Distributed photovoltaic power station construction image video acquisition method, system and equipment based on dynamic optimization, and medium

Abstract

The invention discloses a distributed photovoltaic power station construction image video collection method, system and device based on dynamic optimization and a medium, and the method comprises the steps: carrying out the real-time analysis of collected image and video data through a deep learning model, and generating scene adaptation parameters; adjusting the working mode of acquisition equipment according to the scene adaptation parameters, optimizing the acquisition efficiency of video streams through a dynamic frame rate allocation algorithm, executing high-precision semantic segmentation processing on key frames, executing low-complexity compressed encoding on non-key frames, and respectively generating corresponding feature vectors and compressed data packets; selectively transmitting a key frame feature vector or a non-key frame compressed data packet according to a network bandwidth fluctuation condition; and uploading the transmitted data to a cloud server, carrying out secondary analysis on the data by using a cloud deep learning model, and generating a construction management report. The problems of insufficient intelligent analysis, poor multi-scene adaptability and low data transmission stability are solved, and the construction management efficiency and precision are improved.

Description

Technical Field

[0001] This invention relates to the field of distributed photovoltaic power station construction technology, and in particular to a method, system, equipment and medium for acquiring construction images and videos of distributed photovoltaic power stations based on dynamic optimization. Background Technology

[0002] Currently, the core element of achieving intelligent construction management is the acquisition of image and video data at distributed photovoltaic power station construction sites. The quality of the acquired data directly affects the efficiency of construction progress monitoring, safety hazard identification, and quality issue tracing. As distributed photovoltaic power stations expand into complex terrains, construction sites exhibit characteristics such as variable lighting conditions, highly dynamic construction equipment, and unstable network environments. Existing technologies are no longer sufficient to meet the demands for high-precision, low-latency, and highly reliable data acquisition.

[0003] Existing construction site data acquisition technologies have several limitations. For example, current technologies monitor the status of construction fences using video acquisition and dynamic detection modules. However, this technology is designed only for the specific scenario of construction fences and does not consider the multi-scenario switching needs of distributed photovoltaic power station construction. This results in a loss rate of over 20% of image details in scenarios with sudden changes in lighting or high-speed equipment movement, making it impossible to extract key construction features such as photovoltaic module installation angles and cable routing. Furthermore, existing technologies use hazard identification model libraries to identify violations, but this system focuses on the static scenario of transmission line construction and does not address the data transmission problems caused by network fluctuations in distributed photovoltaic power station construction: when bandwidth is below 2Mbps, the video stream packet loss rate exceeds 30%, and the key frame transmission failure rate reaches 45%; while with sufficient bandwidth, non-key frames account for over 60%, leading to an increase in storage costs of over 50%. In addition, existing technologies generally adopt fixed parameter acquisition strategies, lacking adaptability to different construction scenarios. For example, in static equipment scenarios, a fixed frame rate results in 72% data redundancy; while in high-motion scenarios involving component installation, insufficient frame rate leads to motion blur, reducing the accuracy of key information extraction to below 60%. Furthermore, current technologies rely on manual cloud-based analysis for data processing, with an average delay of over 5 minutes from data acquisition to anomaly alerts, failing to meet the demands of real-time construction safety monitoring.

[0004] Therefore, current data acquisition technologies for distributed photovoltaic power station construction sites have significant shortcomings in terms of dynamic adaptation to multiple scenarios, transmission stability under network fluctuations, and real-time intelligent analysis. It is necessary to improve the efficiency and quality of data acquisition in complex environments by using dynamically optimized distributed photovoltaic power station construction image and video acquisition methods. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a method, system, equipment, and medium for acquiring construction images and videos of distributed photovoltaic power plants based on dynamic optimization, which solves the problems of insufficient intelligent analysis, poor adaptability to multiple scenarios, and low data transmission stability in existing technologies for acquiring images and video data at construction sites of distributed photovoltaic power plants.

[0007] 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 acquiring construction images and videos of distributed photovoltaic power plants based on dynamic optimization, comprising: Based on the construction site environment feature extraction module, the acquisition equipment is adaptively configured, and the acquired image and video data are analyzed in real time using a deep learning model to generate scene adaptation parameters. The working mode of the acquisition device is adjusted according to the scene adaptation parameters, the acquisition efficiency of the video stream is optimized through a dynamic frame rate allocation algorithm, and the acquired data is divided into key frames and non-key frames. High-precision semantic segmentation is performed on the key frames, and low-complexity compression encoding is performed on the non-key frames, generating corresponding feature vectors and compressed data packets respectively. The network environment at the construction site is monitored in real time by a dynamic network optimization module, and key frame feature vectors or non-key frame compressed data packets are selectively transmitted based on network bandwidth fluctuations. The transmitted data is uploaded to a cloud server, where a cloud-based deep learning model is used to perform secondary analysis and generate a construction management report.

[0008] As a preferred embodiment of the dynamically optimized distributed photovoltaic power station construction image and video acquisition method described in this invention, the generated scene adaptation parameters include: The collected image and video data are frame-normalized, and the pixel values ​​of RGB images are normalized to unify the image size. In view of the characteristics of construction site data, data augmentation strategies are adopted to enhance the normalized image and video data. The scene type, lighting conditions, and motion state are respectively encoded using one-hot encoding or embedded with labels. The enhanced image and video data are then preliminarily classified using a convolutional neural network model to obtain scene type, lighting conditions, and motion state information. The convolutional neural network model includes a feature extraction backbone network and three parallel classification branches added after the feature extraction backbone network, including: The scene type branch is used to output the scene category probability; The illumination conditional branch is used to output the illumination conditional probability. Motion state branch, used to output motion state probabilities; Long Short-Term Memory (LSTM) networks are used to model the temporal relationships between consecutive frames and generate scene adaptation parameters.

[0009] The beneficial effects of this preferred technical solution are: it can adapt to the environmental characteristics of different construction stages and greatly improve adaptability to multiple scenarios.

[0010] As a preferred embodiment of the dynamically optimized distributed photovoltaic power station construction image and video acquisition method described in this invention, the generated scene adaptation parameters further include: The high-dimensional features extracted from each frame by the convolutional neural network model are concatenated with the preliminary classification results to form a temporal input sequence. The video frames are then interpolated over time to ensure that the sequence length input to the long short-term memory network model is fixed. The long short-term memory network model includes a basic long short-term memory network layer, an attention layer, and an output layer; The basic long short-term memory network layer adopts a two-layer bidirectional long short-term memory network with a hidden layer dimension of 256, capturing the short-term dependencies between consecutive frames. The attention layer adds a temporal attention module after the output of the long short-term memory network, assigns higher weights to key frames, and enhances the model’s attention to important temporal features. The output layer maps the hidden states of the Long Short-Term Memory network to a scene adaptation parameter vector through a fully connected layer, which includes sensor parameters and algorithm adaptation parameters. The time-series input sequence is input into the long short-term memory network model, the scene adaptation parameters are output, and the scene adaptation parameters are fed back to the front-end acquisition device or the back-end algorithm module in real time.

[0011] As a preferred embodiment of the dynamic optimization-based distributed photovoltaic power station construction image and video acquisition method of the present invention, wherein: the optimization of video stream acquisition efficiency through a dynamic frame rate allocation algorithm includes: Calculate the information entropy value of each frame of data; Based on the information entropy value, the frame data is divided into two categories: high information frames and low information frames. A higher frame rate is set for the high information frames, and a lower frame rate is set for the low information frames.

[0012] The beneficial effects of this preferred technical solution are: significantly reducing storage and transmission resource consumption and optimizing acquisition efficiency.

[0013] As a preferred embodiment of the dynamically optimized distributed photovoltaic power station construction image and video acquisition method of the present invention, wherein: the high-precision semantic segmentation processing of the key frames and the low-complexity compression encoding of the non-key frames include: The high-precision semantic segmentation process includes performing pixel-by-pixel classification on keyframe data, and the classification result includes the category label and bounding box information of the target object; post-processing of the classification result is performed using a conditional random field to optimize the segmentation accuracy of the boundary region. The low-complexity compression coding is achieved through discrete cosine transform, which converts spatial domain data into frequency domain data while retaining the main frequency domain components.

[0014] As a preferred embodiment of the dynamic optimization-based distributed photovoltaic power station construction image and video acquisition method of the present invention, the step of selectively transmitting key frame feature vectors or non-key frame compressed data packets according to network bandwidth fluctuations includes: The available bandwidth is calculated by monitoring network bandwidth fluctuations at the construction site in real time. When the current available bandwidth is lower than a preset threshold, key frame feature vectors are transmitted first. When the current available bandwidth is higher than a preset threshold, key frame feature vectors and non-key frame compressed data packets are transmitted simultaneously.

[0015] The beneficial effects of this preferred technical solution are: by combining low-complexity compression coding of non-key frames, the stable transmission of core information is ensured while reducing the amount of data.

[0016] As a preferred embodiment of the dynamically optimized distributed photovoltaic power station construction image and video acquisition method described in this invention, the secondary analysis of the data using a cloud-based deep learning model includes: The uploaded keyframe feature vectors and non-keyframe compressed data packets are decoded and reassembled to generate complete image and video data. The data is analyzed in a second phase using a pre-trained deep learning model to generate a construction management report; The construction management report includes construction progress assessment, early warning of safety hazards, and quality control recommendations.

[0017] Secondly, the present invention provides a dynamically optimized distributed photovoltaic power station construction image and video acquisition system, comprising: The data acquisition unit is used to adaptively configure the acquisition equipment based on the construction site environment feature extraction module, and to use a deep learning model to analyze the acquired image and video data in real time to generate scene adaptation parameters. The data partitioning unit is used to adjust the working mode of the acquisition device according to the scene adaptation parameters, optimize the acquisition efficiency of the video stream through a dynamic frame rate allocation algorithm, and divide the acquired data into key frames and non-key frames. The data processing unit is used to perform high-precision semantic segmentation processing on the key frames, perform low-complexity compression encoding on the non-key frames, and generate corresponding feature vectors and compressed data packets respectively. The monitoring and transmission unit is used to monitor the network environment of the construction site in real time through the dynamic network optimization module, and selectively transmit key frame feature vectors or non-key frame compressed data packets according to network bandwidth fluctuations. The report generation unit is used to upload the transmitted data to the cloud server, and then use a cloud-based deep learning model to perform secondary analysis on the data to generate a construction management report.

[0018] Thirdly, the present invention provides an electronic device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of a method for acquiring construction images and videos of a distributed photovoltaic power station based on dynamic optimization.

[0019] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a dynamically optimized method for acquiring construction images and videos of distributed photovoltaic power plants.

[0020] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention utilizes a deep learning model combining convolutional neural networks and long short-term memory networks to perform real-time analysis and scene adaptation parameter generation on construction site images and video data, achieving accurate identification of scene type, lighting conditions, and motion state. High-precision semantic segmentation of keyframes can extract category labels and bounding box information of target objects. Combined with secondary analysis in the cloud, it can generate management reports containing construction progress, safety hazards, and quality issues, solving the problem of insufficient intelligent analysis in existing technologies and significantly improving the accuracy of construction management. This invention addresses the complex scenarios of variable lighting and frequent equipment movement in distributed photovoltaic power station construction. It dynamically adjusts the exposure time, gain, and frame rate of the acquisition equipment through scene adaptation parameters and utilizes data augmentation strategies to improve model robustness. Compared with the limitations of existing technologies that only target specific scenarios, this method can adapt to the environmental characteristics of different construction stages, greatly improving multi-scenario adaptability. Furthermore, this invention monitors bandwidth fluctuations in real time through a dynamic network optimization module. When the network is congested, it prioritizes the transmission of key frame feature vectors, and when bandwidth is sufficient, it synchronously transmits compressed data of non-key frames. Combined with low-complexity compression encoding of non-key frames, it reduces data volume while ensuring stable transmission of core information. The dynamic frame rate allocation algorithm of this invention distinguishes between key and non-key frames based on information entropy values. It uses a high frame rate for high-information frames and a lower frame rate for low-information frames. Combined with differentiated processing, this significantly reduces storage and transmission resource consumption and optimizes acquisition efficiency. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0022] Figure 1 This is a schematic diagram of the overall process logic of a method for acquiring construction images and videos of a distributed photovoltaic power station based on dynamic optimization, provided as an embodiment of the present invention. Detailed Implementation

[0023] 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.

[0024] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for acquiring construction images and videos of distributed photovoltaic power stations based on dynamic optimization is provided, such as... Figure 1 The specific steps shown are as follows: S100: Based on the construction site environment feature extraction module, the acquisition equipment is adaptively configured, and the acquired image and video data are analyzed in real time using a deep learning model to generate scene adaptation parameters. S200: Adjusts the working mode of the acquisition device according to the scene adaptation parameters, optimizes the acquisition efficiency of the video stream through the dynamic frame rate allocation algorithm, and divides the acquired data into key frames and non-key frames; S300: Performs high-precision semantic segmentation on key frames and low-complexity compression encoding on non-key frames, generating corresponding feature vectors and compressed data packets respectively; S400: The dynamic network optimization module monitors the network environment at the construction site in real time and selectively transmits key frame feature vectors or non-key frame compressed data packets according to network bandwidth fluctuations. S500: The transmitted data is uploaded to the cloud server, and the cloud-based deep learning model is used to perform secondary analysis on the data to generate a construction management report.

[0025] It should be noted that, to address the shortcomings of existing distributed photovoltaic power station construction site image and video data acquisition technologies, such as insufficient intelligent analysis, poor adaptability to multiple scenarios, and low data transmission stability, steps S100-S500 utilize a deep learning model combining convolutional neural networks and long short-term memory networks to perform real-time analysis and scene adaptation parameter generation on construction site images and video data. This enables accurate identification of scene type, lighting conditions, and motion status. High-precision semantic segmentation of keyframes extracts category labels and bounding box information of target objects. Combined with secondary analysis in the cloud, this generates management reports containing information on construction progress, safety hazards, and quality issues, addressing the limitations of existing intelligent analysis technologies and significantly improving the accuracy of construction management.

[0026] It should be noted that this invention addresses the complex scenarios of distributed photovoltaic power station construction, characterized by variable sunlight and frequent equipment movement. It dynamically adjusts the exposure time, gain, and frame rate of the acquisition equipment through scenario adaptation parameters and utilizes data augmentation strategies to enhance model robustness. Compared to existing technologies that are limited to specific scenarios, this method can adapt to the environmental characteristics of different construction stages, significantly improving multi-scenario adaptability. Furthermore, this invention uses a dynamic network optimization module to monitor bandwidth fluctuations in real time. When the network is congested, it prioritizes the transmission of key frame feature vectors; when bandwidth is sufficient, it synchronously transmits compressed data of non-key frames. Combined with low-complexity compression encoding of non-key frames, it reduces data volume while ensuring stable transmission of core information. The dynamic frame rate allocation algorithm of this invention distinguishes between key and non-key frames based on information entropy values. It uses a high frame rate for high-information frames and a lower frame rate for low-information frames. Combined with differentiated processing, this significantly reduces storage and transmission resource consumption and optimizes acquisition efficiency.

[0027] In this embodiment of the invention, step S100 adaptively configures the acquisition device based on the construction site environment feature extraction module, and uses a deep learning model to perform real-time analysis on the acquired image and video data to generate scene adaptation parameters, including the following sub-steps A1~A3: In A1: The collected image and video data are frame-normalized, and the pixel values ​​of RGB images are normalized to unify the image size; In view of the characteristics of construction site data, data augmentation strategies are used to enhance the normalized image and video data. Specifically, the implementation process of this invention begins with the construction site environmental feature extraction module, which uses a convolutional neural network to perform preliminary classification of image and video data captured by the acquisition device. Specifically, cameras and sensors work together to acquire image and video data of the current scene and transmit it to the convolutional neural network for processing.

[0028] Specifically, considering the characteristics of construction site data, the collected image and video data undergo frame standardization processing, including RGB image pixel value normalization and uniform image size (e.g., 224*224). Data augmentation strategies are also employed to improve model robustness, including random cropping, rotation, lighting perturbation, and motion blur.

[0029] In A2: Scene type, lighting conditions, and motion state are individually encoded or labeled and embedded. A convolutional neural network model is used to perform preliminary classification of the enhanced image and video data to obtain scene type, lighting conditions, and motion state information. In this embodiment of the invention, the convolutional neural network model includes a feature extraction backbone network and three parallel classification branches added after the feature extraction backbone network, including: The scene type branch, consisting of a fully connected layer and Softmax, is used to output the scene category probability. The illumination condition branch, consisting of a fully connected layer and Softmax, is used to output the illumination condition probability. The motion state branch extracts optical flow features from the video frame sequence, combines them with the CNN features of the current frame, and outputs three motion state probabilities through a fully connected layer.

[0030] Specifically, scene type (e.g., "foundation pit," "scaffolding," "material storage area"), lighting conditions (e.g., "strong light," "normal," "low light," "backlight"), and motion state (e.g., "stationary," "slowly moving," "fast moving") are respectively one-hot encoded or labeled and embedded. A CNN model is then used to classify and obtain scene type, lighting conditions, and motion state information. One-hot encoding is a feature engineering method that converts categorical variables into binary vectors. Its core idea is to create an independent binary column for each category; when a sample belongs to a certain category, the corresponding column has a value of 1, and the values ​​of other columns are 0. This encoding method avoids the model misinterpreting the numerical magnitude of categorical variables as a semantic order relationship, and also adapts to the numerical input requirements of machine learning algorithms.

[0031] In A3: Long Short-Term Memory (LSTM) networks are used to model the temporal relationships between consecutive frames, generating scene adaptation parameters; detailed steps include: The high-dimensional features extracted from each frame by the convolutional neural network model are concatenated with the preliminary classification results to form a temporal input sequence. Temporal interpolation is then performed on the video frames to ensure that the sequence length input to the long short-term memory network model is fixed. The long short-term memory network model includes a basic long short-term memory network layer, an attention layer, and an output layer; The basic Long Short-Term Memory (LSTM) network layer employs a two-layer bidirectional LSM network with a hidden layer dimension of 256, capturing short-term dependencies between consecutive frames. The attention layer adds a temporal attention module after the LSM network output, assigning higher weights to keyframes and enhancing the model's attention to important temporal features. The output layer maps the hidden states of the LSM network to scene adaptation parameter vectors through a fully connected layer, including sensor parameters and algorithm adaptation parameters, specifically including camera exposure time, gain settings, and parameters required by the dynamic frame rate allocation algorithm. The time-series input sequence is input into the Long Short-Term Memory network model, which outputs scene adaptation parameters and feeds these parameters back to the front-end acquisition device or the back-end algorithm module in real time.

[0032] Specifically, the expression for the scene adaptation parameter is:

[0033] in, Indicates the first i Scene adaptation parameters for frames, Indicates the first j Weighting factors for each scenario type Let α represent the weight factor of the k-th temporal feature, and β represent the normalization coefficients of the scene type and the temporal feature, respectively.

[0034] It should be noted that step S100 above performs real-time analysis and scene adaptation parameter generation on construction site images and video data, achieving accurate identification of scene type, lighting conditions, and motion state. By dynamically adjusting parameters such as exposure time and gain of the acquisition equipment, the problem of insufficient adaptability of traditional fixed parameter strategies in varying lighting and complex motion scenarios is effectively solved.

[0035] In this embodiment of the invention, step S200 adjusts the working mode of the acquisition device according to the scene adaptation parameters, optimizes the acquisition efficiency of the video stream through a dynamic frame rate allocation algorithm, and divides the acquired data into key frames and non-key frames, including: Specifically, the dynamic frame rate allocation algorithm optimizes the acquisition efficiency of the video stream. The core of the dynamic frame rate allocation algorithm lies in the calculation of the information entropy value, and the formula for calculating the information entropy value is:

[0036] in, Indicates the first n The information entropy value of the frame, p ( m ) indicates the first m Gray-level probability distribution of each pixel M This represents the total number of pixels.

[0037] Specifically, based on the information entropy value, frame data is divided into two categories: high-information frames and low-information frames. A higher frame rate is set for high-information frames to ensure data integrity, while the frame rate is reduced for low-information frames to reduce redundant data.

[0038] It should be noted that step S200 above, based on scene adaptation parameters and a dynamic frame rate allocation algorithm driven by information entropy, intelligently divides the video stream into key frames and non-key frames, and adaptively adjusts the acquisition frame rate according to the content complexity. This mechanism increases the frame rate in high-motion scenes to capture details and decreases the frame rate in static scenes to reduce redundancy, significantly optimizing storage and transmission efficiency.

[0039] In this embodiment of the invention, step S300 includes the following sub-steps C1 and C2: In C1: High-precision semantic segmentation is performed on keyframes to generate corresponding feature vectors; Specifically, high-precision semantic segmentation processing includes performing pixel-by-pixel classification on keyframe data, with the classification results containing the category label and bounding box information of the target object; and using conditional random fields to post-process the classification results to optimize the segmentation accuracy of the boundary region.

[0040] In C2: Low-complexity compression encoding is performed on non-critical frames to generate corresponding compressed data packets; Specifically, low-complexity compression coding is achieved through discrete cosine transform, which converts spatial domain data into frequency domain data while retaining the main frequency domain components.

[0041] It should be noted that step S300 above, based on scene adaptation parameters and a dynamic frame rate allocation algorithm driven by information entropy, intelligently divides the video stream into key frames and non-key frames, and adaptively adjusts the acquisition frame rate according to the content complexity. This mechanism increases the frame rate in high-motion scenes to capture details, and decreases the frame rate in static scenes to reduce redundancy.

[0042] In this embodiment of the invention, step S400 involves real-time monitoring of the network environment at the construction site using a dynamic network optimization module, and selectively transmitting key frame feature vectors or non-key frame compressed data packets based on network bandwidth fluctuations. The available bandwidth is calculated by monitoring network bandwidth fluctuations at the construction site in real time. When the available bandwidth is lower than a preset threshold, priority is given to transmitting key frame feature vectors; When the available bandwidth is higher than a preset threshold, key frame feature vectors and non-key frame compressed data packets are transmitted simultaneously.

[0043] It should be noted that the dynamic network optimization module is responsible for monitoring the network environment at the construction site in real time and selectively transmitting data based on network bandwidth fluctuations. The dynamic network optimization module calculates the currently available bandwidth...B ( t )= R ( t ) / T ,in, R ( t () represents the actual transmission rate at time t. T This indicates the preset time window length, which determines the priority of data transmission.

[0044] It should be noted that the determination of the preset threshold is mainly based on the typical network bandwidth fluctuation range (0.5-20Mbps) at the construction site, the basic transmission requirements of key frame feature vectors (usually not less than 2Mbps), and the incremental transmission load of non-key frame compressed data packets. Through statistical analysis of historical transmission success rate and latency data, the threshold is set at the minimum bandwidth limit to ensure stable transmission of key frames, so as to ensure the reliable uploading of core construction features when network conditions deteriorate.

[0045] In this embodiment of the invention, step S500 involves uploading the transmitted data to a cloud server, using a cloud-based deep learning model to perform secondary analysis on the data, and generating a construction management report, including: The uploaded keyframe feature vectors and non-keyframe compressed data packets are decoded and reassembled to generate complete image and video data. The data is analyzed in a second phase using a pre-trained deep learning model to generate a construction management report; The construction management report includes construction progress assessment, early warning of safety hazards, and quality control recommendations.

[0046] For example, when a keyframe shows that the installation angle of the photovoltaic module deviates from the design value, the construction management report will propose specific adjustment suggestions. This information is ultimately fed back to the terminal equipment, providing comprehensive support for the management of the construction site.

[0047] It should be noted that this invention exhibits strong flexibility in multi-scene adaptation. For example, in nighttime construction scenarios, the scene adaptation parameters automatically adjust the camera's exposure time and gain settings based on light intensity to ensure image quality. In scenarios with complex motion states, the dynamic frame rate allocation algorithm dynamically adjusts the frame rate based on the information entropy value of moving objects, thereby capturing more detailed information. This multi-scene adaptation mechanism enables this invention to flexibly address data acquisition needs in different construction environments.

[0048] Throughout the implementation process, the various modules collaborate closely through software interfaces and hardware devices. For example, the construction site environmental feature extraction module collects data using cameras and sensors and transmits the data to the processor unit where the dynamic frame rate allocation algorithm resides. The dynamic frame rate allocation algorithm adjusts the camera's frame rate setting based on the information entropy value calculation result, while simultaneously sending keyframe data to the high-precision semantic segmentation processing module and non-keyframe data to the low-complexity compression encoding module. The output data from these two modules is then transmitted to the dynamic network optimization module, which determines the data transmission strategy based on the network environment. Finally, the cloud-based deep learning model receives and processes the transmitted data, generates a construction management report, and feeds it back to the terminal devices at the construction site.

[0049] It should be noted that in step S500 above, the received keyframe feature vectors and non-keyframe compressed data are decoded, reassembled, and subjected to secondary analysis in the cloud. A pre-trained deep learning model is then used to generate a management report containing information on construction progress, safety hazards, and quality issues. Through this edge-cloud collaborative intelligent analysis mechanism, a closed loop from data collection to management decision-making is achieved, significantly improving the automation level and response speed of construction management.

[0050] Example 2, based on the previous example, provides an application example of a dynamically optimized method for acquiring construction images and videos of distributed photovoltaic power stations, to verify and illustrate the technical effects of the method.

[0051] This embodiment uses a simulation platform to conduct a simulation experiment on the performance of the dynamic frame rate allocation algorithm. The experimental conditions simulate three typical construction scenarios (component installation, cable laying, and equipment stationary), and a 1-hour video stream is collected. The simulation data results are shown in Table 1 below.

[0052] Table 1: Simulation results of dynamic frame rate allocation algorithm performance.

[0053]

[0054] As shown in Table 1, in high-motion scenarios (component installation), the data volume is reduced by only 28%, but the key information retention rate reaches 99.2%, balancing detail capture and efficiency. In medium-motion scenarios (cable laying), the data volume is reduced by 45%, with an information retention rate of 97.5%, adapting to the needs of medium-complexity construction scenarios. In low-motion scenarios (equipment stationary), the data volume is reduced by 72%, while still maintaining a key information retention rate of 95.8%, greatly reducing storage and transmission pressure. Therefore, in actual deployment, the information entropy threshold can be dynamically adjusted according to the construction stage (e.g., the threshold is set to 6.0 bits / pixel in the component installation stage and 4.0 bits / pixel in the equipment stationary stage) to further optimize the data allocation strategy. Finally, combined with the results of secondary cloud analysis, semantic filtering rules can be introduced for scenarios that frequently exhibit "high information entropy but low value" to avoid invalid data consuming bandwidth.

[0055] Based on the above analysis, the method provided by this invention offers a lightweight and robust video acquisition optimization solution for the construction of distributed photovoltaic power stations through data-driven dynamic frame rate adjustment, which is especially suitable for outdoor scenarios with unstable networks and limited storage resources.

[0056] Example 3: This example provides a dynamically optimized distributed photovoltaic power station construction image and video acquisition system, including: The data acquisition unit is used to adaptively configure the acquisition equipment based on the construction site environment feature extraction module, and to use a deep learning model to analyze the acquired image and video data in real time to generate scene adaptation parameters. The data partitioning unit is used to adjust the working mode of the acquisition device according to the scene adaptation parameters, optimize the acquisition efficiency of the video stream through the dynamic frame rate allocation algorithm, and divide the acquired data into key frames and non-key frames. The data processing unit is used to perform high-precision semantic segmentation on key frames, perform low-complexity compression encoding on non-key frames, and generate corresponding feature vectors and compressed data packets respectively. The monitoring and transmission unit is used to monitor the network environment of the construction site in real time through the dynamic network optimization module, and selectively transmit key frame feature vectors or non-key frame compressed data packets according to network bandwidth fluctuations. The report generation unit is used to upload the transmitted data to the cloud server, and then use a cloud-based deep learning model to perform secondary analysis on the data to generate a construction management report.

[0057] It should be noted that the technical solution of the distributed photovoltaic power station construction image and video acquisition system based on dynamic optimization is based on the same concept as the above-mentioned technical solution of the distributed photovoltaic power station construction image and video acquisition method based on dynamic optimization. For details not described in detail in the technical solution of the distributed photovoltaic power station construction image and video acquisition system based on dynamic optimization in this embodiment, please refer to the description of the above-mentioned technical solution of the distributed photovoltaic power station construction image and video acquisition method based on dynamic optimization.

[0058] This embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a dynamically optimized method for acquiring construction images and videos of distributed photovoltaic power stations. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0059] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method proposed in the above embodiments.

[0060] The storage medium proposed in this embodiment belongs to the same inventive concept as the method 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.

[0061] 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, random access memory, flash memory, hard disk, or optical disk, and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute the method of the embodiments of the present invention.

[0062] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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 distributed photovoltaic power station construction image video acquisition method based on dynamic optimization, characterized in that, The method comprises the following steps: Based on the construction site environment feature extraction module, the acquisition device is adaptively configured, the collected image and video data are analyzed in real time by using a deep learning model, and scene adaptation parameters are generated; According to the scene adaptation parameters, the working mode of the acquisition device is adjusted, the video stream acquisition efficiency is optimized by a dynamic frame rate allocation algorithm, and the collected data is divided into key frames and non-key frames; The high-precision semantic segmentation processing is performed on the key frames, and the low-complexity compression encoding is performed on the non-key frames, and corresponding feature vectors and compressed data packets are generated respectively; The network environment of the construction site is monitored in real time by a dynamic network optimization module, and the key frame feature vectors or non-key frame compressed data packets are selectively transmitted according to the network bandwidth fluctuation; The transmitted data is uploaded to a cloud server, and the data is analyzed again by using a cloud deep learning model to generate a construction management report.

2. The dynamic optimization-based distributed photovoltaic power station construction image video acquisition method according to claim 1, characterized in that, The scene adaptation parameters include: The collected image and video data are frame standardized, and the RGB image pixel value is normalized to unify the image size; according to the data characteristics of the construction site, the normalized image and video data are enhanced by using a data enhancement strategy; The scene type, illumination condition and motion state are respectively one-hot encoded or label embedded, the enhanced image and video data are preliminarily classified by using a convolutional neural network model, and the scene type, illumination condition and motion state information are obtained; The convolutional neural network model includes a feature extraction backbone network and three parallel classification branches added after the feature extraction backbone network, including: The scene type branch is used for outputting the scene category probability; The illumination condition branch is used for outputting the illumination condition probability; The motion state branch is used for outputting the motion state probability; The long short-term memory network is used for modeling the time sequence relationship between the continuous frames to generate the scene adaptation parameters. 3.The method of claim 2, wherein, The scene adaptation parameters also include: The high-dimensional features of each frame extracted by the convolutional neural network model are spliced with the preliminary classification results to form a time sequence input sequence, the video frames are time interpolated to ensure that the sequence length input into the long short-term memory network model is fixed; The long short-term memory network model includes a basic long short-term memory network layer, an attention layer and an output layer; The basic long short-term memory network layer adopts a two-layer bidirectional long short-term memory network with a hidden layer dimension of 256 to capture the short-term dependence relationship between the previous and subsequent frames; The attention layer adds a time attention module after the long short-term memory network output, gives higher weight to the key frames, and improves the attention degree of important time sequence features of the model; The output layer maps the hidden state of the long short-term memory network to a scene adaptation parameter vector through a fully connected layer, which includes sensor parameters and algorithm adaptation parameters; The time sequence input sequence is input into the long short-term memory network model, the scene adaptation parameters are output, and the scene adaptation parameters are fed back to the front-end acquisition device or the back-end algorithm module in real time. 4.The method of claim 3, wherein, The dynamic frame rate allocation algorithm for optimizing the video stream acquisition efficiency includes: The information entropy value of each frame of data is calculated; According to the information entropy value, the frame data is divided into high information frame and low information frame, a higher frame rate is set for the high information frame, and the frame rate is reduced for the low information frame. 5.The dynamic optimization-based distributed photovoltaic power station construction image video acquisition method according to claim 4, characterized in that, The high-precision semantic segmentation processing on the key frame and the low-complexity compression encoding on the non-key frame include: The high-precision semantic segmentation processing includes performing a pixel-by-pixel classification operation on the key frame data, and the classification result contains the category label and the bounding box information of the target object; and the classification result is post-processed by using a conditional random field to optimize the segmentation accuracy of the boundary region; The low-complexity compression encoding is realized by discrete cosine transform, which converts spatial domain data into frequency domain data and retains main frequency components. 6.The dynamic optimization-based distributed photovoltaic power station construction image video acquisition method according to claim 5, characterized in that, The selective transmission of the key frame feature vector or the non-key frame compressed data packet according to the network bandwidth fluctuation includes: The current available bandwidth is calculated by monitoring the network bandwidth fluctuation of the construction site in real time; When the current available bandwidth is lower than a preset threshold, the key frame feature vector is preferentially transmitted; When the current available bandwidth is higher than the preset threshold, the key frame feature vector and the non-key frame compressed data packet are simultaneously transmitted.

7. The dynamic optimization-based distributed photovoltaic power station construction image video acquisition method according to claim 6, characterized in that, The secondary analysis of the data by using the cloud deep learning model includes: The uploaded key frame feature vector and non-key frame compressed data packet are decoded and recombined to generate complete image and video data; The pre-trained deep learning model is used for secondary analysis of the data to generate a construction management report; The construction management report includes construction progress evaluation, safety hazard early warning and quality control suggestion.

8. The distributed photovoltaic power station construction image video acquisition system based on dynamic optimization, applying the distributed photovoltaic power station construction image video acquisition method based on dynamic optimization according to any one of claims 1-7, characterized in that, It includes: The data acquisition unit is used for adaptive configuration of the acquisition equipment based on the construction site environment feature extraction module, real-time analysis of the collected image and video data by using the deep learning model, and generation of scene adaptation parameters; The data division unit is used for adjusting the working mode of the acquisition equipment according to the scene adaptation parameters, optimizing the acquisition efficiency of the video stream by using a dynamic frame rate allocation algorithm, and dividing the collected data into key frames and non-key frames; The data processing unit is used for performing high-precision semantic segmentation processing on the key frame and low-complexity compression encoding on the non-key frame, and generating corresponding feature vectors and compressed data packets respectively; The monitoring and transmission unit is used for real-time monitoring of the network environment of the construction site by using a dynamic network optimization module, and selectively transmitting the key frame feature vector or the non-key frame compressed data packet according to the network bandwidth fluctuation; The report generation unit is used for uploading the transmitted data to the cloud server, and generating a construction management report by using a cloud deep learning model for secondary analysis of the data. 9.An electronic device comprising a memory and a processor, the electronic device characterized by: The memory is used for storing computer executable instructions, and the processor executes the computer executable instructions to realize the steps of the distributed photovoltaic power station construction image and video acquisition method based on dynamic optimization in any one of claims 1-7.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: The computer executable instructions are executed by the processor to realize the steps of the distributed photovoltaic power station construction image and video acquisition method based on dynamic optimization in any one of claims 1-7.

Images