FTTR installation specification quality inspection method and system based on AI image recognition

Abstract

The invention provides an FTTR installation specification quality inspection method and system based on AI image recognition, electronic equipment, a storage medium and a program product, and aims to solve the problem that FTTR construction automatic quality inspection cannot be carried out, the method comprises the following steps: obtaining images to be subjected to quality inspection of an installation construction site, including a first type of images for displaying an installation environment and a second type of images for displaying an equipment identifier; the first model identifies the first type of image and judges whether the first type of image contains related equipment; if so, the second model identifies the second type of images, and locates an MAC address pattern area on the device; obtaining the MAC address of the equipment, comparing the MAC address with a preset database, and verifying the compliance of the equipment; detecting the installation environment of the equipment by using a third model, and evaluating whether the installation position meets a preset specification based on feature detection and a monocular distance measurement algorithm; and generating a quality inspection conclusion according to the two identification results. The construction quality management level and efficiency are remarkably improved.

Description

Technical Field

[0001] This disclosure relates to the field of communication network construction quality management, and in particular to a quality inspection method and system for FTTR installation specifications based on AI image recognition, electronic equipment, computer-readable storage media, and computer program products. Background Technology

[0002] With the large-scale deployment of FTTR (Fiber to the Room) technology, efficiently and accurately inspecting the quality of massive in-home installations has become a key challenge for operators to ensure user experience and reduce maintenance costs. Currently, the mainstream quality inspection method still heavily relies on manual labor: construction workers take photos on-site, which are then manually reviewed by backend quality inspectors to determine whether the equipment installation is correct and whether the environment is compliant. This method has the following prominent problems: The lack of specialized recognition capabilities leads to low efficiency: Existing general-purpose image recognition AI models (such as YOLO and SSD) are primarily trained on everyday objects (people, vehicles, animals, etc.) and cannot recognize the numerous specialized communication devices involved in FTTR services (such as specific models of optical modems, splitters, IPTV (Internet Protocol Television) / OTT (Over-The-Top) terminals, etc.), let alone distinguish subtle differences between equipment from different operators (such as brand logos). This makes it virtually impossible to directly apply AI technology in this specialized field, and quality inspection relies entirely on manual labor, resulting in extremely low efficiency.

[0003] Contextual information cannot be captured, resulting in weak anti-counterfeiting capabilities: Installation specifications require verification of device identity (such as MAC (Media Access Control) addresses) to prevent device replacement. However, device MAC addresses typically exist as "labels," complex patterns containing text and barcodes, rather than a single entity. Existing image recognition technology cannot understand such "contexts," relying solely on manual visual search and input, or the use of separate barcode scanning devices. This process is cumbersome, easily circumvented, and cannot effectively automate anti-counterfeiting verification.

[0004] Environmental compliance assessments are subjective and lack quantifiable standards: Installation specifications have clear requirements for the equipment installation environment, such as keeping it away from high-power electrical appliances, avoiding obstructions, and maintaining specific heights and distances. Currently, the inspection of these requirements relies entirely on the personal experience of quality inspectors. Estimating distances and spatial relationships from two-dimensional photographs is extremely difficult, leading to inconsistent evaluation standards and making misjudgments and disputes more likely.

[0005] Limitations of existing related technical solutions: Although there are solutions that apply image recognition to the appearance quality inspection of industrial parts and the inspection of telecommunications external line boxes (such as the published document CN116012366A and the academic paper "Application of Image Recognition Technology in Telecommunications Operation and Maintenance Quality Inspection"), these solutions are all designed for their specific scenarios and have single recognition targets (such as whether the box door is closed or whether the grounding wire is connected). They do not involve the recognition of complex indoor multi-equipment systems, nor do they have the ability to perform spatial relationship quantitative analysis of the installation environment. Therefore, they cannot solve the automated quality inspection needs of complex indoor construction scenarios such as FTTR.

[0006] Therefore, there is an urgent need for a quality inspection method and system that can automatically and accurately identify FTTR professional equipment, intelligently verify the equipment identity, and quantitatively assess the compliance of the installation environment, in order to fill the technological gap in this field and promote the transformation of construction quality management towards automation and intelligence. Summary of the Invention

[0007] To address the limitations of existing technologies in automating quality inspection for complex indoor FTTR (Fiber to the Reinforced Transmission) installation scenarios, this disclosure provides an AI-based image recognition-based FTTR installation specification quality inspection method and system, electronic equipment, computer-readable storage medium, and computer program product. It achieves accurate identification of professional FTTR equipment, automated anti-counterfeiting verification based on MAC addresses, and quantitative intelligent assessment of installation environment compliance, significantly improving the level and efficiency of construction quality management.

[0008] In a first aspect, this disclosure provides a quality inspection method for FTTR installation specifications based on AI image recognition, the method comprising: Acquire images of the FTTR installation site to be inspected, the images to be inspected including at least: a first type of image for showing the equipment installation environment, and a second type of image for showing equipment identification information; The first neural network model is used to identify the first type of image and determine whether the image contains preset FTTR service-related equipment. If a device related to the FTTR service is identified, the second type of image is selected from the images to be inspected, and the second neural network model is used to identify it and locate the MAC address pattern area on the device. The MAC address pattern area is processed and recognized to obtain the device's MAC address information, and the MAC address information is compared with a preset database to verify the device's compliance. The first type of image is identified using a third neural network model to detect the installation environment of the FTTR service-related equipment. Based on feature detection and monocular ranging algorithms, the installation location is evaluated to determine whether it meets the preset specifications. The preset specifications include at least one or more of the following: distance threshold from interference sources, installation height range, and minimum spacing between equipment. Based on the results of equipment compliance and whether the installation location meets the preset specifications, a quality inspection conclusion for the FTTR installation specification is generated.

[0009] Furthermore, the first neural network model replaces the standard convolutional layer with a transposed convolutional layer to enhance the ability to recognize distant, small-sized target features in images, thereby distinguishing between similar devices from different operators and devices with similar appearances.

[0010] Furthermore, the second neural network model is trained to recognize MAC address pattern scenarios consisting of one or more combinations of text, barcodes, and QR codes; The image processing and recognition of the MAC address pattern area includes: performing image enhancement processing on the MAC address pattern area, and extracting MAC address information using optical character recognition technology or barcode recognition technology.

[0011] Furthermore, the evaluation of whether the installation location meets the preset specifications based on feature detection and monocular ranging algorithms includes: The shooting focal length is obtained based on the EXIF ​​(Exchangeable Image File Format) information of the first type of image; Identify the pixel dimensions of the FTTR service-related equipment and at least one reference object in the first type of image; Based on the known actual size of the reference object, the shooting focal length, and the pixel size, calculate the first distance between the reference object and the shooting point, and the second distance between the FTTR service-related equipment and the shooting point. Based on the difference between the first distance and the second distance, it is determined whether the actual distance between the FTTR service-related equipment and the reference object meets the preset distance threshold requirement.

[0012] Furthermore, the reference objects include high-power electrical appliances, walls, furniture, or other FTTR business-related equipment.

[0013] Furthermore, the preset specifications include: the distance between the FTTR service-related equipment and the interference source should be greater than 1.2 meters, the installation height should be between 0.2 meters and 1.5 meters, and the installation distance between multiple FTTR service-related devices in the same room should be greater than 3 meters.

[0014] Furthermore, the first type of image is captured at a distance of not less than 1.5 meters, and the second type of image is a close-up image containing identification information on the bottom, back, or side of the device.

[0015] Secondly, this disclosure provides an AI image recognition-based FTTR installation specification quality inspection system, the system comprising: The acquisition module is configured to acquire images of the FTTR installation site to be inspected, the images to be inspected including at least: a first type of image for displaying the equipment installation environment, and a second type of image for displaying equipment identification information; The judgment module is configured to use a first neural network model to identify the first type of image and determine whether the image contains a preset FTTR service-related device. The identification module is configured to select the second type of image from the images to be inspected if the judgment module identifies that the device contains the FTTR service-related device, and use the second neural network model to identify it and locate the MAC address pattern area on the device. The MAC address verification module is configured to perform image processing and recognition on the MAC address pattern area to obtain the device's MAC address information, and compare the MAC address information with a preset database to verify the device's compliance. The installation location detection module is configured to use a third neural network model to identify the first type of image, detect the installation environment of the FTTR service-related equipment, and evaluate whether the installation location meets the preset specifications based on feature detection and monocular ranging algorithm. The preset specifications include at least one or more of the following: distance threshold from interference source, installation height range, and minimum spacing between devices. The generation module is configured to generate a quality inspection conclusion for the FTTR installation specification based on the results of equipment compliance and whether the installation location meets the preset specifications.

[0016] Thirdly, this disclosure provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores one or more computer programs executable by the at least one processor, the one or more computer programs being executed by the at least one processor to enable the at least one processor to perform the aforementioned FTTR installation specification quality inspection method based on AI image recognition.

[0017] Fourthly, this disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned AI image recognition-based FTTR installation specification quality inspection method.

[0018] Fifthly, this disclosure provides a computer program product that includes computer-readable code or a non-volatile computer-readable storage medium carrying computer-readable code. When the computer-readable code is run in a processor of an electronic device, the processor in the electronic device executes the above-described FTTR installation specification quality inspection method based on AI image recognition.

[0019] Beneficial effects: This disclosure presents an AI-based image recognition-based FTTR installation specification quality inspection method and system, electronic equipment, computer-readable storage medium, and computer program product. By acquiring two types of images—one for displaying the environment and the other for identification—and designing three neural network models, each with its own function (equipment identification, MAC pattern localization, and environmental assessment), a complete and feasible automated quality inspection logic loop is constructed. This method utilizes the first type of image for professional equipment identification and the second type of image for automated anti-counterfeiting comparison based on MAC addresses, effectively preventing equipment tampering. Furthermore, it uses monocular ranging technology to quantitatively evaluate key indicators such as installation distance and height, transforming the previous reliance on subjective human judgment into objective and accurate algorithmic judgment, greatly improving the uniformity and accuracy of quality inspection standards. Ultimately, this solution integrates discrete manual inspection steps into an efficient end-to-end intelligent process, achieving a leap from "human judgment" to "AI analysis" in the field of communication professional construction quality inspection, significantly reducing operation and maintenance costs, and ensuring and improving the deployment quality and user experience of FTTR networks.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the embodiments of the present disclosure to explain the disclosure and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the detailed description of exemplary embodiments with reference to the accompanying drawings, in which: Figure 1 This is a flowchart illustrating a quality inspection method for FTTR installation specifications based on AI image recognition, provided in Embodiment 1 of this disclosure. Figure 2 This is a schematic diagram illustrating the implementation process of a three-model cascaded application provided in Embodiment 2 of this disclosure; Figure 3 This is an example diagram of an optical modem from a different operator provided in Embodiment 2 of this disclosure; Figure 4This is an example diagram of a method for capturing and recognizing a carrier's logo from a distance, as provided in Embodiment 2 of this disclosure; Figure 5 A comparison diagram of ordinary convolution and transposed convolution provided for Embodiment 2 of this disclosure; Figure 6 This is a schematic diagram of a device that uses a transposed convolution mechanism to distinguish between similar-looking devices, as provided in Embodiment 2 of this disclosure. Figure 7 This is an example diagram illustrating how to determine whether there is interference from high-power electrical appliances near an optical modem, as provided in Embodiment 2 of this disclosure. Figure 8 This is a block diagram of an FTTR installation specification quality inspection system based on AI image recognition, provided in Embodiment 3 of this disclosure; Figure 9 This is a block diagram of an electronic device provided in Embodiment 4 of this disclosure. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions of this disclosure, exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of this disclosure to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0023] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0024] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded. Words such as “connected” or “linked” are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect.

[0026] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning unless expressly so defined herein. Those skilled in the art will understand that the specific order of execution of the steps in the methods described above in the specific embodiments should be determined by their function and possible internal logic.

[0027] The FTTR installation specification quality inspection method based on AI image recognition according to embodiments of this disclosure can be executed by electronic devices such as terminal devices or servers. Terminal devices can be in-vehicle devices, user equipment (UE), mobile devices, user terminals, terminals, cellular phones, cordless phones, personal digital assistants (PDAs), handheld devices, computing devices, in-vehicle devices, wearable devices, etc. The method can be implemented by a processor calling computer-readable program instructions stored in memory. Alternatively, the method can be executed by a server.

[0028] Example 1

[0029] Figure 1 This is a flowchart illustrating a quality inspection method for FTTR installation specifications based on AI image recognition, provided in Embodiment 1 of this disclosure. (Refer to...) Figure 1 The method includes: Step S101: Obtain images of the FTTR installation site to be inspected, the images to be inspected including at least: a first type of image for displaying the equipment installation environment, and a second type of image for displaying equipment identification information; Step S102: Use the first neural network model to identify the first type of image and determine whether the image contains a preset FTTR service-related device; Step S103: If a device related to the FTTR service is identified, the second type of image is selected from the images to be inspected, and the second neural network model is used to identify it and locate the MAC address pattern area on the device. Step S104: Perform image processing and recognition on the MAC address pattern area to obtain the MAC address information of the device, and compare the MAC address information with a preset database to verify the compliance of the device; Step S105: Use the third neural network model to identify the first type of image, detect the installation environment of the FTTR service related equipment, and evaluate whether the installation location meets the preset specifications based on feature detection and monocular ranging algorithm. The preset specifications include at least one or more of the following: distance threshold from interference source, installation height range, and minimum spacing between equipment. Step S106: Based on the equipment compliance results and whether the installation location meets the preset specifications, generate the quality inspection conclusion of the FTTR installation specification.

[0030] The purpose of this disclosure is to provide an automated quality inspection method based on AI image recognition and multi-model collaboration to solve a series of technical problems in the existing FTTR (Fiber to the Room) installation and construction quality inspection, such as low efficiency, inconsistent standards, strong reliance on manual labor, and inability to automatically verify the authenticity of equipment and environmental compliance.

[0031] To achieve this objective, the present invention constructs three neural network models: (1) a neural network model for detecting targets of related equipment involved in FTTR service, namely the first neural network model; (2) a neural network model for detecting targets of MAC address patterns of various equipment in FTTR service, namely the second neural network model; and (3) a quality inspection model for the installation of optical modems in FTTR service, namely the third neural network model.

[0032] The construction steps for the above three models are largely the same, and the specific steps are as follows: 1) Formulation of quality inspection / photo taking specifications: In order to better realize the construction quality inspection of FTTR service, it is necessary to formulate corresponding specifications, including what types of photos need to be taken for the quality inspection of FTTR service, the shooting angle and distance of each type of photo, what the problem characteristic points of FTTR service quality inspection are, and the manifestation of problem characteristic points. 2) Model design: Select from existing open-source neural network models, choosing those with strong applicability to this business scenario, and then tailor and improve them for different scenarios of FTTR business. 3) Dataset construction: Collect a large number of photos uploaded from the FTTR construction site, carry out sample screening, image preprocessing, target feature labeling and other work to construct a sample dataset for model training; 4) Model training and optimization: After multiple iterations, an application model for the corresponding scenario is trained.

[0033] Specifically, in step S101, after the FTTR equipment installation is completed, installation and maintenance personnel can use a dedicated APP to take and upload a set of construction site photos according to the built-in "FTTR Construction Photography Specifications". This set of photos is collectively referred to as "images to be inspected," and its core components include at least two categories: one category is images designed to show the overall placement and orientation of the equipment in the user's room, as well as its relationship with surrounding furniture, appliances, and other environmental elements (usually panoramic or medium-range photos); the other category is images specifically targeting the equipment itself, clearly capturing its nameplate, labels, and other areas (usually close-up or extreme close-up photos). The system's backend service receives and associates these images under the same work order, providing structured input data for subsequent analysis. By standardizing the categories and purposes of image collection, the completeness and relevance of the visual information required for AI analysis are ensured, providing a reliable data foundation for subsequent specialized recognition tasks and guaranteeing the feasibility of the quality inspection process from the source.

[0034] In step S102, the system invokes a deployed first neural network model (e.g., a model based on improved architectures such as Faster R-CNN and YOLO) to automatically analyze the received first type of images. This model is pre-trained using a massive number of construction site environment images labeled with various FTTR devices (such as different models of optical modems, set-top boxes, etc.), and has the ability to detect and identify specific device categories from complex backgrounds. The model output is whether the target device exists in the image and its location in the image. This achieves automated visual discovery of professional communication equipment, replacing the first step of manually searching for and identifying equipment in photos, greatly improving the efficiency and accuracy of the initial screening.

[0035] In step S103, after step S102 confirms the existence of a target device, the system logic is automatically triggered to retrieve and select the corresponding second-type image (close-up photo) from the uploaded image group. Subsequently, a dedicated second neural network model is invoked to process the close-up image. This model is specifically optimized for the "device identification area" detection task and can accurately locate the label or printed area (i.e., the "MAC address pattern area") containing information such as MAC address and serial number, regardless of whether the area is plain text, a one-dimensional barcode, or a QR code. This achieves intelligent connection from "finding the device" to "preparing to read the device identity," accurately locating key information areas through a dedicated model, creating conditions for subsequent accurate information extraction, and avoiding the inefficiency and errors of full-image scanning.

[0036] In step S104, after obtaining the coordinates of the MAC address image area, the system preprocesses the image of that area (e.g., cropping, perspective correction, noise reduction, contrast enhancement, etc.), and then calls the corresponding recognition engine (e.g., OCR engine, barcode decoding library) to extract information based on the image features (e.g., plain text, barcode, QR code) to obtain the MAC address in string format. Subsequently, the system automatically compares this MAC address with a pre-stored database of legitimate MAC addresses of the equipment to be installed in the work order system or resource management system. This completes the automated verification of the compliance of construction materials, seamlessly integrating the traditional anti-counterfeiting steps that rely on manual visual inspection or separate barcode scanning into the AI ​​process, effectively preventing equipment substitution and incorrect installation, and strengthening the closed-loop control of supply chain management and construction quality.

[0037] Step S105: This step is executed in parallel or sequentially with step S102, and is also based on the first type of image. The system calls the third neural network model, which not only identifies the device but also detects potential interference sources (such as televisions, microwave ovens, and refrigerators), obstructions, and other installation references in the environment. Combining the focal length in the image EXIF ​​information, the known device size obtained through device identification, and the pixel distance in the image, a monocular vision ranging algorithm is applied to calculate quantitative indicators such as the actual distance between the device and the interference source, the device's height above the ground, and the spacing between devices. Finally, these indicators are automatically compared with the installation specification thresholds preset in the rule engine. This transforms abstract installation specifications (such as "away from" and "sufficiently high") into objective, measurable spatial data, achieving intelligent and quantitative assessment of the compliance of the installation process environment, and eliminating the subjectivity and inconsistency of manual judgment.

[0038] Finally, in step S106, the system summarizes and logically synthesizes the equipment identity compliance verification results ("pass" / "fail" and specific reasons) from step S104 with the quantitative evaluation results of the installation location ("qualified" / "unqualified" for each indicator) from step S105. Based on preset decision rules (e.g., if the equipment identity fails, the entire system fails; or a weighted score is applied to each environmental indicator), a structured quality inspection conclusion report is automatically generated. This report may include textual conclusions, screenshots of key indicators, and annotations of non-conforming items, and is automatically populated back into the installation and maintenance work order system. This achieves automated generation and archiving of quality inspection conclusions, completing an end-to-end closed loop from image input to quality judgment. It provides immediate and objective digital evidence for construction acceptance, personnel assessment, and problem tracing, greatly improving the efficiency and precision of operation and maintenance management.

[0039] This disclosure, through the construction of a multi-step, multi-model collaborative automated quality inspection process, achieves standardization and closed-loop management of the quality inspection process. By clearly defining the acquisition specifications for "Category 1" and "Category 2" images and applying three types of dedicated neural network models (equipment identification, identifier positioning, and environmental analysis) in series, a complete and repeatable automated technical process for FTTR construction quality inspection is established for the first time. This process integrates the originally discrete and manual-dependent steps of taking photos, inspecting, comparing, and judging into an organic whole, ensuring the standardization of quality inspection inputs (images), the consistency of processing logic, and the structured nature of output conclusions. It fundamentally solves the pain points of inconsistent standards and arbitrary processes in traditional manual quality inspection. It achieves intelligent evaluation of construction quality across all dimensions: decomposing construction quality into two core dimensions, "equipment identity compliance" and "installation environment compliance," and automatically evaluating them through efficient technical means. On the one hand, by locating and identifying equipment identifiers (MAC addresses) through a proprietary model and linking with the backend database, automated anti-counterfeiting verification of equipment authenticity and origin is achieved, eliminating loopholes in material management. On the other hand, by analyzing installation environment images and using visual ranging technology, abstract installation specifications (such as "away from" and "appropriate height") are transformed into quantifiable distance and dimension measurements, enabling objective and accurate automatic evaluation of the installation process. This shifts quality inspection conclusions from subjective qualitative to objective quantitative, resulting in a leapfrog improvement in operation and maintenance efficiency and quality: this method transforms quality inspection work, which heavily relies on the experience of senior engineers and intensive manpower, into a standardized service automatically completed 24 / 7 by an AI system. This greatly frees up human resources, reduces operating costs, and avoids false or missed inspections caused by human fatigue and negligence. The systematic process also makes quality analysis of massive construction data possible, providing data support for continuous optimization of installation specifications and construction processes, thereby promoting the continuous improvement of the overall deployment quality and user service experience of the FTTR network. This disclosed embodiment deeply integrates artificial intelligence technology into professional construction quality inspection scenarios. Through process standardization, intelligent evaluation, and automated execution, it ultimately achieves the core values ​​of controllable quality, increased efficiency, and reduced costs, which has significant practical implications for the digital transformation of operation and maintenance in the communications industry.

[0040] Furthermore, the first neural network model replaces the standard convolutional layer with a transposed convolutional layer to enhance the ability to recognize distant, small-sized target features in images, thereby distinguishing between similar devices from different operators and devices with similar appearances.

[0041] The construction and training of the first neural network model employs specific network structure improvements to address the problem that key small-sized target features, such as operator logos, are difficult to effectively identify in images taken at standard shooting distances (typically not less than 1.5 meters) due to low pixel count and blurred details. A key implementation detail of this disclosure is that the standard convolutional layer in the selected basic convolutional neural network (such as YOLOv5, Faster R-CNN, etc.) is replaced with a transposed convolutional layer, also known as a deconvolutional layer.

[0042] The specific implementation process is as follows: Network architecture adjustment: During the model design phase, the selected basic network architecture is modified. The ordinary convolutional operation originally used for downsampling (reducing the size of feature maps and expanding the receptive field) is replaced with transposed convolutional operation. This change is not limited to a single layer, but is carried out in multiple key layers (especially shallow and intermediate layers) according to the network depth and task requirements, in order to build a backbone network with stronger feature resolution preservation capabilities.

[0043] Transposed convolution operation principle: Transposed convolution is an upsampling technique. During forward propagation, zero values ​​are inserted between elements of the input feature map, and regular convolution operations are performed, thereby increasing the spatial resolution (width and height) of the output feature map. Unlike ordinary convolution (formula: width_out = (W-K+2×P) / S + 1), which keeps the feature map size unchanged or reduces it, the size calculation formula for transposed convolution is: width_out = (W-1)×S-2×P + K. The upsampling factor can be controlled by adjusting the stride (S) and padding (P) parameters.

[0044] Advantages for small target recognition: In this application scenario, operator logos (such as "China Unicom" or "China Telecom") occupy only a very small number of pixels in images taken at a distance. In standard convolutional networks, after multiple downsampling operations, these tiny features almost disappear in deep feature maps, preventing the model from learning their discriminative features. By introducing transposed convolutional layers, the network can retain and gradually recover finer spatial details during forward propagation, ensuring that deep feature maps still contain enhanced representations of small-sized logos. This effectively increases the model's "perceptual weight" for small targets, significantly improving the model's ability to extract and distinguish features from distant, small-sized logos.

[0045] Training and Results: The improved model was trained using a large, labeled dataset of images from different carriers (China Unicom, China Telecom, and China Mobile) and devices with similar appearances (such as IPTV and OTT set-top boxes). During training, the model learned to effectively distinguish between different device categories based solely on logo differences or subtle appearance variations by utilizing the enhanced detail features of the transposed convolutional layers. For example, even if a "China Unicom optical modem" and a "China Telecom optical modem" have completely identical main appearances, the model can accurately classify them by magnifying and enhancing the features of the logo area.

[0046] By introducing the structural innovation of transposed convolutional layers, the first neural network model possesses a powerful ability to recognize small targets at long distances. This not only solves the problem of low recognition rate of general models in specific categories of communication equipment, but also enables accurate differentiation of equipment from different operators and effective identification of devices with highly similar appearances (such as IPTV and OTT), laying an accurate foundation for device identification in subsequent compliance verification.

[0047] Furthermore, the second neural network model is trained to recognize MAC address pattern scenarios consisting of one or more combinations of text, barcodes, and QR codes; The image processing and recognition of the MAC address pattern area includes: performing image enhancement processing on the MAC address pattern area, and extracting MAC address information using optical character recognition technology or barcode recognition technology.

[0048] In practice, the construction and training of the second neural network model focuses on the detection of a specific, non-physical target—the "MAC address pattern scene." This model does not recognize a single object, but is trained to recognize a composite pattern composed of visual elements such as text, barcodes (1D), and QR codes in a specific layout. This pattern typically appears on device labels or nameplates to carry MAC address information.

[0049] The specific implementation process is as follows: Model Task Definition and Training: The training dataset for the second neural network model (e.g., a detection model improved using an attention mechanism) consists of a large number of close-up images of device labels. Instead of labeling individual characters or barcodes, these images are labeled as a single "scene" target, encompassing the entire region containing complete MAC address information (whether in text, barcode, QR code, or a combination thereof). This allows the model to learn to recognize common features of such information patterns (e.g., specific text and graphic arrangements, high-contrast areas, common locations on labels), enabling robust localization of "MAC address pattern regions" from complex backgrounds or damaged labels, even when the region's visual representation is diverse.

[0050] Image sharpening and enhancement processing: After the model locates the "MAC address pattern region" and crops out the sub-image, the system immediately performs a series of image sharpening and enhancement processes on it. This is typically an automated preprocessing pipeline, which may include, but is not limited to: Perspective correction: Corrects label distortion caused by shooting angle; Rotate and straighten: Ensure the text or barcode is horizontal; Adaptive contrast and brightness adjustment: Enhances the distinction between foreground (text / graphics) and background; Denoising and sharpening: Reduce image noise and enhance edge features; Super-resolution reconstruction (optional): For extremely blurry small images, a lightweight super-resolution model can be used to improve the resolution.

[0051] The core purpose of these processes is to maximize the success rate of subsequent recognition engines by transforming on-site images of varying quality into standardized, clear images suitable for machine recognition.

[0052] Multimodal information extraction technology: Based on the features of the sharpened image, the system intelligently selects or calls the corresponding recognition engine in parallel. Optical Character Recognition (OCR) technology: used to extract MAC addresses from plain text or printed numbers and letters. It employs an OCR engine specifically trained for hardware label fonts to improve recognition rates for potential issues such as blurry printing, glare, and character overlap.

[0053] Barcode recognition technology: used to decode MAC addresses in one-dimensional or two-dimensional barcode format. It utilizes a mature barcode parsing library to handle common formats such as Code 128 and QR Code.

[0054] The system design has simple decision-making logic, such as prioritizing barcode recognition (because it is more fault-tolerant), and switching to OCR recognition if it fails; or automatically determining which technology to use based on the morphological features of the region image.

[0055] This specific implementation cleverly solves the challenge of automatically and reliably extracting key equipment identification information (MAC addresses) from unstructured site photos through a combined technical approach of "scene detection + enhanced preprocessing + multimodal recognition." It overcomes the low recognition rate caused by unclear target areas and poor image quality when directly applying general OCR / scanning technologies to the entire image. This provides accurate and robust data input for the automated comparison with the pre-set database in step S104, enabling anti-counterfeiting verification of equipment identity, and is the core technical link in this solution for achieving automated management and control of construction materials.

[0056] Furthermore, the evaluation of whether the installation location meets the preset specifications based on feature detection and monocular ranging algorithms includes: The shooting focal length is obtained based on the EXIF ​​information of the first type of image; Identify the pixel dimensions of the FTTR service-related equipment and at least one reference object in the first type of image; Based on the known actual size of the reference object, the shooting focal length, and the pixel size, calculate the first distance between the reference object and the shooting point, and the second distance between the FTTR service-related equipment and the shooting point. Based on the difference between the first distance and the second distance, it is determined whether the actual distance between the FTTR service-related equipment and the reference object meets the preset distance threshold requirement.

[0057] The step of evaluating whether the installation location meets the preset specifications based on feature detection and monocular ranging algorithms includes the following specific implementation process: Parameter acquisition, the parameters include: Shooting focal length (f): The camera's focal length parameter is read directly from the EXIF ​​(Exchangeable Image File Format) information of the first type of image. This parameter is an inherent property of the camera lens and is usually automatically recorded in the image file by the phone or camera during shooting.

[0058] Pixel size: The first type of image is analyzed using a third neural network model (or combined with the first neural network model) to detect and locate the FTTR service-related equipment (such as optical modems) to be evaluated, as well as selected reference objects (such as refrigerators, televisions, another optical modem, or objects of standard size such as socket panels). The model outputs their bounding boxes in the image, and their pixel size (such as pixel height h_device, h_reference) can be obtained from the height or width pixel values ​​of the bounding boxes.

[0059] Actual dimensions of the reference object: The known actual dimensions of the reference object (such as the actual height H_reference) are derived from a pre-set database of general object dimensions (for example, the height of a standard double-door refrigerator is usually 1.6 meters), or obtained by querying the equipment model specification database after identifying the specific model of the reference object through the equipment identification model. For the equipment itself related to FTTR services, its standard dimensions (such as the height of the optical modem H_device) can also be obtained from the equipment specification library.

[0060] Distance calculation: Based on the principle of pinhole camera model, the distances from the reference object and the device to be evaluated to the camera shooting point are calculated respectively.

[0061] Distance from the reference object to the camera (D_reference): D_reference = f×H_reference / h_reference; Distance from the device to be evaluated to the camera (D_device): D_device = f×H_device / h_device; In the formula, the focal length f and the actual size H must use the same physical unit (such as millimeters), and the calculated distance D is a value in the same unit.

[0062] Distance difference estimation and compliance assessment: In a typical construction photography scenario where the shooting point remains constant and the camera's optical axis is approximately horizontal, the actual horizontal distance difference (ΔD) between the device being evaluated and the reference object can be approximated by the absolute difference in their distances to the camera: ΔD≈|D_reference-D_device| The system compares the calculated ΔD with the distance threshold in the preset specifications. For example, if the specifications require the optical modem to be at least 1.2 meters away from the refrigerator, then it is considered compliant if ΔD ≥ 1.2 meters, otherwise it is not compliant. For the evaluation of installation height, the ground or tabletop can be selected as the reference surface. The pixel distance from the bottom of the device to the reference surface is calculated and converted into the actual height using the principle of similarity.

[0063] Implementation Considerations and Enhancement Strategies: To improve ranging accuracy, strategies such as averaging multiple reference objects and using known geometric relationships in the image (such as corner lines and floor tiles) for auxiliary correction can be adopted in actual deployment.

[0064] The system can be designed with a verification mechanism. For example, if the calculated distance value exceeds a reasonable range (such as less than 0 or abnormally large), the image is determined to be unsuitable for distance assessment and a re-capture or manual processing is required.

[0065] This specific implementation provides a practical distance quantification assessment method based on a single on-site environmental photograph, without the need for additional depth sensors. It cleverly applies the monocular ranging principle from computer vision to the specific field of communication construction quality inspection, achieving automated, objective, and digital inspection of spatial distance requirements in installation specifications. This solves the long-standing problems of inaccurate judgments and inconsistent standards caused by reliance on manual visual estimation.

[0066] Furthermore, the reference objects include high-power electrical appliances, walls, furniture, or other FTTR business-related equipment.

[0067] To achieve a comprehensive and flexible assessment of the installation locations of FTTR-related equipment, the system selects and utilizes diverse and practical references, mainly covering the following categories, each serving a specific compliance inspection dimension. The specific implementation process and reference utilization strategy are as follows: High-power appliances are used as reference points: primarily to evaluate the "keep away from interference sources" requirement in installation specifications. Common high-power appliances used as reference points include refrigerators, microwave ovens, electric ovens, and air conditioner indoor units.

[0068] Implementation: When analyzing the first type of image, the third neural network model simultaneously detects appliances of that type. The system retrieves typical dimensions (such as the standard height and width of a refrigerator) from a pre-set library of general appliance dimensions, or obtains precise dimensions through appliance model identification. By calculating the distance difference ΔD between the FTTR device (such as an optical modem) and the reference object of the appliance, and comparing it with a preset safe distance threshold (e.g., requiring a distance of >1.2 meters from the refrigerator), the system automatically determines whether it complies with anti-interference specifications.

[0069] Walls as a reference point: Primarily used to assess the degree of proximity of equipment installation to the wall or the extent of obstruction. As a large planar reference point, walls help determine whether the equipment is excessively obstructed.

[0070] Implementation: The planar area of ​​the wall is determined through image recognition. By analyzing the relative pixel positions of the device's bounding box and the wall edge, combined with perspective principles, it is possible to qualitatively or semi-quantitatively determine whether the device is surrounded by multiple walls (such as when installed inside a closed low-voltage box). Although accurately measuring the distance to the wall is complex, it is possible to determine whether the device is installed close to the wall by correlating its position and depth in the image with its calculated distance D_device from the camera (in which case D_device is approximately equal to the distance from the camera to the wall).

[0071] Furniture as a reference point: primarily used to assess the installation height or relative position of equipment. Common furniture includes TV cabinets, desks, and sofas.

[0072] Implementation: For example, determining whether the optical modem is installed within the specified height range of 0.2-1.5 meters from the ground. This can be done by comparing the device's vertical position at the pixel level with nearby furniture of known height (such as a standard-height TV cabinet countertop) and using monocular ranging principles for conversion, thereby estimating the device's height from the ground. Typical furniture dimensions can be obtained from the built-in furniture size template library.

[0073] Other FTTR service-related equipment serves as a reference: mainly used to evaluate whether the installation spacing between multiple FTTR devices in the same scenario meets the specifications (e.g., the master and slave optical modems should not be too close).

[0074] Implementation: When the first neural network model identifies multiple FTTR devices (such as a master optical modem and a slave optical modem) in the same image, the system uses them as reference points. Using their known standard dimensions and calculated distances to the camera, D_device1 and D_device2, the horizontal distance between them is approximated using the formula ΔD=|D_device1-D_device2|. The system then checks if the distance is less than the minimum distance threshold between devices (e.g., >3 meters) to avoid signal interference.

[0075] In actual operation, the system will automatically select the most suitable combination of reference objects based on the image content. For example, in an image containing a modem, a refrigerator, and a TV cabinet, the system may simultaneously calculate the distance between the modem and the refrigerator (to assess interference), the vertical relative position of the modem and the TV cabinet (to assess height), and output multiple compliance results.

[0076] Furthermore, the preset specifications include: the distance between the FTTR service-related equipment and the interference source should be greater than 1.2 meters, the installation height should be between 0.2 meters and 1.5 meters, and the installation distance between multiple FTTR service-related devices in the same room should be greater than 3 meters.

[0077] The preset specifications are specified as a series of quantifiable and automatically judged threshold parameters, and integrated into the rule engine of the quality inspection system. These thresholds are formulated based on the technical characteristics of FTTR devices, the laws of wireless signal propagation, and best practices in typical home environments, forming objective standards for automated evaluation.

[0078] For example, regarding the distance specification to the interference source (>1.2 meters): Based on the aforementioned monocular ranging algorithm, the system calculates the estimated distance ΔD between FTTR service-related equipment (mainly wireless transmitting equipment, such as the main optical modem and WiFi equipment) and identified high-power electrical reference objects (such as refrigerators and microwave ovens). The rule engine then compares ΔD with the 1.2-meter threshold.

[0079] Judgment logic: If ΔD ≥ 1.2 meters, the item is judged as "qualified"; if ΔD < 1.2 meters, it is judged as "unqualified", and the quality inspection conclusion will indicate "The equipment is too close to [a certain electrical appliance], posing a risk of signal interference". This threshold aims to reduce the potential interference of electromagnetic radiation from high-power electrical appliances on WiFi signals.

[0080] Installation height specifications (0.2 meters to 1.5 meters): The assessment of installation height is also based on the monocular ranging principle, but the reference object is usually chosen as the ground or its extension (such as a floor), or by identifying the surface of the furniture on which the device is installed (such as a tabletop) as a reference. The system calculates the vertical pixel offset of the bottom of the device relative to this reference surface, and combines the distance from the device to the camera D_device and the geometric relationship to estimate its actual height above the ground H_install.

[0081] Judgment Logic: The rule engine checks whether H_install satisfies 0.2m ≤ H_install ≤ 1.5m. This range is designed to prevent the equipment from being easily affected by moisture and dust due to being placed too low (e.g., on the ground), or from being placed too high (beyond the normal line of sight and operating range), which would affect heat dissipation, aesthetics, and ease of maintenance. If the height exceeds this range, it is determined to be unqualified.

[0082] Equipment installation distance specification (>3 meters): When multiple FTTR devices of the same type or requiring collaborative operation are identified in the same Class I image (e.g., a master optical modem and one or more slave optical modems), the system uses them as reference points. Using their known dimensions and distances to the camera (D1, D2...), the horizontal spacing between them is approximately estimated by calculating the distance difference ΔD_between = |D1-D2| between any two.

[0083] Judgment Logic: The rule engine requires that the ΔD_between distance between any two such devices within the same room (determined by image scene or work order information) should be greater than 3 meters. This threshold is mainly used to prevent multiple WiFi transmitters from being too close together, causing co-channel interference and affecting wireless roaming experience and overall network performance. If the distance does not meet the requirement, the device spacing is deemed unqualified.

[0084] System integration and comprehensive evaluation: In the actual quality inspection process, the system performs parallel or sequential checks on all applicable specifications mentioned above. Each check generates a sub-conclusion (pass / fail). The final comprehensive quality inspection conclusion will summarize all sub-conclusions. Typically, a "failure" in any critical specification (such as equipment identity verification) can directly lead to an overall quality inspection conclusion of "fail".

[0085] Furthermore, the first type of image is captured at a distance of not less than 1.5 meters, and the second type of image is a close-up image containing identification information on the bottom, back, or side of the device.

[0086] To ensure the effectiveness of subsequent AI model recognition and analysis, clear and operable specific acquisition specifications were proposed for the first and second types of images taken on-site by installation and maintenance personnel. These specifications are ensured to be implemented through shooting guidance functions (such as viewfinder prompts, distance detection, and example images) integrated into the mobile terminal APP.

[0087] Shooting guidelines for Category 1 images (environmental images): Core requirement: The shooting distance for Category 1 images is generally required to be no less than 1.5 meters. This requirement is assisted by the app using the phone's proximity sensor or by providing distance reference lines on the shooting interface.

[0088] Purpose and Effects: This distance specification ensures that the image covers a sufficiently wide field of view, thus fully demonstrating the global installation location of the FTTR device in the user's room, its orientation, and its spatial relationship with surrounding key environmental elements (such as appliances, furniture, and walls). Sufficient distance avoids narrow field of view and excessive perspective distortion caused by close-up shooting, providing high-quality input images for subsequent environmental detection and monocular vision-based spatial relationship evaluation (such as distance measurement) by the third neural network model. It effectively solves problems such as incomplete environmental information, missing distance measurement benchmarks, or excessive errors caused by excessively close shooting distances.

[0089] Shooting guidelines for the second type of image (logo image): Core requirement: The second type of image must be a close-up image, and its content must clearly include the specific area on the device body that carries the logo information. Typically, these areas are located on the bottom, back, or side of the device, depending on the device model. The APP can intelligently suggest the parts to be photographed based on the device type.

[0090] Purpose and Effect: Close-up photography ensures that key information on nameplates and labels (such as MAC addresses, serial numbers, model numbers, and QR codes) has sufficiently high pixel resolution and clarity. This directly meets the second neural network model's requirement for high-precision positioning of the "MAC address pattern area" and provides high-quality source images for subsequent OCR or barcode recognition. Clear location guidance (bottom / back / side) solves the installation and maintenance personnel's confusion about "where to photograph," ensuring that what is photographed is what is needed, and guaranteeing the feasibility of the automated equipment identity verification process from the source.

[0091] This disclosure, through defining acquisition specifications for first-class (environmental) and second-class (identification) images and connecting three dedicated neural network models, constructs a complete automated quality inspection process for FTTR installation. This solution achieves accurate identification and operator differentiation of communication equipment, completes automated anti-counterfeiting verification based on MAC address patterns, and utilizes monocular ranging technology to quantitatively assess compliance of key indicators such as installation distance and height. This transforms the traditional, manual, subjective, and inefficient quality inspection model into a highly efficient, objective, and standardized intelligent quality inspection closed loop, significantly improving construction quality control capabilities and operation and maintenance efficiency.

[0092] Example 2

[0093] Embodiment 2 of this disclosure provides an AI-based image recognition-based quality inspection method for FTTR installation specifications. It primarily enables image recognition of professional equipment involved in FTTR services within the telecommunications industry, including master / slave optical modems, IPTV, OTT, indoor security cameras, outdoor security cameras, smart doorbells, splitters, memory cards, and Wi-Fi. By introducing transposed convolution algorithms into each layer of existing traditional neural network models, it achieves effective recognition of long-distance, small targets, thus solving the problems of ineffective differentiation between similar devices from different operators, and between IPTV and OTT devices, which appear similar. Furthermore, by combining feature detection and monocular ranging techniques, it calculates the distance differences between different objects in the image, thereby enabling automatic image checks for obstructions around the optical modem and interference from high-power electrical appliances; it also automatically checks whether multiple optical modems are installed too close together.

[0094] The present invention discloses three neural network models: (1) a neural network model for detecting targets of related equipment involved in FTTR service; (2) a neural network model for detecting targets of MAC address patterns of various equipment in FTTR service; and (3) a quality inspection model for the installation of optical modems in FTTR service.

[0095] The construction steps for the above three models are largely the same, and the specific steps are as follows: Quality Inspection / Photo Capture Standards: To better achieve quality inspection of FTTR service construction, it is necessary to formulate corresponding standards, including which types of photos need to be taken for FTTR service quality inspection, the shooting angle and distance for each type of photo, what the problem characteristic points of FTTR service quality inspection are, and the manifestation of problem characteristic points.

[0096] Model design: Select from existing open-source neural network models that are highly suitable for this business scenario, and then tailor and improve them for different scenarios of FTTR business.

[0097] Dataset Construction: Collect a large number of photos uploaded from the FTTR construction site, carry out sample screening, image preprocessing, target feature labeling and other work to construct a sample dataset for model training.

[0098] Model training and optimization: After multiple iterations, an application model for the corresponding scenario is trained.

[0099] Model deployment and application: The model is implemented by linking the three models together.

[0100] like Figure 2 As shown, the application and implementation process of FTTR installation specification quality inspection based on AI image recognition using three models includes: First, Model 1 is used to identify whether there are any devices required for FTTR services in the photo (including master / slave optical modems, splitters, IPTV devices, OTT devices, indoor security cameras, outdoor security cameras, smart doorbells, memory cards, WIFI devices, etc.), and whether the photo is open enough to clearly show the construction site situation.

[0101] Then, Model 2 is used to identify the MAC address pattern in the photo, and the MAC address pattern area is extracted. After high-definition processing, it is handed over to the AI ​​barcode scanning / OCR model to identify the corresponding MAC address value, and compared with the MAC value in the background database to determine whether the equipment installed on site is the equipment issued from the warehouse (to prevent the substitution of old for new or inferior equipment); including whether the optical modem panel has a QR code for reporting repairs.

[0102] Finally, Model 3 is used to identify whether the installation process and environment of the FTTR equipment are compliant. The identifiable items include: whether the optical modem is blocked on more than three sides, thus affecting the propagation of the wireless signal; whether there are high-power electrical appliances / televisions / monitors or other facilities around the optical modem that interfere with it; whether the installation height of the optical modem meets the requirement of 0.2~1.5 meters; whether multiple optical modems are placed too close together; and whether the installation environment of the optical modem meets the requirements of the design specifications.

[0103] (1) Equipment identification in FTTR related professional fields

[0104] Technical requirements for AI models: They must not only identify the various types of equipment involved in FTTR services in the telecommunications industry, but also distinguish the different operators (China Unicom, China Telecom, China Mobile) affiliation of similar equipment, and only identify equipment of the target operator, while not identifying equipment of other operators. Figure 3 Examples of optical modems from different operators.

[0105] from Figure 3As can be seen, apart from the operator's logo, the rest of the devices and their appearance are completely identical. For close-up photos, the operator's logo can be identified by observing the relationship between the optical modem's outline and the logo. However, due to certain requirements regarding the shooting distance (no less than 1.5 meters), the operator's logo cannot be identified in photos taken at a distance (the target pixels are too small and blurry). See [link / reference needed]. Figure 4 As shown.

[0106] The details of the operator's logo in the image have disappeared, leaving only the outline. This makes it impossible for the model's ordinary convolution algorithm to characterize the logo's features, thus failing to effectively identify the logos of different operators. The solution is to introduce a transposed convolution mechanism to increase the resolution of the feature map, thereby amplifying the weights of the smaller logo features and improving the model's robustness.

[0107] Transposed convolution, also known as deconvolution, is a type of upsampling. It fills the feature map with zeros to increase its resolution.

[0108] In the specific implementation process, this embodiment changes each layer of the selected neural network model from a regular convolutional layer to a transposed convolutional layer. The algorithm comparison between the two is shown in Table 1 below: Parameter description: W: Width of the input feature map; H: Height of the input feature map; K: kernel size, the width and height of the convolution kernel; P: padding (the number of zeros that need to be padded in the feature map); S: stride; width_out: The width of the output feature map after convolution; height_out: The height of the output feature map after convolution; Table 1: Comparison of Ordinary Convolution and Transposed Convolution

[0109] In the table above, this embodiment transforms the original ordinary convolution algorithm into a transposed convolution algorithm, which increases the resolution of the operator's logo image features and improves the recognition rate of small targets at a distance.

[0110] The same technical method can also be used to effectively distinguish between IPTV and OTT devices, which look similar. See Figure 6 As shown, the transposed convolution mechanism effectively distinguishes devices that look similar.

[0111] (2) Technical solution for determining whether there is interference from high-power electrical appliances near the optical modem

[0112] FTTR equipment installation specifications require that the optical modem be installed away from high-power electrical appliances. Refrigerators, microwave ovens, and electric ovens are all considered high-power appliances. Figure 7 There are two optical modems. The upper optical modem is very close to the refrigerator, which is considered non-compliant installation. The lower optical modem is about 1.4 meters away from the refrigerator, which is considered compliant installation.

[0113] This problem was solved by using a combination of feature detection and monocular ranging technology to determine the distance between the optical modem and high-power appliances from photographs.

[0114] Basic principle: When the actual dimensions (such as height or width) of one of the objects are known, the distance can be estimated using pixel ratios. For example, using the camera's focal length f (which can be extracted from the EXIF ​​attribute information of the uploaded photo) and the object's pixel height h in the image, the distance D from the object to the camera can be calculated as D = f × H / h (where H is the actual height). Then, the difference between the estimated distances of the two objects is calculated; if the difference is less than a certain threshold, it indicates that the distance is relatively short.

[0115] by Figure 7 For example, estimate the distance between the two optical modems and the refrigerator.

[0116] Basic parameters: The actual height of the optical modem is approximately 160mm, the actual height of the double-door refrigerator is 1600mm (average value based on experience), the focal length extracted from the photo's EXIF ​​attribute information is 5mm, the pixel height of the upper optical modem is 140, the pixel height of the lower optical modem is 185, and the pixel height of the refrigerator is 1403. Based on these parameters, the following calculations are performed: The distance between the optical modem and the camera is D1 = 5 × 160 / 140 = 5.71. The distance between the optical modem and the camera is D2 = 5 × 160 / 185 = 4.32. The distance between the refrigerator and the camera, Dr = 5 × 1600 / 1403 = 5.70; The calculated distance difference between the optical modem and the refrigerator is |5.71-5.70|=0.01. The distance to the refrigerator is close to 0, indicating an improper installation. The calculated distance difference between the optical modem and the refrigerator is |4.32-5.70|=1.38. The distance to the refrigerator is >1.2 meters, so the installation is qualified.

[0117] (3) Determine whether the installation locations of multiple optical modems are too close together.

[0118] The method combining feature detection and monocular ranging used in this embodiment is also applicable to compliance assessment of whether multiple optical modems are installed too close together. Still using... Figure 7For example, the calculated distance difference between the two optical modems in the diagram is |5.71-4.32|=1.39. The installation distance is less than 3 meters (in the same room). The installation distance between multiple optical modems is too close, which is not up to standard.

[0119] Example 3

[0120] Embodiment 3 of this disclosure provides an FTTR installation specification quality inspection system based on AI image recognition, such as... Figure 8 As shown, the system includes: The acquisition module 11 is configured to acquire images to be inspected at the FTTR installation site. The images to be inspected include at least: a first type of image for displaying the equipment installation environment and a second type of image for displaying equipment identification information. The judgment module 12 is configured to use a first neural network model to identify the first type of image and determine whether the image contains a preset FTTR service-related device. The identification module 13 is configured to select the second type of image from the image to be inspected if the judgment module identifies that the device contains the FTTR service-related device, and use the second neural network model to identify it and locate the MAC address pattern area on the device. The MAC address verification module 14 is configured to perform image processing and recognition on the MAC address pattern area, obtain the MAC address information of the device, and compare the MAC address information with a preset database to verify the compliance of the device. The installation location detection module 15 is configured to use a third neural network model to identify the first type of image, detect the installation environment of the FTTR service-related equipment, and evaluate whether the installation location meets the preset specifications based on feature detection and monocular ranging algorithm. The preset specifications include at least one or more of the following: distance threshold from interference source, installation height range, and minimum spacing between equipment. The generation module 16 is configured to generate a quality inspection conclusion for the FTTR installation specification based on the results of equipment compliance and whether the installation location meets the preset specifications.

[0121] Furthermore, the first neural network model replaces the standard convolutional layer with a transposed convolutional layer to enhance the ability to recognize distant, small-sized target features in images, thereby distinguishing between similar devices from different operators and devices with similar appearances.

[0122] Furthermore, the second neural network model is trained to recognize MAC address pattern scenarios consisting of one or more combinations of text, barcodes, and QR codes; The MAC address verification module 14 is specifically configured to perform image enhancement processing on the MAC address pattern area and extract MAC address information using optical character recognition technology or barcode recognition technology.

[0123] Furthermore, the installation position detection module 15 is specifically configured as follows: The shooting focal length is obtained based on the EXIF ​​information of the first type of image; Identify the pixel dimensions of the FTTR service-related equipment and at least one reference object in the first type of image; Based on the known actual size of the reference object, the shooting focal length, and the pixel size, calculate the first distance between the reference object and the shooting point, and the second distance between the FTTR service-related equipment and the shooting point. Based on the difference between the first distance and the second distance, it is determined whether the actual distance between the FTTR service-related equipment and the reference object meets the preset distance threshold requirement.

[0124] Furthermore, the reference objects include high-power electrical appliances, walls, furniture, or other FTTR business-related equipment.

[0125] Furthermore, the preset specifications include: the distance between the FTTR service-related equipment and the interference source should be greater than 1.2 meters, the installation height should be between 0.2 meters and 1.5 meters, and the installation distance between multiple FTTR service-related devices in the same room should be greater than 3 meters.

[0126] Furthermore, the first type of image is captured at a distance of not less than 1.5 meters, and the second type of image is a close-up image containing identification information on the bottom, back, or side of the device.

[0127] The AI ​​image recognition-based FTTR installation specification quality inspection system of this disclosure is used to implement the AI ​​image recognition-based FTTR installation specification quality inspection method in Embodiment 1 and Embodiment 2. Therefore, the description is relatively simple. For details, please refer to the relevant descriptions in the previous method embodiments, which will not be repeated here.

[0128] Figure 9 This is a block diagram of an electronic device provided in Embodiment 4 of this disclosure.

[0129] Reference Figure 9This disclosure provides an electronic device comprising: at least one processor 701; at least one memory 702; and one or more I / O interfaces 703 connected between the processor 701 and the memory 702; wherein the memory 702 stores one or more computer programs executable by the at least one processor 701, the one or more computer programs being executed by the at least one processor 701 to enable the at least one processor 701 to perform the aforementioned FTTR installation specification quality inspection method based on AI image recognition.

[0130] This disclosure also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the aforementioned AI image recognition-based FTTR installation specification quality inspection method. The computer-readable storage medium may be volatile or non-volatile.

[0131] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code. When the computer-readable code is run in the processor of an electronic device, the processor in the electronic device executes the above-described FTTR installation specification quality inspection method based on AI image recognition.

[0132] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0133] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable program instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0134] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0135] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0136] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0137] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0138] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0139] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0140] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0141] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in connection with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in connection with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. An AI image recognition-based FTTR installation specification quality inspection method, characterized in that, The method includes: Acquire images of the Fiber to the Room (FTTR) installation site to be inspected, wherein the images to be inspected include at least: a first type of image for displaying the equipment installation environment, and a second type of image for displaying equipment identification information; The first neural network model is used to identify the first type of image and determine whether the image contains preset FTTR service-related equipment. If a device related to the FTTR service is identified, the second type of image is selected from the images to be inspected, and the second neural network model is used to identify it and locate the media access control MAC address pattern area on the device. The MAC address pattern area is processed and recognized to obtain the device's MAC address information, and the MAC address information is compared with a preset database to verify the device's compliance. The first type of image is identified using a third neural network model to detect the installation environment of the FTTR service-related equipment. Based on feature detection and monocular ranging algorithms, the installation location is evaluated to determine whether it meets the preset specifications. The preset specifications include at least one or more of the following: distance threshold from interference sources, installation height range, and minimum spacing between equipment. Based on the results of equipment compliance and whether the installation location meets the preset specifications, a quality inspection conclusion for the FTTR installation specification is generated.

2. The method of claim 1, wherein, The first neural network model replaces the standard convolutional layer with a transposed convolutional layer to enhance the ability to recognize features of distant, small targets in images, so as to distinguish between similar devices from different operators and devices with similar appearances.

3. The method of claim 1, wherein, The second neural network model was trained to recognize MAC address pattern scenarios consisting of one or more combinations of text, barcodes, and QR codes; The image processing and recognition of the MAC address pattern area includes: performing image enhancement processing on the MAC address pattern area, and extracting MAC address information using optical character recognition technology or barcode recognition technology.

4. The method of claim 1, wherein, The algorithm based on feature detection and monocular ranging is used to evaluate whether the installation location meets the preset specifications, including: The shooting focal length is obtained based on the EXIF ​​information of the exchangeable image file format of the first type of image; Identify the pixel dimensions of the FTTR service-related equipment and at least one reference object in the first type of image; Based on the known actual size of the reference object, the shooting focal length, and the pixel size, calculate the first distance between the reference object and the shooting point, and the second distance between the FTTR service-related equipment and the shooting point. Based on the difference between the first distance and the second distance, it is determined whether the actual distance between the FTTR service-related equipment and the reference object meets the preset distance threshold requirement.

5. The method of claim 4, wherein, The reference objects include high-power electrical appliances, walls, furniture, or other equipment related to FTTR operations.

6. The method according to claim 1, characterized in that, The preset specifications include: the distance between the FTTR service-related equipment and the interference source should be greater than 1.2 meters, the installation height should be between 0.2 meters and 1.5 meters, and the installation distance between multiple FTTR service-related devices in the same room should be greater than 3 meters.

7. The method according to claim 1, characterized in that, The first type of image is captured at a distance of not less than 1.5 meters, and the second type of image is a close-up image containing markings on the bottom, back, or side of the device.

8. A quality inspection system for FTTR installation specifications based on AI image recognition, characterized in that, The system includes: The acquisition module is configured to acquire images of the FTTR installation site to be inspected, the images to be inspected including at least: a first type of image for displaying the equipment installation environment, and a second type of image for displaying equipment identification information; The judgment module is configured to use a first neural network model to identify the first type of image and determine whether the image contains a preset FTTR service-related device. The identification module is configured to select the second type of image from the images to be inspected if the judgment module identifies that the device contains the FTTR service-related device, and use the second neural network model to identify it and locate the MAC address pattern area on the device. The MAC address verification module is configured to perform image processing and recognition on the MAC address pattern area to obtain the device's MAC address information, and compare the MAC address information with a preset database to verify the device's compliance. The installation location detection module is configured to use a third neural network model to identify the first type of image, detect the installation environment of the FTTR service-related equipment, and evaluate whether the installation location meets the preset specifications based on feature detection and monocular ranging algorithm. The preset specifications include at least one or more of the following: distance threshold from interference source, installation height range, and minimum spacing between devices. The generation module is configured to generate a quality inspection conclusion for the FTTR installation specification based on the results of equipment compliance and whether the installation location meets the preset specifications.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores one or more computer programs that can be executed by the at least one processor, and the one or more computer programs are executed by the at least one processor to enable the at least one processor to perform the FTTR installation specification quality inspection method based on AI image recognition as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the FTTR installation specification quality inspection method based on AI image recognition as described in any one of claims 1-7.

11. A computer program product, characterized in that, Includes computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the FTTR installation specification quality inspection method based on AI image recognition as described in any one of claims 1-7.

Images