. The total carbon emissions of all individuals are summed up to obtain the total carbon emissions of the group, and the results are output in a visual format.

[0007] As a preferred embodiment of the industrial park group carbon emission detection method described in this invention, the step of obtaining the identification result, which includes the category, location, and corresponding multi-scale object features of the identified objects, includes: The basic visual features of the image are extracted layer by layer by the convolutional layers of the convolutional neural network to obtain multiple feature maps containing the basic features. Pooling layers are used to downsample multiple feature maps obtained from convolutional layers to obtain low-dimensional feature maps; Based on the low-dimensional feature map output by the pooling layer, the low-level detailed features are extracted through the bottom-up path of the feature pyramid network, and the high-level semantic features and low-level detailed features are fused through the top-down path to obtain a multi-scale feature map. Based on multi-scale feature maps, the feature information is integrated and fused through a fully connected layer. A classification operation is performed to determine the object category, and a localization operation is performed to generate the bounding box position. The output is a recognition result containing the object category, bounding box position, and corresponding multi-scale feature vectors.

[0008] The beneficial effects of this preferred technical solution are that, by extracting the basic visual features of the image layer by layer through a convolutional neural network and reducing the dimensionality using a pooling layer, and then fusing high-level semantics and low-level detail features at multiple scales through a feature pyramid network, the feature information is finally integrated to achieve accurate object classification and localization. This method effectively improves the recognition accuracy and localization accuracy of multi-size objects in complex scenes.

[0009] As a preferred embodiment of the industrial park group carbon emission detection method described in this invention, the step of obtaining optimized identification results including the finally confirmed object categories and quantities includes: The multi-scale feature map output by the feature pyramid network is input into the self-attention module integrated into the object recognition model; Within the self-attention module, the multi-scale object features in the recognition results are mapped into query vectors, key vectors, and value vectors; The attention weight distribution is obtained by calculating the dot product between the query vector and the key vector and then normalizing it. Based on the attention weight distribution, a weighted summation operation is performed on the value vector to obtain the optimized feature representation; Based on the optimized feature representation, the optimized recognition result, which includes the final confirmed object category and quantity, is obtained through classification and quantity statistics operations.

[0010] The beneficial effects of this preferred technical solution are that by optimizing the multi-scale feature map output by the feature pyramid network through the self-attention module, and focusing on important features according to the attention weight, the accuracy of object classification and quantity statistics and the overall recognition performance of the model are improved.

[0011] As a preferred embodiment of the industrial park group carbon emission detection method described in this invention, the step of obtaining the individual carbon emission of each object includes: Based on the object category determined in the optimized identification results, a carbon emission calculation model that matches the current object category is called from the preset model library; Get the running status parameters corresponding to the current object category; Input the object category, quantity, and operating status parameters into the corresponding carbon emission calculation model; The model calculation process is executed to obtain the individual carbon emissions of each object.

[0012] As a preferred embodiment of the industrial park group carbon emission detection method described in this invention, the step of summing up all individual carbon emissions to obtain the total group carbon emissions and outputting the results in a visual format includes: Perform an summation operation on all calculated individual carbon emissions to generate the total group carbon emissions for the target scenario; The total carbon emissions of the group will be presented in numerical form. Generate data visualizations and output visualization reports that include both numerical values ​​and graphics.

[0013] As a preferred embodiment of the industrial park group carbon emission detection method described in this invention, the preprocessing includes: The color image is converted to a grayscale image using grayscale processing. The Gaussian filtering algorithm is applied to suppress noise in grayscale images. Histogram equalization is used to enhance image contrast, resulting in a preprocessed image.

[0014] As a preferred embodiment of the industrial park group carbon emission detection method described in this invention, the step of obtaining the operating status parameters corresponding to the current object category includes: Based on the object category determined in the optimized recognition results, search and retrieve the preset parameter set associated with that category from the preset parameter database; The preset parameter set includes power and usage time for electrical equipment, as well as mileage and fuel consumption per unit mileage for transportation vehicles. The invocation of the carbon emission calculation model and the acquisition of operating status parameters are executed simultaneously.

[0015] Secondly, the present invention provides a group carbon emission detection system for industrial parks, comprising: The image acquisition module is used to acquire raw images of the target scene; The object recognition module is used to input the preprocessed image into an object recognition model based on convolutional neural networks and feature pyramid networks, extract and fuse multi-scale features, and obtain recognition results that include the category, location and corresponding multi-scale object features of the recognized objects. The feature optimization module is used to calculate the global association weights between pixels in the feature map based on the multi-scale object features in the recognition results through a self-attention mechanism, and to perform weighted optimization on the features to obtain an optimized recognition result that includes the finally confirmed object category and quantity. The carbon emission calculation module is used to obtain the individual carbon emission of each object based on the object category and quantity in the optimized identification results, by calling the carbon emission calculation model that matches the object type, and combining the running status parameters corresponding to the object category. The carbon emissions aggregation and output module is used to sum up the carbon emissions of all individuals to obtain the total carbon emissions of the group, and output the results in a visual format.

[0016] Thirdly, the present invention provides an electronic device, comprising: Memory, used to store programs; A processor is configured to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the industrial park group carbon emission detection method.

[0017] Fourthly, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the steps of implementing the industrial park group carbon emission detection method.

[0018] The beneficial effects of this invention are as follows: By constructing an object recognition model based on convolutional neural networks and feature pyramid networks, this invention extracts basic visual features layer by layer and fuses multi-scale information to achieve accurate detection and positioning of objects of different sizes and categories in industrial parks; by integrating a self-attention module after the feature pyramid network, attention weights are used to perform global correlation modeling and weighted optimization of multi-scale feature maps, thereby strengthening key object features and suppressing background interference, further improving the accuracy of object category and quantity recognition; by calling a carbon emission calculation model that matches the object type based on the optimized recognition results, and combining it with the corresponding operating status parameters in the preset parameter database, the invention achieves automated and refined calculation of carbon emissions of individual equipment; by summing up the carbon emissions of all individuals and outputting the total carbon emissions of the group in the form of data visualization graphics, the invention achieves intuitive display and dynamic monitoring of the overall carbon emission level of the industrial park, providing data support for carbon management decisions. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the basic process of a method for detecting carbon emissions from industrial parks, provided as an embodiment of the present invention.

[0020] Figure 2 A comparison chart of carbon emissions in an industrial park, provided as an embodiment of the present invention, illustrates a method for detecting carbon emissions from industrial parks. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0022] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for detecting carbon emissions from a cluster of industrial parks is provided, comprising: S100: Acquire the original image of the target scene; S200: Input the preprocessed image into an object recognition model based on convolutional neural networks and feature pyramid networks, extract and fuse multi-scale features, and obtain recognition results that include the category, location and corresponding multi-scale object features of the recognized objects. S300: Based on the multi-scale object features in the recognition results, the global association weights between each pixel in the feature map are calculated through a self-attention mechanism, and the features are weighted and optimized to obtain an optimized recognition result that includes the finally confirmed object category and quantity. S400: Based on the object category and quantity in the optimized identification results, the individual carbon emissions of each object are obtained by calling the carbon emission calculation model that matches the object type and combining the running status parameters corresponding to the object category. S500: The total carbon emissions of all individuals are summed up to obtain the total carbon emissions of the group, and the results are output in a visual form.

[0023] It should be noted that existing methods for detecting group carbon emissions face numerous challenges during operation. These include reliance on manual surveys and static record-keeping, making it difficult to achieve real-time, dynamic monitoring of numerous mobile and dispersed stationary sources within industrial parks. Data updates are also delayed and coverage is limited. Acquiring equipment operating parameters (such as usage time and mileage) relies on manual reporting or simple sensors, leading to inaccurate data and omissions, resulting in insufficient accuracy in carbon emission accounting. Most methods use general emission factors for coarse-grained estimation, lacking refined modeling for different equipment types and operating conditions, making it difficult to reflect true emission levels. Furthermore, existing technologies generally lack deep integration with spatial information and visual recognition, failing to automatically identify the type and quantity of objects in a scene. This results in low automation, high labor costs, and slow response times, making it difficult to meet the demands of modern industrial parks for efficient, accurate, and intelligent carbon emission monitoring.

[0024] Therefore, addressing the issues of low detection efficiency and poor accuracy in existing group carbon emission detection methods, this paper proposes a comprehensive technical solution through steps S100-S500. This solution integrates image acquisition and preprocessing, intelligent recognition based on CNN-FPN and self-attention mechanisms, and matching carbon emission calculation and visualization output. This effectively overcomes the problems of manual labor, low accuracy, and poor real-time performance in traditional group carbon emission detection methods. It enables automatic identification, accurate classification, and dynamic calculation of multi-source heterogeneous emission objects within industrial parks, significantly improving the automation level, spatial coverage, and data accuracy of carbon emission detection. This provides industrial parks with an efficient, intuitive, and scalable tool for carbon emission quantification and management.

[0025] Example 2, this is an embodiment of the present invention, which provides a method for detecting group carbon emissions in industrial parks based on the previous embodiment, including: In this embodiment, step S100 uses devices such as cameras and drones to acquire multi-angle, high-resolution images of the target scene, ensuring that the original image data covers multiple objects. Cameras are suitable for long-term, continuous image acquisition of fixed areas or key nodes (such as entrances and exits, energy stations, and production line passages) within industrial parks, and are particularly suitable for monitoring the operating status of frequently active mobile sources (such as transport vehicles) or fixed equipment. They have the advantages of stable deployment, low maintenance costs, and strong real-time data. Drones, on the other hand, are suitable for large-scale, complex terrain or areas where it is difficult to deploy fixed equipment. They are used for periodic panoramic inspections, rooftop equipment inspections, or emergency investigations. They can acquire high-resolution images from multiple angles in the air, effectively covering the blind spots of cameras and improving the comprehensiveness and flexibility of monitoring. By selecting cameras or drones according to the needs of the scenario, efficient and complementary image acquisition of multi-scale and multi-type objects in industrial parks can be achieved, ensuring the integrity and representativeness of the original data.

[0026] In this embodiment of the application, the preprocessing in step S200 includes: The color image is converted to a grayscale image using grayscale processing. The Gaussian filtering algorithm is applied to suppress noise in grayscale images. Histogram equalization is used to enhance image contrast, resulting in a preprocessed image.

[0027] In this embodiment of the application, the denoising process in the image preprocessing in step S200 includes using a Gaussian filtering algorithm to suppress noise in the grayscale image. Gaussian filtering smooths the image by weighted averaging, effectively removing Gaussian white noise while preserving edge information to a certain extent.

[0028] In an optional implementation, the denoising process in the image preprocessing in step S200 can also be performed after grayscale processing by applying a median filtering algorithm to the grayscale image. This involves traversing the image through a sliding window and replacing the center pixel value with the median of the neighboring pixels to remove salt-and-pepper noise and retain edge information, thereby obtaining a denoised image.

[0029] In an optional implementation, the denoising process in the image preprocessing in step S200 can also be performed by applying the NLM algorithm to the grayscale image after grayscale processing. This algorithm searches for similar pixel blocks in the image and performs a weighted average to effectively suppress noise while preserving texture and details, thus obtaining a denoised image.

[0030] In this embodiment of the application, the image contrast enhancement in step S200 includes applying a histogram equalization method to the grayscale image after denoising, thereby expanding the dynamic range of the image by redistributing pixel grayscale values, thus improving the overall contrast and obtaining a preprocessed image with enhanced contrast.

[0031] In an optional implementation, the image contrast enhancement in step S200 can also apply adaptive histogram equalization (CLAHE) to the denoised grayscale image. By dividing the image into local regions and performing histogram equalization independently in each region, while limiting the contrast gain to avoid noise amplification, the local contrast is improved and details are preserved, resulting in an enhanced preprocessed image.

[0032] In an optional implementation, the image contrast enhancement in step S200 can also apply a Retinex algorithm (such as MSRCR) to the denoised grayscale image. By estimating and separating the illumination component and the reflection component, the shadows and uneven illumination are corrected, and the true brightness and details are restored, thereby improving the image contrast and visual quality and obtaining the enhanced preprocessed image.

[0033] In this embodiment of the application, the feature extraction backbone network in the object recognition model in step S200 includes using a convolutional neural network (CNN) as the backbone network (such as ResNet, VGG, etc.), extracting basic visual features such as edges and textures of the image layer by layer through convolutional layers, and reducing dimensionality by combining pooling layers to form a hierarchical feature representation.

[0034] In an optional implementation, the feature extraction backbone network in the object recognition model in step S200 can further segment the preprocessed image into multiple image blocks, convert them into feature vectors through linear projection and add position encoding, input them into the Transformer encoder for multi-layer self-attention calculation to extract global features, and then combine them with the feature pyramid network to fuse the output multi-scale features for subsequent object classification and localization.

[0035] In an optional implementation, the feature extraction backbone network in the object recognition model in step S200 can first extract local basic features of the image through a convolutional neural network, and then input the feature map into the Transformer module for global context modeling. By alternately stacking convolutional layers and self-attention layers, local details and global semantic information are fused, and multi-scale features are output and fed into the feature pyramid network for further fusion, which is used for subsequent object classification and localization.

[0036] In this embodiment of the application, step S200 obtains a recognition result including the category, location, and corresponding multi-scale object features of the identified object, including: The basic visual features of the image are extracted layer by layer by the convolutional layers of the convolutional neural network to obtain multiple feature maps containing the basic features. Pooling layers are used to downsample multiple feature maps obtained from convolutional layers to obtain low-dimensional feature maps; Based on the low-dimensional feature map output by the pooling layer, the low-level detailed features are extracted through the bottom-up path of the feature pyramid network, and the high-level semantic features and low-level detailed features are fused through the top-down path to obtain a multi-scale feature map. Based on multi-scale feature maps, the feature information is integrated and fused through a fully connected layer. A classification operation is performed to determine the object category, and a localization operation is performed to generate the bounding box position. The output is a recognition result containing the object category, bounding box position, and corresponding multi-scale feature vectors.

[0037] In this embodiment, convolutional neural networks (CNNs) are used to extract hierarchical visual features from the input image layer by layer. Shallow convolutional layers capture local details such as edges and corners, while deep convolutional layers extract high-level semantic features to distinguish different object categories. The Feature Pyramid Network (FPN) is built on top of the CNN backbone. It retains high-resolution low-level detail features through a bottom-up path and uses a top-down path to pass high-level semantic information to low-level feature maps through upsampling. Then, it fuses high-level semantics and low-level details of the same spatial scale through lateral connections, thereby generating multi-scale feature maps with rich semantic information and high resolution. This structure effectively enhances the model's ability to detect objects of different distances and sizes (such as small equipment in the distance and large vehicles nearby) in industrial parks, and solves the problems of missed detection and inaccurate localization of small targets caused by single-scale feature extraction.

[0038] In this embodiment of the application, step S300, which yields an optimized recognition result including the final confirmed object category and quantity, includes: The multi-scale feature map output by the feature pyramid network is input into the self-attention module integrated into the object recognition model; Within the self-attention module, the multi-scale object features in the recognition results are mapped into query vectors, key vectors, and value vectors; The attention weight distribution is obtained by calculating the dot product between the query vector and the key vector and then normalizing it. Based on the attention weight distribution, a weighted summation operation is performed on the value vector to obtain the optimized feature representation; Based on the optimized feature representation, the optimized recognition result, which includes the final confirmed object category and quantity, is obtained through classification and quantity statistics operations.

[0039] In this embodiment of the application, the method of calling the carbon emission calculation model and obtaining parameters in step S400 includes calling the matching carbon emission calculation model from the preset model library according to the object category in the optimization identification result, and searching for the corresponding operating status parameters (such as power and driving mileage) from the preset parameter database. The two are executed synchronously to realize automated calculation.

[0040] In an optional implementation, the method of calling the carbon emission calculation model and acquiring parameters in step S400 can also be based on the object category determined by the optimized identification results, and then the operating status parameters are collected in real time by external sensors such as electricity meters, fuel level sensors or GPS associated with it. The collected real-time data is combined with the called carbon emission calculation model to perform dynamic calculation of individual carbon emissions.

[0041] In an optional implementation, the method of calling the carbon emission calculation model and obtaining parameters in step S400 can also be based on the optimized identification results to determine the object category, inputting historical operating data, environmental factors and equipment usage patterns into a pre-trained machine learning prediction model to generate the predicted value of the operating status parameters at the current moment, and then combining the predicted value with the matching carbon emission calculation model to calculate the individual carbon emissions of each object.

[0042] In this embodiment of the application, step S400, which obtains the individual carbon emissions of each object, includes: Based on the object category determined in the optimized identification results, a carbon emission calculation model that matches the current object category is called from the preset model library; Get the running status parameters corresponding to the current object category; Input the object category, quantity, and operating status parameters into the corresponding carbon emission calculation model; The model calculation process is executed to obtain the individual carbon emissions of each object.

[0043] In this embodiment of the application, a preset model library stores carbon emission calculation models for different object categories. The carbon emission calculation models are pre-established according to the object type, including an energy consumption-based emission calculation model applicable to stationary sources and an emission calculation model based on mileage and fuel consumption applicable to mobile sources. The preset model library is mapped to the category labels in the object recognition results, so that the system can automatically match the corresponding carbon emission accounting model according to the recognized object category, thereby realizing the automation and standardization of model calling.

[0044] In this embodiment of the application, step S400, obtaining the running state parameters corresponding to the current object category, includes: Based on the object category determined in the optimized recognition results, search and retrieve the preset parameter set associated with that category from the preset parameter database; The preset parameter set includes power and usage time for electrical equipment, as well as mileage and fuel consumption per unit mileage for transportation vehicles. The carbon emission calculation model is invoked and the operational status parameters are obtained simultaneously.

[0045] In this embodiment of the application, the preset parameter database stores a set of operating status parameters corresponding to various objects. The parameter set is obtained through historical operating data, equipment nameplate information, industry standards or external sensors, and is stored in a structured manner according to object category. The parameter database supports dynamic updates and expansion to adapt to the management needs of changes in equipment types or new emission sources in industrial parks, and to ensure the timeliness and accuracy of carbon emission accounting parameters.

[0046] In this embodiment, the object recognition model undergoes a training phase before deployment. The training dataset contains image samples of industrial parks labeled with object categories, bounding boxes, and multi-scale features. The model parameters are optimized through supervised learning to enable the model to identify typical emission sources in industrial scenarios (such as diesel forklifts, boilers, air conditioner outdoor units, power transformers, etc.). The training process includes joint optimization of convolutional neural networks, feature pyramid networks, and integrated self-attention modules to improve the model's robustness to complex backgrounds, occlusions, and multi-scale targets.

[0047] In this embodiment of the application, step S500 involves summing up the carbon emissions of all individuals to obtain the total carbon emissions of the group, and outputting the result in a visual format, including: Perform an summation operation on all calculated individual carbon emissions to generate the total group carbon emissions for the target scenario; The total carbon emissions of the group will be presented in numerical form. Generate data visualizations and output visualization reports that include both numerical values ​​and graphics.

[0048] In this embodiment, the data visualization graphics include, but are not limited to, bar charts, line charts, pie charts, or heat maps, used to display the carbon emission distribution characteristics of different regions, different time periods, or different types of objects; the visualization report is output through a graphical interface or in the form of a report, supporting multi-perspective analysis by time dimension, spatial dimension, or device type, which makes it easier for managers to intuitively grasp the dynamics of carbon emissions and formulate emission reduction strategies.

[0049] In this embodiment, the total carbon emissions of the group can be statistically analyzed and compared by time period (such as day, week, month). The system can automatically generate trend analysis reports and trigger an early warning mechanism when the carbon emissions exceed a preset threshold, pushing alarm information to the management personnel, thereby realizing dynamic monitoring and proactive management of the carbon emission level of the industrial park.

[0050] Example 3, referring to Tables 1-2 and Figure 2 This is one embodiment of the present invention, which provides a method for detecting carbon emissions from industrial parks. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through specific implementation methods and implementation effects.

[0051] The specific details of this embodiment are as follows: I. Experimental Environment This embodiment selects a medium-sized logistics park as the experimental scenario. The park has an average daily vehicle traffic of approximately 300 vehicles, including heavy-duty diesel trucks, light-duty fuel trucks, fuel-powered cars, and a small number of electric vehicles. Fixed sources mainly include warehouse lighting, air conditioning systems, and ventilation equipment. Ten high-position fixed cameras are deployed within the park, covering the main passages, loading and unloading areas, and areas with concentrated equipment, enabling 24-hour continuous video acquisition. The experimental hardware platform uses a high-performance server equipped with a GPU acceleration unit for running deep learning models. The software system is built based on the PyTorch framework, integrating image preprocessing, object recognition, feature optimization, and carbon emission calculation modules. The preset model library contains carbon emission calculation models for different vehicle types and fixed equipment, while the parameter database is constructed based on equipment nameplates, industry standards, and historical operating data. Specific parameter settings are shown in Tables 1 and 2.

[0052] Table 1 Comparison of Key Visual Recognition Parameters and Model Calculation Parameters for Mobile Sources

[0053] Table 2 Comparison of Key Visual Recognition Parameters and Model Calculation Parameters for Fixed Sources

[0054] II. Experimental Procedure 1. Image acquisition and preprocessing: Multi-angle, high-resolution video data of the park was collected by a high-position camera within 24 hours, and key frames were extracted as the original images. The images were then subjected to grayscale conversion, Gaussian filtering for noise reduction, and histogram equalization enhancement to improve image quality, resulting in pre-processed images.

[0055] 2. Object recognition and feature extraction: The preprocessed image is input into an object recognition model based on ResNet-50 as the backbone network and FPN as the feature fusion structure. The model extracts basic visual features (such as edges and textures) through convolutional layers, and after downsampling by pooling layers, the FPN fuses high-level semantics and low-level details from top to bottom to generate multi-scale feature maps. The fully connected layer performs classification and localization based on the feature maps, and outputs preliminary recognition results containing object category, bounding box position and multi-scale feature vectors.

[0056] 3. Feature optimization and result confirmation: The multi-scale feature map output by FPN is integrated into the self-attention module in the model; the module maps the features into query, key, and value vectors, calculates global attention weights and performs weighted fusion, and outputs the optimized feature representation; based on the optimized features, through classification and quantity statistics operations, the optimized recognition result containing the finally confirmed object category and quantity is obtained.

[0057] 4. Calculation of individual carbon emissions: Based on the object category in the optimized identification results, the matching carbon emission calculation model in the preset model library is called synchronously, and the corresponding operating status parameters (such as the average speed, idling time, and driving emission factor of heavy-duty diesel trucks) are searched from the parameter database. The category, quantity, and parameters are input into the model to calculate the individual carbon emissions of each vehicle and stationary equipment.

[0058] 5. Group carbon emissions aggregation and output: The carbon emissions of all individuals are summed up to obtain the total carbon emissions of the park over 24 hours. The total is presented in numerical form and data visualization graphics such as bar charts and line charts are generated to form a visualization report containing numerical values ​​and charts, which intuitively shows the spatiotemporal distribution and dynamic changes of carbon emissions.

[0059] III. Experimental Results This embodiment successfully achieved 24 / 7 automated monitoring of carbon emissions from a logistics park. Experimental results show that the system can effectively identify major mobile sources (such as heavy-duty diesel trucks and light-duty fuel-powered trucks) and stationary sources (such as air conditioning units and lighting areas) within the park, and dynamically capture their operational status. Carbon emission accounting results show that carbon emissions from the park exhibit significant temporal fluctuations, with obvious peaks during peak operating periods such as early morning, noon, and evening. Compared with traditional estimation methods that use static empirical values, this solution can more accurately reflect the emission dynamics caused by changes in actual operating intensity, avoiding the problem of overestimation of emissions during non-operating periods.

[0060] The generated carbon emission time series data, such as Figure 2 As shown, the monitoring results of this scheme exhibit significant dynamic fluctuations in the early morning, noon, and evening, which are significantly different from traditional static estimation methods. This further verifies that this method can accurately capture emission fluctuations caused by changes in actual operating intensity.

[0061] The generated visualization report clearly presents the fluctuation trend of carbon emissions, providing reliable data support for the park to implement time-of-use electricity pricing for energy conservation and optimize vehicle scheduling, and verifying the significant advantages of this method in terms of accuracy, real-time performance and practicality.

[0062] Example 4 is an embodiment of the present invention. This embodiment differs from the first embodiment in that it provides a group carbon emission detection system for industrial parks.

[0063] It should be noted that the technical solution of the industrial park group carbon emission detection system is based on the same concept as the technical solution of the aforementioned industrial park group carbon emission detection method. For details not described in detail in the technical solution of the industrial park group carbon emission detection system in this embodiment, please refer to the description of the technical solution of the aforementioned industrial park group carbon emission detection method.

[0064] This embodiment of a carbon emission monitoring system for industrial parks includes: The image acquisition module is used to acquire raw images of the target scene; The object recognition module is used to input the preprocessed image into an object recognition model based on convolutional neural networks and feature pyramid networks, extract and fuse multi-scale features, and obtain recognition results that include the category, location and corresponding multi-scale object features of the recognized objects. The feature optimization module is used to calculate the global association weights between pixels in the feature map based on the multi-scale object features in the recognition results through a self-attention mechanism, and to perform weighted optimization on the features to obtain an optimized recognition result that includes the finally confirmed object category and quantity. The carbon emission calculation module is used to obtain the individual carbon emission of each object based on the object category and quantity in the optimized identification results, by calling the carbon emission calculation model that matches the object type, and combining the running status parameters corresponding to the object category. The carbon emissions aggregation and output module is used to sum up the carbon emissions of all individuals to obtain the total carbon emissions of the group, and output the results in a visual format.

[0065] This embodiment also provides an electronic device applicable to a method for detecting group carbon emissions in an industrial park, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a method for detecting carbon emissions from industrial parks, as described in the above embodiments.

[0066] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for detecting carbon emissions from industrial parks as proposed in the above embodiments.

[0067] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for detecting carbon emissions from industrial parks proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0068] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting carbon emissions from industrial parks, characterized in that, include: Obtain the original image of the target scene; The preprocessed image is input into an object recognition model based on convolutional neural networks and feature pyramid networks. Multi-scale features are extracted and fused to obtain recognition results that include the category, location and corresponding multi-scale object features of the recognized objects. Based on the multi-scale object features in the recognition results, the global association weights between each pixel in the feature map are calculated through a self-attention mechanism, and the features are weighted and optimized to obtain an optimized recognition result that includes the finally confirmed object category and number. Based on the object categories and quantities in the optimized identification results, the individual carbon emissions of each object are obtained by calling the carbon emission calculation model that matches the object type and combining the running status parameters corresponding to the object category. The total carbon emissions of all individuals are summed up to obtain the total carbon emissions of the group, and the results are output in a visual format.

2. The method for detecting carbon emissions from industrial parks as described in claim 1, characterized in that: The obtained recognition result, which includes the category, location, and corresponding multi-scale object features of the identified object, includes: The basic visual features of the image are extracted layer by layer by the convolutional layers of the convolutional neural network to obtain multiple feature maps containing the basic features. Pooling layers are used to downsample multiple feature maps obtained from convolutional layers to obtain low-dimensional feature maps; Based on the low-dimensional feature map output by the pooling layer, the low-level detailed features are extracted through the bottom-up path of the feature pyramid network, and the high-level semantic features and low-level detailed features are fused through the top-down path to obtain a multi-scale feature map. Based on multi-scale feature maps, the feature information is integrated and fused through a fully connected layer. A classification operation is performed to determine the object category, and a localization operation is performed to generate the bounding box position. The output is a recognition result containing the object category, bounding box position, and corresponding multi-scale feature vectors.

3. The method for detecting carbon emissions from industrial park clusters as described in claim 2, characterized in that: The process of obtaining optimized recognition results that include the finally confirmed object categories and quantities includes: The multi-scale feature map output by the feature pyramid network is input into the self-attention module integrated into the object recognition model; Within the self-attention module, the multi-scale object features in the recognition results are mapped into query vectors, key vectors, and value vectors; The attention weight distribution is obtained by calculating the dot product between the query vector and the key vector and then normalizing it. Based on the attention weight distribution, a weighted summation operation is performed on the value vector to obtain the optimized feature representation; Based on the optimized feature representation, the optimized recognition result, which includes the final confirmed object category and quantity, is obtained through classification and quantity statistics operations.

4. The method for detecting carbon emissions from industrial park clusters as described in claim 3, characterized in that: The individual carbon emissions of each object are obtained, including: Based on the object category determined in the optimized identification results, a carbon emission calculation model that matches the current object category is called from the preset model library; Get the running status parameters corresponding to the current object category; Input the object category, quantity, and operating status parameters into the corresponding carbon emission calculation model; The model calculation process is executed to obtain the individual carbon emissions of each object.

5. The method for detecting carbon emissions from industrial park clusters as described in claim 4, characterized in that: The process involves summing up the carbon emissions of all individuals to obtain the total carbon emissions of the group, and then outputting the results in a visual format, including: Perform an summation operation on all calculated individual carbon emissions to generate the total group carbon emissions for the target scenario; The total carbon emissions of the group will be presented in numerical form. Generate data visualizations and output visualization reports that include both numerical values ​​and graphics.

6. The method for detecting carbon emissions from industrial park clusters as described in claim 5, characterized in that: The preprocessing includes: The color image is converted to a grayscale image using grayscale processing. The Gaussian filtering algorithm is applied to suppress noise in grayscale images. Histogram equalization is used to enhance image contrast, resulting in a preprocessed image.

7. The method for detecting carbon emissions from industrial park clusters as described in claim 6, characterized in that: The step of obtaining the running status parameters corresponding to the current object category includes: Based on the object category determined in the optimized recognition results, search and retrieve the preset parameter set associated with that category from the preset parameter database; The preset parameter set includes power and usage time for electrical equipment, as well as mileage and fuel consumption per unit mileage for transportation vehicles. The invocation of the carbon emission calculation model and the acquisition of operating status parameters are executed simultaneously.

8. A group carbon emission monitoring system for industrial parks, using the method described in any one of claims 1-7, characterized in that, include: The image acquisition module is used to acquire raw images of the target scene; The object recognition module is used to input the preprocessed image into an object recognition model based on convolutional neural networks and feature pyramid networks, extract and fuse multi-scale features, and obtain recognition results that include the category, location and corresponding multi-scale object features of the recognized objects. The feature optimization module is used to calculate the global association weights between pixels in the feature map based on the multi-scale object features in the recognition results through a self-attention mechanism, and to perform weighted optimization on the features to obtain an optimized recognition result that includes the finally confirmed object category and quantity. The carbon emission calculation module is used to obtain the individual carbon emission of each object based on the object category and quantity in the optimized identification results, by calling the carbon emission calculation model that matches the object type, and combining the running status parameters corresponding to the object category. The carbon emissions aggregation and output module is used to sum up the carbon emissions of all individuals to obtain the total carbon emissions of the group, and output the results in a visual format.

9. An electronic device, characterized in that, include: Memory, used to store programs; A processor for loading the program to perform the steps of the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images