Image recognition-based cable core wire eccentricity automatic detection and compensation method and system

Abstract

The application discloses a cable core wire eccentricity automatic detection compensation method and system based on image recognition, belongs to the quality control technical field in cable manufacturing, and realizes real-time collection of the image of the cable in operation in the cross-section direction, extraction of the core wire and the sheath boundary profile, and calculation of the core wire eccentricity vector; the eccentricity vector is input into a neural network recognition model, the eccentricity trend at the next moment is predicted, and comparison with a preset threshold value is carried out; if the eccentricity is out of limit, then based on the difference value of the current and the predicted eccentricity vector, combined with the equipment structure fluctuation characteristics and displacement response characteristics, the compensation displacement is calculated for adjusting the die core or the guide wheel position, and then the extrusion path is updated and the closed loop control is continued; the application has the advantages of strong predictability, high response accuracy and adaptation to complex working conditions, can realize high-precision real-time regulation and control of the core wire eccentricity, and improves the consistency and yield of the micro cable product.

Description

Technical Field

[0001] This invention relates to the field of quality control technology in cable manufacturing, specifically to an automatic detection and compensation method and system for cable core eccentricity based on image recognition. Background Technology

[0002] In automated cable manufacturing, the misalignment of the cable core (conductor) relative to the outer sheath or insulation layer is a key factor affecting the electrical performance and mechanical strength of the product. Especially in high-speed extrusion production, uncontrollable factors such as mechanical vibration, raw material fluctuations, and die offset often cause the core to shift off-center in the cross-section, affecting its tensile strength and dielectric uniformity, ultimately leading to a decrease in yield.

[0003] Existing methods for detecting cable eccentricity mostly employ physical sensors such as laser displacement and ultrasonic thickness measurement to obtain the boundary positions of each layer of the cable. However, these methods have the following significant drawbacks:

[0004] Limited spatial resolution: It is difficult to detect minute deviations (such as within 0.05mm); Strong environmental dependence: It is sensitive to differences in cable surface finish and color, and is prone to false detection; Lack of real-time feedback mechanism: Detection and correction are disconnected, and production errors cannot be compensated in time; It cannot adapt to different cable structures: Changes in core material, sheath color, and cross-section will interfere with identification.

[0005] Especially when producing ultra-fine signal cables with a diameter of ≤1.2mm and core conductor material of multi-strand stranded copper wire, traditional methods cannot achieve high-precision eccentricity detection and effective compensation, resulting in a product yield decrease of more than 20%. Summary of the Invention

[0006] The purpose of this invention is to provide an automatic detection and compensation method and system for cable core eccentricity based on image recognition, so as to solve the shortcomings in the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an automatic detection and compensation method for cable core eccentricity based on image recognition, comprising:

[0008] Real-time images of the cable in operation are acquired in the cross-sectional direction, including grayscale or color information of the core boundary and the sheath boundary;

[0009] Edge extraction and boundary segmentation are performed on the real-time image to identify the core wire region contour and the sheath region contour respectively. The centroid coordinates of the core wire and the center coordinates of the sheath are calculated to obtain the core wire eccentricity vector ΔP.

[0010] The core wire eccentricity vector ΔP is input into the neural network recognition model to fit and predict the eccentricity trend, output the eccentricity vector ΔP' expected at the next moment, and determine whether it exceeds the preset threshold ε.

[0011] If ΔP' > ε, then a compensation displacement ΔX is calculated based on ΔP and ΔP', which is used to adjust the current position of the cable extruder mandrel or guide wheel to offset the core wire deviation trend;

[0012] After the compensation displacement ΔX is executed, the extrusion path is updated, and a new image is continuously collected.

[0013] Preferably, the core wire eccentricity vector ΔP is obtained by:

[0014] The real-time image is subjected to grayscale conversion and adaptive filtering processing.

[0015] Based on the processed image, a first closed contour of the core wire region and a second closed contour of the sheath region are obtained respectively.

[0016] The centroid coordinates of the core wire region are calculated according to the first closed contour, the center coordinates of the sheath region are calculated according to the second closed contour, and the core wire eccentricity vector ΔP is determined by the vector difference between the centroid coordinates and the center coordinates.

[0017] Preferably, the first closed contour of the core wire region and the second closed contour of the sheath region are obtained by further comprising: identifying two closed boundaries in the image, wherein the closed contour with a small area is defined as the first closed contour of the core wire region, and the closed contour with a large area and approximately circular is defined as the second closed contour of the sheath region.

[0018] Preferably, the predicted next time eccentricity vector ΔP' is outputted by:

[0019] The plurality of groups of core wire eccentricity vectors ΔP collected in time sequence are subjected to sequential processing, a data sequence containing at least N historical eccentricity vectors is constructed, and the data sequence is subjected to normalization processing as an input sample of the neural network identification model;

[0020] A neural network identification model for eccentricity trend prediction is constructed based on the input sample, the neural network identification model comprises an input layer, at least one hidden layer and an output layer, and the model weight is trained by a back propagation algorithm;

[0021] The latest eccentricity vector data sequence is inputted into the trained neural network identification model, and a predicted core wire eccentricity vector ΔP' corresponding to the next sampling time is outputted.

[0022] And the modulus value of the predicted core wire eccentricity vector ΔP' is compared with a preset threshold ε.

[0023] Preferably, the compensation displacement ΔX is calculated based on ΔP and ΔP' by:

[0024] If ΔP' > ε, the core wire eccentricity vector ΔP corresponding to the current sampling time and the next sampling time eccentricity vector ΔP' output by the neural network recognition model are obtained;

[0025] A vector difference operation is performed on the core wire eccentricity vector ΔP and the eccentricity vector ΔP' to obtain a core wire eccentricity change vector ΔD;

[0026] According to the core wire eccentricity change vector ΔD, the structure fluctuation characteristics and displacement response characteristics of the cable extrusion equipment are combined to perform a proportional mapping calculation on the core wire eccentricity change vector ΔD, so as to obtain a compensation displacement ΔX for offsetting the core wire eccentricity change trend.

[0027] Preferably, the structure fluctuation characteristic parameters of the cable extrusion equipment in a stable running state are obtained, and the structure fluctuation characteristic parameters include a periodic offset amplitude of the mold core relative to the extrusion axis and a corresponding fluctuation frequency; the displacement response characteristic parameters of the cable extrusion equipment are obtained, and the displacement response characteristic parameters represent the actual displacement tolerance amount generated by the extrusion equipment under the action of a unit control input.

[0028] Preferably, the periodic offset amplitude is defined as the distance between the maximum and minimum positions of the mold core center in a complete fluctuation period; and the fluctuation frequency is defined as the number of complete reciprocating offsets of the mold core per unit time.

[0029] The application also provides a cable core wire eccentricity automatic detection and compensation system based on image recognition, which comprises:

[0030] An image acquisition module: acquiring real-time images of the running cable in the cross-sectional direction, containing gray scale or color information of the core wire boundary and the sheath boundary;

[0031] A feature extraction module: performing edge extraction and boundary segmentation on the real-time images, respectively identifying the core wire region contour and the sheath region contour, calculating the core wire centroid coordinates and the sheath center coordinates, and obtaining the core wire eccentricity vector ΔP;

[0032] A neural network prediction module: inputting the core wire eccentricity vector ΔP into a neural network recognition model to fit and predict the eccentricity trend, outputting the predicted eccentricity vector ΔP' at the next time, and judging whether it exceeds a preset threshold ε;

[0033] An eccentricity compensation calculation module: if ΔP' > ε, a compensation displacement ΔX is calculated based on ΔP and ΔP', and the compensation displacement ΔX is used to adjust the current position of the cable extruder mold core or the guide wheel to offset the core wire offset trend;

[0034] A path updating module: after performing the compensation displacement ΔX, updating the extrusion path and continuing to acquire new images.

[0035] In the above technical solution, the application provides technical effects and advantages:

[0036] 1、The application realizes high-precision identification and prediction control of the eccentricity change trend of the core wire relative to the sheath by real-time acquisition of cable images in the cross-sectional direction, combined with edge recognition, geometric analysis and neural network prediction. Compared with the traditional method relying on fixed threshold or open-loop detection method, the application can dynamically obtain the core wire offset behavior, and output the compensation displacement in advance before the eccentricity trend exceeds the forming tolerance, effectively inhibiting the accumulation and amplification of eccentricity error in the production process, and improving the real-time performance of cable concentricity control.

[0037] 2、The application establishes a proportional compensation model facing the physical characteristics of the device by introducing the mapping relationship between the core wire eccentricity change vector and the device structure response characteristics, so that the calculation of the compensation displacement ΔX is more accurate and controllable. Combined with the structure fluctuation parameters and displacement response parameters, the compensation calculation not only considers the change amount of the eccentricity trend, but also takes into account the control sensitivity and stability of the device itself, improving the compensation accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings described below are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0039] Figure 1 The method flowchart of the present application.

[0040] Figure 2 The system module flowchart of the present application. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0042] Embodiment 1, please refer to Figure 1 The cable core wire eccentricity automatic detection and compensation method based on image recognition described in this embodiment includes:

[0043] In the cross-sectional direction, a real-time image of the running cable is collected, containing gray or color information of the core wire boundary and the sheath boundary; edge extraction and boundary segmentation are performed on the real-time image, the core wire region contour and the sheath region contour are identified respectively, the core wire centroid coordinates and the sheath center coordinates are calculated, and the core wire eccentricity vector ΔP is obtained; the core wire eccentricity vector ΔP is input into a neural network identification model, the eccentricity trend is fitted and predicted, and the predicted next moment eccentricity vector ΔP' is output, and it is judged whether it exceeds the preset threshold ε; if ΔP'>ε, the compensation displacement ΔX is calculated based on ΔP and ΔP', and the compensation displacement ΔX is used to adjust the current position of the cable extruder die core or guide wheel to offset the core wire offset trend; after the compensation displacement ΔX is executed, the extrusion path is updated, and the new image is continuously collected.

[0044] In the cross-sectional direction, a real-time image of the running cable is collected, containing gray or color information of the core wire boundary and the sheath boundary, specifically including:

[0045] In this embodiment, first, a non-contact image of the cable in high-speed operation is obtained. The image collection direction is perpendicular to the cable axis, that is, the cross-sectional direction, so as to completely capture the relative position relationship of the core wire and the sheath in the cross section.

[0046] The image collection device can be a high-frame-rate industrial camera, which is matched with a telecentric lens and a high-brightness coaxial light source, so as to ensure that clear and distortion-free cable cross-sectional images are obtained in the motion state. The collection frequency can be dynamically adjusted according to the cable running speed, so as to ensure that equidistant images are maintained under different speed conditions.

[0047] In the collected image, the core wire usually appears as an inner high-brightness or low-brightness area, and the sheath appears as an outer closed circular area. Due to the difference in color, reflectivity or texture between the core wire and the sheath material, there is a clear gray or color boundary in the image. This information can be used as the basis for boundary identification in subsequent image processing algorithms.

[0048] In order to enhance the boundary recognition effect, the image can be preprocessed, including gray scale, denoising, contrast enhancement, etc., so that the brightness gradient of the core wire edge and the sheath edge is more obvious, thereby improving the accuracy of subsequent edge detection and geometric fitting.

[0049] In this embodiment, the cross-sectional image collected by the above-mentioned method can not only reflect the core wire eccentricity, but also assist in judging key quality parameters such as core wire shape and sheath roundness.

[0050] Edge extraction and boundary segmentation are performed on the real-time image, the core wire region contour and the sheath region contour are identified respectively, the core wire centroid coordinates and the sheath center coordinates are calculated, and the core wire eccentricity vector ΔP is obtained.

[0051] In this embodiment, in order to accurately extract the boundary features of the core wire region and the sheath region in the cable cross-section image, first, the acquired real-time image is subjected to gray scale conversion and adaptive filtering processing. The gray scale conversion adopts a standard RGB weighted average method to convert the three-channel color image into a single-channel gray scale image. Subsequently, an adaptive median filtering algorithm based on local pixel mean is used to smooth the gray scale image, and a 5x5 window is constructed with each pixel as the center, and the median calculation range is adjusted to effectively remove random noise and retain the edge gradient change region.

[0052] After completing the image preprocessing, the Canny edge detection algorithm is used to perform edge extraction on the gray scale image. The algorithm determines the gradient amplitude and direction by calculating the first derivative gradient of the image in the horizontal and vertical directions, and combining a Gaussian filter to preliminarily smooth the image, and then performs edge screening with a set of double thresholds.

[0053] After edge extraction, two closed boundaries in the image are identified by a contour tracking algorithm, wherein a small-area closed contour is defined as a first closed contour of the core wire region, and a large-area, approximately circular closed contour is defined as a second closed contour of the sheath region. Specifically, it includes:

[0054] The binary image after edge extraction is regarded as a two-dimensional grid composed of foreground pixels (value 1) and background pixels (value 0), and an 8-neighborhood pixel connectivity judgment is adopted, i.e. each foreground pixel and its surrounding 8-direction pixels are connected to identify the continuous edge structure.

[0055] Starting from the top left corner of the image, scan by row, and when the first foreground pixel is encountered, record its position as the initial edge point P0, and initialize the current contour point set to empty.

[0056] Starting from the initial point P0, search for the next foreground pixel in its neighborhood in the 8 directions clockwise , record its position and mark it as "visited". Continue this process with P1 as the center until the search returns to the starting point P0, forming a closed boundary. All tracking points are stored in the current contour point set in order.

[0057] Continue to scan the entire image to find unvisited foreground pixels as new starting points, and perform step 3 for a second round of contour tracking. Each identified closed contour is stored as an independent boundary.

[0058] Sort all the extracted closed contours according to the contour area, and the one with small area and high boundary curvature is determined as the first closed contour of the core wire region, and the one with large area and approximately circular edge is determined as the second closed contour of the sheath region.

[0059] According to the obtained first closed contour, the barycentric coordinates of the core wire region are determined by a barycentric calculation method. This method averages all pixel point coordinates in the closed region, where the average of the horizontal coordinates is the barycentric horizontal coordinate, and the average of the vertical coordinates is the barycentric vertical coordinate, finally forming the barycentric coordinates (X1, Y1) of the core wire. For the second closed contour of the sheath region, a least squares circle fitting algorithm is used to construct a fitting circle model. This algorithm is based on the general equation of a circle, constructs a three-element nonlinear equation set containing the center coordinates (X2, Y2) and the radius R, and uses the Levenberg-Marquardt optimization method for iterative solution, finally determines the center coordinates (X2, Y2) of the sheath region.

[0060] The core wire eccentricity vector ΔP is represented by the coordinate difference between the two geometric centers, that is, the two-dimensional vector difference composed of the barycentric coordinates (X1, Y1) of the core wire and the center coordinates (X2, Y2) of the sheath circle, . This eccentricity vector not only reflects the spatial displacement of the core wire relative to the sheath, but also can be used to evaluate the direction and amplitude of the eccentricity, providing quantitative input for subsequent compensation control.

[0061] The core wire eccentricity vector ΔP is input into the neural network recognition model to fit and predict the eccentricity trend, output the predicted eccentricity vector ΔP' at the next time, and judge whether it exceeds the preset threshold ε.

[0062] According to the sampling time sequence, the continuously collected core wire eccentricity vectors ΔP are grouped into a historical data sequence, denoted as S={ΔP1, ΔP2, …, ΔPn}, where n is the sequence length, and each ΔP contains two components in the X and Y directions. In order to improve the adaptability of the neural network recognition model to different orders of magnitude of input, the minimum-maximum normalization processing is performed on each component in S, and the normalization formula is: normalized value=(original value-minimum value) / (maximum value-minimum value). The data range after normalization is 0 to 1, and the normalized sequence is used as the input sample of the neural network recognition model.

[0063] The neural network recognition model is a feedforward fully connected neural network, including an input layer, two hidden layers and an output layer. The number of nodes in the input layer is equal to the total number of components contained in each normalized sample; the hidden layer uses the ReLU activation function, and the output layer uses the linear activation function to generate the predicted value. The network uses mean square error as the loss function, and uses the back propagation algorithm combined with the gradient descent method for parameter training. The training sample is the normalized historical eccentricity sequence, and the label is the value of the actual observed eccentricity vector ΔP at the next time. The training process is alternately performed on the training set and the validation set until convergence, and the final model with a prediction accuracy that meets the specified error range is obtained.

[0064] The newly collected n sets of core line eccentricity vector data sequence is input to the trained neural network recognition model, and the model automatically outputs the predicted core line eccentricity vector ΔP' corresponding to the next sampling time, which is a two-dimensional vector ΔP'= (X', Y'), corresponding to the predicted displacement of the core line in the X direction and the Y direction, respectively.

[0065] The predicted core line eccentricity vector ΔP' is calculated in modulus, that is, the Euclidean norm formula is used to calculate the modulus of ΔP' = √(X'2+ Y'2). The modulus value is compared with the preset threshold value ε. The threshold value ε is the maximum allowed core line displacement amplitude set according to the product quality standard, and the unit is millimeter. If the modulus value of ΔP' is greater than the threshold value ε, it is determined that the current core line eccentricity trend has exceeded the allowed range, and the subsequent compensation displacement calculation step needs to be entered.

[0066] If ΔP' < ε, the compensation displacement ΔX is calculated based on ΔP and ΔP', and the compensation displacement ΔX is used to adjust the current position of the cable extruder die or guide wheel to offset the core line displacement trend.

[0067] When the modulus value of the predicted core line eccentricity vector ΔP' output by the neural network recognition model is greater than the set preset threshold value ε, it is considered that there is a significant core line eccentricity change trend, and compensation displacement calculation needs to be performed for regulation and control. In this embodiment, the calculation process of the compensation displacement ΔX includes the following operation steps:

[0068] The core line eccentricity vector ΔP corresponding to the current sampling time and the predicted core line eccentricity vector ΔP' corresponding to the next sampling time output by the neural network recognition model are collected, wherein ΔP and ΔP' are both two-dimensional vector forms, representing the position displacement of the core line in the transverse (X direction) and longitudinal (Y direction), respectively, and the unit is millimeter.

[0069] The predicted eccentricity vector ΔP' is subtracted from the current eccentricity vector ΔP to obtain the core line eccentricity change vector ΔD, that is, The ΔD is a two-dimensional vector, and the X direction component represents the predicted displacement increment of the core line in the transverse direction, and the Y direction component represents the predicted displacement increment of the core line in the longitudinal direction.

[0070] Under the condition of stable operation of the equipment, based on high-precision displacement sensors and structural analysis methods, the displacement behavior of the die relative to the extrusion axis within a period of time is collected, the periodic displacement amplitude and the corresponding fluctuation frequency are extracted, and a set of structural fluctuation characteristic parameters is constructed. The periodic displacement amplitude is defined as the distance between the maximum and minimum positions of the die center within a complete fluctuation period; the fluctuation frequency is defined as the number of complete reciprocating displacements of the die per unit time, and the unit is hertz.

[0071] The displacement response of the extrusion device is measured under known control input conditions to establish a mapping between the input and the response. The displacement response characteristic parameter is defined as the actual displacement fault tolerance, i.e. the average displacement, generated by the moving part of the device under the action of a unit control instruction (such as a unit voltage or a step pulse), and is measured in millimeters per unit input.

[0072] Each component of the core wire eccentricity change vector ΔD is divided by the structural fluctuation characteristic amplitude in its corresponding direction, multiplied by the unit displacement gain coefficient defined in the response characteristic parameter, to obtain the compensation displacement ΔXx and ΔXy in the corresponding direction, and finally form the compensation displacement ΔX = (ΔXx, ΔXy). This displacement is used to adjust the position of the guide device or the mold core to offset the expected deviation trend of the core wire.

[0073] It is noted that in this embodiment, the unit displacement gain coefficient is denoted as Kdir, where dir represents the direction (X or Y direction), and is defined as the stable effective displacement generated by the guide part of the extrusion device in the direction under the action of a unit control input. Its calculation formula is: unit displacement gain coefficient Kdir = effective displacement amount / control input intensity, where: the effective displacement amount is the stable state displacement value generated by the extrusion device in the corresponding direction under the action of a single control input, recorded by a measuring device (such as a high-precision displacement sensor), and is measured in millimeters; the control input intensity is the amplitude of the control signal applied to the servo driver or the stepper motor, which can be voltage (volts), pulse number or PWM duty cycle percentage, depending on the specific control method.

[0074] After performing the compensation displacement ΔX, the extrusion path is updated and new images are continuously collected.

[0075] After the calculation of the compensation displacement ΔX is completed, the position of the core wire in the cross-sectional direction is fine-tuned by applying ΔX as a control input to the guide mechanism or the mold core position adjustment mechanism of the extrusion device. The compensation execution process adopts a closed-loop servo control method to ensure that the actual execution displacement is consistent with the calculated value, avoiding secondary eccentricity caused by execution errors. After the compensation action is completed, the spatial position of the core wire in the cross-sectional direction tends to return to the vicinity of the sheath center, thereby effectively reducing the eccentricity error.

[0076] After the compensation displacement ΔX is executed, the real-time extrusion path of the cable is immediately updated, the relative position relationship between the guide position and the mold core center line is reset, and the image collection process is restarted in the next sampling period. The cross-sectional image collected at this time will reflect the new position of the compensated core wire, and the relative relationship between the core wire and the sheath boundary in the image will be used as new input into the edge extraction and eccentricity vector calculation process, forming a continuous closed-loop image recognition-prediction-compensation control link to ensure stable operation of the eccentricity control under dynamic working conditions.

[0077] The application realizes high-precision identification and prediction control of the eccentricity change trend of the core wire relative to the sheath by collecting cable images in the cross-sectional direction in real time, combining edge recognition, geometric analysis and neural network prediction. Compared with the traditional method relying on fixed threshold or open-loop detection method, the application can dynamically obtain the core wire offset behavior, and output the compensation displacement in advance before the eccentricity trend exceeds the forming tolerance, effectively inhibiting the accumulation and amplification of eccentricity error in the production process, and improving the real-time performance of cable concentricity control.

[0078] Embodiment 2, please refer to Figure 2 The cable core wire eccentricity automatic detection and compensation system based on image recognition described in the embodiment comprises:

[0079] The image acquisition module collects real-time images of the running cable in the cross-sectional direction, including gray scale or color information of the core wire boundary and the sheath boundary;

[0080] The feature extraction module performs edge extraction and boundary segmentation on the real-time image, identifies the core wire region contour and the sheath region contour respectively, calculates the core wire centroid coordinates and the sheath center coordinates, and obtains the core wire eccentricity vector ΔP;

[0081] The neural network prediction module inputs the core wire eccentricity vector ΔP into the neural network identification model, fits and predicts the eccentricity trend, outputs the predicted eccentricity vector ΔP' at the next moment, and judges whether it exceeds the preset threshold ε;

[0082] The eccentricity compensation calculation module calculates the compensation displacement ΔX based on ΔP and ΔP' if ΔP' > ε, and the compensation displacement ΔX is used to adjust the current position of the cable extruder die core or guide wheel to offset the core wire offset trend;

[0083] The path update module updates the extrusion path after executing the compensation displacement ΔX and continues to collect new images.

[0084] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. An automatic detection and compensation method for cable core eccentricity based on image recognition, characterized in that: include: Real-time images of the cable in operation are acquired in the cross-sectional direction, including grayscale or color information of the core boundary and the sheath boundary; Edge extraction and boundary segmentation are performed on the real-time image to identify the core wire region contour and the sheath region contour respectively. The centroid coordinates of the core wire and the center coordinates of the sheath are calculated to obtain the core wire eccentricity vector ΔP. The core wire eccentricity vector ΔP is input into the neural network recognition model to fit and predict the eccentricity trend, output the eccentricity vector ΔP' expected at the next moment, and determine whether it exceeds the preset threshold ε. The step of outputting the predicted eccentricity vector ΔP' for the next moment includes: serializing multiple sets of continuously acquired core line eccentricity vectors ΔP in chronological order to construct a data sequence containing at least N historical eccentricity vectors, and normalizing the data sequence as input samples for a neural network recognition model; constructing a neural network recognition model for eccentricity trend prediction based on the input samples, wherein the neural network recognition model includes an input layer, at least one hidden layer, and an output layer, and training the model weights using a backpropagation algorithm; inputting the latest eccentricity vector data sequence into the trained neural network recognition model, outputting the predicted core line eccentricity vector ΔP' corresponding to the next sampling moment, and comparing the magnitude of the predicted core line eccentricity vector ΔP' with a preset threshold ε; If ΔP'>ε, then the compensation displacement ΔX is calculated based on ΔP and ΔP'. The compensation displacement ΔX is used to adjust the current position of the die core or guide wheel of the cable extruder to counteract the core wire offset trend. The calculation of the compensation displacement ΔX based on ΔP and ΔP' includes: if ΔP'>ε, obtaining the core wire eccentricity vector ΔP corresponding to the current sampling time and the eccentricity vector ΔP' output by the neural network recognition model for the next sampling time; performing vector difference calculation based on the core wire eccentricity vector ΔP and the eccentricity vector ΔP' to obtain the core wire eccentricity change vector ΔD; based on the core wire eccentricity change vector ΔD, combined with the structural fluctuation characteristics and displacement response characteristics of the cable extrusion equipment, performing proportional mapping calculation on the core wire eccentricity change vector ΔD to obtain the compensation displacement ΔX used to offset the core wire eccentricity change trend; wherein, the structural fluctuation characteristic parameters of the cable extrusion equipment under stable operating conditions are obtained, including the periodic offset amplitude of the die core relative to the extrusion axis and the corresponding fluctuation frequency; the displacement response characteristic parameters of the cable extrusion equipment are obtained, the displacement response characteristic parameters characterizing the actual displacement tolerance of the extrusion equipment under unit control input. After performing the compensation displacement ΔX, update the extrusion path and continue acquiring new images.

2. The automatic detection and compensation method for cable core eccentricity based on image recognition according to claim 1, characterized in that: The process of obtaining the core wire eccentricity vector ΔP includes: Perform grayscale conversion and adaptive filtering on real-time images; Based on the processed image, the first closed contour of the core wire region and the second closed contour of the sheath region are obtained respectively. The centroid coordinates of the core wire region are calculated based on the first closed contour, the center coordinates of the sheath region are calculated based on the second closed contour, and the core wire eccentricity vector ΔP is determined by the vector difference between the centroid coordinates and the center coordinates.

3. The automatic detection and compensation method for cable core eccentricity based on image recognition according to claim 2, characterized in that: Obtaining a first closed contour of the core wire region and a second closed contour of the sheath region further includes: identifying two closed boundaries in the image, wherein a small-area closed contour is defined as the first closed contour of the core wire region, and a large-area, approximately circular closed contour is defined as the second closed contour of the sheath region.

4. The automatic detection and compensation method for cable core eccentricity based on image recognition according to claim 1, characterized in that: The periodic offset amplitude is defined as the distance between the maximum and minimum positions of the mold core center within a complete fluctuation cycle; The fluctuation frequency is defined as the number of times the mold core undergoes a complete reciprocating offset per unit time.

5. An image recognition-based automatic detection and compensation system for cable core eccentricity, used to implement the image recognition-based automatic detection and compensation method for cable core eccentricity as described in any one of claims 1-4, characterized in that: include: Image acquisition module: Acquires real-time images of the cable in operation in the cross-sectional direction, including grayscale or color information of the core wire boundary and the sheath boundary; Feature extraction module: performs edge extraction and boundary segmentation on the real-time image, identifies the core wire region contour and the sheath region contour respectively, calculates the centroid coordinates of the core wire and the center coordinates of the sheath, and obtains the core wire eccentricity vector ΔP; Neural network prediction module: Input the core wire eccentricity vector ΔP into the neural network recognition model, fit and predict the eccentricity trend, output the eccentricity vector ΔP' expected at the next moment, and determine whether it exceeds the preset threshold ε. Eccentricity compensation calculation module: If ΔP' > ε, then calculate the compensation displacement ΔX based on ΔP and ΔP'. The compensation displacement ΔX is used to adjust the current position of the wire extruder die core or guide wheel to counteract the core wire offset trend. Path update module: After performing compensation displacement ΔX, update the extrusion path and continue to acquire new images.

Images