An edge-computing-based internet-of-things image data processing method and system

Abstract

This invention relates to the field of data processing technology, specifically to an IoT image data processing method and system based on edge computing. The method includes receiving IoT image data captured by a surveillance camera in an industrial IoT scenario; obtaining a confidence level of image quality recognition obtained by an edge computing node performing image quality recognition processing on the IoT image data; comparing the confidence level of the quality recognition with a preset confidence level to obtain a comparison value; obtaining image fault information determined by the edge computing node based on the comparison value; determining maintenance work order information corresponding to the surveillance camera based on the image fault information and a preset maintenance model; and sending the maintenance work order information to an operation and maintenance management platform to complete the processing of the IoT image data. The purpose of this invention is to address the problem of reduced accuracy in IoT data processing methods after the quality of the original image data deteriorates.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to an IoT image data processing method and system based on edge computing. Background Technology

[0002] In industrial production environments, the quality of raw image data captured by surveillance cameras continuously and subtly declines due to factors such as the accumulation of metal dust and oil mist, changes in lighting conditions, and equipment vibration. Debris adhering to the lens surface of surveillance cameras can blur images and reduce contrast; uneven or insufficient lighting can affect the recognition of image features; and even slight equipment vibration can cause image blurring or ghosting. This decline in image data quality leads to deviations in the training data of existing data processing methods, resulting in decreased accuracy, increased false alarms, and missed alarms. For example, dust may be misidentified as a defect, and subtle defects may fail to be detected.

[0003] Meanwhile, existing IoT data processing methods cannot automatically adjust image preprocessing parameters when they detect image blur. When performance degradation occurs, technicians need to conduct time-consuming manual troubleshooting and intervention, which is inefficient and cannot cope with continuous and subtle changes in industrial environments, thus reducing the reliability of image data processing. Summary of the Invention

[0004] The purpose of this invention is to provide an IoT image data processing method and system based on edge computing, which solves the problem that the accuracy of existing IoT data processing methods decreases after the quality of the original image data deteriorates.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an IoT image data processing method based on edge computing, comprising:

[0006] Receive IoT image data collected by surveillance cameras in industrial IoT scenarios;

[0007] After loading the IoT image data into the edge computing node corresponding to the surveillance camera, the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data is obtained.

[0008] The confidence level of the quality identification is compared with a preset confidence level to obtain a comparison value of the confidence level;

[0009] After transmitting the comparison value of the confidence level to the edge computing node, the image fault information determined by the edge computing node based on the comparison value of the confidence level is obtained;

[0010] Based on the image fault information and the preset maintenance model, the maintenance work order information corresponding to the surveillance camera is determined; wherein, the maintenance work order information includes the surveillance camera's number information, location information, and maintenance suggestion information;

[0011] The maintenance work order information is sent to the operation and maintenance management platform to complete the processing of IoT image data.

[0012] Further, the step of loading the IoT image data into the edge computing node corresponding to the surveillance camera, and then obtaining the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data, includes:

[0013] The IoT image data is subjected to quality assessment processing to obtain the data quality assessment coefficient of the IoT image data;

[0014] Using the aforementioned data quality evaluation coefficients, data correction processing is performed on the IoT image data to obtain corrected IoT image data;

[0015] After loading the corrected IoT image data into the edge computing node corresponding to the surveillance camera, the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data is obtained.

[0016] Further, the step of performing quality assessment processing on the IoT image data to obtain the data quality assessment coefficient of the IoT image data includes:

[0017] The IoT image data is subjected to quality assessment processing to obtain the initial evaluation coefficients of the IoT image data;

[0018] The initial evaluation coefficients are compared using the evaluation coefficient threshold to obtain the coefficient comparison results;

[0019] If the coefficient comparison result indicates that the initial evaluation coefficient is unqualified, an alarm is issued, and IoT image data is reacquired. Then, the quality evaluation processing of the reacquired IoT image data is performed to obtain the initial evaluation coefficient for re-evaluation. The initial evaluation coefficient for re-evaluation is then compared with the evaluation coefficient threshold until a qualified coefficient comparison result is obtained.

[0020] If the coefficient comparison result indicates that the initial evaluation coefficient is qualified, then the initial evaluation coefficient shall be used as the data quality evaluation coefficient.

[0021] Furthermore, the steps of performing data correction processing on the IoT image data using the aforementioned data quality evaluation coefficients to obtain corrected IoT image data include:

[0022] Based on the data quality evaluation coefficients, determine the image correction area for the IoT image data;

[0023] Using a preset correction model, a correction scheme is matched to the image correction area to obtain a data correction scheme;

[0024] Using the aforementioned data correction scheme, IoT image data is processed to obtain corrected IoT image data.

[0025] Furthermore, the step of matching a correction scheme to the image correction region using a preset correction model to obtain a data correction scheme includes:

[0026] Using a preset correction model, the image correction region is matched with correction schemes to obtain all candidate correction schemes;

[0027] All candidate correction schemes are screened to obtain a data correction scheme.

[0028] Furthermore, the steps for receiving IoT image data collected by monitoring cameras in an industrial IoT scenario include:

[0029] Acquire image acquisition parameters from surveillance cameras in an industrial IoT scenario; wherein, the image acquisition parameters include acquisition time, acquisition accuracy, and acquisition area;

[0030] Based on the image acquisition parameters, receive raw IoT image data acquired by a monitoring camera in an industrial IoT scenario;

[0031] Image analysis is performed on the raw IoT image data to obtain the quality score of each frame.

[0032] The IoT image data is determined based on the image acquisition parameters and the quality score of each frame.

[0033] Further, the step of determining the IoT image data based on the image acquisition parameters and the quality score of each frame includes:

[0034] The image acquisition parameters are adjusted using the quality score of each frame to obtain the adjusted image acquisition parameters;

[0035] Using the adjusted image acquisition parameters, the monitoring camera in the industrial IoT scenario is driven to acquire data and obtain IoT image data.

[0036] Further, the step of adjusting the image acquisition parameters using the quality score of each frame to obtain the adjusted image acquisition parameters includes:

[0037] Based on the quality score of each frame and the preset parameter adjustment model, the adjustment direction and adjustment range of the image acquisition parameters are determined;

[0038] By adjusting the direction and magnitude of the image acquisition parameters, the image acquisition parameters are adjusted to obtain the adjusted image acquisition parameters.

[0039] Further, the step of comparing the confidence level of the quality identification with a preset confidence level to obtain a comparison value of the confidence level includes:

[0040] The confidence level of the quality identification is compared with a preset confidence level to obtain an initial comparison value of the confidence level;

[0041] The initial comparison value of the confidence level is corrected to obtain the comparison value of the confidence level.

[0042] The present invention also provides an IoT image data processing system based on edge computing, the system comprising:

[0043] The data receiving module is used to receive IoT image data collected by monitoring cameras in industrial IoT scenarios.

[0044] The first acquisition module is used to load the IoT image data into the edge computing node corresponding to the surveillance camera, and then acquire the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data.

[0045] The confidence comparison module is used to compare the confidence level of the quality identification with a preset confidence level to obtain a comparison value of the confidence level.

[0046] The second acquisition module is used to transmit the comparison value of the confidence level to the edge computing node and then acquire the image fault information determined by the edge computing node based on the comparison value of the confidence level.

[0047] The information determination module is used to determine the maintenance work order information corresponding to the surveillance camera based on the image fault information and the preset maintenance model; wherein, the maintenance work order information includes the surveillance camera's number information, location information, and maintenance suggestion information;

[0048] The information sending module is used to send the maintenance work order information to the operation and maintenance management platform to complete the processing of IoT image data.

[0049] Compared with existing technologies, the IoT image data processing method and system based on edge computing of the present invention have the following advantages:

[0050] This invention receives IoT image data collected by surveillance cameras and loads it onto an edge computing node for image quality recognition, obtaining a confidence level for the quality recognition. Subsequently, this confidence level is compared with a preset confidence level to obtain a comparison value, which is then transmitted to the edge computing node to obtain image fault information. Finally, based on the image fault information and a preset maintenance model, the corresponding maintenance work order information for the surveillance camera is determined and sent to the operation and maintenance management platform. By introducing a real-time image quality recognition and fault diagnosis mechanism, image data quality problems can be detected and located promptly, avoiding a decrease in recognition accuracy due to data quality degradation. Simultaneously, the generated maintenance work order is sent to the operation and maintenance management platform, enabling preventative maintenance and rapid fault response for surveillance cameras, significantly improving automation levels and operation and maintenance efficiency, reducing the need for manual intervention and potential economic losses and safety hazards, while also improving the reliability of IoT image data processing. Attached Figure Description

[0051] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0052] Figure 1 This is a flowchart of an IoT image data processing method based on edge computing according to the present invention.

[0053] Figure 2 This is a structural block diagram of an IoT image data processing system based on edge computing according to the present invention.

[0054] In the diagram: 210, data receiving module; 220, first acquisition module; 230, confidence comparison module; 240, second acquisition module; 250, information determination module; 260, information sending module.

[0055] The implementation and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0056] The following drawings disclose several embodiments of the present invention. For clarity, many practical details will be described in the following description. However, it should be understood that these practical details are not intended to limit the invention. That is, in some embodiments of the invention, these practical details are not essential. Furthermore, for the sake of simplicity, some conventional structures and components will be shown in the drawings in a simple schematic manner.

[0057] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0058] Furthermore, in this invention, the use of terms such as "first" and "second" is for descriptive purposes only and does not specifically refer to any order or sequence, nor is it intended to limit the invention. They are merely used to distinguish components or operations described using the same technical terms, and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If a combination of technical solutions is contradictory or impossible to implement, such a combination should be considered nonexistent and not within the scope of protection claimed by this invention.

[0059] To further understand the content, features, and effects of this invention, the following embodiments are provided, and detailed descriptions are given below in conjunction with the accompanying drawings:

[0060] Please see Figure 1 This invention provides an IoT image data processing method based on edge computing, comprising the following steps:

[0061] S100: Receive IoT image data collected by a surveillance camera in an industrial IoT scenario. Specifically, the surveillance camera directly sends the collected raw image data to the edge computing node. For example, the surveillance camera transmits image data streams or single-frame images to the edge computing node at fixed time intervals or when a specific event is detected.

[0062] S200: After loading the IoT image data into the edge computing node corresponding to the surveillance camera, obtain the confidence level of the image quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data. Specifically, the edge computing node uses a pre-trained image quality assessment model to analyze the loaded image data and outputs a numerical value representing the image quality level, i.e., the confidence level. This model can evaluate based on multiple dimensions such as image sharpness, brightness, contrast, and noise level.

[0063] S300. The confidence level of the quality identification is compared with a preset confidence level to obtain a comparison value of the confidence level. Specifically, a fixed confidence level threshold is set. When the confidence level of the quality identification is lower than the threshold, it is considered that there is a problem with the image quality, and the difference between the two is calculated as the comparison value.

[0064] S400: After transmitting the comparison value of the confidence level to the edge computing node, the edge computing node obtains the image fault information determined by the edge computing node based on the comparison value of the confidence level. Specifically, the edge computing node can determine the degree of image quality degradation based on the magnitude of the comparison value and map it to a specific fault type, such as image blurring, insufficient lighting, and image jitter.

[0065] S500. Based on the image fault information and the preset maintenance model, determine the maintenance work order information corresponding to the surveillance camera; wherein, the maintenance work order information includes the surveillance camera's identification number, location information, and maintenance suggestion information. Specifically, the maintenance model can be a rule base or a decision tree, matching corresponding maintenance suggestions based on different image fault information. For example, when an image blur fault is identified, the maintenance model may suggest cleaning the lens; when insufficient lighting is identified, it may suggest checking the supplementary lighting. The maintenance work order information typically includes the surveillance camera's identification number, location information, and specific maintenance suggestion information to facilitate processing by maintenance personnel.

[0066] S600. The maintenance work order information is sent to the operation and maintenance management platform to complete the processing of IoT image data. Upon receiving the maintenance work order, the operation and maintenance management platform can automatically assign tasks to the corresponding operation and maintenance personnel based on the work order content, or perform further scheduling and management, thereby completing the processing of the IoT image data.

[0067] This invention introduces quality identification of IoT image data at edge computing nodes, enabling real-time automated assessment of image data quality. Based on the confidence level of the quality identification, it can proactively detect declining trends in image quality and further determine specific image fault information. According to the fault information and a preset maintenance model, a work order containing specific maintenance suggestions is automatically generated and sent to the operation and maintenance management platform. This transforms the passive response mode of image data processing into a proactive early warning and automated maintenance mode, significantly improving the reliability, automation level, and operation and maintenance efficiency of industrial IoT image data processing systems, effectively avoiding false alarms, missed alarms, and unnecessary production interruptions and economic losses caused by image quality issues.

[0068] In some embodiments of this application described above, the step of loading the IoT image data into the edge computing node corresponding to the surveillance camera and then obtaining the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data includes:

[0069] The IoT image data undergoes quality assessment processing to obtain data quality assessment coefficients. Specifically, the quality assessment processing analyzes the intrinsic quality of the image data, such as sharpness, brightness, contrast, and noise levels, and generates quantified data quality assessment coefficients based on the assessment results. These data quality assessment coefficients can be comprehensive indicators reflecting the overall quality status of the current IoT image data. The data quality assessment coefficients can be obtained using various image quality assessment algorithms, such as reference image-based assessment methods (e.g., PSNR or SSIM) or reference-free assessment methods (e.g., BRISQUE or NIQE). The aim is to objectively quantify the quality of the image data, providing a basis for subsequent data correction.

[0070] Using the aforementioned data quality evaluation coefficients, IoT image data undergoes data correction processing to obtain corrected IoT image data. Specifically, data correction processing involves targeted optimization of the image data based on the image quality issues reflected by the data quality evaluation coefficients. For example, if the evaluation coefficients indicate that the image is blurry, deblurring is performed; if noise is present, denoising is performed; if brightness is uneven, brightness equalization is performed. Through correction, higher-quality corrected IoT image data can be obtained. Data correction processing may include, but is not limited to, image enhancement, denoising, deblurring, and color correction. Its purpose is to eliminate or mitigate various defects in the original image data, making it closer to the ideal input state, thereby providing high-quality input for subsequent image quality recognition.

[0071] After loading the corrected IoT image data into the edge computing node corresponding to the surveillance camera, the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data is obtained.

[0072] Specifically, in industrial IoT scenarios, surveillance cameras collect a batch of IoT image data. Before loading the IoT image data onto edge computing nodes, the IoT image data undergoes quality assessment processing. For example, the signal-to-noise ratio (SNR) and structural similarity index (SSIM) of the image can be used as evaluation indicators to obtain a data quality assessment coefficient. If the data quality assessment coefficient indicates that the image has obvious noise and blurriness, based on the coefficient, a preset denoising algorithm (such as non-local mean denoising) and deblurring algorithm (such as Wiener filtering) are used to perform data correction processing on the image, generating corrected IoT image data. Subsequently, the optimized image data is transmitted to the edge computing nodes. After receiving the high-quality image data, the edge computing nodes perform image quality recognition processing, such as using deep learning models to determine whether the image is clear and whether there are any anomalies, and output a high-confidence quality recognition result. For example, if the corrected image is identified as clear and without anomalies, the confidence level may be as high as 98%, far higher than the 70% confidence level that may occur when the image is not corrected, thus ensuring the accuracy of subsequent fault diagnosis.

[0073] This invention effectively addresses the problem of unreliable quality identification confidence levels caused by poor quality of raw IoT image data by introducing quality assessment and data correction processes. Specifically, the IoT image data is first subjected to quality assessment processing, which objectively quantifies the image quality and generates a data quality assessment coefficient. This coefficient serves as a guide, enabling targeted data correction processing. Through data correction, defects such as noise, blur, and uneven brightness in the image data are effectively eliminated or mitigated, resulting in higher-quality corrected IoT image data. When this high-quality image data is loaded onto edge computing nodes for image quality identification, the edge computing nodes can analyze the data based on clearer and more accurate input data, thereby obtaining more precise and reliable quality identification confidence levels. This ensures the accuracy of subsequent fault diagnosis and maintenance work order generation, avoiding a chain of errors caused by data source quality issues.

[0074] In some embodiments of this application described above, the step of performing quality assessment processing on the IoT image data to obtain the data quality assessment coefficient of the IoT image data includes:

[0075] The IoT image data undergoes quality assessment processing to obtain initial evaluation coefficients. These initial evaluation coefficients are quantitative indicators obtained after a preliminary quality assessment of the original IoT image data, reflecting its current quality status. This assessment can be comprehensively calculated based on multiple dimensions, including image sharpness, noise level, brightness, and contrast.

[0076] The initial evaluation coefficients are compared using evaluation coefficient thresholds to obtain coefficient comparison results. Specifically, the evaluation coefficient thresholds are a pre-set set of standard values ​​used to determine whether the initial evaluation coefficients meet the expected quality requirements. For example, a minimum passing score or a passing range can be set.

[0077] If the coefficient comparison result indicates that the initial evaluation coefficient is unqualified, an alarm is issued, and the IoT image data is reacquired. The reacquired IoT image data is then subjected to quality assessment to obtain a new initial evaluation coefficient. This process continues, comparing the reacquired initial evaluation coefficient with the evaluation coefficient threshold until a qualified coefficient comparison result is obtained. Specifically, the coefficient comparison result is a judgment derived by comparing the initial evaluation coefficient with the evaluation coefficient threshold, indicating whether the current image data quality is qualified. When the coefficient comparison result indicates that the initial evaluation coefficient is unqualified, an alarm is issued, for example, through audible and visual signals and message notifications, to inform relevant personnel or the system. Simultaneously, the operation of reacquiring IoT image data is triggered to ensure the data quality of subsequent processing. The reacquired IoT image data undergoes another quality assessment to obtain a new initial evaluation coefficient, which is then compared with the evaluation coefficient threshold again until a qualified coefficient comparison result is obtained.

[0078] If the coefficient comparison result indicates that the initial evaluation coefficient is qualified, then the initial evaluation coefficient shall be used as the data quality evaluation coefficient.

[0079] Specifically, in an Industrial Internet of Things (IIoT) scenario, a surveillance camera acquires a batch of IoT image data. First, the image data undergoes quality assessment to obtain an initial assessment coefficient, for example, 0.6. This initial assessment coefficient of 0.6 is then compared to a preset assessment coefficient threshold of 0.8. Since 0.6 is less than 0.8, the comparison result indicates that the initial assessment coefficient is unqualified. An alarm is then issued, for example, displaying a notification on the operation and maintenance management platform indicating unqualified image quality and requesting the camera to be checked or re-acquired, and driving the surveillance camera to re-acquire a batch of IoT image data. The re-acquired image data is then subjected to quality assessment again, yielding a new initial assessment coefficient, for example, 0.9. This 0.9 is again compared to the assessment coefficient threshold of 0.8. This time, 0.9 is greater than 0.8, and the coefficient comparison result indicates that the initial assessment coefficient is qualified. Therefore, this assessment coefficient of 0.9 is determined as the final data quality assessment coefficient for subsequent data correction processing. In this way, even if the initially acquired image quality is poor, a qualified data quality assessment coefficient can be obtained through iterative assessment and re-acquisition, ensuring the effectiveness of subsequent processing.

[0080] This invention effectively addresses the problem of potentially inaccurate or substandard initial quality assessments in IoT image data processing by introducing an evaluation coefficient threshold and a cyclical evaluation mechanism. Specifically, the IoT image data is first subjected to a preliminary quality assessment to obtain initial evaluation coefficients. These initial evaluation coefficients are then compared with a preset evaluation coefficient threshold. If the initial evaluation coefficient is substandard, an alarm is issued, triggering a re-acquisition of the IoT image data. This ensures that only IoT image data that has undergone multiple evaluations and meets the passing standard is used for subsequent data correction and image quality recognition processing. Due to the iterative and filtering process, the final data quality evaluation coefficients used for subsequent processing have higher reliability and accuracy, thereby avoiding erroneous decisions or ineffective processing caused by low-quality data evaluations.

[0081] In some embodiments of this application described above, the step of using the data quality assessment coefficient to perform data correction processing on IoT image data to obtain corrected IoT image data includes:

[0082] Based on the data quality assessment coefficients, the image correction areas of the IoT image data are determined. This step refers to identifying specific areas in the image that have quality defects or need improvement based on the data quality assessment coefficients obtained from the quality assessment processing of the IoT image data. For example, the data quality assessment coefficients may indicate that a certain local area of ​​the image has problems such as blurring, noise, abnormal brightness, or color distortion. In this case, the local area is identified as the image correction area. The purpose is to focus subsequent correction processing on the parts that truly need improvement, avoiding indiscriminate processing of the entire image.

[0083] Using a preset correction model, a correction scheme is matched to the image correction area to obtain a data correction scheme. This step refers to calling a pre-established correction model library after determining the image correction area. This preset correction model stores various correction algorithms or strategies for different image quality problems. By analyzing the specific type and degree of quality problem in the image correction area, the preset correction model can intelligently match the most suitable correction scheme for the current problem. For example, if the image correction area is identified as having Gaussian noise, the preset correction model may match a denoising filter scheme; if there is local overexposure, it may match a local brightness adjustment scheme. This ensures the targetedness and effectiveness of the correction.

[0084] The data correction scheme is used to perform data correction processing on IoT image data to obtain corrected IoT image data. This step refers to applying the matched data correction scheme to the identified image correction area in the IoT image data. This data correction scheme will guide specific image processing operations, such as performing noise reduction, sharpening, color balancing, and brightness adjustment, thereby eliminating or reducing image quality defects and obtaining corrected IoT image data of higher quality.

[0085] Specifically, in an Industrial Internet of Things (IIoT) scenario, a surveillance camera captures a frame of image data. After quality assessment, the data quality evaluation coefficient indicates significant blurring in the upper left corner of the image, while other areas are of good quality. Based on this data quality evaluation coefficient, the upper left corner is identified as the image correction area. A preset correction model is used to analyze this blurred area and match it with an image sharpening algorithm as the data correction scheme. Subsequently, this sharpening algorithm is applied only to the upper left corner, performing localized sharpening on that area while leaving the rest of the image unchanged. After processing, the corrected IoT image data is obtained, where the blurring problem in the upper left corner is effectively resolved without negatively impacting other high-quality areas of the image.

[0086] This invention effectively addresses the potential blindness and inefficiency of traditional data correction processes by introducing a mechanism for determining image correction regions and matching preset correction models. Specifically, firstly, data quality assessment coefficients are used to accurately identify specific regions in IoT image data that require correction, rather than processing the entire image indiscriminately. This regionalized processing allows subsequent correction resources to be concentrated on key areas. Secondly, a preset correction model intelligently matches correction schemes to the identified image correction regions, ensuring that the adopted correction method closely matches the actual image quality problem type. For example, a denoising scheme is matched for noise problems, and a sharpening scheme is matched for blur problems, thus avoiding potential side effects from inappropriate correction operations. Therefore, this invention enables accurate and efficient correction of IoT image data, providing higher-quality input data for subsequent image quality recognition processing.

[0087] In some embodiments of this application described above, the step of matching a correction scheme to the image correction region using a preset correction model to obtain a data correction scheme includes:

[0088] Using a preset correction model, correction schemes are matched to the image correction region to obtain all candidate correction schemes. This step refers to, after determining the image correction region based on the data quality evaluation coefficient, generating a series of potential correction schemes that meet the initial matching conditions by calling the preset correction model, tailored to the characteristics of the image correction region. Candidate correction schemes may include different types of correction algorithms, the same algorithm with different parameter configurations, or dedicated correction strategies for different types of image distortion (such as blur, noise, and uneven brightness). The purpose is to provide a comprehensive and multi-dimensional set of correction schemes for subsequent refined selection.

[0089] All candidate correction schemes are screened to obtain a data correction scheme. This step involves evaluating and comparing these schemes based on preset screening rules or optimization objectives after obtaining all candidate correction schemes, thereby selecting the unique or optimal data correction scheme that best suits the current IoT image data and image correction area. Screening rules can be based on various factors, such as quantitative indicators of correction effect (e.g., image sharpness, signal-to-noise ratio, and color reproduction), computational resource consumption, real-time requirements, and compatibility with subsequent image quality recognition processing. The aim is to ensure that the finally selected data correction scheme can maximize image quality and meet the needs of practical applications.

[0090] Specifically, in industrial IoT scenarios, IoT image data captured by surveillance cameras needs correction due to uneven lighting, slight blurring, or the presence of minor noise. After determining the image correction area based on data quality assessment coefficients, a preset correction model is used to match correction schemes for that area. At this point, the preset correction model does not directly output a fixed correction scheme, but rather generates a series of candidate correction schemes based on the characteristics of the image correction area. For example, candidate scheme A might be an algorithm based on local contrast enhancement, suitable for uneven lighting; candidate scheme B might be an algorithm based on wavelet denoising, suitable for noise removal; candidate scheme C might be an algorithm based on deconvolution, suitable for slight blurring. Furthermore, each scheme can have different parameter configurations, such as denoising intensity and contrast enhancement coefficients, forming more sub-candidate schemes. Subsequently, all candidate correction schemes are screened. For example, an image quality assessment module can be introduced to simulate or test the expected effect of each candidate scheme on the image correction area, and calculate the corresponding image quality indicators, such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), or recognition accuracy in specific application scenarios. Furthermore, the computational complexity of each scheme can be considered to meet the real-time requirements of edge computing nodes. For example, if the screening results show that candidate scheme A performs best in improving local contrast with moderate computational resource consumption, while candidate scheme B has good denoising effect but high computational cost, then candidate scheme A will ultimately be selected as the data correction scheme. This ensures that the selected data correction scheme is optimal for the current image problem and best meets the system performance requirements.

[0091] This invention addresses the problem of incomplete or suboptimal scheme selection that may arise from direct matching by first generating all candidate correction schemes and then performing a screening process. When the characteristics of the image correction region are complex or multiple possible correction methods exist, the preset correction model can provide a set of potential solutions. Subsequently, by introducing a screening mechanism, these candidate schemes can be referenced one by one according to more refined evaluation criteria, thereby avoiding the limitations that may arise from blindly selecting a single scheme. This allows the final data correction scheme to better adapt to the specific circumstances of the image data, thus improving the targeting and effectiveness of data correction.

[0092] In some embodiments of this application described above, the step of receiving IoT image data collected by a monitoring camera in an industrial IoT scenario includes:

[0093] This document describes how to acquire image acquisition parameters from surveillance cameras in an Industrial Internet of Things (IIoT) scenario. These parameters include acquisition time, acquisition accuracy, and acquisition area. The image acquisition parameters are crucial information describing the image acquisition process. For example, acquisition time indicates whether the image was acquired during the day or night, acquisition accuracy reflects the image's resolution and sharpness, and the acquisition area defines the physical range of image capture. These parameters provide important contextual information for subsequent image analysis and data determination.

[0094] Based on the image acquisition parameters, raw IoT image data acquired by a monitoring camera in an industrial IoT scenario is received. This raw IoT image data is an initial image stream or image frame without any processing.

[0095] Image analysis is performed on the raw IoT image data to obtain a quality score for each frame. This step is used to evaluate the intrinsic quality of each frame, for example, by calculating indicators such as image sharpness, contrast, noise level, and brightness uniformity, thereby obtaining a quality score for each frame. This quality score is an objective basis for quantifying image quality; a higher score generally indicates better image quality.

[0096] IoT image data is determined based on the image acquisition parameters and the quality score of each frame. Specifically, the raw data is filtered, optimized, or reconstructed based on the quality score. For example, image frames with too low a quality score can be removed, or images with low quality scores but still valuable can be enhanced to form high-quality IoT image data for subsequent processing.

[0097] Specifically, on an industrial production line, a surveillance camera is used to monitor the operating status of equipment in real time. Due to fluctuations in workshop lighting and slight vibrations of the equipment, the raw image data captured by the camera may contain some blurry or unevenly bright frames. First, the image acquisition parameters of the surveillance camera are obtained, for example, the acquisition time is 10:00-10:10 AM, the acquisition resolution is 1080P, and the acquisition area is the operating area of ​​equipment A. Next, the raw IoT image data stream captured by the camera during this time period is received based on these parameters. Then, image analysis is performed on each frame in the raw data stream. For example, a sharpness assessment algorithm based on Fourier transform can be used to calculate the sharpness score of each frame; simultaneously, histogram equalization analysis is used to analyze the brightness distribution of the image and calculate the brightness uniformity score. These scores are combined to obtain the quality score of each frame. If the analysis results show that the image frames between 10:05-10:07 AM generally have lower quality scores, indicating blurriness or brightness issues, then the final IoT image data is determined based on the image acquisition parameters and the quality scores of each frame. Specifically, a quality score threshold can be set, and image frames below this threshold are marked as low-quality frames. Low-quality frames can be either discarded or processed using image enhancement algorithms (such as deblurring and brightness adjustment) to improve their quality. Image frames with acceptable quality scores are directly included in the final IoT image data. This ensures that the IoT image data transmitted to edge computing nodes for subsequent processing is filtered and optimized, thereby avoiding interference with subsequent fault identification due to raw data quality issues.

[0098] This invention first acquires image acquisition parameters to understand the image generation background and potential quality-influencing factors. Then, it performs image analysis on the raw IoT image data to obtain a quality score for each frame, objectively quantifying the quality level of each frame. These quality scores, combined with the image acquisition parameters, intelligently determine the final IoT image data. For example, high-quality image frames can be selected based on the quality score, or low-quality frames can be processed specifically, ensuring that the data input to the edge computing node for subsequent image quality recognition processing is optimized, avoiding misjudgments or inefficiencies caused by raw data quality issues.

[0099] In some embodiments of this application described above, the step of determining IoT image data based on the image acquisition parameters and the quality score of each frame includes:

[0100] The image acquisition parameters are adjusted using the quality score of each frame to obtain the adjusted image acquisition parameters. The quality score of each frame is a quantitative assessment of the quality of the acquired image data, reflecting multiple dimensions such as image sharpness, brightness, contrast, and noise level. Image acquisition parameters refer to the configuration used by the surveillance camera when acquiring images, such as acquisition time, acquisition accuracy, and acquisition area. By adjusting these parameters, the camera's acquisition settings can be dynamically optimized based on the actual quality performance of the current image. For example, when the quality score is low, it indicates the need to increase acquisition accuracy or extend exposure time; when the quality score is too high and resource consumption is high, the accuracy can be appropriately reduced to save resources.

[0101] The adjusted image acquisition parameters are used to drive surveillance cameras in industrial IoT scenarios to acquire data, resulting in IoT image data. Specifically, the parameters adjusted by quality score feedback are applied to the surveillance cameras in real-time or near real-time, enabling them to perform subsequent data acquisitions according to the new, more optimized parameters. This ensures that the subsequently acquired IoT image data has higher quality and stronger usability, thus providing more reliable input for subsequent image quality identification and fault diagnosis.

[0102] Specifically, in an Industrial Internet of Things (IIoT) scenario, a surveillance camera is deployed to monitor product quality on a production line. Initially, the camera may use a set of preset image acquisition parameters for data collection. However, due to changes in lighting conditions in the production workshop (e.g., day versus night, cloudy versus sunny days), or the gradual accumulation of dust on the camera lens surface, the raw IoT image data may exhibit problems such as blurriness, overexposure, or underexposure, resulting in consistently low quality scores for each frame obtained through image analysis. Using these low quality scores, the current image acquisition parameters are adjusted. For example, if the quality score indicates underexposure, the exposure time parameter might be adjusted to extend it appropriately; if the quality score indicates blurriness, the focal length parameter might be adjusted, or it might be suggested to improve the acquisition accuracy. After these adjustments, a new set of optimized image acquisition parameters is obtained. Subsequently, these adjusted image acquisition parameters are immediately sent and applied to the surveillance camera, driving the camera to acquire data according to the new parameters. As a result, the camera can capture IoT image data with higher clarity, more accurate exposure, and better quality requirements, thereby ensuring that subsequent image quality recognition and processing can be based on high-quality data, effectively improving the accuracy and timeliness of fault diagnosis.

[0103] This invention effectively addresses the problem of poor data quality caused by image acquisition parameters potentially being unsuitable for the actual environment by introducing a feedback mechanism based on image quality scores. Specifically, after receiving raw IoT image data and performing image analysis to obtain a quality score for each frame, this score is no longer merely used for evaluation but is actively used to adjust the image acquisition parameters. This adjustment process allows the image acquisition parameters to be dynamically optimized based on the actual quality of the acquired images, thus avoiding image quality fluctuations or persistently low quality issues caused by fixed parameters. Subsequently, the adjusted parameters are used to drive the surveillance camera to acquire data, ensuring that the subsequently acquired IoT image data is optimized and better meets the needs of subsequent image quality recognition and processing.

[0104] In some embodiments of this application described above, the step of adjusting the image acquisition parameters using the quality score of each frame to obtain the adjusted image acquisition parameters includes:

[0105] Based on the quality score of each frame and the preset parameter adjustment model, the adjustment direction and magnitude of the image acquisition parameters are determined. Specifically, the quality score of each frame is a quantitative indicator obtained by comprehensively evaluating multiple dimensions such as sharpness, brightness, contrast, and noise of a single frame; its value reflects the overall quality level of the image. The preset parameter adjustment model refers to a pre-established intelligent model to guide the adjustment of image acquisition parameters, which can be built based on technologies such as machine learning, expert systems, or rule engines. This model can analyze the current image quality problems based on the quality score of each input frame and output targeted parameter adjustment suggestions. The adjustment direction of the image acquisition parameters refers to whether a specific parameter (such as exposure time, aperture, ISO, and focal length) should be increased or decreased to improve image quality. The adjustment magnitude refers to the specific amount by which the parameter needs to be increased or decreased. For example, if the image is too dark, the adjustment direction might be to increase the exposure time, and the adjustment magnitude might be an increase of 0.1 seconds; if the image is blurry, the adjustment direction might be to adjust the focal length, and the adjustment magnitude might be a fine adjustment of 0.05 millimeters.

[0106] By adjusting the direction and magnitude of the image acquisition parameters, the image acquisition parameters are adjusted to obtain the adjusted image acquisition parameters.

[0107] Specifically, in an Industrial Internet of Things (IIoT) scenario, a surveillance camera is acquiring image data. First, a frame of the image is acquired with a quality score of 0.6 (out of 1.0, indicating average image quality). This quality score is input into a preset parameter adjustment model. This model is based on a deep learning-trained neural network. Its inputs are the image quality score and the current image acquisition parameters (e.g., exposure time 0.05 seconds, aperture F / 2.8, ISO 400), and its output is adjustment suggestions. After analysis, the model determines that the current image is slightly blurry and insufficiently bright. Therefore, the model determines the direction of image acquisition parameter adjustment as follows: increase exposure time, decrease aperture (i.e., increase aperture value), and increase ISO; the adjustment magnitude is: increase exposure time by 0.01 seconds, increase aperture value by 0.2, and increase ISO by 100. Subsequently, using the adjustment direction and magnitude, the exposure time is adjusted to 0.06 seconds, the aperture to F / 3.0, and the ISO to 500. The adjusted image acquisition parameters are used to drive the surveillance camera for the next data acquisition, which is expected to obtain higher quality IoT image data.

[0108] This invention addresses the lack of refined guidance in parameter adjustment by introducing a preset parameter adjustment model and determining the adjustment direction and magnitude of image acquisition parameters based on the quality score of each image frame. Specifically, when the quality score of each image frame is received, it is input into the preset parameter adjustment model. This model uses internal logic or algorithms to perform in-depth analysis of the quality score, identifying the specific reasons for poor image quality (e.g., overexposure, underexposure, blurring, or excessive noise). Based on the identification results, the model can accurately calculate targeted adjustment schemes, determining which image acquisition parameters need adjustment, in which direction (increase or decrease) each parameter should be adjusted, and by what magnitude. This transforms image acquisition parameter adjustment from simple trial and error or empirical judgment into precise and efficient optimization based on the actual image quality, ensuring higher quality IoT image data acquired subsequently.

[0109] In some embodiments of this application described above, the step of comparing the confidence level of the quality identification with a preset confidence level to obtain a comparison value of the confidence level includes:

[0110] The confidence level of the quality identification is compared with a preset confidence level to obtain an initial comparison value of the confidence level. Specifically, the confidence level of the quality identification refers to the numerical value used by the edge computing node to evaluate the reliability of the image quality after performing image quality identification processing on the IoT image data. The preset confidence level is a pre-set benchmark value or threshold used to determine whether the image quality meets the requirements. Comparing the confidence level of the quality identification with the preset confidence level involves calculating the difference, ratio, or mapping between the two through a certain functional relationship to obtain a preliminary evaluation result, i.e., the initial comparison value of the confidence level. This initial comparison value reflects the original gap between the current image quality and the preset standard.

[0111] The initial comparison value of the confidence level is corrected to obtain a new confidence level comparison value. This step refers to further refining and optimizing the original comparison result. The purpose of the correction is to eliminate or reduce any errors, noise, or systematic biases that may exist in the initial comparison value, making it more accurate and reliable. For example, the correction process may apply a smoothing filter algorithm to reduce random noise; perform bias compensation based on historical data or environmental parameters; or transform the initial comparison value through a nonlinear mapping function to better meet the actual fault diagnosis requirements. Through the correction process, a refined and optimized confidence level comparison value can be obtained, which will serve as the basis for subsequent determination of image fault information.

[0112] Specifically, in an Industrial Internet of Things (IIoT) scenario, image data collected by surveillance cameras, after being processed by edge computing nodes, yields a quality identification confidence level of 0.75. The preset confidence threshold is 0.80. First, the quality identification confidence level of 0.75 is compared with the preset confidence level of 0.80 to obtain an initial comparison value. For example, the difference between the two is calculated, resulting in an initial comparison value of -0.05. Further, this initial comparison value of -0.05 is corrected. Specifically, a correction model trained on historical data can be used. This model may consider factors such as current ambient lighting and camera usage time to adjust the original comparison value. For example, if historical data shows that the confidence level is generally low (0.02) under current lighting conditions, the correction model will correct the initial comparison value of -0.05 to -0.03. Alternatively, a moving average filter can be used to weight the current initial comparison value with the initial comparison values ​​from previous time points to smooth data fluctuations and obtain the final confidence comparison value. Through correction processing, the final confidence comparison value (e.g., -0.03) will more accurately reflect the true gap between the image quality and the preset standard than the original -0.05, thus providing a more reliable basis for subsequent determination of image fault information.

[0113] This invention effectively addresses the inaccuracies and lack of robustness that can arise from direct comparison by introducing a correction process for the initial confidence comparison value. Specifically, after the edge computing node obtains the confidence level for image quality recognition, it first performs a preliminary comparison with a preset confidence level to obtain the initial confidence comparison value. While the initial value reflects basic quality differences, it may contain uncertainties. By correcting this initial comparison value—for example, by using statistical methods to remove outliers or by using a machine learning model to fine-tune the comparison value—the final confidence comparison value more accurately reflects the true quality of the image. This ensures the reliability of subsequent image fault information determination and avoids misjudgments or omissions caused by fluctuations in the initial comparison value.

[0114] For an IoT image data processing method based on edge computing based on any of the above embodiments, please refer to [link to relevant documentation]. Figure 2 The present invention also provides an IoT image data processing system based on edge computing, the system comprising a data receiving module 210, a first acquisition module 220, a confidence comparison module 230, a second acquisition module 240, an information determination module 250, and an information sending module 260.

[0115] The data receiving module 210 is used to receive IoT image data collected by monitoring cameras in industrial IoT scenarios.

[0116] The first acquisition module 220 is used to load the IoT image data into the edge computing node corresponding to the surveillance camera, and then acquire the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data.

[0117] The confidence comparison module 230 is used to compare the confidence of the quality identification with a preset confidence level to obtain a comparison value of the confidence level.

[0118] The second acquisition module 240 is used to transmit the comparison value of the confidence level to the edge computing node, and then acquire the image fault information determined by the edge computing node based on the comparison value of the confidence level.

[0119] The information determination module 250 is used to determine the maintenance work order information corresponding to the surveillance camera based on the image fault information and the preset maintenance model; wherein, the maintenance work order information includes the surveillance camera's number information, location information and maintenance suggestion information.

[0120] The information sending module 260 is used to send the maintenance work order information to the operation and maintenance management platform to complete the processing of IoT image data.

[0121] In this embodiment, raw image data is acquired by the data receiving module 210, and then the first acquisition module 220 performs preliminary quality identification processing on the edge computing node to obtain the confidence level of the quality identification. Next, the confidence level comparison module 230 compares this confidence level with a preset threshold to evaluate the reliability of the image quality. When a potential problem is found in the image quality, the second acquisition module 240 further determines the image fault information on the edge computing node. Based on the fault information and a preset maintenance model, the information determination module 250 can automatically generate detailed maintenance work order information, including the camera number, location, and specific maintenance suggestions. Finally, the information sending module 260 sends the work order information to the operation and maintenance management platform, thereby realizing automated monitoring, fault diagnosis, and maintenance scheduling of the IoT image data processing process, avoiding the problems of image quality degradation in the system, inaccurate fault identification, and delayed maintenance response, thus significantly improving the intelligence and reliability of industrial IoT image data processing.

[0122] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention specification.

Claims

1. A method for processing IoT image data based on edge computing, characterized in that, include: Receive IoT image data collected by surveillance cameras in industrial IoT scenarios; After loading the IoT image data into the edge computing node corresponding to the surveillance camera, the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data is obtained. The confidence level of the quality identification is compared with a preset confidence level to obtain a comparison value of the confidence level; After transmitting the comparison value of the confidence level to the edge computing node, the image fault information determined by the edge computing node based on the comparison value of the confidence level is obtained; Based on the image fault information and the preset maintenance model, the maintenance work order information corresponding to the surveillance camera is determined; wherein, the maintenance work order information includes the surveillance camera's number information, location information, and maintenance suggestion information; The maintenance work order information is sent to the operation and maintenance management platform to complete the processing of IoT image data.

2. The IoT image data processing method based on edge computing according to claim 1, characterized in that, After loading the IoT image data into the edge computing node corresponding to the surveillance camera, the step of obtaining the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data includes: The IoT image data is subjected to quality assessment processing to obtain the data quality assessment coefficient of the IoT image data; Using the aforementioned data quality evaluation coefficients, data correction processing is performed on the IoT image data to obtain corrected IoT image data; After loading the corrected IoT image data into the edge computing node corresponding to the surveillance camera, the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data is obtained.

3. The IoT image data processing method based on edge computing according to claim 2, characterized in that, The steps for performing quality assessment processing on the IoT image data to obtain the data quality assessment coefficient of the IoT image data include: The IoT image data is subjected to quality assessment processing to obtain the initial evaluation coefficients of the IoT image data; The initial evaluation coefficients are compared using the evaluation coefficient threshold to obtain the coefficient comparison results; If the coefficient comparison result indicates that the initial evaluation coefficient is unqualified, an alarm is issued, and IoT image data is reacquired. Then, the quality evaluation processing of the reacquired IoT image data is performed to obtain the initial evaluation coefficient for re-evaluation. The initial evaluation coefficient for re-evaluation is then compared with the evaluation coefficient threshold until a qualified coefficient comparison result is obtained. If the coefficient comparison result indicates that the initial evaluation coefficient is qualified, then the initial evaluation coefficient shall be used as the data quality evaluation coefficient.

4. The IoT image data processing method based on edge computing according to claim 2, characterized in that, The steps for performing data correction processing on IoT image data using the aforementioned data quality evaluation coefficients to obtain corrected IoT image data include: Based on the data quality evaluation coefficients, determine the image correction area for the IoT image data; Using a preset correction model, a correction scheme is matched to the image correction area to obtain a data correction scheme; Using the aforementioned data correction scheme, IoT image data is processed to obtain corrected IoT image data.

5. The IoT image data processing method based on edge computing according to claim 4, characterized in that, The steps of matching a correction scheme to the image correction region using a preset correction model to obtain a data correction scheme include: Using a preset correction model, the image correction region is matched with correction schemes to obtain all candidate correction schemes; All candidate correction schemes are screened to obtain a data correction scheme.

6. The IoT image data processing method based on edge computing according to claim 1, characterized in that, The steps for receiving IoT image data captured by monitoring cameras in an industrial IoT scenario include: Acquire image acquisition parameters from surveillance cameras in an industrial IoT scenario; wherein, the image acquisition parameters include acquisition time, acquisition accuracy, and acquisition area; Based on the image acquisition parameters, receive raw IoT image data acquired by a monitoring camera in an industrial IoT scenario; Image analysis is performed on the raw IoT image data to obtain the quality score of each frame. The IoT image data is determined based on the image acquisition parameters and the quality score of each frame.

7. The IoT image data processing method based on edge computing according to claim 6, characterized in that, The steps for determining IoT image data based on the image acquisition parameters and the quality score of each frame include: The image acquisition parameters are adjusted using the quality score of each frame to obtain the adjusted image acquisition parameters; Using the adjusted image acquisition parameters, the monitoring camera in the industrial IoT scenario is driven to acquire data and obtain IoT image data.

8. The IoT image data processing method based on edge computing according to claim 7, characterized in that, The steps for adjusting the image acquisition parameters using the quality score of each frame to obtain the adjusted image acquisition parameters include: Based on the quality score of each frame and the preset parameter adjustment model, the adjustment direction and adjustment range of the image acquisition parameters are determined; By adjusting the direction and magnitude of the image acquisition parameters, the image acquisition parameters are adjusted to obtain the adjusted image acquisition parameters.

9. The IoT image data processing method based on edge computing according to claim 1, characterized in that, The step of comparing the confidence level of the quality identification with a preset confidence level to obtain the comparison value of the confidence level includes: The confidence level of the quality identification is compared with a preset confidence level to obtain an initial comparison value of the confidence level; The initial comparison value of the confidence level is corrected to obtain the comparison value of the confidence level.

10. An IoT image data processing system based on edge computing, characterized in that, The system includes: The data receiving module is used to receive IoT image data collected by monitoring cameras in industrial IoT scenarios. The first acquisition module is used to load the IoT image data into the edge computing node corresponding to the surveillance camera, and then acquire the confidence level of the quality recognition obtained by the edge computing node through image quality recognition processing of the IoT image data. The confidence comparison module is used to compare the confidence level of the quality identification with a preset confidence level to obtain a comparison value of the confidence level. The second acquisition module is used to transmit the comparison value of the confidence level to the edge computing node and then acquire the image fault information determined by the edge computing node based on the comparison value of the confidence level. The information determination module is used to determine the maintenance work order information corresponding to the surveillance camera based on the image fault information and the preset maintenance model; wherein, the maintenance work order information includes the surveillance camera's number information, location information, and maintenance suggestion information; The information sending module is used to send the maintenance work order information to the operation and maintenance management platform to complete the processing of IoT image data.

Images