Data cable adaptive production method and system based on image analysis

Abstract

The invention discloses a data cable self-adaptive production method and system based on image analysis, and the method specifically comprises the steps: synchronously collecting three-mode image data containing visible light, infrared light and polarization at a gas injection section, an extrusion section and a molding section of a data cable through a multispectral imaging unit; performing spatial alignment on the three-mode image data by adopting a sub-pixel registration algorithm to obtain standard image data; based on the standard image data, combined diagnosis is carried out on the cable gas injection structure, the insulation layer quality and the surface defect through a multi-task analysis engine, and a defect diagnosis result is obtained; and based on a defect diagnosis result, dynamically adjusting the traction speed, the extrusion temperature and the gas injection pressure by utilizing a fuzzy PID controller optimized by reinforcement learning to form online process parameter closed-loop control. The defects of single function, static detection, high data dependence and the like of a traditional data cable production detection method are effectively overcome, and a more efficient and intelligent solution is provided for data cable production.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and in particular to an image analysis-based adaptive production method and system for data cables. Background Art

[0002] In the data cable manufacturing industry, ensuring product quality and production efficiency is a core element of sustainable development. As a crucial carrier of information transmission, the quality of data cables is directly related to the stability and reliability of data transmission, which in turn affects the normal operation of various electronic devices. However, traditional data cable production and testing methods have gradually exposed numerous shortcomings in response to increasingly complex and diverse production demands. These shortcomings have seriously restricted the automation level of the production process and the improvement of product quality.

[0003] First, traditional testing methods have significant functional limitations. These methods often only test a specific aspect of data cables, such as focusing solely on the cable's structural integrity or detecting surface defects like scratches and cracks. This single-function testing approach makes it difficult to comprehensively and accurately assess the overall cable quality during production. The inability to promptly detect and correct other potential quality issues, such as uneven insulation thickness and internal bubbles, significantly increases the risk of product failure. Once these issues become apparent during subsequent use, they not only damage the company's reputation but can also lead to customer complaints and product returns, resulting in significant financial losses.

[0004] Secondly, many traditional testing methods rely on static testing, meaning they are performed offline after cable production is complete. This approach has significant drawbacks. First, offline testing requires additional testing equipment and facilities, increasing production costs and space requirements. Second, because testing is performed after production is complete, any quality issues discovered often require rework or scrapping of the entire batch, resulting in low production efficiency and significant resource waste. More critically, static testing cannot reflect quality fluctuations during the production process in real time, making it difficult for companies to adjust production parameters and processes in a timely manner, thus failing to effectively prevent quality issues.

[0005] Finally, with the rapid development of smart manufacturing and the Industrial Internet, some data analysis-based inspection methods are gradually being applied to data cable production. These methods collect and analyze large amounts of historical data to train models, enabling intelligent detection of cable quality. However, in actual production, especially when changing production lines or introducing new products, companies often face the problem of insufficient sample data. Limited production data for new specifications or new products makes it difficult to train models with sufficient generalization capabilities, resulting in reduced detection accuracy and difficulty meeting the requirements for high-precision inspection. Furthermore, even with sufficient historical data, data differences between different production lines and equipment can make model transfer difficult, further limiting the application scope of data analysis-based inspection methods.

[0006] In summary, traditional data cable production and testing methods have many defects in terms of functionality, efficiency, and data dependence, and are unable to meet the needs of modern production for high quality, high efficiency, and high flexibility. Summary of the Invention

[0007] The purpose of the present invention is to provide an adaptive production method and system for data cables based on image analysis, which effectively overcomes the defects of traditional data cable production and detection methods such as single function, static detection and strong data dependence, and provides a more efficient and intelligent solution for data cable production to solve at least one of the above-mentioned existing technical problems.

[0008] In a first aspect, the present invention provides a method for adaptively producing data cables based on image analysis, the method specifically comprising:

[0009] The multispectral imaging unit synchronously collects trimodal image data including visible light, infrared and polarization in the gas injection section, extrusion section and forming section of the data cable, and uses the sub-pixel registration algorithm to spatially align the trimodal image data to obtain standard image data;

[0010] Based on standard image data, a multi-task analysis engine is used to jointly diagnose the cable gas injection structure, insulation layer quality, and surface defects to obtain defect diagnosis results.

[0011] Based on the defect diagnosis results, the fuzzy PID controller optimized by reinforcement learning is used to dynamically adjust the pulling speed, extrusion temperature and gas injection pressure to form an online closed-loop control of process parameters;

[0012] When changing production lines, historical process knowledge is transferred through cross-line federated learning and model distillation technology to perform rapid specification switching under few-sample conditions.

[0013] In a second aspect, the present invention provides a data cable adaptive production system based on image analysis, the system specifically comprising:

[0014] The first production module is used to synchronously collect trimodal image data including visible light, infrared light, and polarization in the gas injection, extrusion, and molding sections of the data cable through a multispectral imaging unit, and spatially align the trimodal image data using a sub-pixel registration algorithm to obtain standard image data;

[0015] The second production module is used to perform joint diagnosis of cable gas injection structure, insulation layer quality and surface defects based on standard image data through a multi-task analysis engine to obtain defect diagnosis results;

[0016] The third production module is used to dynamically adjust the pulling speed, extrusion temperature and gas injection pressure based on the defect diagnosis results using a fuzzy PID controller optimized by reinforcement learning, forming an online closed-loop control of process parameters;

[0017] The fourth production module is used to transfer historical process knowledge through cross-line federated learning and model distillation technology when the production line is changed, and to perform rapid specification switching under few-sample conditions.

[0018] In a third aspect, the present invention provides a computer device comprising: a memory and a processor and a computer program stored in the memory, wherein when the computer program is executed on the processor, the adaptive production method for data cables based on image analysis as described in any one of the above methods is implemented.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the adaptive production method for data cables based on image analysis as described in any one of the above methods.

[0020] Compared with the prior art, the present invention has at least one of the following technical effects:

[0021] 1. The present invention effectively overcomes the defects of traditional data cable production and detection methods such as single function, static detection and strong data dependence, and provides a more efficient and intelligent solution for data cable production.

[0022] 2. The present invention uses a multispectral imaging unit to synchronously collect trimodal image data and utilizes a multitasking analysis engine for joint diagnosis, which can comprehensively and accurately evaluate the quality of cables and promptly discover and address potential problems.

[0023] 3. The present invention uses a fuzzy PID controller optimized by reinforcement learning to dynamically adjust the pulling speed, extrusion temperature and injection pressure according to the defect diagnosis results, thereby achieving real-time optimization of production parameters and improving production efficiency and product quality.

[0024] 4. This invention uses cross-line federated learning and model distillation technology to transfer historical process knowledge, and combines the FE-GAN generator to expand small-sample defect data, thereby achieving high-precision defect detection under small-sample conditions and reducing dependence on new sample data.

[0025] 5. The present invention uses a multispectral imaging unit to synchronously collect three-modal image data and adopts a sub-pixel registration algorithm for spatial alignment, thereby realizing multi-dimensional, high-precision image data acquisition of the gas injection section, extrusion section, and molding section of the data cable, providing a reliable data basis for subsequent joint diagnosis and process parameter adjustment.

[0026] 6. The present invention realizes the synchronous acquisition and adaptive exposure time adjustment of trimodal image data through the coordinated work of rotary encoder, FPGA and trimodal camera. Combined with the sub-pixel registration algorithm, it ensures the spatial alignment accuracy of image data and improves the accuracy of subsequent image analysis.

[0027] 7. Based on standard image data, the present invention conducts joint diagnosis of cable gas injection structure, insulation layer quality and surface defects through a multi-task analysis engine, thereby achieving comprehensive and real-time monitoring of data cable quality and improving the efficiency and accuracy of defect diagnosis.

[0028] 8. The present invention detects the gas core distribution by grayscale-gradient-texture entropy fusion feature vector and corrects it by combining the stress distribution of the polarization image, thereby achieving accurate detection of the gas core distribution and providing strong support for the evaluation of the cable gas injection structure.

[0029] 9. The present invention performs spatiotemporal wavelet decomposition on the infrared image sequence and establishes a thickness-heat conduction regression model in combination with laser thickness measurement data, thereby achieving accurate measurement of the insulation layer thickness and analysis of the heat conduction characteristics, providing a scientific basis for the evaluation of the insulation layer quality.

[0030] 10. The present invention deploys an attention network with infrared temperature correction on visible light images and uses the FE-GAN generator to expand the few-sample defect data, achieving high-precision detection of surface defects, especially maintaining high detection performance under few-sample conditions.

[0031] 11. Based on the defect diagnosis results, the present invention uses a fuzzy PID controller optimized by reinforcement learning to dynamically adjust the pulling speed, extrusion temperature and injection pressure, forming an online closed-loop control of process parameters, realizing adaptive adjustment of the production process, and improving production efficiency and product quality.

[0032] 12. When changing production lines, the present invention transfers historical process knowledge through cross-production line federated learning and model distillation technology, thereby achieving rapid specification switching under conditions of few samples, reducing the cost and time of deploying new specification production lines, and improving the flexibility and efficiency of the production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 1 is a flow chart of a method for adaptively producing data cables based on image analysis according to an embodiment of the present invention;

[0035] Figure 2 1 is a schematic structural diagram of a data cable adaptive production system based on image analysis provided by an embodiment of the present invention;

[0036] Figure 3 It is a structural diagram of a computer device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0037] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0038] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0039] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0040] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0041] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0042] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0043] In the embodiments of this application, Figure 1 A flow chart of a method for adaptively producing data cables based on image analysis according to an embodiment of the present invention is shown, and is described in detail as follows:

[0044] S101, using a multispectral imaging unit to synchronously collect trimodal image data including visible light, infrared and polarization in the gas injection section, extrusion section and forming section of the data cable, and using a sub-pixel registration algorithm to spatially align the trimodal image data to obtain standard image data.

[0045] In this embodiment, a three-modal imaging system consisting of a visible light camera, an infrared camera, and a polarization camera is configured to ensure that each camera has high resolution, high sensitivity, and synchronized acquisition capabilities. The multispectral imaging unit is installed above the gas injection, extrusion, and molding sections of the data cable production line, ensuring coverage of the entire cable production process and clear, unobstructed imaging.

[0046] A rotary encoder is configured to acquire real-time cable position signals, ensuring synchronization of multispectral imaging unit acquisition with cable production. An FPGA (field programmable gate array) is configured as a synchronous trigger controller. When a cable enters the inspection area, the FPGA generates a synchronous trigger pulse signal, controlling the tri-modal camera for adaptive exposure time adjustment and synchronous acquisition.

[0047] Configure a high-performance computer or server to receive, process, and store the trimodal image data collected by the multispectral imaging unit. Install image processing software and a sub-pixel registration algorithm library for subsequent spatial alignment.

[0048] The multispectral imaging unit and synchronous trigger and control system are turned on and initialized, including adjusting camera parameters and generating synchronous trigger pulse signals. To ensure normal operation of the cable production line, the cable passes through the gas injection, extrusion, and molding sections. When the cable enters the inspection area, the rotary encoder acquires the cable's position signal in real time and transmits it to the FPGA. The FPGA generates a synchronous trigger pulse signal based on the cable's position signal, controlling the tri-modal camera to adaptively adjust the exposure time, ensuring that visible light, infrared, and polarization images are captured at the same time. The tri-modal camera transmits the captured image data to the data processing and storage device.

[0049] The collected visible light, infrared, and polarization images are preprocessed, including denoising and contrast enhancement, to improve the accuracy of subsequent registration. Feature points, such as SURF (Speeded Up Robust Features), are extracted from each of the visible light, infrared, and polarization images. The extracted feature points are matched to establish feature point correspondences between the visible light, infrared, and polarization images. Taylor expansion corrections are applied to the matched feature points to obtain sub-pixel position corrections, improving registration accuracy. Based on the feature point correspondences and sub-pixel position corrections, the coordinate mapping is calculated using an affine transformation model. The registration results are verified using methods such as normalized mutual information to ensure accuracy. The registered visible light, infrared, and polarization images are fused to generate spatially aligned standard image data.

[0050] S102, based on the standard image data, a multi-task analysis engine is used to jointly diagnose the cable gas injection structure, insulation layer quality and surface defects to obtain defect diagnosis results.

[0051] In this embodiment, a feature extractor based on a convolutional neural network (CNN) is used to extract multi-scale features from standard image data. Specifically, a network structure consisting of multiple convolutional and pooling layers is constructed, using convolution kernels of varying sizes to capture local and global features in the image. For example, 3×3, 5×5, and 7×7 convolution kernels are used to extract feature maps of different scales. Multi-scale features are then fused through concatenation to form a more discriminative feature representation.

[0052] The data cable quality inspection task is broken down into three subtasks: cable gas injection structure inspection, insulation layer quality inspection, and surface defect inspection. Each subtask corresponds to an independent inspection branch, but shares the initial feature extraction process to improve computational efficiency and model generalization.

[0053] In the cable gas injection structure detection branch, extracted feature maps are used to measure gas injection structural characteristics, including gas core distribution characteristics (such as gas core diameter and gas core spacing) and airtightness-related structural parameters (such as gas core wall thickness uniformity and the tightness of the gas core-insulation layer connection). Specialized convolution kernels and pooling layers are designed to enhance and extract cable gas injection structural features. A classifier is also used to determine whether the cable gas injection structure complies with regulations and output the cable gas injection structure detection results.

[0054] The insulation layer quality inspection branch focuses on detecting defects such as thickness uniformity and internal bubbles. By analyzing the grayscale distribution and texture features in the feature map, we determine whether the insulation layer has quality issues. For example, we use the gray-level co-occurrence matrix (GLCM) to extract the insulation layer's texture features, and then use threshold segmentation or classifiers to determine whether the insulation layer is qualified.

[0055] The surface defect detection branch is responsible for detecting scratches, cracks, and other defects on the surface of data cables. By designing specialized defect detection algorithms, such as those based on edge detection and morphological processing, feature maps are further analyzed. Specifically, the Canny edge detection algorithm is used to extract edge information from the image, and morphological opening and closing operations are used to remove noise and fine edges, ultimately resulting in surface defect detection results.

[0056] The features extracted by the three detection branches are fused to form a comprehensive feature representation. This fusion can be done through a simple concatenation operation or a more complex attention mechanism that dynamically adjusts the feature weights based on the importance of each task.

[0057] Based on the fused features, a joint decision-making mechanism is used to assess the overall quality of data cables. Specifically, a multi-task classifier is designed to comprehensively judge the inspection results of the cable's gas injection structure, insulation quality, and surface defects. The classifier can be a multi-output neural network, with each output node corresponding to the result of a specific inspection task. By training this classifier, it can accurately determine the overall quality of the data cable based on the input features.

[0058] Ultimately, the combined diagnostic results are output as structured data, including information on whether the cable's gas injection structure meets specifications, whether the insulation layer has quality issues, whether there are surface defects, and an overall quality assessment of the data cable. These results are used to adjust process parameters and optimize the production process.

[0059] This embodiment enables simultaneous inspection of the data cable's gas injection structure, insulation quality, and surface defects, effectively improving inspection efficiency and accuracy. Furthermore, because this method utilizes multispectral imaging technology and a multi-task analysis engine, it can capture multiple quality characteristics during the data cable production process, providing strong support for production process optimization.

[0060] S103, based on the defect diagnosis results, the fuzzy PID controller optimized by reinforcement learning is used to dynamically adjust the pulling speed, extrusion temperature and gas injection pressure to form an online closed-loop control of process parameters.

[0061] In this embodiment, by combining the defect diagnosis results with the fuzzy PID controller optimized by reinforcement learning, dynamic adjustment of process parameters is achieved, effectively improving the automation level of the production process and product quality.

[0062] Specifically, the defect diagnosis results output by the multi-task analysis engine are analyzed to identify specific defect types, such as abnormal cable gas injection structure, uneven insulation thickness, and surface scratches. Each defect type corresponds to a different quality issue and possible cause. The severity of each identified defect type is further assessed. Defect severity can be categorized into minor, moderate, and severe levels by setting thresholds or employing methods such as fuzzy logic. The severity assessment results serve as an important basis for subsequent process parameter adjustments.

[0063] Build a fuzzy PID controller that includes three control steps: proportional (P), integral (I), and differential (D). It dynamically adjusts the PID parameters using fuzzy logic. The fuzzy PID controller can automatically adjust the PID parameters based on changes in the input signal to adapt to different production conditions.

[0064] To further improve the performance of the fuzzy PID controller, a reinforcement learning algorithm was introduced to optimize it. The reinforcement learning algorithm continuously learns the optimal control strategy through trial and error through interaction with the production environment. Specifically, the process parameter adjustment process is modeled as a Markov decision process (MDP), where the state represents the current defect diagnosis results and production parameters, the action represents the adjustment to the process parameters, and the reward function is designed based on the adjusted product quality and production efficiency. By maximizing the cumulative reward, the reinforcement learning algorithm learns the optimal process parameter adjustment strategy.

[0065] The pulling speed is dynamically adjusted based on defect diagnosis results, particularly those related to the cable's gas injection structure. For example, if an abnormality is detected in the cable's gas injection structure, the pulling speed can be appropriately reduced to reduce stress during the injection process and improve the quality of the cable's gas injection structure. This adjustment strategy is implemented using a fuzzy PID controller, which automatically calculates and outputs the appropriate adjustment for the pulling speed based on the severity of the defect and the current pulling speed.

[0066] Extrusion temperature significantly impacts the thickness and uniformity of the insulation layer. When defect diagnosis reveals issues such as uneven thickness or internal bubbles, the extrusion temperature is dynamically adjusted using a fuzzy PID controller. For example, increasing the extrusion temperature can improve the flowability of the insulation material and help reduce the formation of internal bubbles. The adjustment strategy is also calculated based on the severity of the defect and the current extrusion temperature.

[0067] Injection pressure is a key factor affecting the stability of cable gas injection structures. When abnormalities such as looseness or overtightening are detected, the fuzzy PID controller dynamically adjusts the injection pressure. This adjustment strategy takes into account the characteristics of the cable gas injection structure material, the cable gas injection speed, and the current injection pressure to ensure stability and consistency of the cable gas injection structure.

[0068] During the production process, sensors collect real-time process parameters such as pulling speed, extrusion temperature, and gas injection pressure, as well as image data collected by a multispectral imaging unit. This data serves as input to a fuzzy PID controller, which adjusts process parameters in real time.

[0069] The fuzzy PID controller calculates the optimal process parameter adjustments based on the input defect diagnosis results and current process parameters. It then uses actuators (such as inverters and heaters) to adjust the pulling speed, extrusion temperature, and gas injection pressure in real time. The adjusted process parameters are then fed back into the production process, forming a closed-loop control system.

[0070] The reinforcement learning algorithm continuously learns and optimizes the parameters of the fuzzy PID controller during the closed-loop control process to adapt to different production conditions and defect types. Through continuous trial and error and adjustment, it ultimately achieves optimal control of process parameters, improving the automation level of the production process and product quality.

[0071] In this embodiment, key process parameters such as pulling speed, extrusion temperature, and gas injection pressure can be dynamically adjusted in real time based on defect diagnosis results, effectively improving the automation level of the production process and product quality. Furthermore, the fuzzy PID controller optimized by reinforcement learning has greater adaptability and robustness, capable of handling different production conditions and defect types.

[0072] S104, when changing production lines, migrates historical process knowledge through cross-production line federated learning and model distillation technology to perform rapid specification switching under few-sample conditions.

[0073] In this embodiment, in the data cable production process, production line changes and new product introductions are common production needs. However, when faced with production line changes, traditional methods often require re-collecting a large amount of sample data to train new detection or control models, which is not only time-consuming and labor-intensive, but may also cause production interruptions. In addition, data differences between different production lines and different equipment may also make model migration difficult, limiting the flexibility and efficiency of the production line. The cross-line federated learning and model distillation technology proposed in this embodiment can effectively solve the above problems and achieve rapid specification switching under conditions of few samples.

[0074] Specifically, federated learning nodes are deployed on each production line, with each node responsible for collecting and processing production data for that line. This data includes trimodal image data collected by the multispectral imaging unit, process parameter records, and defect diagnosis results. A secure and reliable communication protocol is established to ensure data exchange and model updates between federated learning nodes. The communication protocol should utilize encryption technology to protect the security and privacy of data during transmission. A base model is initialized on each federated learning node. This model can be a pretrained convolutional neural network (CNN) or other model suitable for image analysis and quality inspection. The structure and parameters of the base model remain consistent across all nodes.

[0075] The local model updates of each federated learning node are aggregated on a central server periodically (e.g., daily or after each batch of production). The aggregation process can use weighted averaging or other federated learning algorithms to integrate the production experience and knowledge of each production line. The aggregated global model is distributed back to each federated learning node as the starting point for the next round of local training. In this way, each production line can share the production experience and knowledge of other production lines, accelerating model convergence and optimization. During the federated learning process, feature representations of historical data are extracted. These features can include characteristics of cable gas injection structure, insulation layer quality, surface defects, etc. The feature extraction process should be consistent to facilitate migration and comparison between different production lines.

[0076] When a production line is changed, a fully trained, high-performing global model is selected as the teacher model. The teacher model incorporates a wealth of historical process knowledge and production experience. A student model is initialized for the new production line or new product. The student model's structure can be similar to the teacher model, but the parameters need to be reinitialized. The teacher model is used to perform knowledge distillation on the student model. Specifically, the teacher model's predictions on historical data (such as classification probabilities and regression values) are used as soft labels to guide the training of the student model. At the same time, it can also be fine-tuned with a small amount of new sample data (few-shot conditions) to adapt to new production requirements. The student model is evaluated to check its performance on the new sample data. If the performance does not meet the requirements, the knowledge distillation and fine-tuning process can be iterated until the student model reaches a satisfactory performance level.

[0077] According to the new production requirements, the corresponding specification parameters are configured, such as air core pitch, insulation layer thickness, surface quality requirements, etc. These parameters will serve as the targets of model prediction and control. The trained student model is deployed to the production line to collect data from the multispectral imaging unit in real time, and perform defect diagnosis and process parameter prediction. Based on the prediction results, the fuzzy PID controller optimized by reinforcement learning is used to dynamically adjust process parameters such as pulling speed, extrusion temperature and injection pressure to achieve rapid specification switching. After the specification switch, the quality fluctuations and process parameter changes in the production process are continuously monitored. Based on the monitoring results, the model parameters or control strategies are adjusted in a timely manner to ensure the stability of the production process and the consistency of product quality.

[0078] In this embodiment, historical process knowledge can be effectively transferred during production line changeovers, enabling rapid specification switching with limited sample sizes. This not only shortens production interruptions, improves production line flexibility and efficiency, but also reduces the cost of re-collecting sample data and training models.

[0079] In some embodiments, in the above step S101, the multispectral imaging unit is used to synchronously collect trimodal image data including visible light, infrared light, and polarization in the gas injection section, extrusion section, and forming section of the data cable, and a sub-pixel registration algorithm is used to spatially align the trimodal image data to obtain standard image data, specifically including:

[0080] The rotary encoder acquires the cable's position signal in real time. When the cable enters the detection area, the FPGA generates a synchronous trigger pulse signal to control the tri-modal camera to adaptively adjust the exposure time and simultaneously capture visible light images, infrared images, and polarization images.

[0081] Extracting the first SURF feature point set, the second SURF feature point set and the third SURF feature point set from the visible light image, the infrared image and the polarization image respectively;

[0082] Based on the Hessian matrix response value, the key feature area is selected from the first SURF feature point set;

[0083] Perform Taylor expansion correction on the second SURF feature point set and the third SURF feature point set to obtain sub-pixel position correction;

[0084] Based on the key feature areas and sub-pixel position corrections, the coordinate mapping relationship is calculated through the affine transformation model, and the registration results are verified by normalized mutual information to output the spatially aligned standard image data.

[0085] In this embodiment, during the production process of data cables, cable quality inspection is of vital importance. Traditional inspection methods can often only obtain image data of a single modality, which makes it difficult to fully reflect the internal structure and surface quality of the cable. Multispectral imaging technology can simultaneously obtain image data of multiple modalities such as visible light, infrared, and polarization, providing richer information for cable quality inspection. However, there are spatial deviations between image data of different modalities, and precise spatial alignment is required to ensure the accuracy of subsequent analysis. The method proposed in this embodiment can effectively solve this problem and realize the synchronous acquisition and spatial alignment of trimodal image data.

[0086] Specifically, a rotary encoder is installed on the data cable production line to acquire real-time cable position signals. Connected to the cable's transmission mechanism, the rotary encoder accurately measures the cable's travel distance and speed. When a cable enters a predefined detection zone, the rotary encoder transmits the position signal to a field-programmable gate array (FPGA). Based on the received position signal, the FPGA generates a synchronized trigger pulse signal. This pulse signal has precise timing control, ensuring simultaneous exposure by the three-modal cameras.

[0087] The tri-modal camera consists of a visible light camera, an infrared camera, and a polarization camera. Synchronous trigger pulse signals generated by the FPGA simultaneously control the exposure time of all three cameras. The camera exposure time is adaptively adjusted based on the imaging characteristics of the different modal images and the cable's travel speed to obtain clear and accurate image data. For example, infrared images may require a longer exposure time due to their weaker radiation; whereas visible light images can have a relatively shorter exposure time.

[0088] Controlled by synchronized trigger pulse signals, the trimodal camera simultaneously exposes the gas injection, extrusion, and forming sections of the data cable, capturing visible light, infrared, and polarization images. The captured image data is then transferred to a subsequent image processing system for further processing.

[0089] The captured visible light image is processed to extract the first set of SURF (Speeded Up Robust Features) feature points. SURF feature points are scale-invariant and rotation-invariant local feature descriptors that effectively represent key information in the image. During the extraction process, the image scale space is first constructed. Feature points are then detected at different scales, and the main direction and feature descriptor of each feature point are calculated.

[0090] Similarly, the infrared image is processed to extract the second set of SURF feature points. Due to the differences in imaging principles and features between infrared and visible light images, the extraction of SURF feature points requires appropriate parameter adjustments based on the characteristics of the infrared image to ensure accurate and stable feature points are extracted.

[0091] SURF feature point extraction is performed on the polarization image to obtain the third SURF feature point set. The polarization image contains polarization information on the cable surface, reflecting the surface texture and material properties of the cable. When extracting feature points, the specific characteristics of the polarization image must be fully considered to improve the quality and reliability of the feature points.

[0092] For the first set of SURF feature points (visible light image feature points), the Hessian matrix response value is calculated for each feature point. The Hessian matrix response value reflects the local structural information of the region where the feature point is located. A larger response value indicates a more distinct feature in that region. Based on the calculated Hessian matrix response values, key feature regions are screened from the first set of SURF feature points. A response value threshold is set, and regions containing feature points with response values ​​greater than the threshold are designated as key feature regions. These key feature regions play an important role in the subsequent image registration process, improving the accuracy and stability of registration.

[0093] Taylor expansion correction is performed on the second set of SURF feature points (infrared image feature points). Taylor expansion can locally approximate the image near the feature points. By calculating the Taylor expansion coefficients, the sub-pixel position correction of the feature points can be obtained. The calculation of the correction takes into account the local gradient information of the image, which can more accurately determine the position of the feature points.

[0094] Similarly, Taylor expansion correction is performed on the third SURF feature point set (polarization image feature points) to obtain its sub-pixel position correction. By performing sub-pixel correction on the feature points of the infrared and polarization images, the matching accuracy of feature points between images of different modalities can be improved.

[0095] Based on the selected key feature areas and sub-pixel position corrections, the coordinate mapping relationship between the three modal images is calculated using an affine transformation model. The affine transformation model can describe linear transformation relationships between images, such as translation, rotation, and scaling. By solving the affine transformation matrix, image data from different modalities can be mapped to the same coordinate system.

[0096] Normalized Mutual Information (NMI) is used as a validation metric for registration results. NMI measures the statistical correlation between two images; a higher value indicates better registration. Registration results are evaluated and validated by calculating the NMI between the images before and after registration. If the NMI does not meet the preset threshold, the registration parameters can be adjusted and the registration calculation repeated.

[0097] After verification of the spatial alignment and registration results, the spatially aligned standard image data is output. These standard image data are precisely aligned in space, providing an accurate and reliable basis for subsequent cable quality inspection and analysis.

[0098] In this embodiment, a multispectral imaging unit can achieve simultaneous acquisition and spatial alignment of trimodal image data for data cables. Synchronous acquisition ensures temporal consistency of the image data from different modalities, while spatial alignment eliminates spatial deviations between the different modal images, improving the accuracy and usability of the image data. This method provides more comprehensive and accurate image information for data cable quality inspection, helping to improve inspection accuracy and efficiency.

[0099] In some embodiments, in step S102, the defect diagnosis results are obtained by performing a joint diagnosis of the cable gas injection structure, insulation layer quality, and surface defects using a multi-task analysis engine based on the standard image data, specifically including:

[0100] Based on the standard image data, the gas core distribution is detected by grayscale-gradient-texture entropy fusion feature vector to determine the gas core distribution detection result;

[0101] Based on standard image data, the infrared image sequence is decomposed into time and space domains by wavelet decomposition, and the thickness-heat conduction regression model is established in combination with laser thickness measurement data.

[0102] Based on standard image data, an attention network with infrared temperature correction is deployed on visible light images, and the FE-GAN generator is used to expand the few-shot defect data to obtain feature maps.

[0103] Based on the gas core distribution detection results, thickness-heat conduction regression model and characteristic map, the defect diagnosis results including defect type, location and severity level are output.

[0104] In this embodiment, the grayscale, gradient, and texture entropy features of an image are extracted. Grayscale features reflect the brightness distribution of the image; gradient features describe changes in edges and contours within the image; and texture entropy features reflect the texture complexity of local regions of the image. The grayscale, gradient, and texture entropy features are fused to construct a grayscale-gradient-texture entropy fusion feature vector. The fusion process can employ weighted averaging or other feature fusion methods, enabling the fused feature vector to comprehensively reflect the various characteristic information of the gas core distribution.

[0105] Based on the constructed fused feature vector, an image segmentation algorithm is used to segment the cable's air core distribution area. For example, a threshold-based segmentation method can be used to determine an appropriate threshold based on the distribution of each eigenvalue in the fused feature vector to distinguish the air core area from other cable components (such as the insulation layer, conductor, etc.). Alternatively, a segmentation method based on region growing or edge detection can be used to further accurately determine the boundary contours of the air core. The segmentation results are post-processed, such as removing noise areas (eliminating non-air core mis-segmented areas caused by imaging interference) and filling small holes (filling small unsegmented areas within the air core caused by imaging defects) to obtain accurate air core distribution detection results.

[0106] Infrared image sequences are acquired from standard image data. These image sequences reflect the thermal distribution of the data cable at different times. Spatiotemporal wavelet decomposition is performed on these infrared image sequences. In the temporal domain, the image sequence is decomposed chronologically to extract features at different time scales. In the spatial domain, a wavelet transform is performed on each image frame to decompose the image into subbands of different frequencies. This spatiotemporal wavelet decomposition allows for the simultaneous acquisition of both temporal and spatial feature information of the image sequence, providing a more comprehensive description of the thermal conductivity characteristics of the insulation layer.

[0107] During the data cable production process, a laser thickness gauge is used to measure the insulation thickness in real time. Laser thickness gauges offer high precision and resolution, enabling accurate measurement of insulation thickness. The laser thickness measurement data is correlated with the wavelet decomposition results of the infrared image sequence. By using time synchronization, the laser thickness measurement data and infrared image data at the same moment can be matched, establishing a corresponding relationship between the two.

[0108] A thickness-heat conduction regression model is established based on the correlated laser thickness measurement data and the infrared image wavelet decomposition results. This regression model can use linear regression, support vector regression, or other suitable regression algorithms. During model training, the infrared image wavelet decomposition results serve as input features, and the laser thickness measurement data serves as the output target. Training with a large amount of sample data enables the model to learn the relationship between insulation layer thickness and thermal conductivity characteristics.

[0109] An attention network is deployed on visible light images. The attention network automatically learns important regions in the image and assigns more attention weight to these regions, thereby improving the detection of surface defects. An infrared temperature correction mechanism is introduced. Since infrared images can reflect the temperature distribution on the cable surface, and temperature changes can be associated with surface defects, the temperature information in the infrared image is used as a correction factor to modify the output of the attention network. For example, attention weights can be adjusted based on temperature, causing the network to pay more attention to areas with abnormal temperatures, which may indicate surface defects.

[0110] In actual production, there may be limited sample data for certain defect types. To address this issue, the FE-GAN (Feature Enhancement Generative Adversarial Network) generator is used to expand the small sample defect data. The FE-GAN generator learns and analyzes a small number of existing defect samples to extract defect feature information. The generator then generates new samples with similar features to the real defect samples. During the generation of new samples, features can be enhanced and modified to increase sample diversity.

[0111] The infrared temperature-corrected attention network is used to process visible light images, generating a preliminary feature map. This map contains important characteristic information related to surface defects in the image. The attention network is further trained and optimized using the expanded, small-sample defect data, enabling it to better identify various types of surface defects. The optimized network is then used again to process visible light images, generating a more accurate and richer feature map.

[0112] Based on the air core distribution segmentation results, analyze the cable's air core structural parameters, such as the air core diameter and distribution uniformity. If the air core diameter exceeds the normal range (too large or too small) or the air core distribution is uneven, it may affect the cable's inflation effect, electrical performance (such as changes in local electric field distribution due to air core anomalies), and mechanical strength (for example, uneven air core distribution causes uneven local stress on the cable). This will be identified as a defect type and the defect location information will be recorded.

[0113] The thickness and thermal conductivity characteristics of the insulation layer are evaluated using a thickness-heat conduction regression model. Based on the model's predictions, the insulation layer is judged to contain defects such as uneven thickness or abnormal thermal conductivity. For example, if the thickness of a region of the insulation layer is significantly lower than normal, or if the thermal conductivity is abnormal, this region is considered defective and its severity is determined.

[0114] Based on the acquired feature map, surface defects in the visible light image are identified. By analyzing the distribution and intensity of features in the feature map, the type of surface defect, such as scratches, cracks, and bubbles, is determined and the defect location is precisely marked. The severity of the defect is also assessed based on the apparent nature of the defect and its potential impact on cable quality.

[0115] The system integrates the results of cable gas injection structure analysis, insulation layer quality assessment, and surface defect identification to output defect diagnosis results, including defect type, location, and severity level. These results can be presented in the form of a report or visual interface, allowing production personnel to promptly understand the cable's quality status and take appropriate measures.

[0116] This embodiment enables a combined diagnosis of the gas injection structure, insulation quality, and surface defects of data cables. Accurate segmentation of the gas core distribution clearly reflects the structural condition of the cable's gas injection structure. A thickness-heat conduction regression model accurately assesses insulation quality. The combination of an attention network with infrared temperature correction and a FE-GAN generator improves the accuracy of surface defect detection and its adaptability to small sample sizes. The combined diagnostic results provide production personnel with comprehensive and accurate cable quality information, helping to promptly identify and resolve quality issues during production, thereby improving the quality and efficiency of data cable production.

[0117] Furthermore, the gas core distribution is detected based on the standard image data by grayscale-gradient-texture entropy fusion feature vector, and the gas core distribution detection result specifically includes:

[0118] Based on standard image data, the grayscale mean of the visible light image is calculated through a sliding window, the gradient amplitude is calculated using the improved Scharr operator, and the texture entropy is calculated using the grayscale co-occurrence matrix.

[0119] The grayscale mean, gradient amplitude and texture entropy are fused to form a three-dimensional feature vector;

[0120] The three-dimensional feature vector is input into the radial basis kernel support vector machine for classification to obtain the initial detection result of the gas core distribution;

[0121] Based on the initial detection results, the stress distribution is calculated according to the Stokes parameters extracted from the polarization image. The initial detection results are corrected by the stress distribution to obtain the gas core distribution detection results.

[0122] In this embodiment, a visible light image is obtained from standard image data. In order to extract the grayscale features of the image, the visible light image is processed using a sliding window. The sliding window moves on the image at a certain step size to cover various areas of the image. For each pixel within the sliding window, the mean of its grayscale value is calculated. The grayscale mean reflects the average brightness level of the pixels in the area and can reflect the difference in brightness between the gas core distribution and the main part of the gas injection section. For example, the grayscale mean of the gas core distribution may be significantly different from that of the main part of the gas injection section due to reasons such as light reflection.

[0123] An improved Scharr operator is used to calculate the gradient magnitude of visible light images. The Scharr operator is highly effective for edge detection. The improved Scharr operator is better suited to the characteristics of data cable images and improves edge detection accuracy. The improved Scharr operator performs convolution operations on the image in both the horizontal and vertical directions, generating gradient components in both directions. The gradient magnitude is then calculated for each pixel based on these two gradient components. The gradient magnitude reflects the degree of change in pixel values ​​within the image. Pixel values ​​in areas with air core distribution typically experience more dramatic changes, resulting in larger gradient magnitudes.

[0124] The texture entropy of visible light images is calculated using the gray-level co-occurrence matrix. The gray-level co-occurrence matrix describes the spatial dependencies between gray levels in an image and is constructed by counting the frequency of occurrence of different gray-level pairs in the image. When constructing the gray-level co-occurrence matrix, parameters such as the distance and direction between pixel pairs need to be considered. An appropriate combination of parameters is selected based on the actual application scenario and data characteristics. Texture entropy is then calculated based on the gray-level co-occurrence matrix. Texture entropy reflects the texture complexity of a localized image region. The texture characteristics of the gas core distribution and the main portion of the gas injection segment may differ, and texture entropy can help distinguish these two areas.

[0125] The extracted grayscale mean, gradient magnitude, and texture entropy are fused. Specifically, the grayscale mean, gradient magnitude, and texture entropy corresponding to each pixel are combined to form a three-dimensional feature vector. This three-dimensional feature vector combines the grayscale, gradient, and texture information of the image, providing a more comprehensive description of the pixel's characteristics.

[0126] The resulting three-dimensional feature vector is input into a radial basis kernel support vector machine (RBF-SVM) for classification. Support vector machines are a commonly used classification algorithm. The radial basis kernel function can map input features into a high-dimensional space, making it easier to find an optimal classification hyperplane in this high-dimensional space. Before classification, training samples are required. These training samples include the three-dimensional feature vectors and their category labels corresponding to the pixels in the gas core distribution area and the main part of the gas injection segment. The RBF-SVM is trained using a large number of training samples to learn the classification boundaries of the gas core distribution and the main part of the gas injection segment in the three-dimensional feature space. The three-dimensional feature vectors of all pixels in the image to be segmented are input into the trained RBF-SVM, and the initial detection results of the gas core distribution are obtained based on the classification results. The initial detection results classify the pixels in the image into two categories: the gas core distribution and the main part of the gas injection segment.

[0127] Obtain a polarization image from standard image data. The polarization image contains polarization information on the cable surface. By calculating the Stokes parameters of the polarization image, the stress distribution on the cable surface can be further analyzed. Stokes parameters are a set of parameters that describe the polarization state of light. Polarization analysis of the polarization image can obtain the Stokes parameters corresponding to each pixel. Based on the relationship between Stokes parameters and stress, the stress distribution on the cable surface is calculated. Generally speaking, areas of stress concentration may cause changes in the optical properties of the cable surface, which are reflected in the Stokes parameters of the polarization image.

[0128] Analyze the relationship between stress distribution and the initial detection results of air core distribution. In some cases, factors such as stress concentration on the cable surface or material inhomogeneity may cause errors in the initial detection results. For example, stress concentration areas may cause changes in the grayscale, gradient, or texture characteristics of the image, resulting in misjudgment as air core distribution or the main part of the gas injection section. Correct the initial detection results based on the stress distribution information. For example, if a certain area is determined to be an air core distribution in the initial segmentation, but according to the stress distribution analysis, the area may be affected by stress and cause abnormal characteristics, the segmentation result of the area can be adjusted and it can be re-determined as the main part of the gas injection section, or vice versa. In this way, more accurate air core distribution detection results are obtained.

[0129] This embodiment effectively segments the distribution of air cores in data cable images. The fused feature vector of grayscale, gradient, and texture entropy improves the comprehensiveness and accuracy of feature descriptions. The radial basis kernel support vector machine (RBFK) classification method quickly and accurately generates initial detection results. Using stress distribution information from polarization images to correct initial detection results further improves segmentation accuracy and reduces segmentation errors caused by production environment interference and image feature complexity.

[0130] Furthermore, the method of performing spatiotemporal wavelet decomposition on the infrared image sequence based on the standard image data and establishing a thickness-heat conduction regression model in combination with the laser thickness measurement data specifically includes:

[0131] Based on standard image data, a four-layer three-dimensional wavelet decomposition is performed on the infrared image sequence to extract the time domain average temperature change rate and spatial temperature gradient;

[0132] The cable thickness data, time-domain average temperature change rate, and spatial temperature gradient acquired in real time by the laser thickness gauge are combined to form a training sample set.

[0133] Based on the training sample set, the regression coefficient is solved by the least square method, and a thickness-heat conduction mapping model is established.

[0134] In this embodiment, infrared image sequences are acquired from standard image data. These images are continuously captured using an infrared thermal imager during the data cable production process, reflecting the cable's thermal distribution at different times. Preprocessing operations are performed on the acquired infrared image sequences, including denoising to eliminate random noise and interference signals in the images; and image registration to ensure spatial alignment of images captured at different times, facilitating subsequent feature extraction and analysis.

[0135] A four-layer three-dimensional wavelet decomposition method is used to process the preprocessed infrared image sequence. Three-dimensional wavelet decomposition considers information in both the temporal and spatial domains, decomposing the image sequence into subbands of varying frequencies and scales. In the temporal domain, wavelet decomposition breaks down the image sequence chronologically, extracting temperature variation information at different time scales. In the spatial domain, a wavelet transform is performed on each image frame, decomposing the image into approximate subbands and detail subbands. The approximate subbands reflect the general outline of the image, while the detail subbands contain detailed information such as edges and textures. After four layers of decomposition, rich time-frequency feature information is obtained.

[0136] For each pixel, the temperature variation over different time periods is calculated based on the time-domain subband information obtained through wavelet decomposition. Specifically, the average rate of change of pixel temperature within a specific time window is calculated. This rate of change reflects the temporal trend of the cable surface temperature. For example, a rapid temperature increase in a certain area over a short period of time may indicate abnormal heat conduction in that area.

[0137] Using the spatial subband information obtained from wavelet decomposition, the temperature gradient of the area surrounding each pixel is calculated. The temperature gradient describes the degree and direction of temperature variation in space and can reflect the uneven temperature distribution on the cable surface. By calculating the spatial temperature gradient, areas of drastic temperature changes can be identified, which may be related to variations in insulation thickness or defects.

[0138] Install a laser thickness gauge on the data cable production line to obtain real-time insulation thickness data. Laser thickness gauges offer high precision and resolution, accurately measuring cable thickness changes during the production process. Ensure the laser thickness gauge's measurement position aligns with the infrared thermal imager's acquisition area to correlate thickness data with temperature information from the infrared image sequence.

[0139] The cable thickness data acquired in real time by the laser thickness gauge, along with the extracted time-domain average temperature change rate and spatial temperature gradient, are combined. For each time point or image frame, the corresponding thickness value, time-domain average temperature change rate, and spatial temperature gradient form a data sample.

[0140] Collect a large number of data samples at different time points or locations to construct a training sample set. The training sample set should cover a variety of normal and abnormal conditions in the cable production process to ensure that the established regression model has good generalization capabilities. Preprocess the training sample set, such as performing data normalization to convert data of different dimensions to the same numerical range, to minimize the impact of data dimension differences on the regression model.

[0141] This embodiment selects a linear regression model based on the least squares method to establish the thickness-heat conduction mapping relationship. The linear regression model is simple, easy to understand and implement, and is suitable for describing the linear relationship between two or more variables. Assume that there is a linear relationship between the thickness T and the time-domain average temperature change rate R and the spatial temperature gradient G, that is, T=aR+bG+c, where a, b, and c are regression coefficients. Based on the training sample set, the least squares method is used to solve the regression coefficients a, b, and c. The goal of the least squares method is to minimize the sum of squared errors between the predicted values ​​and the actual values ​​of all samples in the training sample set. In the solution process, the estimated value of the regression coefficient is calculated by performing statistical analysis on the training sample set. For example, the optimal solution of the regression coefficient can be obtained by calculating statistics such as the mean and covariance of the sample data.

[0142] The established thickness-heat conduction mapping model is validated using an independent validation sample set. The time-domain average temperature change rate and spatial temperature gradient from the validation sample set are input into the model to obtain predicted thickness values. These values ​​are then compared with the actual thickness values ​​to calculate the prediction error. If the prediction error is large, the model's accuracy needs improvement. In this case, the model can be optimized, such as increasing the number of training samples, adjusting the model's structure or parameters, or employing a more complex regression algorithm. Repeated iterations and optimizations are performed until the model's prediction accuracy meets actual requirements.

[0143] In this embodiment, a regression model can be effectively established between the insulation thickness and thermal conductivity characteristics of data cables. A four-layer three-dimensional wavelet decomposition fully extracts temporal and spatial features from infrared image sequences. The temporal average temperature change rate and spatial temperature gradient accurately reflect the thermal conductivity of the cable. Combined with laser thickness measurement data, the established thickness-thermal conductivity regression model accurately predicts insulation thickness, providing a reliable basis for cable quality testing. Furthermore, this method offers high flexibility and scalability, allowing for adjustment and optimization based on actual production needs.

[0144] Furthermore, based on the standard image data, an attention network with infrared temperature correction is deployed on the visible light image, and the FE-GAN generator is used to expand the few-sample defect data to obtain the feature map, which specifically includes:

[0145] Based on standard image data, the temperature difference map is calculated through the infrared temperature field and linearly fused with the visible light image features to generate a spatial attention weight map;

[0146] In the FE-GAN generator, the style vector and defect code are concatenated as input to generate defect samples with diverse textures.

[0147] Based on the defect samples, the features of the visible light image are dynamically enhanced according to the spatial attention weight map to obtain the infrared temperature corrected feature map.

[0148] In this embodiment, during the data cable production process, an infrared thermal imager is used to collect infrared temperature field data on the cable surface. The infrared thermal imager can obtain temperature information at each point on the cable surface in real time, forming an infrared temperature image. Based on the collected infrared temperature field data, a temperature difference map is calculated. Specifically, a reference temperature value is selected. This reference temperature value can be the average temperature of the cable under normal operating conditions or a threshold temperature set based on experience. Then, the difference between the temperature of each pixel and the reference temperature value is calculated to generate a temperature difference map. The temperature difference map can highlight areas of abnormal cable surface temperature, which may be associated with defects.

[0149] Visible light images are obtained from standard image data. These images reflect the appearance characteristics of the cable surface. Feature extraction methods such as convolutional neural networks (CNNs) are used to process these visible light images and extract their features. CNNs can automatically learn local and global features in images, such as edges, texture, and shape. These features are crucial for describing the appearance of cable surfaces.

[0150] The calculated temperature difference map is linearly fused with the extracted visible light image features. This linear fusion can be performed using a weighted average, assigning different weights to the temperature difference map and visible light image features based on their importance. For example, if temperature difference information is more critical for defect detection, the weight of the temperature difference map can be appropriately increased.

[0151] The fused features are processed through an attention mechanism module to generate a spatial attention weight map. The attention mechanism assigns a weight to each location in the fused features based on its importance. A larger weight indicates that the feature at that location is more important for defect detection. This spatial attention weight map highlights characteristic regions associated with temperature anomalies in the visible light image, providing guidance for subsequent feature enhancement.

[0152] Style information is extracted from a large number of normal cable images to construct a style vector. The style vector can reflect overall characteristics of the cable surface, such as texture and color distribution. For example, methods such as principal component analysis (PCA) can be used to reduce the dimensionality of the features of normal cable images to obtain the main components that represent the image style. These main components form the style vector.

[0153] Analyze a small number of existing defect samples, extract defect feature information, and encode it into a defect code. The defect code can include features such as the defect's shape, size, and location. For example, image segmentation techniques can be used to separate the defect area from the visible light image, and then the features of the defect area can be quantified and encoded.

[0154] In the FE-GAN generator, the acquired style vector and defect code are concatenated and used as input. This concatenated input information incorporates both the style characteristics of a normal cable and the characteristic information of the defect. The FE-GAN generator generates defect samples with diverse textures based on this input information. By learning the style characteristics of normal cables and the characteristic information of defects, the generator can generate defect images that resemble real defect samples but have different textures. For example, the generator can modify the texture details of the defect area while maintaining the basic shape and size of the defect, thereby generating diverse defect samples and expanding the number of defect samples.

[0155] The generated defect samples are associated with the corresponding visible light images. Since the defect samples are generated based on the defect characteristics and style information of the visible light image, the defect samples can be matched with the corresponding areas in the visible light image based on information such as the location and size of the defect.

[0156] Based on the generated spatial attention weight map, the features of the visible light image are dynamically enhanced. Regions with larger weight values ​​in the spatial attention weight map represent feature regions associated with defects or temperature anomalies. For these regions, the contribution of their features in the final feature map can be increased. Specifically, the features of the visible light image are multiplied by the spatial attention weight map, so that the features of regions with larger weight values ​​are enhanced, and the features of regions with smaller weight values ​​are suppressed. At the same time, the enhanced visible light image features are further corrected in combination with the feature information of the generated defect samples. For example, the features of the defect samples can be fused with the enhanced visible light image features to obtain an infrared temperature-corrected feature map. This feature map not only contains the appearance features of the visible light image, but also combines the infrared temperature information and the expanded defect sample features, and can more accurately describe the defects on the cable surface.

[0157] This embodiment effectively improves the accuracy and robustness of data cable surface defect detection. A spatial attention weight map generated by combining infrared temperature information with visible light image features can highlight defect-related feature areas. Using the FE-GAN generator to expand defect samples solves the problem of insufficient model training data in the case of few samples. Dynamically enhancing visible light image features based on the spatial attention weight map produces an infrared temperature-corrected feature map. This feature map more comprehensively reflects cable surface defect information, helping to improve the recall and accuracy of defect detection.

[0158] In some embodiments, in the above step S103, based on the defect diagnosis results, the fuzzy PID controller optimized by reinforcement learning is used to dynamically adjust the pulling speed, extrusion temperature and gas injection pressure to form an online closed-loop control of process parameters, specifically including:

[0159] The air core distribution variance, thickness deviation and defect density are taken as state vectors;

[0160] The optimal PID parameter adjustment amount is calculated by the Double DQN algorithm and used as the action vector;

[0161] Minimizing defect density, energy consumption per unit length, and control quantity change amplitude vector is used as the reward function;

[0162] Based on the state vector, action vector and reward function, the control variables of traction speed, extrusion temperature and gas injection pressure are calculated through reinforcement learning and sent to the PLC actuator;

[0163] The reward function value is calculated based on the control effect of the PLC actuator, and the Q network parameters are updated once every N meters of cable are produced.

[0164] In this example, the variance of the gas core distribution over a given length of cable is calculated. This variance reflects the degree of dispersion in the gas core distribution. A large variance indicates an unstable gas core distribution, potentially affecting the gas injection quality and electrical performance of the cable. Using the gas core distribution variance as part of the state vector helps the controller understand the current fluctuations in the gas core distribution.

[0165] The thickness of the data cable insulation layer is measured in real time using online testing equipment such as a laser thickness gauge. The measured thickness is compared with the preset target thickness to calculate the thickness deviation. Thickness deviation directly reflects the accuracy of the cable insulation thickness and is a key indicator of cable quality. By incorporating thickness deviation into the state vector, the controller can adjust process parameters such as extrusion temperature based on the size and trend of the thickness deviation to ensure that the cable insulation thickness meets the requirements.

[0166] Combined with previous defect diagnosis results, the produced cables are inspected for defects. The number of defects along a certain length of cable is counted, and the defect density is calculated. Defect density provides a direct reflection of cable production quality; higher defect density indicates poorer cable quality. By including defect density as a key component of the state vector, the controller can dynamically adjust parameters such as pulling speed, extrusion temperature, and gas injection pressure based on cable quality to reduce defects.

[0167] The Double DQN (Double Deep Q-Network) algorithm is based on deep reinforcement learning. It uses two neural networks (a target network and an evaluation network) to estimate the state-action value function. This effectively addresses the overestimation problem inherent in traditional DQN algorithms, improving their stability and convergence. In this example, the Double DQN algorithm was selected to calculate the optimal PID parameter adjustment. Because the data cable production process is a complex dynamic system characterized by nonlinearity and time-varying characteristics, the Double DQN algorithm continuously learns and optimizes through interaction with the environment, finding the PID parameter adjustment strategy that optimizes system performance under different conditions.

[0168] The state vector consisting of the core distribution variance, thickness deviation, and defect density is input into the evaluation network of the Double DQN algorithm. Based on the current state vector, the evaluation network outputs the Q value corresponding to each possible PID parameter adjustment. The PID parameter adjustment with the largest Q value is selected as the optimal PID parameter adjustment, and this adjustment constitutes the action vector. The optimal PID parameter adjustment includes adjustments to the proportional, integral, and differential coefficients of the PID controller. By adjusting these parameters, the control performance of the fuzzy PID controller can be improved to better adapt to the current production conditions.

[0169] Defect density is a key indicator of cable quality, and minimizing it is a key component of the reward function. When defect density decreases, the controller is rewarded positively; conversely, when defect density increases, the controller is penalized negatively. This guides the controller to adjust process parameters to reduce cable defects and improve cable quality.

[0170] Energy consumption is a significant cost factor in data cable production. To reduce production costs, energy consumption per unit cable length must be minimized. By incorporating energy consumption per unit length into the reward function, the controller is rewarded when energy consumption decreases. This encourages the controller to optimize process parameters to reduce energy consumption while maintaining cable quality.

[0171] Frequent changes in process parameters (pulling speed, extrusion temperature, and gas injection pressure) can adversely affect production equipment and cable quality. To maintain production stability, the control variable change amplitude vector is incorporated into the reward function. When the control variable change amplitude is small, the controller is rewarded positively; when the control variable change amplitude is large, the controller is penalized negatively. This guides the controller to maintain stable adjustments to process parameters and avoid large fluctuations.

[0172] Based on the defined state vector, action vector, and reward function, the reinforcement learning system (centered on the Double DQN algorithm) continuously interacts with the production environment. During each control cycle, the optimal PID parameter adjustment (action vector) is selected based on the current state vector. Then, based on the principles of a fuzzy PID controller, the control variables for the pulling speed, extrusion temperature, and gas injection pressure are calculated. This fuzzy PID controller combines the advantages of both fuzzy and PID control. It uses fuzzy inference rules to adjust PID parameters in real time based on system errors and error rates, thereby achieving precise control of process parameters. The calculated control variables for pulling speed, extrusion temperature, and gas injection pressure are transmitted to the PLC (Programmable Logic Controller) actuator via a communication interface. Based on the received control variables, the PLC actuator precisely controls the pulling, extrusion, and gas injection equipment on the production line, enabling real-time adjustment of process parameters.

[0173] Based on the PLC actuator's effect on process parameter adjustments, data such as core distribution variance, thickness deviation, and defect density are collected again. Based on the reward function design, the reward function value for this control cycle is calculated. For example, if the process parameter adjustments reduce defect density, energy consumption, and the change in control volume is small, the reward function value will be a large positive value. Conversely, if the adjustment is ineffective, the reward function value will be small or even negative.

[0174] For every N meters of cable produced, the collected state vector, action vector, and calculated reward function value are combined into training samples, which are used to update the Q network parameters in the Double DQN algorithm. By continuously updating the Q network parameters, the reinforcement learning system gradually learns the optimal PID parameter adjustment strategy for different production conditions, thereby continuously improving the control accuracy of pulling speed, extrusion temperature, and gas injection pressure, and achieving closed-loop control of online process parameters in the data cable production process.

[0175] This embodiment significantly improves the production quality and efficiency of data cables. A reinforcement learning-optimized fuzzy PID controller dynamically adjusts process parameters such as pulling speed, extrusion temperature, and gas injection pressure based on defect diagnosis results and real-time production status, effectively reducing cable defects and lowering energy consumption per unit length while maintaining production stability. By continuously updating the Q network parameters, the controller can gradually adapt to various changes in the production process, improving the control system's adaptability and intelligence.

[0176] In some embodiments, in step S104, when the production line is changed, historical process knowledge is transferred through cross-line federated learning and model distillation technology to perform rapid specification switching under the condition of few samples, specifically including:

[0177] When the production line is changed, the cosine-Euclidean mixed similarity between the new specification parameters and the specification parameters in the historical configuration library is calculated, and the top k configurations are selected as the initial parameters;

[0178] Based on the initial parameters, the multi-line process model is aggregated through the federated learning framework. Each production line uses local historical data and a small amount of new samples to update the multi-line process model.

[0179] Knowledge distillation technology is used to compress multi-line process models into lightweight models, retaining key process knowledge through KL divergence and feature alignment loss;

[0180] Channel pruning is implemented on lightweight models. When the mean channel weight is lower than the preset channel weight threshold, redundant parameters are removed to complete the deployment of new specification production lines.

[0181] In this embodiment, in industrial production, production line changeover is a common operation. For example, data cable manufacturers need to produce cables of different specifications according to market demand. However, each production line change faces the problem of insufficient sample data under the new specifications. Traditional model training methods require a large amount of sample data to achieve better performance, which is difficult to achieve in the early stages of the changeover. At the same time, a wealth of historical process knowledge has accumulated between different production lines, but this knowledge is often scattered across various production lines, making it difficult to effectively integrate and utilize. The method proposed in this embodiment can transfer historical process knowledge through cross-line federated learning and model distillation technology, achieve rapid specification switching under few sample conditions, and improve production efficiency and product quality.

[0182] When a production line changes, new specifications and parameters from the historical configuration library are structured. For example, for data cable production, specifications may include cable diameter, insulation thickness, and gas injection method. Each specification parameter is represented as a multidimensional vector, with each dimension corresponding to a characteristic of the parameter.

[0183] Calculate the cosine-Euclidean mixed similarity between the new specifications and the specifications in the historical configuration database. This mixed similarity calculation can combine the advantages of cosine similarity and Euclidean distance. For example, each can be weighted and then summed. Based on the mixed similarity calculation, the top k historical configurations that are most similar to the new specifications are selected as initial parameters. These initial parameters reflect production experience similar to the new specifications and provide a good starting point for subsequent model training.

[0184] Build a cross-production-line federated learning framework that connects computing devices and data storage devices across multiple production lines. In federated learning, each production line is considered a client, with its own local historical data. For example, different production lines may have produced data cables of similar specifications but different models, each accumulating its own production data and process models.

[0185] Based on the top k selected configurations as initial parameters, the multi-line process model is aggregated through a federated learning framework. During the federated learning process, each production line uses local historical data and a small number of new samples to update the multi-line process model. Each production line trains and updates the model on its local data, and then uploads the updated model parameters to the central server. The central server aggregates these parameters to obtain a more powerful multi-line process model. This distributed learning method can fully utilize the historical data of each production line while protecting the data privacy of each production line. Each production line uses local historical data and a small number of new samples to update the multi-line process model. The local historical data contains rich information about the production line's previous production process, while the small number of new samples reflects some characteristics under the new specifications. By combining these two parts of data, the model can better adapt to the new specifications even with a small number of samples.

[0186] Knowledge distillation technology is used to compress the multi-line process model into a lightweight model. Knowledge distillation is a technique that transfers knowledge from a complex model (the teacher model) to a simpler model (the student model). In this example, the multi-line process model serves as the teacher model, containing rich process knowledge; the lightweight model, serving as the student model, needs to learn from the teacher model's knowledge.

[0187] Key process knowledge is retained through KL divergence and feature alignment loss. KL divergence measures the difference between two probability distributions and is used here to measure the similarity between the output distributions of the student and teacher models. Feature alignment loss ensures that the student model extracts key features similar to those of the teacher model during the learning process. For example, in data cable production, key features may include the effect of extrusion temperature on cable insulation thickness and the effect of pulling speed on cable gas injection quality.

[0188] Channel pruning is implemented on lightweight models. Channel pruning is a model compression technique that reduces the number of model parameters and computational complexity by removing unimportant channels. When the mean channel weight falls below a preset channel weight threshold, the channel is considered to have a minimal contribution to the model and can be removed, thereby reducing redundant model parameters. After channel pruning, the lightweight model is capable of rapid deployment to new production lines. This model retains key knowledge from multi-line process models while minimizing computational complexity and storage requirements, enabling it to meet the real-time control requirements of production lines.

[0189] In this example, historical process knowledge can be quickly leveraged during production line changes, enabling specification switching with minimal sample size. Cross-line federated learning breaks down data silos, enabling the sharing of process experience across different production lines. Model distillation compresses complex models into lightweight ones, improving deployment efficiency and responsiveness. In practical applications, this approach can significantly shorten the debugging time for production line changes, improve production efficiency, and ensure the quality stability of new product specifications.

[0190] Reference Figure 2 An embodiment of the present invention provides a data cable adaptive production system 2 based on image analysis, and the system 2 specifically includes:

[0191] The first production module 201 is used to synchronously collect trimodal image data including visible light, infrared light, and polarization at the gas injection section, extrusion section, and forming section of the data cable through a multispectral imaging unit, and spatially align the trimodal image data using a sub-pixel registration algorithm to obtain standard image data;

[0192] The second production module 202 is used to perform a joint diagnosis of the cable gas injection structure, insulation layer quality and surface defects through a multi-task analysis engine based on standard image data to obtain defect diagnosis results;

[0193] The third production module 203 is used to dynamically adjust the pulling speed, extrusion temperature and gas injection pressure based on the defect diagnosis results using a fuzzy PID controller optimized by reinforcement learning to form an online closed-loop control of process parameters;

[0194] The fourth production module 204 is used to transfer historical process knowledge through cross-line federated learning and model distillation technology when the production line is changed, and perform rapid specification switching under few sample conditions.

[0195] It is understandable that if Figure 1 The contents of the embodiment of the data cable adaptive production method based on image analysis shown in the embodiment are all applicable to the embodiment of the data cable adaptive production system based on image analysis. The functions specifically implemented by the embodiment of the data cable adaptive production system based on image analysis are the same as those in the embodiment of the data cable adaptive production method based on image analysis shown in the embodiment of the data cable adaptive production method based on image analysis. Figure 1 The embodiment of the data cable adaptive production method based on image analysis shown is the same as that shown in FIG. Figure 1 The beneficial effects achieved by the embodiment of the data cable adaptive production method based on image analysis shown are also the same.

[0196] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0197] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0198] Reference Figure 3 An embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored on the memory 302. When the computer program 303 is executed on the processor 301, the image analysis-based adaptive production method for data cables as described in any one of the above methods is implemented.

[0199] The computer device 3 may be a desktop computer, a notebook computer, a PDA, a cloud server or other computing devices. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that Figure 3 This is merely an example of the computer device 3 and does not constitute a limitation on the computer device 3 . The computer device 3 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device 3 may also include input and output devices, network access devices, etc.

[0200] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0201] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard drive or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped on the computer device 3. Furthermore, the memory 302 may include both an internal storage unit of the computer device 3 and an external storage device. The memory 302 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 302 may also be used to temporarily store data that has been output or is about to be output.

[0202] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for adaptively producing data cables based on image analysis as described in any one of the above methods is implemented.

[0203] In this embodiment, if the integrated control unit is implemented as a software functional unit and sold or used as a standalone product for machine tool control, it can be stored in a computer-readable storage medium specifically for machine tools. Based on this understanding, the present application implements all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware of the machine tool through a specific computer program. The computer program can be stored in a computer-readable storage medium specifically for machine tools. When executed by the machine tool's processor, the computer program can implement the application steps of each of the above-mentioned method embodiments in machine tool control. The computer program includes computer program code for machine tool control, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code for machine tool control to the machine tool equipment, recording media, machine tool computer memory, read-only memory (ROM), random access memory (RAM), and other media suitable for machine tool software distribution, such as dedicated machine tool control cards and machine tool data storage cards.

[0204] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0205] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0206] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely schematic. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0207] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A data cable adaptive production method based on image analysis, characterized in that: The method specifically includes: The multispectral imaging unit synchronously collects trimodal image data including visible light, infrared and polarization in the gas injection section, extrusion section and forming section of the data cable, and uses the sub-pixel registration algorithm to spatially align the trimodal image data to obtain standard image data; Based on standard image data, a multi-task analysis engine is used to jointly diagnose the cable gas injection structure, insulation layer quality, and surface defects to obtain defect diagnosis results. Based on the defect diagnosis results, the fuzzy PID controller optimized by reinforcement learning is used to dynamically adjust the pulling speed, extrusion temperature and gas injection pressure to form an online closed-loop control of process parameters; When changing production lines, historical process knowledge is transferred through cross-line federated learning and model distillation technology to perform rapid specification switching under few-sample conditions.

2. The method according to claim 1, characterized in that The multispectral imaging unit synchronously collects trimodal image data including visible light, infrared and polarization in the gas injection section, extrusion section and forming section of the data cable, and uses a sub-pixel registration algorithm to spatially align the trimodal image data to obtain standard image data, specifically including: The rotary encoder acquires the cable's position signal in real time. When the cable enters the detection area, the FPGA generates a synchronous trigger pulse signal to control the tri-modal camera to adaptively adjust the exposure time and simultaneously capture visible light images, infrared images, and polarization images. Extracting the first SURF feature point set, the second SURF feature point set and the third SURF feature point set from the visible light image, the infrared image and the polarization image respectively; Based on the Hessian matrix response value, the key feature area is selected from the first SURF feature point set; Perform Taylor expansion correction on the second SURF feature point set and the third SURF feature point set to obtain sub-pixel position correction; Based on the key feature areas and sub-pixel position corrections, the coordinate mapping relationship is calculated through the affine transformation model, and the registration results are verified by normalized mutual information to output the spatially aligned standard image data.

3. The method according to claim 2, characterized in that Based on the standard image data, the cable gas injection structure, insulation layer quality and surface defects are jointly diagnosed through a multi-task analysis engine to obtain defect diagnosis results, specifically including: Based on the standard image data, the gas core distribution is detected by grayscale-gradient-texture entropy fusion feature vector to determine the gas core distribution detection result; Based on standard image data, the infrared image sequence is decomposed into time and space domains by wavelet decomposition, and the thickness-heat conduction regression model is established in combination with laser thickness measurement data. Based on standard image data, an attention network with infrared temperature correction is deployed on visible light images, and the FE-GAN generator is used to expand the few-shot defect data to obtain feature maps. Based on the gas core distribution detection results, thickness-heat conduction regression model and characteristic map, the defect diagnosis results including defect type, location and severity level are output.

4. The method according to claim 3, characterized in that The gas core distribution is detected based on the standard image data by grayscale-gradient-texture entropy fusion feature vector, and the gas core distribution detection result specifically includes: Based on standard image data, the grayscale mean of the visible light image is calculated through a sliding window, the gradient amplitude is calculated using the improved Scharr operator, and the texture entropy is calculated using the grayscale co-occurrence matrix. The grayscale mean, gradient amplitude and texture entropy are fused to form a three-dimensional feature vector; The three-dimensional feature vector is input into the radial basis kernel support vector machine for classification to obtain the initial detection result of the gas core distribution; Based on the initial detection results, the stress distribution is calculated according to the Stokes parameters extracted from the polarization image. The initial detection results are corrected by the stress distribution to obtain the gas core distribution detection results.

5. The method according to claim 3, characterized in that The method of performing spatiotemporal wavelet decomposition on the infrared image sequence based on the standard image data and establishing a thickness-heat conduction regression model in combination with the laser thickness measurement data specifically includes: Based on standard image data, a four-layer three-dimensional wavelet decomposition is performed on the infrared image sequence to extract the time domain average temperature change rate and spatial temperature gradient; The cable thickness data, time-domain average temperature change rate, and spatial temperature gradient acquired in real time by the laser thickness gauge are combined to form a training sample set. Based on the training sample set, the regression coefficient is solved by the least square method, and a thickness-heat conduction mapping model is established.

6. The method according to claim 3, characterized in that Based on standard image data, the attention network with infrared temperature correction is deployed on the visible light image, and the FE-GAN generator is used to expand the few-sample defect data to obtain the feature map, which specifically includes: Based on standard image data, the temperature difference map is calculated through the infrared temperature field and linearly fused with the visible light image features to generate a spatial attention weight map; In the FE-GAN generator, the style vector and defect code are concatenated as input to generate defect samples with diverse textures. Based on the defect samples, the features of the visible light image are dynamically enhanced according to the spatial attention weight map to obtain the infrared temperature corrected feature map.

7. The method according to any one of claims 1 to 6, characterized in that Based on the defect diagnosis results, the fuzzy PID controller optimized by reinforcement learning is used to dynamically adjust the pulling speed, extrusion temperature and injection pressure to form an online closed-loop control of process parameters, specifically including: The air core distribution variance, thickness deviation and defect density are taken as state vectors; The optimal PID parameter adjustment amount is calculated by the Double DQN algorithm and used as the action vector; Minimizing defect density, energy consumption per unit length, and control quantity change amplitude vector is used as the reward function; Based on the state vector, action vector and reward function, the control variables of traction speed, extrusion temperature and gas injection pressure are calculated through reinforcement learning and sent to the PLC actuator; The reward function value is calculated based on the control effect of the PLC actuator, and the Q network parameters are updated once every N meters of cable are produced.

8. The method according to any one of claims 1 to 6, characterized in that When changing production lines, historical process knowledge is transferred through cross-line federated learning and model distillation technology to perform rapid specification switching under low-sample conditions. Specifically, the following steps are involved: When the production line is changed, the cosine-Euclidean mixed similarity between the new specification parameters and the specification parameters in the historical configuration library is calculated, and the top k configurations are selected as the initial parameters; Based on the initial parameters, the multi-line process model is aggregated through the federated learning framework. Each production line uses local historical data and a small amount of new samples to update the multi-line process model. Knowledge distillation technology is used to compress multi-line process models into lightweight models, retaining key process knowledge through KL divergence and feature alignment loss; Channel pruning is implemented on lightweight models. When the mean channel weight is lower than the preset channel weight threshold, redundant parameters are removed to complete the deployment of new specification production lines.

9. An adaptive production system for data cables based on image analysis, characterized in that: The system specifically includes: The first production module is used to synchronously collect trimodal image data including visible light, infrared light, and polarization in the gas injection, extrusion, and molding sections of the data cable through a multispectral imaging unit, and spatially align the trimodal image data using a sub-pixel registration algorithm to obtain standard image data; The second production module is used to perform joint diagnosis of cable gas injection structure, insulation layer quality and surface defects based on standard image data through a multi-task analysis engine to obtain defect diagnosis results; The third production module is used to dynamically adjust the pulling speed, extrusion temperature and gas injection pressure based on the defect diagnosis results using a fuzzy PID controller optimized by reinforcement learning, forming an online closed-loop control of process parameters; The fourth production module is used to transfer historical process knowledge through cross-line federated learning and model distillation technology when the production line is changed, and to perform rapid specification switching under few-sample conditions.

10. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory, which, when executed on the processor, implements the adaptive production method for data cables based on image analysis as claimed in any one of claims 1 to 8.

Images