Optimized scheduling method and system for logistics and warehousing video monitoring in internet of things environment

Abstract

The application discloses a logistics and warehousing video monitoring optimization scheduling method and system in an Internet of Things environment. The method comprises the following steps: acquiring dynamic information of a logistics and warehousing area; inputting the dynamic information into an enhanced image processing model to obtain high-quality image data and generate a computing resource scheduling scheme; predicting a cargo flow trend and planning a monitoring strategy using the computing resource scheduling scheme to ensure that there is sufficient computing resource support in a high-flow area; when an abnormal event is detected, quickly responding according to a pre-set task priority and an optimized resource allocation scheme, adjusting the working state of a camera or increasing the monitoring density of a specific area; collecting operation effect data as feedback to continuously learn and optimize the intelligent computing resource allocation algorithm. By implementing the method of the application, the data center burden can be reduced, the real-time performance can be improved, the automation and abnormal detection capability can be improved, the system resource allocation can be optimized, the storage cost can be reduced, and the system efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and more specifically to a method and system for optimizing and scheduling logistics warehousing video monitoring in an Internet of Things environment. Background Art

[0002] In modern logistics and warehousing environments, video surveillance systems are essential for ensuring cargo safety, improving operational efficiency, and optimizing inventory management. However, with the widespread adoption of high-definition cameras and the rapid growth of video data, traditional video surveillance systems face a series of challenges, particularly in resource management, data storage, real-time monitoring, and intelligent scheduling. The following details these challenges:

[0003] With the widespread adoption of high-definition and ultra-high-definition cameras, the image quality and data volume of video surveillance are growing exponentially. HD cameras generate massive amounts of video data, placing pressure on traditional systems to store and process it. This surge in data volume impacts the system's real-time performance and quality. Large amounts of data require efficient storage and transmission strategies. Failure to do so can result in video delays or loss, impacting monitoring effectiveness. Traditional systems operate on fixed schedules and fail to adjust resources based on real-time demand. Intelligent scheduling can dynamically optimize monitoring frequency based on environmental changes. The lack of such a mechanism can lead to uneven resource allocation, potentially impacting monitoring of key areas and creating blind spots. High-definition video consumes significant storage space and bandwidth requirements. Traditional storage and transmission solutions struggle to handle such large amounts of data. Unoptimized storage and transmission can cause system delays and data loss. Large volumes of surveillance data require efficient management to ensure stability. Traditional systems can only passively record data and lack automatic anomaly detection capabilities. Warehouse environments require the ability to identify potential safety hazards and operational anomalies. Video analytics can automatically identify suspicious activity, reducing manual intervention. Intelligent analysis improves monitoring efficiency and accuracy. Video data requires high real-time performance, and traditional systems can experience latency. Centralized architectures lead to data transmission delays, impacting monitoring effectiveness. Real-time processing and response are crucial to warehouse safety. The system architecture needs to be optimized to improve response speed. The volume of video data is enormous, and manual backtracking is time-consuming and inefficient. Intelligent retrieval technology can quickly locate the time and location of an incident. Video tagging and indexing can improve data management efficiency. Intelligent backtracking reduces manual intervention and improves event processing efficiency. The integration of monitoring systems with other management systems often encounters compatibility issues. Data sharing and collaboration between different systems are not smooth. The system needs to have higher compatibility to promote collaborative work among various systems. Optimized integration can improve overall management efficiency.

[0004] Therefore, it is necessary to design a new method to reduce the burden on data centers, improve real-time performance, enhance automation and anomaly detection capabilities, optimize system resource allocation, reduce storage costs, and improve system efficiency. Summary of the Invention

[0005] The present application aims to overcome the defects of the prior art and provide a logistics warehouse video monitoring optimization scheduling method and system in an Internet of Things environment.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: a logistics warehouse video monitoring optimization scheduling method in an Internet of Things environment, comprising:

[0007] acquiring dynamic information of the logistics warehouse area;

[0008] inputting the dynamic information into an enhanced image processing model to obtain high-quality image data, wherein the enhanced image processing model is trained based on a logistics scene and automatically adjusts RGB color values according to illumination changes in a specific time period;

[0009] establishing a task graph structure according to the task type and priority based on the high-quality image data, and generating a computing resource scheduling scheme using a hierarchical attention graph neural network and a particle swarm optimization model;

[0010] using the computing resource scheduling scheme to predict cargo flow trends and plan monitoring strategies to ensure sufficient computing resources in high-flow areas;

[0011] when an abnormal event is detected, quickly respond according to the pre-set task priority and the optimized resource allocation scheme, adjust the camera working state or increase the monitoring density of the specific area;

[0012] collecting operation effect data as feedback, continuously learning and optimizing the intelligent computing resource allocation algorithm.

[0013] Further technical solutions thereof are that the dynamic information includes video streams captured by each camera, real-time data provided by environmental sensors, and cargo positions determined by tracking cargo movement paths through RFID tags or barcode scanners and combining GPS positioning systems.

[0014] Further technical solutions thereof are that the training process of the enhanced image processing model includes:

[0015] collecting image data in a logistics warehouse environment, which should cover different time periods, weather conditions and different types of lighting conditions;

[0016] annotating the collected image data, including object bounding boxes and class labels;

[0017] generating image samples under synthetic lighting conditions by combining the annotated image data through image enhancement technology to obtain a sample set;

[0018] constructing a basic model;

[0019] defining a loss function and an optimizer;

[0020] training the base model with the sample set by using the loss function and the optimizer to obtain an enhanced image processing model.

[0021] Further technical solutions are as follows: the base model comprises a convolutional neural network, an illumination perception module is added in the convolutional neural network, the illumination perception model receives environmental information such as illumination intensity by introducing an additional input layer, or focuses on an area with large illumination change in an image by using an attention mechanism;

[0022] A color adjustment layer is added in the convolutional neural network, which is responsible for dynamically modifying RGB color values of an image according to information provided by the illumination perception module, and the color adjustment layer directly predicts target color parameters by using an end-to-end learning method based on a traditional color space conversion theory.

[0023] Further technical solutions are as follows: the task graph structure is established according to the high-quality image data analysis task type and priority, a hierarchical attention graph neural network and a particle swarm optimization model are used to generate a computing resource scheduling scheme, which comprises:

[0024] Task classification and priority setting are performed according to the high-quality image data to obtain task characteristics;

[0025] A task graph structure is constructed according to the task characteristics;

[0026] In the task graph structure, a hierarchical attention graph neural network is applied to update task features to obtain a task feature matrix;

[0027] A computing resource scheduling scheme is generated by using the particle swarm optimization model in combination with the task feature matrix.

[0028] Further technical solutions are as follows: in the task graph structure, a hierarchical attention graph neural network is applied to update task features to obtain a task feature matrix, which comprises:

[0029] An initial task feature vector is extracted from the task graph structure, and the task feature vector comprises task type, priority, predicted execution time, and resource demand information;

[0030] A graph neural network with a hierarchical attention mechanism is used to iteratively update features of nodes in the task graph structure according to the initial task feature vector, capture complex interaction patterns between tasks, and highlight task paths that are crucial to the overall goal to obtain a task feature matrix.

[0031] Further technical solutions are as follows: the computing resource scheduling scheme is used to predict cargo flow trends and plan monitoring strategies to ensure sufficient computing resources in high-flow areas, which comprises:

[0032] preprocessing and feature extraction are performed on the dynamic information to obtain an extraction result;

[0033] the extraction result and the high-quality image data are input into a prediction model to identify potential trend patterns, to obtain a traffic trend prediction result;

[0034] According to the traffic trend prediction result, the computing resources required in different regions in the future are estimated to support the effective execution of video monitoring and other related tasks, and the corresponding computing resources are pre-deployed in the predicted high-traffic areas combined with the computing resource scheduling scheme, to ensure that all monitoring devices in these areas can obtain the necessary support for efficient image processing and data analysis.

[0035] The application also provides a logistics warehouse video monitoring optimization scheduling system in an Internet of Things environment, comprising:

[0036] A dynamic information acquisition unit is configured to acquire dynamic information of a logistics warehouse area.

[0037] An image enhancement unit is configured to input the dynamic information into an enhanced image processing model to obtain high-quality image data, wherein the enhanced image processing model is trained based on a logistics scene and automatically adjusts RGB color values according to illumination changes in a specific time period.

[0038] A scheme generation unit is configured to establish a task graph structure according to the high-quality image data, task types and priorities, and generate a computing resource scheduling scheme by using a hierarchical attention graph neural network and a particle swarm optimization model.

[0039] A planning unit is configured to predict cargo traffic trends and plan monitoring strategies by using the computing resource scheduling scheme, to ensure sufficient computing resource support in high-traffic areas.

[0040] An adjustment unit is configured to quickly respond to a pre-set task priority and an optimized resource allocation scheme when an abnormal event is detected, to adjust the working state of the camera or increase the monitoring density of a specific area.

[0041] A collection unit is configured to collect operation effect data as feedback, to continuously learn and optimize the intelligent computing resource allocation algorithm.

[0042] The application also provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the above method when executing the computer program.

[0043] The application further provides a storage medium storing a computer program, which, when executed by a processor, implements the method described above.

[0044] Compared with the prior art, the application has the following beneficial effects: the application optimizes image quality through an enhanced image processing model, automatically adjusts RGB color values under light changes, reduces data redundancy and storage pressure during image processing, and improves image data quality; based on high-quality image data analysis task priority and type, a hierarchical attention graph neural network and a particle swarm optimization model are used to generate a computing resource scheduling scheme, real-time task optimization is realized, and computing resource waste is reduced; the flow trend of goods is predicted and a monitoring strategy is planned to ensure that high-flow areas are supported by sufficient computing resources, avoid excessive consumption of computing resources, and improve system response speed; when an abnormal event occurs, the camera working state is quickly adjusted or the monitoring density is increased to ensure that potential problems are discovered and handled in a timely manner, and the automation and anomaly detection capabilities are improved; by collecting and feeding back operation effect data, the intelligent computing resource allocation algorithm is continuously optimized, the system efficiency is continuously improved, the storage cost is reduced, and the burden on the data center is reduced.

[0045] The application will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0047] Figure 1 An application scenario diagram of the logistics warehouse video monitoring optimization scheduling method in the Internet of Things environment provided by the embodiments of the application is shown.

[0048] Figure 2 A flowchart of the logistics warehouse video monitoring optimization scheduling method in the Internet of Things environment provided by the embodiments of the application is shown.

[0049] Figure 3 A schematic block diagram of the logistics warehouse video monitoring optimization scheduling system in the Internet of Things environment provided by the embodiments of the application is shown.

[0050] Figure 4 A schematic block diagram of the computer device provided by the embodiments of the application is shown. DETAILED DESCRIPTION

[0051] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those ordinarily skilled in the art without creative effort belong to the scope of the present application.

[0052] It should be understood that the terms "comprise" and "include" as used in the specification and the appended claims indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0053] It should also be understood that the terms used in the present application specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the present application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms, unless the context clearly indicates otherwise.

[0054] It should be further understood that the term "and / or" as used in the present application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0055] Please refer to Figure 1 and Figure 2 , Figure 1 The application scenario diagram of the logistics warehouse video monitoring optimization scheduling method provided by the embodiments of the present application in the Internet of Things environment. Figure 2A schematic flowchart of a method for optimizing and scheduling video surveillance in logistics and warehousing under an Internet of Things (IoT) environment, provided by an embodiment of the present invention. This method is applied to a server that interacts with sensors, cameras, and terminals. The server optimizes image quality through an enhanced image processing model, automatically adjusts RGB color values ​​under varying lighting conditions, reduces data redundancy and storage pressure during image processing, and improves image data quality. Based on high-quality image data, the method analyzes task priorities and types, generates a computing resource scheduling scheme using a hierarchical attention graph neural network and a particle swarm optimization model, achieves real-time task optimization, and reduces computing resource waste. The method predicts cargo flow trends and plans monitoring strategies to ensure sufficient computing resource support for high-traffic areas, avoids excessive computing resource consumption, and improves system response speed. When abnormal events occur, the method rapidly responds and adjusts camera operating status or increases monitoring density to ensure timely detection and resolution of potential issues, enhancing automation and anomaly detection capabilities. The method continuously optimizes intelligent computing resource allocation algorithms by collecting and providing feedback on operational performance data, continuously improving system efficiency, reducing storage costs, and alleviating the burden on data centers.

[0056] Figure 2 This is a flow chart of the logistics warehousing video monitoring optimization scheduling method under the Internet of Things environment provided by the embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S160.

[0057] S110. Obtain dynamic information of the logistics storage area.

[0058] In this embodiment, the dynamic information includes video streams captured by various cameras, real-time data provided by environmental sensors, and the location of the goods determined by tracking the movement path of the goods through RFID tags or barcode scanners and combined with the GPS positioning system.

[0059] Specifically, comprehensive and real-time data is collected to reflect the current status within the logistics warehousing environment, providing a basis for subsequent image processing, task analysis and resource scheduling.

[0060] Camera video streams: Multiple high-definition cameras deployed throughout the logistics warehouse capture continuous video footage at key locations. The video content covers activities such as cargo loading and unloading, handling, and storage, as well as worker operations.

[0061] Environmental sensors monitor physical parameters such as temperature, humidity, and light intensity, and upload this information to a central control system in real time. This data is crucial for understanding warehouse conditions, especially when adjusting image quality or performing anomaly detection.

[0062] RFID Tag / Barcode Scanner and GPS Positioning System: Utilize RFID tags or barcodes installed on goods, and read their unique identifiers through scanners during the process of goods entering or leaving the warehouse or moving. Combined with GPS positioning technology, the location change path of each piece of goods can be accurately tracked, realizing the visual management of the entire logistics process.

[0063] All sensors and monitoring devices are connected to a unified Internet of Things platform, which can automatically send updated measurement results at regular intervals. When certain events occur (such as goods arriving at designated locations or abnormal activities being discovered), the system will immediately trigger additional data recording to ensure that important information is not missed.

[0064] The collected data contains noise or incomplete parts, so it needs to be cleaned and formatted before further processing. Preliminary filtering is performed on video streams to remove unnecessary background information; smoothing is performed on environmental sensor data to eliminate the effects of transient fluctuations.

[0065] Data from different sources must be accurately synchronized according to timestamps to ensure the correct time relationship between them. Integrate all types of data into a comprehensive database to facilitate easy access and use in subsequent steps.

[0066] Strict security measures are implemented during transmission and storage to protect sensitive information from unauthorized access or tampering. Comply with relevant laws and regulations to ensure proper protection of personal privacy.

[0067] In summary, step S110 aims to build an efficient and reliable data acquisition system that not only captures rich dynamic details within the logistics and storage area but also lays a solid foundation for subsequent task analysis and intelligent decision-making. This step is the premise and starting point of the entire logistics and storage video monitoring optimization scheduling method, and its accuracy directly affects the performance of the entire system.

[0068] S120, input the dynamic information into an enhanced image processing model to obtain high-quality image data, wherein the enhanced image processing model is trained based on the logistics scene and automatically adjusts the RGB color value according to the light change in a specific time period.

[0069] In this embodiment, the training process of the enhanced image processing model described above includes:

[0070] Collect image data in the logistics and storage environment, which should cover different time periods, weather conditions, and different types of lighting conditions;

[0071] Label the collected image data, including object bounding boxes and class labels;

[0072] Generate image samples under synthetic lighting conditions by combining image enhancement techniques with the labeled image data to obtain a sample set;

[0073] Construct a base model;

[0074] Define a loss function and an optimizer;

[0075] Train the base model using the sample set with the loss function and optimizer to obtain an enhanced image processing model.

[0076] In this embodiment, the original video stream is processed by the enhanced image processing model to adapt to changes in different lighting conditions, thereby outputting clearer and more accurate image data to support subsequent tasks such as object recognition and path planning.

[0077] This model can automatically adjust RGB color values based on lighting changes within a specific time period, ensuring good visual effects even in low-light or high-contrast environments. Advanced machine learning algorithms such as convolutional neural networks (CNN) are used to improve image quality.

[0078] To make the model widely applicable, a large amount of image data needs to be collected from the logistics warehouse environment. These images should cover different time periods (morning, noon, evening), various weather conditions (sunny, cloudy, rainy), and different lighting conditions (natural light, artificial light). Ensure that the collected data is representative and maintains a certain proportion under various conditions to avoid model bias towards a specific situation.

[0079] Detailed bounding box drawing is performed on key elements in each image (such as goods, shelves, personnel, etc.), and corresponding class labels are assigned (e.g., "carton", "pallet", "forklift driver"). This step is crucial for training the model to correctly identify and classify targets. Semi-automatic labeling tools can be used to improve efficiency, but final manual review is still required to ensure accuracy.

[0080] Image enhancement techniques (such as brightness adjustment, contrast enhancement, noise removal, etc.) are used in combination with labeled real image data to generate a series of synthetic image samples under extreme lighting conditions (such as extremely dark or bright environments). This method not only increases the number of training samples, but also improves the model's ability to handle complex lighting environments.

[0081] Select appropriate CNN or other types of deep learning networks as the base model architecture according to task requirements, such as ResNet, EfficientNet, etc., which have shown excellent performance in various computer vision tasks.

[0082] Determine the number of network layers, filter size, activation function, etc. Hyperparameters are prepared for the next training.

[0083] Define a proper loss function to measure the difference between the predicted results and the true labels. Common ones include cross-entropy loss, mean square error loss, etc. For this task, additional constraint terms may be considered to encourage the model to generate more natural image color distribution. Select efficient optimization algorithms (such as Adam, SGD) to minimize the loss function, accelerate the convergence speed and find the global optimal solution.

[0084] Divide the constructed sample set into multiple small batches, input them into the model one by one, calculate the loss value by forward propagation, and then update the weight parameters by back propagation. Regularly evaluate the performance of the model on an independent validation set, adjust the hyperparameters or improve the model structure according to the results, until a satisfactory accuracy is obtained.

[0085] The enhanced image processing model obtained through the above training can receive real-time video streams from the camera and quickly output high-quality image data after optimization. These images have better color restoration and detail expression, which helps to improve the accuracy and reliability of subsequent analysis tasks.

[0086] This step is one of the key technical links in the whole logistics warehouse video monitoring optimization scheduling method, which directly affects whether the system can effectively extract valuable information and make correct decisions.

[0087] In this embodiment, the base model includes a convolutional neural network, and a light perception module is added to the convolutional neural network. The light perception model receives environmental information such as light intensity by introducing an additional input layer, or focuses on areas with large changes in light in the image using an attention mechanism.

[0088] A color adjustment layer is added to the convolutional neural network, which is responsible for dynamically modifying the RGB color values of the image based on the information provided by the light perception module. The color adjustment layer is based on traditional color space conversion theory or directly predicts target color parameters through end-to-end learning.

[0089] Introducing the light perception module and color adjustment layer into the enhanced image processing model is an important improvement for image quality optimization under complex and variable lighting conditions in the logistics scenario. Adding the light perception module improves the understanding of the convolutional neural network (CNN) for images under different lighting conditions, thereby improving the quality of the output images.

[0090] To enable the CNN to perceive information such as the intensity of light in the environment, one or more additional input layers are added to the network architecture. These layers can receive external information such as light intensity data provided by light sensors, timestamps for inferring the time of day, data returned by weather forecast APIs, and so on.

[0091] Another approach is to embed an attention mechanism in the CNN, which allows the network to automatically focus on those image regions that are more affected by changes in lighting. By learning which parts need more attention, the network can adjust its weight distribution during training, more effectively handling local lighting differences.

[0092] The information provided by the light perception module can help the CNN better understand the background conditions of the input image, and thus more accurately extract useful features. Provide necessary parameters and guidance for subsequent color adjustment layers to ensure that the color adjustment process is both scientific and efficient.

[0093] Using classical color theory, such as the conversion of RGB to Lab color space, to achieve dynamic adjustment of color values. This method relies on predefined mathematical formulas and has the advantages of fast calculation speed and good stability.

[0094] Through deep learning technology, the network learns itself how to adjust the color of the image according to the changes in lighting conditions. This way does not need to pre-set complex conversion rules, but lets the model learn the optimal color mapping relationship from a large number of training samples.

[0095] The color adjustment layer can immediately adjust the RGB color values of the image after receiving the information from the light perception module to compensate for the color deviation caused by changes in lighting. Ensure that the adjusted image not only meets the technical requirements, but also has a natural and realistic visual effect without obvious color distortion.

[0096] Integrate the above improvements into the existing CNN framework to form a more intelligent and adaptive enhanced image processing model. This model not only significantly improves the quality of images in the logistics and warehouse environment, but also may be applicable to other fields such as autonomous driving, security monitoring, etc. Any application scenario that has high requirements for image quality and faces challenges of lighting changes can benefit from it.

[0097] In this embodiment, light-sensitive sensors are installed near the image acquisition device (such as a camera) to directly measure the ambient light intensity. These sensors can provide accurate lighting information in real time, including but not limited to brightness, color temperature, and other parameters. Ensure that the data from the light-sensitive sensor is synchronized with the image frame, and perform necessary preprocessing on the collected data, such as filtering and denoising, unit conversion, etc. to facilitate subsequent use.

[0098] A subnetwork dedicated to estimating lighting conditions can be trained, which can extract lighting-related features such as shadow distribution, highlight areas, etc. from the input image, and then infer the current lighting state. Considering that lighting changes may be reflected at different scales, multi-scale convolution kernels can be used to capture lighting patterns from local details to the overall scene.

[0099] Attention mechanisms are introduced to enable the CNN to automatically identify and emphasize the attention of areas that are more affected by lighting during the learning process. By adjusting the attention weights, the network can focus more on the lighting characteristics of key parts, thereby improving its understanding of complex lighting environments. The dynamic feature map generated in combination with the lighting perception results guides the subsequent color adjustment layer or other processing steps to respond more accurately to lighting changes.

[0100] Data provided by external resources such as weather forecasting services, geographic location information, etc. are used as supplementary information sources to help the network better predict and adapt to different lighting conditions. For example, the solar elevation angle can be estimated based on geographic location and time stamp, thereby indirectly understanding the lighting conditions.

[0101] If external sensors are chosen to obtain lighting information, one or more additional input channels can be added to the existing CNN architecture to specifically receive this information. This has the advantage that lighting data can be involved in the entire network's calculation process along with other image features. For cases where lighting estimation relies entirely on image content, the lighting perception function can be embedded in a specific layer of the CNN to form an independent but closely linked sub-module. This module is responsible for analyzing the input image and outputting estimates of lighting intensity and distribution for use by subsequent layers.

[0102] Specifically, the collaborative work of the lighting perception module and the color adjustment layer is to ensure that the image maintains good visual effects under different lighting conditions, which is achieved through the following aspects:

[0103] The lighting perception module is responsible for analyzing the lighting conditions in the environment or image, including but not limited to brightness, color temperature, direction of light, etc. It can extract relevant information based on external sensor data or directly from image content. The obtained lighting information is converted into a form suitable for subsequent processing, such as standardized numerical values, feature vectors, etc., so that it can be understood and utilized by the color adjustment layer.

[0104] Based on the information provided by the light perception module, the color adjustment layer can apply predefined rules to adjust the colors of the image. For example, increasing brightness in low-light environments, or adjusting white balance to compensate for the blueish phenomenon in high color temperature (such as daylight). More advanced systems may use machine learning models to dynamically decide how to adjust the colors. These models can learn the best color adjustment scheme from historical data and automatically adjust parameters for different lighting conditions.

[0105] After completing an adjustment, the color adjustment layer can evaluate the adjustment effect in some way and feed back the results to the light perception module. This helps to continuously improve the accuracy of light estimation and the effect of color adjustment. Over time and with more data accumulation, the entire system can continuously optimize its performance through online learning or offline training, enabling it to provide more accurate color adjustment services in various complex lighting environments.

[0106] Some applications may require special treatment for specific types of scenes (such as indoor, outdoor, night scenes, etc.). At this time, the light perception module can help identify the current scene type and inform the color adjustment layer to adopt the corresponding strategy. Allow users to set some basic color adjustment parameters according to personal preferences, such as contrast, saturation, etc. The light perception module and the color adjustment layer should consider these preferences together to ensure image quality while meeting the user's individual needs.

[0107] In this embodiment, the light perception module can extract relevant information from external sensors or directly from image content. To improve the accuracy of light estimation, the following method is recommended to integrate the two information sources:

[0108] First, by analyzing the brightness, contrast and other visual features in the input image, the system makes a preliminary assessment of the current lighting conditions. This step can quickly provide a general overview of the lighting environment. Then, use the precise light intensity measurement provided by the light-sensitive sensor installed near the camera to calibrate and refine the preliminary estimate results. The light-sensitive sensor can directly capture the actual lighting level of the environment, including brightness, color temperature and direction, etc., providing a more accurate basis for subsequent color adjustment. Finally, the light perception module will integrate information from image analysis and light-sensitive sensors to generate a comprehensive lighting status report. The color adjustment layer adjusts the RGB color values of the image dynamically based on this report to ensure good visual effects even in complex and changing lighting conditions.

[0109] Suppose in a smart camera system, the light perception module detects that the current environment is at dusk and there is a significant backlight situation. It will pass this information to the color adjustment layer, which will perform the following operations accordingly:

[0110] Increase overall brightness to improve visibility of dark details; adjust white balance to reduce color deviation caused by changes in color temperature; apply HDR (High Dynamic Range) technology to balance the difference between bright and dark areas; if there are objects in the shadows, it may also be necessary to locally enhance the contrast to make the foreground object clearer.

[0111] Through such a collaborative working mechanism, the lighting perception module and color adjustment layer can effectively cope with complex and changing lighting conditions and provide users with a high-quality visual experience.

[0112] S130. Establish a task graph structure according to the type and priority of the high-quality image data analysis task, and generate a computing resource scheduling plan using a hierarchical attention graph neural network and a particle swarm optimization model.

[0113] In one embodiment, the aforementioned step S130 may include steps S131 to S134 .

[0114] S131, performing task classification and priority setting based on the high-quality image data to obtain task characteristics;

[0115] S132, constructing a task graph structure according to the task characteristics;

[0116] S133. In the task graph structure, apply a hierarchical attention graph neural network to update task features to obtain a task feature matrix.

[0117] In one embodiment, the aforementioned step S133 may include steps S1331 - S1332 .

[0118] S1331. Extracting an initial task feature vector from the task graph structure, wherein the task feature vector includes task type, priority, estimated execution time, and resource requirement information;

[0119] S1332. Based on the initial task feature vector, a graph neural network with a hierarchical attention mechanism is used to iteratively update the features of the nodes in the task graph structure, capture the complex interaction patterns between tasks, and highlight the task paths that are critical to the overall goal, so as to obtain a task feature matrix.

[0120] S134. Generate a computing resource scheduling solution using the particle swarm optimization model in combination with the task feature matrix.

[0121] In this embodiment, the input high-quality image data is first analyzed to determine the type of each task (e.g., edge detection, color correction, object recognition, etc.) and its relative importance or urgency (i.e., priority). This process results in the definition of unique characteristics of each task, which are crucial for subsequent resource allocation.

[0122] Based on the task characteristics and their dependencies obtained in S131, a task graph structure representing all the tasks to be executed and their interrelationships is created. This helps to visualize the logical order among different tasks and possible concurrent operations. Specifically, in units of each image analysis task, corresponding nodes are created in the task graph. For tasks involving multi-stage processing, further sub-task nodes are subdivided. Through directed edges connecting task nodes, the order or dependency between tasks is expressed. For example, only after completing object recognition in a certain area, can the number of goods in that area be started. Assign appropriate weights to nodes and edges to reflect the time required for tasks, resource consumption, and their importance in the entire workflow. This helps to more accurately plan the task execution path later.

[0123] From the constructed task graph, the initial task feature vector is extracted, which contains information such as task type, priority, estimated execution time, and resource requirements. A hierarchical attention mechanism graph neural network is used to iteratively update the features of nodes in the task graph. HAGNN can capture complex interaction patterns between tasks and highlight those paths that are critical to the overall goal, generating a more optimized task feature matrix. Specifically, the basic attributes of each task are extracted from the task graph structure, such as task type, priority, estimated execution time, and required computing resource information. These information constitutes the basis of the task feature vector. Using hierarchical attention mechanism graph neural network (HAGNN), the node features in the task graph structure are iteratively updated multiple times. This not only captures complex interaction patterns between tasks, but also highlights those paths or nodes that are critical to completing the entire task. The final output is an optimized task feature matrix.

[0124] The last step is to combine the task feature matrix obtained earlier and apply the Particle Swarm Optimization (PSO) algorithm to determine how to optimally allocate available computing resources to individual tasks. PSO is a heuristic search algorithm that mimics the behavior of bird foraging in the solution space to find the global optimal solution. In this context, PSO is used to find the best resource allocation strategy to maximize efficiency and effectiveness while meeting all task requirements. Specifically, under the PSO framework, each "particle" represents a possible task scheduling scheme. The position of the particle corresponds to which edge computing device each task is assigned to and its start time point. A well-designed fitness function is designed to quantify the pros and cons of each scheme. The fitness function considers multiple dimensions such as task completion efficiency and resource utilization to ensure the selection of the optimal solution. The PSO model receives the task feature matrix from the HAGNN as one of the inputs, uses these features to guide the search direction of the particle swarm, speeds up the convergence speed and improves the probability of finding the global optimal solution. After multiple iterations, the particle swarm continuously adjusts its position to ultimately determine a scheduling scheme that can meet the task priority and efficiently utilize existing resources.

[0125] The best task scheduling scheme obtained by the PSO model is applied to the actual environment to reasonably allocate computing resources to each image analysis task. The system remains flexible and can quickly respond to new tasks joining or changes in the status of existing tasks by recalculating the optimal scheduling scheme. Continuously track task execution, collect performance indicator data, and form a feedback mechanism to guide the optimization of resource allocation strategies in similar scenarios in the future.

[0126] In summary, the above process not only considers the characteristics of image data analysis tasks themselves, but also fully combines advanced machine learning algorithms and intelligent optimization techniques, aiming to achieve efficient computing resource management and ensure the rapid and accurate completion of image processing tasks within the logistics and warehousing system.

[0127] In this embodiment, although the PSO model mimics the behavior of bird foraging, its core advantage lies in its parallel search capability and ability to avoid getting stuck in local optima. Specifically:

[0128] The PSO algorithm allows multiple "particles" (representing different task scheduling schemes) to explore different parts of the solution space simultaneously. This parallelism increases the likelihood of finding a globally optimal solution, especially when dealing with large-scale or complex optimization problems. PSO helps the system escape local optimal traps and continue to search for better solutions through the interaction and information sharing mechanism between particles. For example, in a logistics warehouse scenario, when considering both task priority and resource utilization as objectives, PSO can help the system balance the relationship between the two and find the best allocation strategy that can meet the urgent task demand and efficiently utilize existing resources. Suppose there are multiple image processing tasks that need to be completed within a limited time. The PSO model will quickly determine an optimal task scheduling scheme based on task characteristics (such as type, priority, estimated execution time, and resource requirements) and existing computing resource conditions. This scheme not only ensures the timely processing of critical tasks but also maximizes the overall system efficiency.

[0129] In addition to deep learning algorithms such as LSTM and GRU, considering the seasonality and periodicity in the logistics scenario, Prophet and ARIMA can also be introduced, which are specifically designed for time series prediction. In addition, a regular update mechanism should be established to adapt to the changing business environment:

[0130] For cargo flow prediction tasks with obvious periodicity, a dedicated time series prediction tool such as Prophet can be selected. It is good at handling datasets containing trends, seasonality, and holiday effects, and can more accurately capture long-term patterns. For non-periodic short-term fluctuations, neural network models such as LSTM or GRU can be used.

[0131] To ensure that the prediction results are always close to the actual situation, the system should set a fixed period (such as every week or every month) to retrain the prediction model with the latest collected data. This not only reflects the latest business dynamics but also gradually improves the expressiveness of the model.

[0132] Throughout the process, the system needs to continuously track the prediction performance and feed back the actual running data to the model for further optimization and adjustment. For example, if it is found that the prediction error is large in some specific time period, the learning intensity of this part of data can be targeted to improve the prediction accuracy.

[0133] S140, using the computing resource scheduling scheme to predict cargo flow trends and plan monitoring strategies to ensure sufficient computing resource support in high-flow areas.

[0134] In an embodiment, the above-mentioned step S140 can include steps S141-S143.

[0135] S141, preprocessing and feature extraction are performed on the dynamic information to obtain an extraction result;

[0136] S142, the extraction result and the high-quality image data are input into a prediction model to identify potential trend patterns to obtain a traffic trend prediction result;

[0137] S143, according to the traffic trend prediction result, the computing resources needed in different areas in the future are estimated to support the effective execution of video monitoring and other related tasks, and the corresponding computing resources are pre-deployed in the predicted high-traffic areas combined with the computing resource scheduling scheme, ensuring that all monitoring devices in these areas can obtain the necessary support for efficient image processing and data analysis.

[0138] In this embodiment, first, the system needs to accumulate high-quality image data within a certain time period, which includes but is not limited to the timestamp of goods entering and leaving the warehouse, the type of goods, the transportation path, etc.

[0139] From the collected data, features that help analyze the trend of goods traffic are extracted, such as the throughput of goods in different time periods, the activity level in certain areas, etc. This step may also involve preliminary processing of image data, such as object recognition, trajectory tracking, etc.

[0140] Using machine learning or deep learning algorithms (such as LSTM, GRU, etc.), a goods traffic prediction model is built based on historical data and extracted features. This model can make predictions about future goods traffic at a certain time or over a certain period based on observed patterns in the past.

[0141] The trained model regularly outputs prediction results of goods traffic trends in the next few days or even weeks. This will help decision-makers understand in advance which time periods may experience traffic peaks and adjust monitoring strategies accordingly.

[0142] Combined with the traffic trend prediction result, areas that are expected to become traffic hotspots are identified. These areas may be goods loading and unloading areas, intersection points, or other places prone to congestion. For the identified hotspot areas, sufficient computing resources are pre-allocated according to the previously developed computing resource scheduling scheme. This means ensuring that there are enough edge computing devices available in these areas and they have been configured to prioritize tasks from these areas. Considering the uncertainty in actual operations, a dynamic adjustment mechanism needs to be established. When real-time monitoring data shows that the actual situation deviates from the prediction, a quick response can be made to re-evaluate resource needs and make corresponding adjustments.

[0143] Implement the new monitoring strategy as planned, increasing the intensity of monitoring during expected high-traffic periods while ensuring all relevant equipment is functioning properly. Continuously track the performance of various metrics, including task completion efficiency and resource utilization. The data collected is used not only for improvements within the current cycle but also to provide a more accurate foundation for future traffic predictions. As more data accumulates and technology advances, continually optimize the prediction models and resource scheduling algorithms to make the entire system more accurate in predictions and more reasonable in resource utilization.

[0144] For key areas identified as high-traffic, ensure they have the highest priority support from computing resources. This may mean deploying additional high-performance computing nodes in these areas or temporarily reallocating idle resources from other areas if necessary. In addition to daily resource planning, be prepared to respond to unexpected situations. For example, when an unexpected event causes a sudden surge in traffic in a certain area, an emergency plan can be quickly activated to supplement the required computing power immediately.

[0145] In summary, through the above process, not only can we effectively predict the trend of goods traffic, but also can reasonably plan the monitoring strategy to ensure that key areas get sufficient computing resource support during peak periods, thereby improving the operation efficiency and service quality of the entire logistics and warehouse system.

[0146] In addition, based on historical data, feature engineering is performed, and a goods traffic prediction model is trained using machine learning or deep learning algorithms. In this step, the "computing resource scheduling scheme" ensures that the computing resources used for data processing and model training are sufficient and efficiently utilized.

[0147] Use the trained model to predict areas that may become traffic hotspots in the future. This step relies on strong computing power to quickly process large amounts of data and generate prediction results. According to the prediction results, the "computing resource scheduling scheme" will allocate additional computing resources to these hotspot areas in advance. This means that edge servers, cloud services, etc. will be optimally configured to ensure that sufficient processing power and storage space can be provided immediately when needed. When the actual traffic pattern does not match the prediction, the "computing resource scheduling scheme" allows the system to dynamically adjust resource allocation. For example, if a high traffic suddenly appears in an unexpected area, the system can quickly transfer part of the idle resources to that area through an automated scheduling program to ensure that its monitoring and processing needs are met.

[0148] For predicted high-traffic areas, the Computing Resource Scheduling Plan guides the optimal deployment of monitoring equipment and technologies, such as high-definition cameras and smart sensors. This ensures these devices have the necessary computing power to perform tasks such as image recognition and data analysis. The performance of various indicators is continuously monitored, and actual operational data is fed back to the Computing Resource Scheduling Plan to refine future resource allocation decisions. This also helps continuously optimize the prediction model and improve prediction accuracy.

[0149] The computing resource scheduling solution incorporates a flexible resource expansion mechanism that can rapidly increase computing resources in specific areas in response to emergencies. For example, during promotional events or unexpected events, additional resources can be promptly deployed to cope with surges in workload. To prevent any single point of failure from impacting overall system stability, the "Computing Resource Scheduling Solution" also incorporates redundancy designs and disaster recovery plans, ensuring that critical tasks can continue to execute even if some resources fail.

[0150] In this embodiment, the collected dynamic information is first preprocessed, such as removing noise and filling missing values, before useful features are extracted. These features may include, but are not limited to, time period, weather conditions, special events (such as holidays), and historical traffic patterns. They are crucial for understanding current traffic conditions and predicting future trends.

[0151] The feature extraction results from S141, along with the high-quality image data, are fed into a pre-trained prediction model. This model, which can be based on machine learning or deep learning algorithms, identifies potential trend patterns. By analyzing this data, the model outputs a forecast of future cargo flow trends, helping to identify areas likely to experience high traffic volumes.

[0152] Based on the traffic trend forecast results from S142, the amount of computing resources required for different areas over the next period of time is estimated. To support the effective execution of video surveillance and other related tasks, sufficient computing resources must be pre-deployed in areas expected to experience high traffic based on the forecast results. This requires not only considering existing infrastructure conditions but also flexibly adjusting resource allocation to ensure that all monitoring devices have the necessary support for efficient image processing and data analysis. Furthermore, this can be combined with the previously established computing resource scheduling plan to optimize resource utilization and avoid resource waste or shortages.

[0153] This predictive maintenance and resource management approach not only improves the response speed and service quality of the logistics system, but also enhances the adaptability and stability of the system in the face of unexpected situations. In this way, enterprises can more effectively manage and optimize their logistics network, improve operational efficiency, reduce costs, and provide better service experiences for customers.

[0154] When an abnormal event is detected, the system quickly responds according to the pre-set task priority and optimized resource allocation scheme, adjusting the working state of the camera or increasing the monitoring density in a specific area.

[0155] In this embodiment, when an abnormal event (such as a traffic accident, illegal intrusion, etc.) is detected, the system can quickly adjust the configuration of the monitoring system according to the pre-set task priority and optimized resource allocation scheme, to ensure the safety and operational efficiency of key areas.

[0156] Each monitoring task is assigned a priority, which determines which tasks should be prioritized in the case of limited resources. For example, when an emergency is detected, the working state of the relevant camera will be immediately adjusted to the highest priority to provide the clearest real-time picture. Based on previous predictions of cargo flow trends and other dynamic information, the system has developed an optimized resource allocation scheme. This means that when an abnormal event occurs, the system can quickly mobilize necessary computing resources to where they are needed, such as increasing the number of cameras in a specific area or increasing video resolution to obtain more detailed monitoring information. In order to better capture and analyze abnormal events, the angle, focal length or sensitivity settings of certain cameras are changed, or the recording function is started. In addition, mobile monitoring devices can be temporarily deployed to enhance coverage in specific areas.

[0157] In this embodiment, after detecting an abnormal event, not only the working state of the camera should be adjusted, but also other types of sensors should be considered to form a multi-modal monitoring system. This can provide more comprehensive information support when different types of emergencies occur:

[0158] When an abnormal event is detected, the system not only relies on visual information provided by video streams, but also combines physical signals provided by devices such as smoke detectors and thermometers. For example, in a fire warning scenario, in addition to observing the video pictures of flames and smoke, the temperature change curve recorded by the thermometer and the alarm signal sent by the smoke detector can be used to jointly confirm whether there is a fire.

[0159] Based on the comprehensive analysis of multi-modal data, the system can flexibly adjust the monitoring strategy. For example, if a large number of people suddenly gather in a certain area, in addition to increasing the resolution and frame rate of the camera, the infrared thermal imager in the nearby area can also be started to obtain more dimensional human heat distribution map, so as to more accurately evaluate the on-site situation.

[0160] A complete emergency plan is prepared in advance, covering various possible abnormal situations. Once the corresponding conditions are triggered, the system can immediately execute the preset operation process, such as notifying relevant personnel, blocking dangerous areas, dispatching rescue forces, etc., to ensure that safety measures are quickly in place.

[0161] S160, collect operation effect data as feedback, continuously learn and optimize intelligent computing resource allocation algorithm.

[0162] By collecting and analyzing the data of the system operation effect as feedback, the intelligent computing resource allocation algorithm is continuously improved, so that the system can become more intelligent and efficient over time.

[0163] Operation effect data includes but is not limited to system response time, image quality, recognition accuracy, resource utilization rate, etc. These data not only reflect the current performance of the system, but also provide important reference for subsequent optimization. Using machine learning technology, especially reinforcement learning algorithm, the system learns from each operation and automatically adjusts parameters to adapt to new challenges. With the accumulation of more data, the prediction ability and decision level of the system will continue to improve. According to the collected data, the effectiveness of the existing computing resource allocation algorithm is evaluated regularly, and new models or improved existing models are introduced to ensure that it is always in the best state. This may involve updating the prediction model, adjusting the task priority rules, optimizing the resource scheduling strategy, etc.

[0164] This self-optimization process is the key to building an adaptive and intelligent logistics monitoring system. It not only improves the flexibility and response speed of the system, but also promotes the maximization of resource use, thereby bringing greater value to the enterprise. At the same time, it also reflects the important role of modern technology in improving safety and operational efficiency.

[0165] In this embodiment, the decision-making ability of the system is gradually improved through the definition of a clear reward function and a trial-and-error process:

[0166] In the logistics and warehouse environment, the design of the reward function should focus on task completion quality, resource utilization efficiency, and service response speed, etc. For example, successfully completing a high-priority task on time can be rewarded positively; while failing to respond to an emergency request or wasting too many computing resources will be punished negatively. In this way, the system is encouraged to make better decisions.

[0167] The system adjusts its internal parameters based on actual results after each operation, constantly trying new resource allocation strategies. As experience accumulates, the system gradually learns how to better balance the relationship between various indicators and find the most suitable solution. For example, when encountering a traffic peak, the system may try to increase the number of edge servers, observe the impact on task completion time and resource consumption, and then decide whether to promote this practice.

[0168] The entire process constitutes a self-optimizing closed loop, where the system learns from each operation and automatically adjusts parameters to adapt to new challenges. Over time, the system's predictive ability and decision-making level will continuously improve, bringing greater value to the enterprise. For example, after multiple iterations, the system can significantly reduce energy consumption and hardware costs while almost not affecting service quality.

[0169] For example, suppose there is a large logistics warehouse with multiple high-definition cameras, environmental sensors (monitoring temperature, humidity, etc.), RFID tag reading devices, and GPS positioning systems deployed. The warehouse receives a large number of goods in and out every day, forming a complex and dynamic working environment.

[0170] The system collects data from various devices around the clock, including video streams, environmental parameters, and cargo location information. All data is synchronously uploaded to the central server and pre-processed to remove noise and redundant parts. Then, these data are integrated into a comprehensive database, providing a solid foundation for subsequent steps.

[0171] For images taken under different time periods and weather conditions, the system uses a pre-trained enhanced image processing model to automatically adjust RGB color values, ensuring high-quality images even in low-light or high-contrast environments. During this process, the light perception module analyzes the intensity of ambient light in real-time and guides the color adjustment layer to make corresponding modifications.

[0172] Based on the optimized image data, the system identifies the types of tasks currently being performed and their priorities. Then, it constructs a task graph structure to represent the dependency relationships between tasks and iteratively updates task features through a hierarchical attention graph neural network. Finally, it uses a particle swarm optimization algorithm to determine the best computational resource allocation scheme, ensuring that critical tasks are processed in a timely manner.

[0173] Using historical data and machine learning algorithms, the system predicts future trends in cargo flow over the next few days. Based on the prediction results, additional computing resources are deployed in areas where high traffic is expected, ensuring that monitoring devices in these areas can operate efficiently.

[0174] When the system detects an abnormal event (such as a sudden gathering of too many people in a certain area), it will immediately activate the emergency response program. Specifically, the working state of the relevant camera is adjusted to the highest priority, and the monitoring density of the area is increased to quickly obtain more detailed information for subsequent analysis.

[0175] The system regularly assesses its performance and adjusts internal algorithms based on actual operational effect data. For example, if it finds that false positives often occur in a certain area, it can reduce such occurrences by fine-tuning the parameters of the light perception module or the color adjustment layer. Over time, the entire system will become more intelligent and efficient.

[0176] The above-mentioned logistics warehouse video monitoring optimization scheduling method in the Internet of Things environment optimizes image quality through an enhanced image processing model, automatically adjusts RGB color values under varying light conditions, reduces data redundancy and storage pressure during image processing, and improves image data quality; based on high-quality image data analysis task priority and type, a hierarchical attention graph neural network and a particle swarm optimization model are used to generate a computing resource scheduling scheme, realizing real-time task optimization and reducing computing resource waste; predicting cargo flow trends and planning monitoring strategies ensures that high-flow areas receive adequate computing resource support, avoiding excessive consumption of computing resources and improving system response speed; in the event of an abnormal event, quickly respond and adjust the camera working state or increase the monitoring density to ensure timely detection and handling of potential problems and improve automation and anomaly detection capabilities; by collecting and feeding back operational effect data, continuously optimizing intelligent computing resource allocation algorithms, continuously improving system efficiency, reducing storage costs, and reducing data center burden.

[0177] Figure 3 is a schematic block diagram of a logistics warehouse video monitoring optimization scheduling system 300 in the Internet of Things environment provided by an embodiment of the present application. As Figure 3 shown, corresponding to the above-mentioned logistics warehouse video monitoring optimization scheduling method in the Internet of Things environment, the present application also provides a logistics warehouse video monitoring optimization scheduling system 300 in the Internet of Things environment. The logistics warehouse video monitoring optimization scheduling system 300 in the Internet of Things environment includes units for executing the above-mentioned logistics warehouse video monitoring optimization scheduling method in the Internet of Things environment, and the system can be configured in a server. Specifically, please refer to Figure 3 , the logistics warehouse video monitoring optimization scheduling system 300 in the Internet of Things environment includes a dynamic information acquisition unit 301, an image enhancement unit 302, a scheme generation unit 303, a planning unit 304, an adjustment unit 305, and a collection unit 306.

[0178] The dynamic information acquisition unit 301 is configured to acquire dynamic information of a logistics storage area; the image enhancement unit 302 is configured to input the dynamic information into an enhanced image processing model to acquire high-quality image data, wherein the enhanced image processing model is trained based on a logistics scene and automatically adjusts RGB color values according to illumination changes in a specific time period; the scheme generation unit 303 is configured to analyze a task type and a priority according to the high-quality image data to establish a task graph structure, and generate a computing resource scheduling scheme by using a hierarchical attention graph neural network and a particle swarm optimization model; the planning unit 304 is configured to predict a cargo flow trend and plan a monitoring strategy by using the computing resource scheduling scheme, so as to ensure that there is sufficient computing resource support in a high-flow area; the adjustment unit 305 is configured to quickly respond to a pre-set task priority and an optimized resource allocation scheme when an abnormal event is detected, and adjust a camera working state or increase a monitoring density of a specific area; and the collection unit 306 is configured to collect running effect data as feedback, and continuously learn and optimize an intelligent computing resource allocation algorithm.

[0179] In an embodiment, the system further comprises a training unit configured to:

[0180] Collect image data in a logistics storage environment, which should cover different time periods, weather conditions and different types of lighting conditions; label the collected image data, including object bounding boxes and class labels; generate image samples under synthetic lighting conditions by combining image enhancement techniques with the labeled image data to obtain a sample set; construct a basic model; define a loss function and an optimizer; and train the basic model by using the loss function, the optimizer and the sample set to obtain an enhanced image processing model.

[0181] In an embodiment, the scheme generation unit 303 is configured to:

[0182] Classify tasks and set priorities according to the high-quality image data to obtain task characteristics; construct a task graph structure according to the task characteristics; update task features in the task graph structure by using a hierarchical attention graph neural network to obtain a task feature matrix; and generate a computing resource scheduling scheme by using a particle swarm optimization model in combination with the task feature matrix.

[0183] In an embodiment, the scheme generation unit 303 is further configured to:

[0184] From the task graph structure, an initial task feature vector is extracted, the task feature vector including task type, priority, estimated execution time and resource requirement information; a hierarchical attention mechanism graph neural network is used to iteratively update the features of the nodes in the task graph structure according to the initial task feature vector, capture complex interaction patterns between tasks, and highlight task paths that are critical to the overall goal, to obtain a task feature matrix.

[0185] In an embodiment, the planning unit 304 is configured to:

[0186] The dynamic information is preprocessed and feature extracted to obtain an extraction result; the extraction result and the high-quality image data are input into a prediction model to identify potential trend patterns to obtain a traffic trend prediction result; the required computing resources of different regions in the future are estimated according to the traffic trend prediction result to support the effective execution of video monitoring and other related tasks, and the corresponding computing resources are pre-deployed in the predicted high-traffic area in combination with the computing resource scheduling scheme to ensure that all monitoring devices in these areas can obtain the necessary support for efficient image processing and data analysis work.

[0187] It should be noted that the specific implementation process of the above-mentioned logistics warehouse video monitoring optimization scheduling system 300 and each unit in the Internet of Things environment can be clearly understood by those skilled in the art, and can refer to the corresponding description in the foregoing method embodiments. For the convenience and brevity of description, it will not be repeated here.

[0188] The above-mentioned logistics warehouse video monitoring optimization scheduling system 300 can be realized in the form of a computer program, which can run on a computer device as shown in the computer device. Figure 4

[0189] Please refer to Figure 4 , Figure 4 is a schematic block diagram of a computer device provided by an embodiment of the present application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0190] Referring to Figure 4 , the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501, wherein the memory can include a non-volatile storage medium 503 and an internal memory 504.

[0191] ​The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, can cause the processor 502 to perform a logistics warehouse video monitoring optimization scheduling method in an Internet of Things environment.

[0192] The processor 502 is configured to provide computing and control capabilities to support the operation of the entire computer device 500.

[0193] The non-volatile storage medium 503 provides an environment for the computer program 5032 in the non-volatile storage medium 503 to run, and the computer program 5032, when executed by the processor 502, can cause the processor 502 to perform a logistics warehouse video monitoring optimization scheduling method in an Internet of Things environment.

[0194] The network interface 505 is configured to communicate with other devices via a network. Those skilled in the art can understand that the network interface 505 can be configured to support wired communication or wireless communication, or both wired and wireless communication. Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device 500 to which the scheme of the present application is applied. The specific computer device 500 can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0195] The processor 502 is configured to run the computer program 5032 stored in the memory to implement the following steps:

[0196] Obtain dynamic information of a logistics warehouse area; input the dynamic information to an enhanced image processing model to obtain high-quality image data, wherein the enhanced image processing model is trained based on a logistics scene and automatically adjusts RGB color values according to changes in illumination in a specific time period; analyze the task type and priority based on the high-quality image data to establish a task graph structure, and use a hierarchical attention graph neural network and a particle swarm optimization model to generate a computing resource scheduling scheme; use the computing resource scheduling scheme to predict cargo flow trends and plan monitoring strategies to ensure sufficient computing resources in high-flow areas; when an abnormal event is detected, quickly respond according to the pre-set task priority and the optimized resource allocation scheme to adjust the working state of the camera or increase the monitoring density in a specific area; collect operation effect data as feedback to continuously learn and optimize the intelligent computing resource allocation algorithm.

[0197] It should be understood that, in the embodiments of the present application, the processor 502 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, or the like. The general-purpose processor can be a microprocessor or any conventional processor.

[0198] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by instructing relevant hardware by a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the above-mentioned embodiments.

[0199] Therefore, the present application also provides a storage medium. The storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program is executed by a processor to make the processor execute the following steps:

[0200] obtain dynamic information of a logistics storage area; input the dynamic information to an enhanced image processing model to obtain high-quality image data, wherein the enhanced image processing model is trained based on a logistics scene and automatically adjusts RGB color values according to light changes in a specific time period; analyze task types and priorities based on the high-quality image data to establish a task graph structure, and generate a computing resource scheduling scheme by using a hierarchical attention graph neural network and a particle swarm optimization model; predict cargo flow trends and plan monitoring strategies by using the computing resource scheduling scheme to ensure sufficient computing resources in high-flow areas; when an abnormal event is detected, quickly respond according to a pre-set task priority and an optimized resource allocation scheme, adjust the working state of a camera or increase the monitoring density of a specific area; collect operation effect data as feedback to continuously learn and optimize the intelligent computing resource allocation algorithm.

[0201] The storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk, and various computer-readable storage media that can store program codes.

[0202] It should be noted that the above-mentioned functions or steps that can be achieved by the storage medium or the computer device can correspond to the related descriptions of the server side and the client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0203] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the foregoing description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0204] In several embodiments provided by the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each unit is only a logical function division, and actual implementation can have another division manner. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0205] The steps in the method embodiments of the present application can be adjusted, combined and deleted in sequence according to actual needs. The units in the system embodiments of the present application can be combined, divided and deleted according to actual needs. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0206] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or say the parts that make contributions to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a terminal or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0207] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements shall be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A logistics warehousing video surveillance optimization scheduling method under the Internet of Things environment, characterized by: include: Obtain dynamic information of logistics and warehousing areas; Inputting the dynamic information into an enhanced image processing model to obtain high-quality image data, wherein the enhanced image processing model is trained based on logistics scenarios and automatically adjusts RGB color values ​​based on lighting changes within a specific time period; Establishing a task graph structure based on the type and priority of the high-quality image data analysis task, and generating a computing resource scheduling solution using a hierarchical attention graph neural network and a particle swarm optimization model; Use the computing resource scheduling solution to predict cargo flow trends and plan monitoring strategies to ensure sufficient computing resources in high-flow areas; When an abnormal event is detected, the system quickly responds based on pre-set task priorities and optimized resource allocation plans, adjusting the camera's working status or increasing the monitoring density in a specific area. Collect operational performance data as feedback to continuously learn and optimize intelligent computing resource allocation algorithms; The method of establishing a task graph structure according to the type and priority of the high-quality image data analysis task and generating a computing resource scheduling solution using a hierarchical attention graph neural network and a particle swarm optimization model includes: performing task classification and priority setting based on the high-quality image data to obtain task characteristics; Constructing a task graph structure according to the task characteristics; In the task graph structure, a hierarchical attention graph neural network is applied to update the task features to obtain a task feature matrix; Generate a computing resource scheduling solution using the particle swarm optimization model combined with the task feature matrix; In the task graph structure, a hierarchical attention graph neural network is applied to update the task features to obtain a task feature matrix, including: Extracting an initial task feature vector from the task graph structure, wherein the task feature vector includes task type, priority, estimated execution time, and resource requirement information; Iteratively updating the features of nodes in the task graph structure using a graph neural network with a hierarchical attention mechanism based on the initial task feature vector, capturing complex interaction patterns between tasks and highlighting task paths that are critical to the overall goal, thereby obtaining a task feature matrix; The computing resource scheduling scheme is used to predict cargo flow trends and plan monitoring strategies to ensure sufficient computing resources in high-flow areas, including: Preprocessing and feature extraction are performed on the dynamic information to obtain extraction results; Inputting the extraction results and the high-quality image data into a prediction model to identify potential trend patterns to obtain a flow trend prediction result; Based on the traffic trend prediction results, the computing resources required in different areas in the future are estimated to support the effective execution of video surveillance and other related tasks. In combination with the computing resource scheduling plan, the corresponding computing resources are pre-deployed in the predicted high-traffic areas to ensure that all monitoring equipment in these areas can obtain the necessary support for efficient image processing and data analysis. The dynamic information includes video streams captured by various cameras, real-time data provided by environmental sensors, and the location of goods determined by tracking their movement paths through RFID tags or barcode scanners combined with GPS positioning systems; The training process of the enhanced image processing model includes: Collect image data in logistics warehousing environments, covering different time periods, weather conditions, and different types of lighting conditions; Annotate the collected image data, including object bounding boxes and category labels; generating image samples under synthetic lighting conditions by combining the annotated image data with an image enhancement technique to obtain a sample set; Build a basic model; Define the loss function and optimizer; Using the loss function and the optimizer to train the basic model with the sample set to obtain an enhanced image processing model; The basic model includes a convolutional neural network, and a light perception module is added to the convolutional neural network. The light perception model receives light intensity environment information by introducing an additional input layer, or uses an attention mechanism to focus on areas with large light changes in the image; A color adjustment layer is added to the convolutional neural network, which is responsible for dynamically modifying the RGB color values ​​of the image according to the information provided by the illumination perception module. The color adjustment layer is based on traditional color space conversion theory or directly predicts the target color parameters through an end-to-end learning method.

2. A logistics warehousing video surveillance optimization and scheduling system under the Internet of Things environment, wherein the system uses the logistics warehousing video surveillance optimization and scheduling method under the Internet of Things environment according to claim 1, characterized in that: include: Dynamic information acquisition unit, used to obtain dynamic information of the logistics storage area; An image enhancement unit, configured to input the dynamic information into an enhanced image processing model to obtain high-quality image data, wherein the enhanced image processing model is trained based on logistics scenarios and automatically adjusts RGB color values ​​based on lighting changes within a specific time period; A solution generation unit is used to establish a task graph structure according to the type and priority of the high-quality image data analysis task, and generate a computing resource scheduling solution using a hierarchical attention graph neural network and a particle swarm optimization model; a planning unit, configured to use the computing resource scheduling scheme to predict cargo flow trends and plan monitoring strategies to ensure sufficient computing resources in high-flow areas; The adjustment unit is used to quickly respond to abnormal events based on pre-set task priorities and optimized resource allocation plans, adjusting the camera working status or increasing the monitoring density of specific areas; The collection unit is used to collect operation effect data as feedback, and continuously learn and optimize the intelligent computing resource allocation algorithm.

3. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to claim 1 when executing the computer program.

4. A storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to claim 1 is implemented.

Images