A multi-dimensional fusion-based installation quality evaluation and abnormal installation work order detection method and device

Abstract

The application discloses a kind of based on multi-dimensional fusion's installation and maintenance quality evaluation and abnormal installation and maintenance work order detection method and device, the method includes, obtaining the installation and maintenance image corresponding to broadband installation and maintenance work order;Based on the installation and maintenance image, the characteristic information of each quality detection dimension is extracted respectively, and the dimension quality value of each quality detection dimension is calculated based on the characteristic information;Based on each dimension quality value, and the dimension quality weight value corresponding to each dimension quality value, the installation and maintenance comprehensive quality value is calculated, and whether installation and maintenance quality is qualified based on the installation and maintenance comprehensive quality value;The image similarity and space-time consistency coefficient between several installation and maintenance images obtained in the preset time period are calculated, and whether there is abnormal installation and maintenance work order based on the image similarity and the space-time consistency coefficient is judged.The application improves the accuracy and timeliness of installation and maintenance quality evaluation, and simultaneously realizes the effective identification to abnormal installation and maintenance work order.

Description

Technical Field

[0001] This invention relates to the field of broadband installation and maintenance testing technology, and in particular to a method and device for quality assessment of installation and maintenance and detection of abnormal installation and maintenance work orders based on multi-dimensional fusion. Background Technology

[0002] In current installation and maintenance quality inspections, comprehensive construction quality inspections largely rely on backend-configured quality inspectors manually reviewing installation and maintenance images uploaded by installation and maintenance personnel. However, on the one hand, manual quality inspection is inefficient and depends on the subjective judgment of quality inspectors, making it susceptible to differences in experience and workload, resulting in low accuracy of inspection results. Furthermore, since manual quality inspection is usually conducted through periodic sampling, it lacks real-time monitoring capabilities, making it difficult to promptly detect quality problems in the installation and maintenance process. On the other hand, existing manual quality inspection methods typically only independently check images from a single or a small number of work orders, lacking the ability to conduct cross-sectional comparative analysis of historical images from multiple work orders. This makes it difficult to effectively identify abnormal installation and maintenance work orders involving forged images, duplicate image uploads, or image reuse across work orders. Summary of the Invention

[0003] The purpose of this invention is to propose a method and apparatus for evaluating installation and maintenance quality and detecting abnormal installation and maintenance work orders based on multi-dimensional fusion. By extracting multi-dimensional quality features from installation and maintenance images and calculating dimensional quality values, the overall quality of installation and maintenance is quantitatively evaluated by combining dimensional quality weights. Anomaly detection is performed on installation and maintenance images based on image similarity and spatiotemporal consistency coefficients. This achieves accurate and timely evaluation of installation and maintenance quality, improves the accuracy and real-time performance of installation and maintenance quality detection, and enhances the ability to identify abnormal installation and maintenance work orders such as forged images, repeatedly uploaded images, and images reused across work orders.

[0004] To achieve the above objectives, a first aspect of the present invention provides a method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection, the method comprising: Obtain the installation and maintenance images corresponding to the broadband installation and maintenance work orders; Based on the image, feature information of each quality detection dimension is extracted, and the dimensional quality value of each quality detection dimension is calculated based on the feature information. Based on the quality values ​​of each dimension and the corresponding dimension quality weight values, the overall installation and maintenance quality value is calculated, and the installation and maintenance quality is determined to be qualified based on the overall installation and maintenance quality value. Calculate the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images obtained within a preset time period, and determine whether there are any abnormal installation and maintenance work orders based on the image similarity and spatiotemporal consistency coefficient.

[0005] In this embodiment, multi-dimensional quality feature fusion is used to achieve quantitative evaluation of installation and maintenance quality, and image similarity and spatiotemporal consistency are combined to jointly determine abnormal work orders, thereby improving the accuracy and timeliness of quality detection and effectively identifying abnormal installation and maintenance work orders.

[0006] Furthermore, each of the quality detection dimensions includes several sub-quality detection dimensions. The step of extracting feature information for each quality detection dimension based on the assembled image, and calculating the dimensional quality value for each quality detection dimension based on the feature information, includes: From the installation and maintenance image, the feature information of each quality detection dimension is extracted, and based on the feature information, the sub-dimensional compliance confidence of each sub-quality detection dimension is obtained; Based on the compliance confidence scores of all the sub-dimensions of the i-th quality inspection dimension and the sub-dimension compliance weight values ​​corresponding to the compliance confidence scores of the sub-dimensions, calculate the quality review value of the i-th quality inspection dimension; Calculate the quality adjustment factor based on broadband installation and maintenance time and workload. The dimension quality value of the i-th quality inspection dimension is calculated based on the quality review value of the i-th quality inspection dimension, the compliance confidence level of the sub-dimension, the compliance weight value of the sub-dimension, the quality adjustment factor, and the decay function.

[0007] In this embodiment, by calculating the confidence level of multi-dimensional sub-quality detection dimensions and introducing a quality adjustment factor and a decay function, the installation and maintenance quality under different working conditions and time conditions can be comprehensively reflected, thereby improving the robustness and accuracy of the quality assessment results.

[0008] Further, the step of extracting feature information for each quality detection dimension from the installation and dimension image, and obtaining the sub-dimensional compliance confidence score for each sub-quality detection dimension based on the feature information, includes: The feature information of each sub-quality detection dimension in the image is extracted using a convolutional neural network, and the compliance confidence score of each sub-quality detection dimension is output based on the feature information.

[0009] In this embodiment, image features of the sub-quality detection dimension are extracted by a convolutional neural network and confidence scores are output, thereby improving the accuracy and consistency of feature extraction.

[0010] Further, the step of calculating the overall installation and maintenance quality value based on each of the said dimension quality values ​​and the dimension quality weight values ​​corresponding to each of the said dimension quality values, and determining whether the installation and maintenance quality is qualified based on the overall installation and maintenance quality value, includes: The overall quality value of the equipment is obtained by weighted summing of the quality values ​​of each dimension and the corresponding dimension quality weight values. If the overall installation and maintenance quality value is less than the preset quality assessment value, the installation and maintenance quality is determined to be unqualified; otherwise, the installation and maintenance quality is determined to be qualified.

[0011] In this embodiment, by weighting and fusing the quality values ​​of each dimension and setting a unified quality assessment threshold, the standardized judgment of installation and maintenance quality is achieved, thereby improving the consistency of quality assessment.

[0012] Further, the image similarity includes a first image similarity and a second image similarity, and the spatiotemporal consistency coefficient includes a temporal consistency coefficient and a spatial consistency coefficient. Therefore, calculating the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images acquired within a preset time period, and determining whether there are any abnormal installation and maintenance work orders based on the image similarity and the spatiotemporal consistency coefficient, includes: The first image similarity between the assembled images is calculated using the average hash algorithm; When the similarity of the first image is greater than or equal to the preset image similarity threshold, the corresponding installation and maintenance image is determined to be an abnormal image; when the similarity of the first image is less than the preset image similarity threshold, the corresponding installation and maintenance image is determined to be a normal image, and the corresponding installation and maintenance image is added to the first installation and maintenance image set. Calculate the second image similarity among the dimensional images in the first dimensional image set; Based on the temporal correlation parameters of the installation and maintenance images in the first installation and maintenance image set, the temporal consistency coefficient among the installation and maintenance images in the first installation and maintenance image set is calculated; Based on the spatial correlation parameters of the installation and maintenance images in the first installation and maintenance image set, the spatial consistency coefficient among the installation and maintenance images in the first installation and maintenance image set is calculated; based on the first image similarity, the second image similarity, the temporal consistency coefficient, and the spatial consistency coefficient, the final similarity among the installation and maintenance images in the first installation and maintenance image set is calculated. When the final similarity is greater than the first preset similarity threshold, the corresponding installation and maintenance image is determined to be abnormal; when the final similarity is less than the first preset similarity threshold but greater than the second preset similarity threshold, the corresponding installation and maintenance image is determined to be awaiting manual review; when the final similarity is less than the second preset similarity threshold, the corresponding installation and maintenance image is determined to be normal.

[0013] In this embodiment, a joint judgment mechanism based on first image similarity, second image similarity, and temporal and spatial consistency coefficients is constructed to achieve hierarchical recognition of abnormal installation and maintenance images, thereby improving the accuracy of abnormal installation and maintenance work order detection.

[0014] Further, the step of calculating the first image similarity between the stacked-dimensional images using the average hash algorithm includes: Each of the installation and maintenance images is scaled to a preset size, and the scaled installation and maintenance images are converted to grayscale to obtain the corresponding grayscale images. Calculate the average gray value of all pixels in each of the grayscale images; The grayscale value of each pixel in each grayscale image is compared with the corresponding average grayscale value. When the grayscale value of the target pixel is greater than the corresponding average grayscale value, the hash bit corresponding to the target pixel is assigned a first preset value; when the grayscale value of the target pixel is less than or equal to the corresponding average grayscale value, the hash bit corresponding to the target pixel is assigned a second preset value, so as to generate the image hash value corresponding to each of the two images. Calculate the Hamming distance between the image hash values ​​of each of the assembled images; Based on the Hamming distance, the first image similarity between each of the assembled images is calculated.

[0015] In this embodiment, the average hash algorithm is used to quickly calculate the similarity of images, thereby achieving efficient screening of highly similar images, reducing the computational complexity of subsequent anomaly screening, and improving the efficiency of identifying abnormal installation and maintenance work orders.

[0016] Further, calculating the second image similarity between the dimensional images in the first dimensional image set includes: Each dimensional image in the first dimensional image set is processed by grayscale and intervalization to obtain the grayscale interval feature vector of each dimensional image, and the histogram similarity between each dimensional image is calculated based on the grayscale interval feature vector. The edge features of each dimensional image in the first dimensional image set are extracted to obtain the corresponding edge density features, and the edge feature similarity between each dimensional image is calculated based on the edge density features. Obtain the aspect ratio and file size of each dimensional image in the first dimensional image set, and calculate the metadata similarity between each dimensional image based on the difference between the aspect ratio and file size of each dimensional image; The second image similarity is calculated based on the histogram similarity, the edge feature similarity, and the metadata similarity.

[0017] In this embodiment, the second image similarity is calculated by fusing grayscale distribution features, edge structure features, and metadata features, thereby improving the accuracy and reliability of image similarity determination.

[0018] Furthermore, the time-related parameters include the installation and maintenance image capture time and the installation and maintenance work order creation time; the spatial-related parameters include the installation and maintenance image capture address and the installation and maintenance work order operation address. The time consistency coefficient and the spatial consistency coefficient are then calculated through the following steps: Based on the image capture time, the work order creation time, and a preset time threshold, the time consistency coefficient between the installation and maintenance images in the first set of installation and maintenance images is calculated. Based on the location where the installation and maintenance image was captured, the location of the installation and maintenance work order, and a preset spatial distance threshold, the spatial consistency coefficient between the installation and maintenance images in the first set of installation and maintenance images is calculated.

[0019] In this embodiment, by introducing time consistency coefficients and spatial consistency coefficients to constrain the images, abnormal installation and maintenance images that do not conform to the time or spatial logic of the work order can be effectively identified.

[0020] Furthermore, the method also includes: Based on the quality values ​​and weight values ​​of each dimension, calculate the quality improvement priority coefficient for each quality inspection dimension, and determine the rectification priority for each quality inspection dimension based on the quality improvement priority coefficient.

[0021] In this embodiment, by calculating the priority coefficient for installation and maintenance quality improvement and determining the priority of rectification, the system can sort and manage installation and maintenance quality issues, thereby improving the efficiency and targeting of rectification.

[0022] To achieve the above objectives, a second aspect of the present invention further provides a device for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection, used to implement the method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection described in any of the first aspects above. The device includes: The data acquisition module is used to acquire installation and maintenance images corresponding to broadband installation and maintenance work orders; The dimension quality value calculation module is used to extract feature information of each quality detection dimension based on the assembled dimension image, and calculate the dimension quality value of each quality detection dimension based on the feature information. The installation and maintenance quality assessment module is used to calculate the comprehensive installation and maintenance quality value based on the quality values ​​of each dimension and the dimension quality weight values ​​corresponding to each dimension quality value, and to determine whether the installation and maintenance quality is qualified based on the comprehensive installation and maintenance quality value. The abnormal installation and maintenance work order determination module is used to calculate the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images obtained within a preset time period, and to determine whether there is an abnormal installation and maintenance work order based on the image similarity and the spatiotemporal consistency coefficient. Attached Figure Description

[0023] Figure 1 This is a flowchart of a preferred embodiment of a method for evaluating installation and maintenance quality and detecting abnormal installation and maintenance work orders based on multi-dimensional fusion, provided in the first aspect of the present invention. Figure 2 This is a structural block diagram of a preferred embodiment of a multi-dimensional fusion-based device for evaluating installation and maintenance quality and detecting abnormal installation and maintenance work orders, provided in the second aspect of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It should be noted that the data involved in this invention (including but not limited to data used for analysis, data stored, data displayed, etc.) are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0026] In this embodiment of the invention, the words "exemplarily" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplarily" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of the words "exemplarily" or "for example" is intended to present the relevant concepts in a specific manner.

[0027] In this invention description, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In this invention description, unless otherwise stated, "a plurality of" means two or more. In this invention description, the term "comprising" and its variations are open-ended, meaning "including but not limited to." The term "based on" means "at least partially based on." The term "according to" means "at least partially according to." The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments."

[0028] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0029] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0030] The first aspect of this invention provides a method for evaluating installation and maintenance quality and detecting abnormal installation and maintenance work orders based on multi-dimensional fusion. See [link to relevant documentation]. Figure 1 The diagram shown is a flowchart of a preferred embodiment of a method for evaluating installation and maintenance quality and detecting abnormal installation and maintenance work orders based on multi-dimensional fusion, provided by the first aspect of the present invention. The method includes steps S1 to S4, as follows: Step S1: Obtain the installation and maintenance image corresponding to the broadband installation and maintenance work order; In one example, after completing broadband installation and maintenance, the installation and maintenance personnel use an installation and maintenance terminal to take pictures of the construction site, obtaining at least one installation and maintenance image. Subsequently, the installation and maintenance terminal sends the installation and maintenance image to the installation and maintenance business system. After receiving the installation and maintenance image, the installation and maintenance business system associates the installation and maintenance image with the corresponding broadband installation and maintenance work order, and sends the associated image data to the background processing platform for subsequent installation and maintenance quality assessment and abnormal installation and maintenance work order detection.

[0031] It should be noted that the installation and maintenance terminal can be a mobile phone, tablet computer, dedicated installation and maintenance terminal or other device with image acquisition capabilities.

[0032] It should be noted that the installation image can be a single image, multiple images, a target frame from a series of images, or an image extracted from video data.

[0033] This embodiment provides a data foundation for subsequent installation and maintenance quality assessment and abnormal installation and maintenance work order detection.

[0034] Step S2: Based on the assembled image, extract the feature information of each quality detection dimension, and calculate the dimensional quality value of each quality detection dimension based on the feature information. In one example, after receiving the installation and maintenance image, the backend processing platform sends the installation and maintenance image to the quality detection module. The quality detection module analyzes the installation and maintenance image for each quality detection dimension, extracts the corresponding feature information, and outputs the detection results for each quality detection dimension. Subsequently, the dimension quality value calculation module receives the detection results for each quality detection dimension, obtains the dimension quality value corresponding to each quality detection dimension, and sends each dimension quality value to the comprehensive quality assessment module.

[0035] It should be noted that the quality inspection dimensions may include at least one of the following: cable laying, equipment fixing, labeling specifications, equipment identification information, environmental specifications, or other inspection dimensions that can characterize the standardization of construction, and dynamic expansion is supported.

[0036] It should be noted that the feature information can be image content features, target state features, text recognition features, or other feature information that can characterize the quality state.

[0037] In this embodiment, by extracting multi-dimensional features from the installation and maintenance images and calculating the dimensional quality values ​​of each dimension, construction quality problems that originally relied on human experience for judgment can be transformed into quantifiable and comparable detection results, thereby improving the objectivity and consistency of installation and maintenance quality assessment.

[0038] Step S3: Calculate the overall installation and maintenance quality value based on the quality values ​​of each dimension and the corresponding dimension quality weight values, and determine whether the installation and maintenance quality is qualified based on the overall installation and maintenance quality value. In one example, the comprehensive quality assessment module receives the dimensional quality values ​​of each quality detection dimension and obtains the dimensional quality weight values ​​corresponding to each dimensional quality value from the rule configuration module or parameter management module. Then, the comprehensive quality assessment module performs fusion processing based on each dimensional quality value and each dimensional quality weight value to obtain the comprehensive installation and maintenance quality value. Subsequently, the quality judgment module compares the comprehensive installation and maintenance quality value with preset quality judgment conditions and outputs a judgment result on whether the installation and maintenance quality is qualified.

[0039] It should be noted that when the judgment result is unqualified, the guidance information generation module can also generate corresponding rectification prompt information and feed the rectification prompt information back to the installation and maintenance terminal or installation and maintenance business system.

[0040] It should be noted that the dimensional quality weight values ​​can be pre-configured according to different business types, different construction scenarios or different quality standards, and can be automatically updated based on historical data.

[0041] It should be noted that the quality judgment conditions may include preset thresholds, graded intervals, or other rules used to determine whether the installation and maintenance quality meets the requirements.

[0042] In this embodiment, by introducing dimensional quality weight values, the results of multiple quality inspection dimensions are fused and calculated to output an accurate comprehensive installation and maintenance quality value in real time. At the same time, the comprehensive installation and maintenance quality is qualified, thereby achieving an effective assessment of construction quality.

[0043] Step S4: Calculate the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images obtained within a preset time period, and determine whether there are any abnormal installation and maintenance work orders based on the image similarity and spatiotemporal consistency coefficient.

[0044] In one example, the background processing platform retrieves installation and maintenance images and associated work order information corresponding to several archived work orders from the work order database within a preset time period, and sends the installation and maintenance images to the installation and maintenance work order anomaly detection module. The installation and maintenance work order anomaly detection module first performs image similarity analysis on the installation and maintenance images and filters out installation and maintenance images with high similarity. Then, it calculates the corresponding spatiotemporal consistency coefficient by combining the time information and spatial information of the corresponding work order. Finally, the judgment module outputs the abnormal installation and maintenance work order detection result based on the image similarity and the spatiotemporal consistency coefficient, and sends the detection result to the management system.

[0045] It should be noted that the preset time period can be a non-busy period after the work order is archived, or it can be any preset batch processing time period.

[0046] It should be noted that the image similarity can be calculated based on image content features, image structure features, metadata features, or a combination thereof.

[0047] It should be noted that the spatiotemporal consistency coefficient can be determined based on the installation and maintenance image capture time, work order creation time, installation and maintenance image capture location, and work order operation address.

[0048] It is understandable that the anomaly detection module adopts a layered detection architecture, first performing rapid screening, then performing fine comparison, and finally combining spatiotemporal information for rationality verification, so as to balance detection efficiency and detection accuracy.

[0049] In this embodiment, by combining image similarity analysis with spatiotemporal consistency verification, historical work order images can be compared horizontally to effectively identify abnormal installation and maintenance work orders, thereby improving the work order compliance detection capability.

[0050] In another preferred embodiment, each of the quality detection dimensions includes several sub-quality detection dimensions. The step of extracting feature information for each quality detection dimension based on the dimensional image, and calculating the dimensional quality value for each quality detection dimension based on the feature information, includes: From the installation and maintenance image, the feature information of each quality detection dimension is extracted, and based on the feature information, the sub-dimensional compliance confidence of each sub-quality detection dimension is obtained; Based on the compliance confidence scores of all the sub-dimensions of the i-th quality inspection dimension and the sub-dimension compliance weight values ​​corresponding to the compliance confidence scores of the sub-dimensions, calculate the quality review value of the i-th quality inspection dimension; Calculate the quality adjustment factor based on broadband installation and maintenance time and workload. The dimension quality value of the i-th quality inspection dimension is calculated based on the quality review value of the i-th quality inspection dimension, the compliance confidence level of the sub-dimension, the compliance weight value of the sub-dimension, the quality adjustment factor, and the decay function.

[0051] In one example, after receiving installation and maintenance images uploaded by the installation and maintenance terminal, the backend quality inspection server first processes the images according to preset quality inspection dimensions. For example, the quality inspection dimension of equipment fixation can be further broken down into four sub-quality inspection dimensions: screw tightness, bracket levelness, anti-fall-off measures, and grounding wire connection; the quality inspection dimension of cable laying can be further broken down into three sub-quality inspection dimensions: bending radius, cable straightness, and binding standardization; and the quality inspection dimension of label standardization can be further broken down into three sub-quality inspection dimensions: whether labels are missing, whether label text is clear, and whether label content is standardized. Subsequently, the system calls the corresponding sub-model for each sub-quality inspection dimension and outputs the sub-dimensional compliance confidence score for each sub-quality inspection dimension. Based on this, the system first calculates the... Weighted confidence mean of sub-dimensions of each quality inspection dimension:

[0052] in, Indicates the first The total number of sub-dimensions corresponding to each quality inspection dimension Indicates the first The first quality inspection dimension The weights of each sub-quality detection dimension, and satisfying:

[0053] Subsequently, the system determines the first threshold based on the multi-level threshold determination rules for key items. Quality verification values ​​for each quality inspection dimension :

[0054] in, Indicates the first The first-level threshold of key items in each quality inspection dimension Indicates the first Secondary thresholds for key items in each quality inspection dimension ( ), Represents the verification coefficient (default) ).

[0055] The primary and secondary thresholds can also be dynamically updated based on historical compliance rates.

[0056]

[0057] in, and These represent the initial baseline thresholds, Represents the iteration coefficient, Indicates the first 3 months The historical compliance rate of each quality inspection dimension, and:

[0058] in, Indicates the first three months The historical number of compliances across each quality inspection dimension Indicates the first three months Total number of historical inspections across all quality inspection dimensions.

[0059] Next, the system calculates the quality adjustment factor based on broadband installation and maintenance time and workload. :

[0060] in, Indicates the month of construction. This indicates the scene adjustment factor.

[0061] Finally, the system combines the sub-dimension compliance confidence score, sub-dimension compliance weight value, quality adjustment factor, and uncertainty decay function to calculate the first... Dimensional quality values ​​for each quality inspection dimension :

[0062] in, Indicates the first The first quality inspection dimension The confidence variance of each sub-quality testing dimension This represents the attenuation coefficient.

[0063] It should be noted that the number of sub-quality detection dimensions... It is not a fixed value, but can be dynamically expanded according to different business scenarios and different installation and maintenance standards.

[0064] It should be noted that the rights mentioned It can be pre-configured by the operator or iteratively updated based on historical quality inspection results.

[0065] It should be noted that the verification coefficients mentioned above... It can be used to reduce the contribution of intermediate states in the pending review period to the final score of sub-dimensions, so as to highlight the issues that need to be rectified first.

[0066] It should be noted that the attenuation function It is mainly used to reduce the interference of high uncertainty detection results on dimensional quality values.

[0067] In this embodiment, by calculating the confidence level of multi-dimensional sub-quality detection dimensions and introducing a threshold verification mechanism, a time factor, and an uncertainty decay mechanism, the dimensional quality values ​​can more accurately reflect the actual construction quality, thereby improving the robustness and accuracy of the quality assessment results.

[0068] In another preferred embodiment, the step of extracting the feature information of the quality inspection dimension from the installation and dimension image, and obtaining the sub-dimensional compliance confidence score of each of the sub-quality inspection dimensions based on the feature information, includes: The feature information of each sub-quality detection dimension in the image is extracted using a convolutional neural network, and the compliance confidence score of each sub-quality detection dimension is output based on the feature information.

[0069] In one example, after acquiring the installation and maintenance quality inspection module, the image is first preprocessed by unifying the size, normalizing the brightness, and suppressing noise. Then, the processed image is input into a convolutional neural network model. For equipment compliance inspection scenarios, the system can call a target detection network containing a convolutional feature extraction backbone to identify indicator light status, grounding wire connection status, installation location and orientation, and component integrity. For cable compliance inspection scenarios, the system can call a detection model based on a convolutional neural network to identify cable bending level, binding status, and wiring trajectory. For label compliance inspection scenarios, the system can call a combination of a convolutional feature extraction network and an OCR recognition module to identify label areas, label text, and key field matching results. For the first... For each quality inspection dimension, the system outputs the image recognition confidence score as follows:

[0070] in, Indicates that for the first A dedicated model trained for each quality detection dimension. This refers to on-site photos uploaded by installation and maintenance personnel.

[0071] Furthermore, for the first The first quality inspection dimension For each sub-dimensional quality inspection dimension, the system extracts local feature maps from the intermediate feature layers of the convolutional neural network and outputs the corresponding sub-dimensional compliance confidence score through the classification head or regression head. For example, in the equipment fixing dimension, the system can output the corresponding probability values ​​for whether the screws are tight, whether the bracket is horizontal, and whether the grounding wire is connected; in the cable laying dimension, the system can output the corresponding probability values ​​for whether the bending degree is compliant, whether there is obvious entanglement, and whether the binding specifications are met; in the label specification dimension, the system can output the corresponding probability values ​​for whether the label exists, whether the label text is clear, and whether the label content matches the rules.

[0072] It should be noted that the convolutional neural network can be a single-stage detection network, a two-stage detection network, or other deep learning networks that include convolutional feature extraction structures.

[0073] It should be noted that the installation and maintenance image can be a single image or multiple images corresponding to the same work order, and the detection results of multiple images can be fused in subsequent processing.

[0074] Understandably, different sub-quality detection dimensions can correspond to different detection heads or different task branches to improve the targeting of feature extraction.

[0075] Understandably, when the target area corresponding to a certain sub-quality detection dimension is small or severely occluded, the system can also improve the recognition effect based on image enhancement strategies.

[0076] It is understandable that although this embodiment uses a convolutional neural network as an example, it does not exclude the use of other deep learning feature extraction networks to output the corresponding confidence scores.

[0077] In this embodiment, a convolutional neural network is used to automatically extract various construction features from the installation and maintenance image and output the sub-dimension compliance confidence score, which reduces the reliance on manual rules and improves the consistency and accuracy of feature extraction.

[0078] In another preferred embodiment, the step of calculating the overall installation and maintenance quality value based on each of the dimensional quality values ​​and the dimensional quality weight values ​​corresponding to each of the dimensional quality values, and determining whether the installation and maintenance quality is qualified based on the overall installation and maintenance quality value, includes: The overall quality value of the equipment is obtained by weighted summing of the quality values ​​of each dimension and the corresponding dimension quality weight values. If the overall installation and maintenance quality value is less than the preset quality assessment value, the installation and maintenance quality is determined to be unqualified; otherwise, the installation and maintenance quality is determined to be qualified.

[0079] In one example, the background comprehensive quality assessment module obtains the dimensional quality values ​​for each quality detection dimension. Then, retrieve the corresponding dimension quality weight values ​​from the parameter configuration library. Then, the quality values ​​of each dimension are weighted and summed according to the comprehensive compliance calculation formula to obtain the comprehensive quality value of installation and maintenance. :

[0080] in, This represents the total number of quality inspection dimensions, and: .

[0081] For example, in a broadband installation and maintenance scenario, you can set up... These correspond to four quality inspection dimensions: equipment fixation, cable routing, label specifications, and equipment serial number (SN) code. If the configured weights are respectively... , , , The corresponding dimension quality values ​​are as follows: , , , Then, the comprehensive quality assessment module can calculate the comprehensive quality value of installation and maintenance based on the above formula. Subsequently, the quality assessment module will... Compared with the preset quality assessment value Compare and determine eligibility according to the following criteria. Is the output assembly and maintenance quality up to standard?

[0082] in, This indicates that the test is successful. This indicates that the value is unqualified. Preferably, the preset quality assessment value... It can be set to 0.85.

[0083] It should be noted that the aforementioned dimensional quality weight values It can be flexibly configured according to different business types. For example, in government and enterprise leased line business, the weight of equipment fixing and equipment identification dimensions can be increased, and in home broadband business, the weight of cable laying and labeling specifications dimensions can be increased.

[0084] It should be noted that the preset quality assessment value Different settings can be made for different regions, different construction standards, and different product types.

[0085] It is understandable that the aforementioned qualification determination conclusion It can be directly fed back to the installation and maintenance terminal to remind installation and maintenance personnel to continue rectification or execute the return order operation.

[0086] In this embodiment, a comprehensive quality value for installation and maintenance with overall representation significance is obtained by uniformly weighting and fusing the quality values ​​of each dimension, and standardized judgment is achieved based on a unified threshold, thereby improving the consistency and executability of the installation and maintenance quality assessment results.

[0087] In another preferred embodiment, the image similarity includes a first image similarity and a second image similarity, and the spatiotemporal consistency coefficient includes a temporal consistency coefficient and a spatial consistency coefficient. Then, calculating the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images acquired within a preset time period, and determining whether there are any abnormal installation and maintenance work orders based on the image similarity and the spatiotemporal consistency coefficient, includes: The first image similarity between the assembled images is calculated using the average hash algorithm; When the similarity of the first image is greater than or equal to the preset image similarity threshold, the corresponding installation and maintenance image is determined to be an abnormal image; when the similarity of the first image is less than the preset image similarity threshold, the corresponding installation and maintenance image is determined to be a normal image, and the corresponding installation and maintenance image is added to the first installation and maintenance image set. Calculate the second image similarity among the dimensional images in the first dimensional image set; Based on the temporal correlation parameters of the installation and maintenance images in the first installation and maintenance image set, the temporal consistency coefficient among the installation and maintenance images in the first installation and maintenance image set is calculated; Based on the spatial correlation parameters of the installation and maintenance images in the first installation and maintenance image set, the spatial consistency coefficient among the installation and maintenance images in the first installation and maintenance image set is calculated; based on the first image similarity, the second image similarity, the temporal consistency coefficient, and the spatial consistency coefficient, the final similarity among the installation and maintenance images in the first installation and maintenance image set is calculated. When the final similarity is greater than the first preset similarity threshold, the corresponding installation and maintenance image is determined to be abnormal; when the final similarity is less than the first preset similarity threshold but greater than the second preset similarity threshold, the corresponding installation and maintenance image is determined to be awaiting manual review; when the final similarity is less than the second preset similarity threshold, the corresponding installation and maintenance image is determined to be normal.

[0088] In one example, during off-peak hours after work orders are archived, the backend compliance inspection module batch-extracts installation and maintenance images corresponding to archived work orders of the day and historical work orders from the work order database, and constructs a set of image pairs to be compared. The system first performs a first-level detection on each image pair and calculates the first image similarity. .when When the image similarity is greater than or equal to a preset image similarity threshold (e.g., 0.9), the system can mark the corresponding work order for that image pair as a first-level anomaly candidate; when If the direct judgment threshold is not met, the system adds the image pair to the first set of images and proceeds to the second-level detection. Subsequently, the system calculates the second image similarity for the image pairs in the first set of images. Simultaneously, by combining the image capture time, work order creation time, image capture address, and work order job address, a time consistency coefficient is calculated. Spatial consistency coefficient Thus, the spatiotemporal verification coefficients are obtained:

[0089] In obtaining , and Then, the system calculates the final similarity score according to the comprehensive judgment model:

[0090] Then, the anomaly detection result is output according to the following judgment rules: when... When, it is judged as an abnormal image; when When, mark as pending manual review; when At that time, it was determined to be a normal image.

[0091] It should be noted that the first image similarity is mainly used for rapid screening, while the second image similarity is mainly used for fine comparison.

[0092] It should be noted that the temporal consistency coefficient and spatial consistency coefficient are used to constrain the rationality of the image from the business logic level, so as to avoid misjudgment caused by relying solely on visual features.

[0093] It should be noted that the preset time period can be any batch processing cycle set by day, week, or month.

[0094] It is understandable that the first set of images can be understood as the set of images that need to enter the deep feature comparison stage.

[0095] In this embodiment, by constructing a joint judgment mechanism based on first image similarity, second image similarity, and temporal and spatial consistency coefficients, it not only takes into account the efficiency of large-scale work order processing, but also improves the accuracy of identifying forged construction images, repeatedly uploaded images, and images reused across work orders.

[0096] In yet another preferred embodiment, the step of calculating the first image similarity between the stacked-dimensional images using an average hash algorithm includes: Each of the installation and maintenance images is scaled to a preset size, and the scaled installation and maintenance images are converted to grayscale to obtain the corresponding grayscale images. Calculate the average gray value of all pixels in each of the grayscale images; The grayscale value of each pixel in each grayscale image is compared with the corresponding average grayscale value. When the grayscale value of the target pixel is greater than the corresponding average grayscale value, the hash bit corresponding to the target pixel is assigned a first preset value; when the grayscale value of the target pixel is less than or equal to the corresponding average grayscale value, the hash bit corresponding to the target pixel is assigned a second preset value, so as to generate the image hash value corresponding to each of the two images. Calculate the Hamming distance between the image hash values ​​of each of the assembled images; Based on the Hamming distance, the first image similarity between each of the assembled images is calculated.

[0097] In one example, after receiving two images of the structure to be compared, the primary detection module first resizes each image to a uniform size. Pixels are processed and converted to grayscale to obtain a grayscale image. Then, the system calculates the average gray value of all pixels in the grayscale image. :

[0098] in, This indicates the pixel position in a grayscale image.

[0099] Subsequently, the system compares the grayscale value of each pixel with the average grayscale value. Compare and construct a 256-bit hash value Specifically, each hash bit can be determined according to the following rules:

[0100] After obtaining the hash values ​​of the two dimensional images respectively and Then, the system further calculates the Hamming distance between the two. The similarity of the first image is calculated using the following formula:

[0101] For example, given two suspected duplicate uploaded device installation images, the system generates two sets of 256-bit hash values ​​after uniform scaling and grayscale conversion. If the two differ by only 18 bits, then:

[0102] The result is greater than 0.9, so the system can directly determine that the image pair is highly similar.

[0103] It should be noted that the preset size is not limited to... In other implementations, it can also be set to other sizes suitable for rapid screening.

[0104] Understandably, the first-level detection module can The image is directly compared with a preset threshold, and only image pairs that are not directly excluded or directly identified are sent to the subsequent secondary detection.

[0105] In this embodiment, by uniformly scaling, grayscale, and hash encoding the installation and maintenance images, and calculating the first image similarity based on Hamming distance, a rapid initial screening of similarity for a large number of historical work order images is achieved, thereby effectively reducing the computational burden of subsequent fine comparison.

[0106] In yet another preferred embodiment, calculating the second image similarity among the dimensional images in the first dimensional image set includes: Each dimensional image in the first dimensional image set is processed by grayscale and intervalization to obtain the grayscale interval feature vector of each dimensional image, and the histogram similarity between each dimensional image is calculated based on the grayscale interval feature vector. The edge features of each dimensional image in the first dimensional image set are extracted to obtain the corresponding edge density features, and the edge feature similarity between each dimensional image is calculated based on the edge density features. Obtain the aspect ratio and file size of each dimensional image in the first dimensional image set, and calculate the metadata similarity between each dimensional image based on the difference between the aspect ratio and file size of each dimensional image; The second image similarity is calculated based on the histogram similarity, the edge feature similarity, and the metadata similarity.

[0107] In one example, the secondary detection module performs deep feature comparison on the pairs of images to be compared that have entered the first set of images. First, the system performs grayscale and interval processing on each image, dividing each image into 16 grayscale intervals, and calculates the pixel percentage within each grayscale interval to form a 16-dimensional grayscale interval feature vector. Among them, the first The pixel proportion characteristics of each grayscale range are:

[0108] in, Indicates the first The number of pixels in each grayscale range This represents the total number of pixels in the image. The system then calculates the histogram similarity between the two images using the following formula:

[0109] Next, the system uses the Sobel operator to extract edge features, calculating the edge density in the horizontal, vertical, and diagonal directions respectively, to form an edge feature vector. For example, the horizontal edge density features can be obtained separately. Vertical edge density features and edge density features in the diagonal direction After obtaining the edge feature vectors of the two images, the system calculates the edge feature similarity according to the following formula:

[0110] Furthermore, the system also extracts metadata features for each image, including the image aspect ratio. File size And other optional metadata information. Among them:

[0111] Based on this, metadata similarity can be expressed as:

[0112] in, and For weighting coefficients, preferably, , .

[0113] Finally, the system fuses histogram similarity, edge feature similarity, and metadata similarity to obtain the second image similarity:

[0114] For example, for two cropped and slightly compressed images of cable laying, although the first-level hash similarity did not reach the high similarity threshold, further calculations by the system revealed that their grayscale distributions were similar. High; edge traces are oriented in similar directions. It is also relatively high; at the same time, the differences in file size and aspect ratio are small. It remains at a high level. At this point, after weighted fusion using the above formula, It can reach a relatively high value.

[0115] It should be noted that the weights of histogram similarity, edge feature similarity, and metadata similarity are not fixed and can be adjusted based on business experience.

[0116] Understandably, the second image similarity is a supplement to the first image similarity; the former focuses on fine-grained comparison, while the latter focuses on rapid screening.

[0117] It is understood that the extraction method of the edge density features is not limited to the Sobel operator, and other edge detection methods can also be used.

[0118] Understandably, metadata features can also be extended to include information such as resolution, camera identification, and watermark attributes.

[0119] In this embodiment, by fusing grayscale range features, edge structure features, and metadata features to calculate the second image similarity, similar images under deformation scenarios such as intersection, cropping, compression, and rotation can be identified more accurately, thereby improving the accuracy of abnormal installation and maintenance image detection.

[0120] In another preferred embodiment, the time correlation parameters include the installation and maintenance image capture time and the installation and maintenance work order creation time, and the spatial correlation parameters include the installation and maintenance image capture address and the installation and maintenance work order operation address. The time consistency coefficient and the spatial consistency coefficient are then calculated through the following steps: Based on the image capture time, the work order creation time, and a preset time threshold, the time consistency coefficient between the installation and maintenance images in the first set of installation and maintenance images is calculated. Based on the location where the installation and maintenance image was captured, the location of the installation and maintenance work order, and a preset spatial distance threshold, the spatial consistency coefficient between the installation and maintenance images in the first set of installation and maintenance images is calculated.

[0121] In one example, after completing the first and second level checks, the background spatiotemporal verification module further reads the corresponding work order time information and spatial location information for image pairs that have entered the third level verification stage. For time correlation parameters, the system obtains the timestamp of the installation and maintenance image capture. and work order creation timestamp First, calculate the time difference:

[0122] Then based on the preset time threshold Calculate the time consistency coefficient:

[0123] in, The preferred setting is 72 hours.

[0124] For spatial correlation parameters, the system obtains the location coordinates corresponding to the location of the maintenance image capture address and the location coordinates corresponding to the work order operation address, and calculates the straight-line distance between the two. Then, based on the preset spatial distance threshold Calculate the spatial consistency coefficient:

[0125] in, The preferred setting is 5 meters.

[0126] Subsequently, the system fuses the temporal consistency coefficient and the spatial consistency coefficient to obtain the spatiotemporal verification coefficient:

[0127] For example, for a business-to-business service order, if the time the image was taken differs from the time the service order was created by only 1.5 hours, then... Smaller, time consistency coefficient The spatial consistency coefficient is close to 1; if the straight-line distance between the image capture location and the work order's work address is approximately 2 meters, then the spatial consistency coefficient is close to 1. It is also relatively high, ultimately resulting in a high spatiotemporal verification coefficient. Conversely, if a photo was taken several days before the work order was created, and the location of the photo is tens of meters away from the work order's address, then... and All decreased significantly, leading to If the value is low, the system can determine that the image has obvious abnormalities.

[0128] It should be noted that the time of the installation and maintenance image capture can be extracted from the image information or obtained from the metadata attached when the installation and maintenance terminal uploads the image.

[0129] It should be noted that the location where the installation and maintenance image was captured can be determined by the capture location coordinates, base station location results, or address resolution results.

[0130] It should be noted that the time threshold... and spatial distance threshold It can be dynamically adjusted according to different business types and different installation and maintenance specifications.

[0131] In this embodiment, by quantitatively modeling the temporal and spatial rationality of installation and maintenance images, it is possible to further filter suspected similar images from the business logic level, thereby improving the credibility of abnormal work order identification results.

[0132] In yet another preferred embodiment, the method further includes: Based on the quality values ​​and weight values ​​of each dimension, calculate the quality improvement priority coefficient for each quality inspection dimension, and determine the rectification priority for each quality inspection dimension based on the quality improvement priority coefficient.

[0133] In one example, after the system completes the comprehensive quality assessment of installation and maintenance, the rectification guidance module further reads the dimensional quality values ​​of each quality inspection dimension. and the corresponding dimensional quality weight values And calculate the first according to the following formula Quality improvement priority coefficient for each quality inspection dimension :

[0134] in, The larger the value, the higher the importance of that quality inspection dimension and the lower the current compliance rate, therefore it should be prioritized for rectification.

[0135] It should be noted that the quality improvement priority coefficient takes into account both the importance of the dimension itself and the current level of achievement, and is therefore more instructive than simply ranking based on quality scores.

[0136] It is understood that the quality improvement priority coefficient can be calculated in real time and fed back to the installation and maintenance terminal immediately to support on-site rectification.

[0137] It is understandable that the rectification priority can be used not only for on-site guidance of front-line installation and maintenance personnel, but also for back-end management personnel to conduct problem attribution statistics and regional quality analysis.

[0138] In this embodiment, by introducing time consistency coefficients and spatial consistency coefficients to constrain the images, abnormal installation and maintenance images that do not conform to the time or spatial logic of the work order can be effectively identified.

[0139] A second aspect of this invention provides a device for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection, used to implement the multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection method described in any of the first aspects above. See also... Figure 2The diagram shown is a structural block diagram of a preferred embodiment of a multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection device provided in the second aspect of the present invention. The device includes: Data acquisition module 11 is used to acquire installation and maintenance images corresponding to broadband installation and maintenance work orders; The dimension quality value calculation module 12 is used to extract feature information of each quality detection dimension based on the assembled dimension image, and calculate the dimension quality value of each quality detection dimension based on the feature information. The installation and maintenance quality assessment module 13 is used to calculate the comprehensive installation and maintenance quality value based on the quality values ​​of each dimension and the dimension quality weight values ​​corresponding to each dimension quality value, and to determine whether the installation and maintenance quality is qualified based on the comprehensive installation and maintenance quality value. The abnormal installation and maintenance work order determination module 14 is used to calculate the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images obtained within a preset time period, and to determine whether there is an abnormal installation and maintenance work order based on the image similarity and the spatiotemporal consistency coefficient.

[0140] Preferably, each of the quality inspection dimensions includes several sub-quality inspection dimensions, then the dimension quality value calculation module 12 is further used to calculate the dimension quality value of each quality inspection dimension through the following steps: From the installation and maintenance image, the feature information of each quality detection dimension is extracted, and based on the feature information, the sub-dimensional compliance confidence of each sub-quality detection dimension is obtained; Based on the compliance confidence scores of all the sub-dimensions of the i-th quality inspection dimension and the sub-dimension compliance weight values ​​corresponding to the compliance confidence scores of the sub-dimensions, calculate the quality review value of the i-th quality inspection dimension; Calculate the quality adjustment factor based on broadband installation and maintenance time and workload. The dimension quality value of the i-th quality inspection dimension is calculated based on the quality review value of the i-th quality inspection dimension, the compliance confidence level of the sub-dimension, the compliance weight value of the sub-dimension, the quality adjustment factor, and the decay function.

[0141] Preferably, the dimension quality value calculation module 12 is further configured to calculate the sub-dimensional compliance confidence level of each of the sub-quality detection dimensions through the following steps: The feature information of each sub-quality detection dimension in the image is extracted using a convolutional neural network, and the compliance confidence score of each sub-quality detection dimension is output based on the feature information.

[0142] Preferably, the installation and maintenance quality assessment module 13 is further used to determine whether the installation and maintenance quality is qualified through the following steps: The overall quality value of the equipment is obtained by weighted summing of the quality values ​​of each dimension and the corresponding dimension quality weight values. If the overall installation and maintenance quality value is less than the preset quality assessment value, the installation and maintenance quality is determined to be unqualified; otherwise, the installation and maintenance quality is determined to be qualified.

[0143] Preferably, the image similarity includes a first image similarity and a second image similarity, the spatiotemporal consistency coefficient includes a temporal consistency coefficient and a spatial consistency coefficient, and the abnormal installation and maintenance work order determination module specifically includes: The first abnormal installation and maintenance work order determination unit is used to calculate the first image similarity between the installation and maintenance images using an average hash algorithm. The second abnormal installation and maintenance work order determination unit is used to determine that the corresponding installation and maintenance image is an abnormal image when the similarity of the first image is greater than or equal to the preset image similarity threshold; and to determine that the corresponding installation and maintenance image is a normal image when the similarity of the first image is less than the preset image similarity threshold, and to add the corresponding installation and maintenance image to the first installation and maintenance image set. The third abnormal installation and maintenance work order determination unit is used to calculate the second image similarity between installation and maintenance images in the first installation and maintenance image set; The fourth abnormal installation and maintenance work order determination unit is used to calculate the time consistency coefficient between installation and maintenance images in the first installation and maintenance image set based on the time correlation parameter of the installation and maintenance images in the first installation and maintenance image set; and to calculate the spatial consistency coefficient between installation and maintenance images in the first installation and maintenance image set based on the spatial correlation parameter of the installation and maintenance images in the first installation and maintenance image set. The fifth abnormal installation and maintenance work order determination unit is used to calculate the final similarity between installation and maintenance images in the first installation and maintenance image set based on the first image similarity, the second image similarity, the temporal consistency coefficient, and the spatial consistency coefficient. The sixth abnormal installation and maintenance work order determination unit is used to determine that the corresponding installation and maintenance image is abnormal when the final similarity is greater than the first preset similarity threshold; to determine that the corresponding installation and maintenance image is pending manual review when the final similarity is less than the first preset similarity threshold but greater than the second preset similarity threshold; and to determine that the corresponding installation and maintenance image is normal when the final similarity is less than the second preset similarity threshold.

[0144] Preferably, the first abnormal installation and maintenance work order determination unit is used to calculate the first image similarity between the installation and maintenance images through the following steps: Each of the installation and maintenance images is scaled to a preset size, and the scaled installation and maintenance images are converted to grayscale to obtain the corresponding grayscale images. Calculate the average gray value of all pixels in each of the grayscale images; The grayscale value of each pixel in each grayscale image is compared with the corresponding average grayscale value. When the grayscale value of the target pixel is greater than the corresponding average grayscale value, the hash bit corresponding to the target pixel is assigned a first preset value; when the grayscale value of the target pixel is less than or equal to the corresponding average grayscale value, the hash bit corresponding to the target pixel is assigned a second preset value, so as to generate the image hash value corresponding to each of the two images. Calculate the Hamming distance between the image hash values ​​of each of the assembled images; Based on the Hamming distance, the first image similarity between each of the assembled images is calculated.

[0145] Preferably, the third abnormal installation and maintenance work order determination unit is used to calculate the second image similarity between installation and maintenance images in the first installation and maintenance image set through the following steps: Each dimensional image in the first dimensional image set is processed by grayscale and intervalization to obtain the grayscale interval feature vector of each dimensional image, and the histogram similarity between each dimensional image is calculated based on the grayscale interval feature vector. The edge features of each dimensional image in the first dimensional image set are extracted to obtain the corresponding edge density features, and the edge feature similarity between each dimensional image is calculated based on the edge density features. Obtain the aspect ratio and file size of each dimensional image in the first dimensional image set, and calculate the metadata similarity between each dimensional image based on the difference between the aspect ratio and file size of each dimensional image; The second image similarity is calculated based on the histogram similarity, the edge feature similarity, and the metadata similarity.

[0146] Preferably, the time correlation parameters include the installation and maintenance image capture time and the installation and maintenance work order creation time, and the spatial correlation parameters include the installation and maintenance image capture address and the installation and maintenance work order operation address. Then, the fourth abnormal installation and maintenance work order determination unit calculates the time consistency coefficient and the spatial consistency coefficient through the following steps: Based on the image capture time, the work order creation time, and a preset time threshold, the time consistency coefficient between the installation and maintenance images in the first set of installation and maintenance images is calculated. Based on the location where the installation and maintenance image was captured, the location of the installation and maintenance work order, and a preset spatial distance threshold, the spatial consistency coefficient between the installation and maintenance images in the first set of installation and maintenance images is calculated.

[0147] Preferably, the device further includes: The rectification priority determination module is used to calculate the quality improvement priority coefficient of each quality inspection dimension based on the quality value and the quality weight value of each dimension, and to determine the rectification priority of each quality inspection dimension based on the quality improvement priority coefficient.

[0148] It should be noted that the multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection device provided in the second aspect embodiment of the present invention can realize all the processes of the multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection method described in the first aspect. The functions and technical effects of each module and unit in the device are the same as those of the multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection method described in the first aspect embodiment, and will not be repeated here.

[0149] Through the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary hardware platforms, and of course, it can also be implemented entirely by hardware. Based on this understanding, all or part of the technical solution of the present invention that contributes to the background technology can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM (Read-Only Memory) / RAM (Random Access Memory), magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0150] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for quality assessment of installation and maintenance and detection of abnormal installation and maintenance work orders based on multi-dimensional fusion, characterized in that, include: Obtain the installation and maintenance images corresponding to the broadband installation and maintenance work orders; Based on the image, feature information of each quality detection dimension is extracted, and the dimensional quality value of each quality detection dimension is calculated based on the feature information. Based on the quality values ​​of each dimension and the corresponding dimension quality weight values, the overall installation and maintenance quality value is calculated, and the installation and maintenance quality is determined to be qualified based on the overall installation and maintenance quality value. Calculate the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images obtained within a preset time period, and determine whether there are any abnormal installation and maintenance work orders based on the image similarity and spatiotemporal consistency coefficient.

2. The method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in claim 1, characterized in that, Each of the quality detection dimensions includes several sub-quality detection dimensions. The step of extracting feature information for each quality detection dimension based on the assembled image, and calculating the dimensional quality value for each quality detection dimension based on the feature information, includes: From the installation and maintenance image, the feature information of each quality detection dimension is extracted, and based on the feature information, the sub-dimensional compliance confidence of each sub-quality detection dimension is obtained; Based on the compliance confidence scores of all the sub-dimensions of the i-th quality inspection dimension and the sub-dimension compliance weight values ​​corresponding to the compliance confidence scores of the sub-dimensions, calculate the quality review value of the i-th quality inspection dimension; Calculate the quality adjustment factor based on broadband installation and maintenance time and workload. The dimension quality value of the i-th quality inspection dimension is calculated based on the quality review value of the i-th quality inspection dimension, the compliance confidence level of the sub-dimension, the compliance weight value of the sub-dimension, the quality adjustment factor, and the decay function.

3. The method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in claim 2, characterized in that, The step of extracting feature information for each quality detection dimension from the installation and maintenance image, and obtaining the compliance confidence score of each sub-quality detection dimension based on the feature information, includes: The feature information of each sub-quality detection dimension in the image is extracted using a convolutional neural network, and the compliance confidence score of each sub-quality detection dimension is output based on the feature information.

4. The method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in claim 1, characterized in that, The step of calculating the overall installation and maintenance quality value based on the quality values ​​of each dimension and the corresponding dimension quality weight values, and determining whether the installation and maintenance quality is qualified based on the overall installation and maintenance quality value, includes: The overall quality value of the equipment is obtained by weighted summing of the quality values ​​of each dimension and the corresponding dimension quality weight values. If the overall installation and maintenance quality value is less than the preset quality assessment value, the installation and maintenance quality is determined to be unqualified; otherwise, the installation and maintenance quality is determined to be qualified.

5. The method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in claim 1, characterized in that, The image similarity includes a first image similarity and a second image similarity, and the spatiotemporal consistency coefficient includes a temporal consistency coefficient and a spatial consistency coefficient. The calculation of the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images acquired within a preset time period, and the determination of whether there are abnormal installation and maintenance work orders based on the image similarity and the spatiotemporal consistency coefficient, includes: The first image similarity between the assembled images is calculated using an average hash algorithm; When the similarity of the first image is greater than or equal to the preset image similarity threshold, the corresponding installation and maintenance image is determined to be an abnormal image; when the similarity of the first image is less than the preset image similarity threshold, the corresponding installation and maintenance image is determined to be a normal image, and the corresponding installation and maintenance image is added to the first installation and maintenance image set. Calculate the second image similarity among the dimensional images in the first dimensional image set; Based on the temporal correlation parameters of the installation and maintenance images in the first installation and maintenance image set, the temporal consistency coefficient among the installation and maintenance images in the first installation and maintenance image set is calculated; based on the spatial correlation parameters of the installation and maintenance images in the first installation and maintenance image set, the spatial consistency coefficient among the installation and maintenance images in the first installation and maintenance image set is calculated. Based on the first image similarity, the second image similarity, the temporal consistency coefficient, and the spatial consistency coefficient, the final similarity between the dimensional images in the first dimensional image set is calculated; When the final similarity is greater than the first preset similarity threshold, the corresponding installation and maintenance image is determined to be abnormal; when the final similarity is less than the first preset similarity threshold but greater than the second preset similarity threshold, the corresponding installation and maintenance image is determined to be awaiting manual review; when the final similarity is less than the second preset similarity threshold, the corresponding installation and maintenance image is determined to be normal.

6. The method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in claim 5, characterized in that, The step of calculating the first image similarity between the dimensional images using the average hash algorithm includes: Each of the installation and maintenance images is scaled to a preset size, and the scaled installation and maintenance images are converted to grayscale to obtain the corresponding grayscale images. Calculate the average gray value of all pixels in each of the grayscale images; The grayscale value of each pixel in each grayscale image is compared with the corresponding average grayscale value. When the grayscale value of the target pixel is greater than the corresponding average grayscale value, the hash bit corresponding to the target pixel is assigned a first preset value; when the grayscale value of the target pixel is less than or equal to the corresponding average grayscale value, the hash bit corresponding to the target pixel is assigned a second preset value, so as to generate the image hash value corresponding to each of the two images. Calculate the Hamming distance between the image hash values ​​of each of the assembled images; Based on the Hamming distance, the first image similarity between each of the assembled images is calculated.

7. The method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in claim 5, characterized in that, The calculation of the second image similarity between the dimensional images in the first dimensional image set includes: Each dimensional image in the first dimensional image set is processed by grayscale and intervalization to obtain the grayscale interval feature vector of each dimensional image, and the histogram similarity between each dimensional image is calculated based on the grayscale interval feature vector. The edge features of each dimensional image in the first dimensional image set are extracted to obtain the corresponding edge density features, and the edge feature similarity between each dimensional image is calculated based on the edge density features. Obtain the aspect ratio and file size of each dimensional image in the first dimensional image set, and calculate the metadata similarity between each dimensional image based on the difference between the aspect ratio and file size of each dimensional image; The second image similarity is calculated based on the histogram similarity, the edge feature similarity, and the metadata similarity.

8. The method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in claim 5, characterized in that, The time correlation parameters include the installation and maintenance image capture time and the installation and maintenance work order creation time; the spatial correlation parameters include the installation and maintenance image capture address and the installation and maintenance work order operation address. The time consistency coefficient and the spatial consistency coefficient are then calculated using the following steps: Based on the image capture time, the work order creation time, and a preset time threshold, the time consistency coefficient between the installation and maintenance images in the first set of installation and maintenance images is calculated. Based on the location where the installation and maintenance image was captured, the location of the installation and maintenance work order, and a preset spatial distance threshold, the spatial consistency coefficient between the installation and maintenance images in the first set of installation and maintenance images is calculated.

9. The method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in claim 1, characterized in that, Also includes: Based on the quality values ​​and weight values ​​of each dimension, calculate the quality improvement priority coefficient for each quality inspection dimension, and determine the rectification priority for each quality inspection dimension based on the quality improvement priority coefficient.

10. A device for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection, used to implement the method for multi-dimensional fusion-based installation and maintenance quality assessment and abnormal installation and maintenance work order detection as described in any one of claims 1 to 9, the device comprising: The data acquisition module is used to acquire installation and maintenance images corresponding to broadband installation and maintenance work orders; The dimension quality value calculation module is used to extract feature information of each quality detection dimension based on the assembled dimension image, and calculate the dimension quality value of each quality detection dimension based on the feature information. The installation and maintenance quality assessment module is used to calculate the comprehensive installation and maintenance quality value based on the quality values ​​of each dimension and the dimension quality weight values ​​corresponding to each dimension quality value, and to determine whether the installation and maintenance quality is qualified based on the comprehensive installation and maintenance quality value. The abnormal installation and maintenance work order determination module is used to calculate the image similarity and spatiotemporal consistency coefficient among several installation and maintenance images obtained within a preset time period, and to determine whether there is an abnormal installation and maintenance work order based on the image similarity and the spatiotemporal consistency coefficient.

Images