A cable production process real-time monitoring method and system

Abstract

The present application relates to the technical field of image processing, more particularly, the present application relates to a kind of cable production process real-time monitoring method and system, comprising: the image sequence of cable is collected, each frame of image in the image sequence is carried out optical enhancement processing, comprising: each frame of image is carried out low-pass filtering and obtains reflection structure model, each frame of image and the reflection structure model are carried out pixel level division operation and obtain the anti-reflection light enhancement image;The local information entropy of each pixel point in the anti-reflection light enhancement image is calculated, the luminance value of the anti-reflection light enhancement image is multiplied by the gain coefficient based on the local information entropy, and the final optical enhancement image is obtained.The present application is judged by comprehensive intensity, cross-frame stability and area and other multidimensional information, effectively excludes instantaneous interference such as flash noise, significantly reduces false positive rate and miss rate, and greatly improves the detection precision of micro defect.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method and system for real-time monitoring of the cable production process. Background Technology

[0002] As cable manufacturing evolves towards high-speed, continuous production, higher demands are placed on the real-time performance and accuracy of quality control. These demands not only affect the product's performance and lifespan but also directly impact the safe and stable operation of critical infrastructure such as power and communications. As the core carrier of electrical energy and data, cables are precisely composed of conductor cores, insulation layers, and outer sheaths. The integrity of the outer sheath directly determines the protective capabilities and transmission performance of the internal structure. However, the high gloss of the cable's surface exposes image acquisition during transmission to a complex optical environment. Its mirror-like properties, under industrial light sources, produce strong specular reflections, easily leading to large-scale continuous reflective bands and transient flash noise, among other interference.

[0003] However, early defects in cables, such as minor scratches and slight discoloration, are often masked by these optical artifacts, resulting in extremely low contrast and a severely inadequate signal-to-noise ratio. Although automated vision systems have been introduced, they currently rely primarily on simple fixed-parameter filtering and feature extraction. These methods lack the ability to perceive local image characteristics and cannot distinguish between dynamic optical interference and static, real defects. Because early cable defects have inconspicuous visual features and low contrast, and traditional methods struggle to adaptively remove optical interference, they are prone to filtering out equally weak defect signals when attempting to suppress strong reflections. This leads to a low recognition rate for minor defects and a risk of misjudgment and missed detection. Summary of the Invention

[0004] This invention provides a method and system for real-time monitoring of the cable production process, aiming to solve the problem in related technologies that, when attempting to suppress strong reflections, it is easy to filter out equally weak defect signals, resulting in a low identification rate of minor defects and the risk of misjudgment and missed detection.

[0005] In a first aspect, the present invention provides a real-time monitoring method for cable production processes, comprising: acquiring an image sequence of the cable; performing optical enhancement processing on each frame of the image sequence, including: performing low-pass filtering on each frame of the image to obtain a reflective structure model; performing pixel-level division operation on each frame of the image and the reflective structure model to obtain a de-reflective enhanced image; calculating the local information entropy of each pixel in the de-reflective enhanced image; multiplying the brightness value of the de-reflective enhanced image by a gain coefficient constructed based on the local information entropy to obtain a final optically enhanced image; and performing feature extraction on the final optically enhanced image to obtain a color anomaly index feature map and a multi-scale texture energy. The system generates a comprehensive defect index feature map. Based on the variance of the color anomaly index feature map and the multi-scale texture energy feature map over time, dynamic weights corresponding to color and texture are determined. The color anomaly index feature map and the multi-scale texture energy feature map are then weighted and fused to obtain a comprehensive defect index feature map. The comprehensive defect index feature map is segmented to obtain candidate defect regions. The comprehensive defect degree of each candidate defect region is calculated. The calculation method of the comprehensive defect degree includes: weighting and combining the intensity information, cross-frame stability, and area information of the candidate defect region, and judging whether the candidate defect region is a real defect region based on the comprehensive defect degree. By combining the reflective structure model with local information entropy for optical enhancement, reflection is effectively suppressed and defects are highlighted. Subsequently, dynamic weighted fusion is performed based on the temporal activity of color and texture features to adaptively focus on key features. Finally, a joint judgment is made based on multi-dimensional information such as comprehensive intensity, cross-frame stability, and area, thereby accurately identifying real defects and significantly improving the accuracy and robustness of detection.

[0006] Furthermore, the gain coefficient is constructed by multiplying the local information entropy by a preset weight coefficient and adding 1 to obtain the gain coefficient.

[0007] Furthermore, the method for obtaining the color anomaly index feature map includes: converting the final optically enhanced image to the CIE-Lab color space; calculating the color difference between each pixel in the final optically enhanced image and its local background color; calculating the neighborhood manifold consistency of each pixel in the final optically enhanced image; multiplying the color difference by the neighborhood manifold consistency to obtain the color anomaly index, thus obtaining the color anomaly index feature map constructed from the color anomaly indices of all pixels. By multiplying the color difference by this consistency index, it is possible to effectively distinguish between genuine color spot defects and pseudo-color difference regions caused by reflection. Because the color of a genuine color spot is inconsistent with its surroundings (low manifold consistency), while the color inside a reflective region is highly consistent (high manifold consistency), this method can accurately identify color-abnormal and locally uneven regions, greatly reducing color misjudgment caused by reflection.

[0008] Furthermore, the calculation method for neighborhood manifold consistency is as follows: The color distances between all pixels in the local neighborhood of each pixel in the final optically enhanced image and the center pixel are exponentially decayed and summed. By exponentially decaying and summing the color distances between all pixels in the neighborhood and the center point, the color smoothness of the local region is quantified. Compared to simple neighborhood mean or variance calculations, this exponential decay method is more sensitive to the color similarity of neighboring pixels and can more accurately reflect the local color consistency.

[0009] Furthermore, the method for obtaining the texture energy feature map includes: performing multi-scale edge detection on the final optically enhanced image to obtain texture energy at multiple scales; performing weighted summation on the texture energy at multiple scales to obtain the final multi-scale texture energy, thereby obtaining the multi-scale texture energy feature map. By performing multi-scale detection and weighted summation, texture defects of different sizes can be captured simultaneously. For example, small-scale detection can highlight minute scratches, while large-scale detection can capture long ripples. This multi-scale fusion strategy significantly broadens the detection range, enabling the system to have a high detection capability for various types of texture defects.

[0010] Furthermore, the method for determining the dynamic weights includes: using the ratio of the variance of the color anomaly index feature map or multi-scale texture energy feature map over time to the sum of the variances of all feature maps over time as the dynamic weight of the corresponding feature map. By calculating the variance of each feature map over time to allocate weights, adaptive adjustment of the fusion process is achieved. Compared to the information loss caused by fixed-weight fusion in existing technologies, this method can automatically increase the weight of corresponding features based on the main manifestations of recent defects, such as drastic color changes or drastic texture changes. This allows the system to dynamically focus on the most critical defect information, thereby maintaining optimal detection performance when facing different types of defects.

[0011] Furthermore, the segmentation method for the comprehensive defect index feature map includes: calculating a dynamic threshold for the comprehensive defect index feature map, wherein the dynamic threshold is calculated by adding the mean and standard deviation of the comprehensive defect index feature map by a preset multiple; identifying pixels in the comprehensive defect index feature map that are greater than the dynamic threshold as candidate defect pixels; and performing connected component analysis on the candidate defect pixels to obtain the candidate defect region. Compared with the method using a fixed threshold, this dynamic threshold method has stronger robustness to fluctuations in the production environment such as changes in illumination and differences in cable materials, ensuring that candidate defect regions can be accurately separated from the background under different working conditions.

[0012] Furthermore, the method for calculating the cross-frame stability is as follows: obtain the area sequence of the candidate defect region in multiple consecutive frames of images, calculate the variance of the area sequence, and exponentially decay the variance to obtain the cross-frame stability of the candidate defect region. By calculating the variance of the area sequence of the candidate region in multiple consecutive frames of images and performing exponential decay, the temporal persistence of the target is effectively quantified.

[0013] Furthermore, the low-pass filter employs a Gaussian filter.

[0014] In a second aspect, the present invention also provides a real-time monitoring system for cable production processes, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the real-time monitoring method for cable production processes described in any of the above claims.

[0015] Beneficial effects: The first step, optical enhancement, through low-pass filtering and gain adjustment based on information entropy, can adaptively suppress strong reflections while enhancing weak defect signals, solving the problem that traditional methods easily filter out defect information when de-reflecting. The second step, through dynamic weighted fusion of color and texture features, enables the algorithm to automatically focus on the most significant defect type, improving the flexibility and robustness of detection. The third step, by integrating multi-dimensional information such as intensity, cross-frame stability, and area, effectively eliminates instantaneous interference such as flash noise, significantly reducing the false alarm rate and false negative rate, and greatly improving the detection accuracy of small defects. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating a cable defect monitoring process according to an embodiment of the present invention;

[0017] Figure 2 This is a flowchart illustrating the calculation of the comprehensive defect index of a pixel according to an embodiment of the present invention;

[0018] Figure 3 This is a schematic illustration of a cable surface according to an embodiment of the present invention;

[0019] Figure 4 This is a schematic illustration of the result after light and dark field compensation according to an embodiment of the present invention;

[0020] Figure 5 This is a schematic diagram illustrating the local information entropy distribution according to an embodiment of the present invention;

[0021] Figure 6 This is a schematic illustration of the final optically enhanced image according to an embodiment of the present invention;

[0022] Figure 7 This is a schematic diagram illustrating the annotation of candidate defect areas according to an embodiment of the present invention. Detailed Implementation

[0023] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] like Figures 1 to 7 As shown, S101: Acquire cable images and perform optical enhancement processing on the images.

[0025] Industrial cameras are used to continuously and at high speed acquire images of the cable production line, obtaining images of the cable surface (such as...). Figure 2 (As shown). To accurately focus on the effective detection area of ​​the cable, a rolling window sampling method is used. The size of this window matches the actual width of the cable, and it slides and captures images in one-frame increments. This method allows extraction of the region containing the complete cable cross-section from each frame, forming a time-sequential frame sequence. ,in The time frame number, These are the pixel coordinates within the window.

[0026] Considering that the high reflectivity of the cable surface will create large-scale continuous reflective bands in the image, and that these reflective bands are low-frequency information that will severely affect the subsequent extraction of high-frequency defect features, this embodiment constructs a local brightness model to estimate and compensate for this reflection.

[0027] First, a local brightness model is constructed to estimate the large-scale reflective structure. This model is based on the fact that large-scale reflections are low-frequency components in the image, which can be effectively estimated by low-pass filtering the cable surface image. The specific calculation formula is as follows: In the formula, In order to be in Large-scale reflective structures estimated in real time. Indicates in Images of the cable surface captured at all times. To represent a low-frequency smoothing kernel, a large Gaussian kernel can be used as a preferred option. For an image with a resolution of 2048×2048 pixels, the preferred range for the Gaussian kernel size σ is 100 to 400 pixels. For example, an empirical value of 200 pixels can be selected as the initial calibration value. The principle for selecting this parameter is that the kernel size should be large enough to smooth out real local defects (high-frequency signals), but not too large to accurately fit the contour of the reflective band. If the value is too small, defect information may be incorrectly introduced into the local brightness model, resulting in weakened contrast of defects during subsequent compensation.

[0028] After obtaining the reflective structure, a de-reflective enhanced image is generated by performing pixel-level division operations on it. The calculation formula is based on the principle that dividing the brightness value of the cable surface image by its corresponding low-frequency background brightness value normalizes the brightness, thereby effectively suppressing the intensity of the reflective area. The formula is as follows: In the formula, where, for The de-reflection enhanced image corresponding to the cable surface image at time [time]. for Image of the cable surface at any given time. In order to be in Large-scale reflective structures estimated at any given time. For a very small positive number, such as Its function is to prevent the denominator from being zero.

[0029] From this formula, we can see that, In areas with larger values, i.e., areas with stronger reflection, this division operation can significantly reduce pixel intensity, thereby weakening the effect of reflective bands; while in... In smaller areas, the image intensity remains essentially unchanged. This process effectively removes reflections while preserving local color difference information, thus stabilizing the color gamut.

[0030] After light and dark field compensation (such as) Figure 3 As shown in the image, there may still be transient flickering bright spots caused by factors such as ambient light or electrical sparks. This type of noise is characterized by high brightness in localized areas but low information entropy. This embodiment utilizes this characteristic to construct an optical entropy suppression model to further purify the image.

[0031] First, calculate the local information entropy of the image (e.g., Figure 4 The image shown is a local information entropy distribution map. Its construction is based on the fact that information entropy can measure the randomness or complexity of pixel grayscale distribution within a region. Real defects or textured areas typically have more complex grayscale distributions and higher information entropy; while reflective or flickering noise areas are usually large, uniform bright areas with lower information entropy. The formula for calculating local information entropy is: In the formula, This represents the local information entropy of a pixel (x, y) in an image. This indicates that within the local neighborhood of pixel (x, y), the gray value is... The probability distribution of pixels.

[0032] As the formula shows, the more chaotic the distribution of gray values ​​in a local neighborhood and the higher the amount of information, the better the calculated value will be, such as in the case of a real defect. The larger the value, the lower the information content, such as in cases of reflection or noise. The smaller the value.

[0033] Subsequently, an optical entropy suppression map is constructed based on local information entropy, and the final optically enhanced image (e.g.) is then built. Figure 5 (As shown). Its construction is based on the principle that adjusting image brightness through a gain factor positively correlated with information entropy can suppress low-entropy regions and enhance high-entropy regions. The formula is as follows: In the formula, for The final optically enhanced image at that moment, for The de-reflection enhanced image corresponding to the cable surface image at time [time]. It represents the local information entropy of a pixel (x,y) in an image. This is a weighting coefficient used to adjust the enhancement strength of information entropy. Its value can be determined empirically through offline experiments, and the preferred value can be 1.0, 1.5 or 2.0.

[0034] As can be seen from the formula, when the independent variable... The smaller, The closer to 1, the better. The smaller the increase, the more effectively the brightness of the reflective band is suppressed; while when When the value is large, the enhancement effect is improved, achieving a true increase in the brightness of localized defects. Dependent variable The higher the value, the more likely the area is to be a real defect.

[0035] The above steps yield the final optically enhanced images at each time point, which are then used for subsequent processing.

[0036] S102: Perform dual feature extraction and dynamic fusion of color domain and texture domain on the final optically enhanced image, and calculate the comprehensive defect index of each pixel.

[0037] This step aims to extract features in both color and texture dimensions from the final optically enhanced image in parallel and fuse them through an adaptive dynamic weighting mechanism to form a comprehensive defect index that can fully characterize various types of defects.

[0038] S1021: Calculate the color difference of each pixel.

[0039] To quantify color anomalies caused by defects such as blemishes, this embodiment employs an improved ΔE model. First, the final optically enhanced image is converted to the CIE-Lab color space, as the Euclidean distance in this space better corresponds to the human eye's perception of color difference. The formula for calculating color difference is based on the Euclidean distance between two points in the CIE-Lab space, used to measure the difference between a pixel and its local background color. The formula is as follows: In the formula, For pixels Color difference , and These represent the brightness, red-green hue, and blue-yellow hue of the pixel (x, y) in the CIE-Lab color space. , , It is a pixel. Mean brightness, mean green-red intensity, and mean blue-yellow intensity within a local neighborhood. The larger the value, the more significant the color anomaly of that pixel.

[0040] S1022: Calculate the neighborhood manifold uniformity of each pixel.

[0041] However, using only the color difference of each pixel can misclassify the uniform color areas remaining in the reflective band as anomalies. Therefore, a neighborhood manifold consistency of each pixel is introduced as a constraint. This is based on the fact that the color of a true defect is usually inconsistent with its surrounding environment, while the color of the reflective band region exhibits high local consistency. The formula for calculating the neighborhood manifold consistency is: In the formula, Represents pixels The neighborhood manifold uniformity. This represents all pixels in a local neighborhood centered at pixel (x, y). This represents the local neighborhood. The total number of pixels within. A vector representing pixel color, i.e. In the formula, , , Each pixel Lightness, red-green hue, and yellow-blue hue in the CIE-Lab color space. This represents the pixels within the local neighborhood. With the center pixel The color vector distance between them.

[0042] In the above formula, the independent variable The smaller the value, the more consistent the color within the neighborhood, and the smoother the manifold. The larger the value, the greater the likelihood that it belongs to a reflective zone. At this point, the exponential function... The closer to 1, the better. When color consistency is high, such as in reflective areas, The larger. It reflects the smoothness of color within the local neighborhood of a pixel. If The larger the value, the more uniform the local color distribution, the greater the likelihood that it belongs to a reflective zone, and the lower the likelihood that it is a real defect.

[0043] S1023: Calculate the color anomaly index for each pixel.

[0044] Finally, the color difference of each pixel is combined with the manifold consistency of the neighborhood to obtain the corrected color anomaly index, thus obtaining the color anomaly index feature map. Its construction is based on utilizing (1) As an inhibitory factor, when When the level is high, the inhibitory effect is strong; when At low levels, the inhibitory effect is weak. The formula is as follows: In the formula, This represents the color anomaly index of pixel (x, y). Represents pixels Color difference Represents pixels The neighborhood manifold consistency. Only when a region simultaneously satisfies both the conditions of color anomaly and neighborhood color inconsistency will its color anomaly index be enhanced, thereby effectively distinguishing true color spots from reflective band artifacts.

[0045] S1024: Calculate texture energy at each scale.

[0046] To capture texture defects such as scratches and ripples, this embodiment employs a multi-scale local energy model. This model achieves this by detecting texture abrupt changes or edge intensity at different scales.

[0047] The formula for calculating texture energy at a single scale is based on obtaining local energy through convolution of the image with an edge / gradient extraction operator. The formula is as follows: In the formula, Represents the texture energy at scale s. for The final optically enhanced image at that moment, For scale The derivative of the Gaussian kernel, This represents the convolution operation. The larger the value, the greater the texture abruptness or edge intensity at scale s, thus obtaining the texture energy of each pixel in the image.

[0048] To simultaneously detect texture defects of different sizes, such as fine scratches and wider ripples, the texture energies at multiple different scales are weighted and summed to obtain the final multi-scale texture energy, thus yielding a multi-scale texture energy feature map. The calculation formula is as follows: In the formula, Represents multi-scale texture energy. Representing a set of scales, in the embodiments, . Representing scale Texture energy. Representing scale The weights are set, for example, the weights of all scales are set to the same weight, such as 0.8 for each scale. Smaller scales are given higher weights to highlight minor scratches, while larger scales are given higher weights to capture long wavy defects.

[0049] S1025: Calculate the overall defect index of each pixel.

[0050] After obtaining the color anomaly index and multi-scale texture energy, they need to be fused into a single comprehensive defect index to obtain the comprehensive defect index feature map. Since the dominant features of defects may switch between color and texture at different times, using fixed weights for fusion would lead to information loss. Therefore, this embodiment constructs a dynamic weighted fusion model. The fusion formula is: In the formula, For pixels The comprehensive defect index. and The dynamic weights are for color and texture. For pixels Color abnormality index, For pixels Multiscale texture energy.

[0051] The dynamic weights are calculated based on feature activity, meaning that the feature domain (color or texture) that changes more dramatically over a recent period is assigned a higher weight. This is achieved by calculating the variance of the feature map sequence. The formula is as follows: In the formula, where, yes The dynamic weight of the i-th feature map at time i. In order to be in Within the motion time window of the 20 frames prior to time, the pixel on the i-th feature map The sequence of values ​​at each position. It is the variance of the sequence composed of the i-th feature map. When i=1 is the color anomaly index feature map and i=2 is the multi-scale texture energy feature map, This represents the number of feature maps. It is the sum of the variances of all the constituent sequences.

[0052] It is important to note that the moving time window uses a sliding method with a step size of 1 frame. Each time a new image is acquired, the window automatically removes the earliest frame data and incorporates the new frame data to ensure that the weight calculation is based on the latest 20 frame feature sequences.

[0053] As shown by this formula, if color features change drastically in the near future, the dynamic weight corresponding to color will increase, causing the system to pay more attention to color anomalies; conversely, the dynamic weight corresponding to texture will decrease. This method ensures that the fusion model can adaptively focus on the most significant defect features at present.

[0054] S103: Candidate defect region determination and multi-dimensional constraint scoring.

[0055] The purpose of this step is to accurately segment candidate defect regions from the comprehensive defect index feature map, and to filter out residual noise and artifacts by introducing multi-dimensional physical constraints such as cross-frame stability and region area. Finally, a comprehensive defect degree reflecting the true severity of each candidate defect region is calculated.

[0056] To initially extract potential defect regions from the comprehensive defect index feature map, an adaptive dynamic threshold is employed. This threshold is based on the premise that the overall brightness and noise level of the image are dynamically changing, and a fixed threshold cannot adapt. The threshold should be adjusted according to the statistical characteristics of the image content. The calculation formula is as follows: In the formula, This indicates a dynamic threshold. This represents the mean value of the feature map of the comprehensive defect index of the current frame. This represents the standard deviation of the feature map of the comprehensive defect index of the current frame. This represents the adaptive adjustment coefficient, which is applied when overall reflectivity enhancement is detected. The dynamic threshold is also increased accordingly, thereby improving the dynamic threshold and avoiding a large number of false alarms caused by strong reflection. In this embodiment, The value is 3. By extracting all pixels greater than the dynamic threshold from the comprehensive defect index feature map, a preliminary set of candidate defect pixels is formed.

[0057] Connectivity analysis is performed on the candidate defect pixel set to segment it into several independent candidate defect regions. For each candidate defect region, the following key descriptors are calculated: Region Area: the total number of pixels in the region; Average Intensity: the average of the comprehensive defect indices of all pixels in the region; Maximum Intensity: the maximum value of the comprehensive defect indices of all pixels in the region.

[0058] Transient flicker noise is characterized by its highly unstable position and area across consecutive frames, while true defects exhibit high temporal stability. Utilizing this prior knowledge, a cross-frame stability index is constructed to filter out noise. Its construction is based on the fact that the variance of the region's area over time can effectively measure its stability. The formula is as follows: In the formula, It is a candidate defect region Cross-frame stability. It is a candidate defect region exist The area of ​​the region at any given time. Candidate defect region The variance of the area of ​​the region in the past Q consecutive frames, where Q is 5. It represents an exponential function with the natural constant e as its base.

[0059] Finally, by integrating the intensity, stability, and scale information of the region, a comprehensive defect severity assessment is constructed. This model is based on the premise that real defects typically exhibit high intensity, high stability, and small area characteristics, while large, low-intensity areas are more likely to be residual reflections. The formula is as follows: ; Candidate defect region The degree of overall defects, Candidate defect region Maximum local intensity Candidate defect region average strength, Candidate defect region Cross-frame stability, Candidate defect region Area , and These represent the maximum value of the maximum local intensity, the maximum value of the average intensity, and the maximum value of the region area, respectively, among all candidate defect regions in the current frame. to These are the weighting coefficients. , , For positive weighting coefficients, satisfying + + ≤1; This is the area penalty coefficient. ≥0. In this embodiment, the values ​​of each weight coefficient are: =0.3、 =0.2、 =0.3、 =0.2.

[0060] S104: Determine whether a region is a real defect region based on the comprehensive defect severity of each candidate defect region.

[0061] A final judgment threshold is preset, and in this embodiment, the value of the final judgment threshold is 0.3. For any candidate defect area, if its overall defect level is greater than or equal to the final judgment threshold, the candidate defect area is confirmed as a real defect area, the real defect area is marked, and a prompt is issued to the staff.

[0062] The present invention also provides a real-time monitoring system for cable production processes. The system includes a processor and a memory, the memory storing computer program instructions. When the processor executes the computer program instructions, it implements a real-time monitoring method for cable production processes according to the first aspect of the present invention.

[0063] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0064] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0065] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for real-time monitoring of a cable production process, characterized by, The method comprises the following steps: An image sequence of the cable is collected, and each frame of image in the image sequence is subjected to optical enhancement processing, which comprises the following steps: low-pass filtering each frame of image to obtain a reflection structure model, and performing pixel-level division operation on each frame of image and the reflection structure model to obtain a reflection-removed enhanced image; Local information entropy of each pixel point in the reflection-removed enhanced image is calculated, and a brightness value of the reflection-removed enhanced image is multiplied by a gain coefficient constructed based on the local information entropy to obtain a final optical enhanced image; Feature extraction is performed on the final optical enhanced image to obtain a color anomaly index feature map and a multi-scale texture energy feature map; Based on variances of the color anomaly index feature map and the multi-scale texture energy feature map in a time sequence, a dynamic weight corresponding to color and a dynamic weight corresponding to texture are determined, and the color anomaly index feature map and the multi-scale texture energy feature map are fused to obtain a comprehensive defect index feature map; The comprehensive defect index feature map is segmented to obtain a candidate defect region, and a comprehensive defect degree of each candidate defect region is calculated, wherein the method for calculating the comprehensive defect degree comprises the following steps: intensity information, cross-frame stability and area information of the candidate defect region are combined by weighting, and whether the candidate defect region is a real defect region is determined according to the comprehensive defect degree; The method for obtaining the color anomaly index feature map comprises the following steps: The final optical enhanced image is converted to a CIE-Lab color space, and color difference between each pixel point in the final optical enhanced image and a local background color of the pixel point is calculated; Neighborhood manifold consistency of each pixel point in the final optical enhanced image is calculated; The color difference is multiplied by the neighborhood manifold consistency to obtain a color anomaly index, and a color anomaly index feature map constructed by color anomaly indexes of all pixel points is obtained; The method for obtaining the texture energy feature map comprises the following steps: Multi-scale edge detection is performed on the final optical enhanced image to obtain texture energy in multiple scales; 2. The cable production process real-time monitoring method according to claim 1, characterized in that, The texture energy in multiple scales is summed by weighting to obtain multi-scale texture energy, thereby obtaining a multi-scale texture energy feature map. The method for constructing the gain coefficient comprises the following steps:

3. The real-time monitoring method of a cable production process according to claim 1, characterized in that, The product of the local information entropy and a preset weight coefficient is added by 1 to obtain a gain coefficient. The method for calculating the neighborhood manifold consistency comprises the following steps:

4. The cable production process real-time monitoring method of claim 1, wherein Color distance between all pixel points in a local neighborhood of each pixel point in the final optical enhanced image and the center pixel point is exponentially attenuated and summed. The method for determining the dynamic weight comprises the following steps:

5. The cable production process real-time monitoring method of claim 1, wherein A ratio of a variance of the color anomaly index feature map or the multi-scale texture energy feature map in a time sequence to a sum of variances of all feature maps in a time sequence is taken as a dynamic weight of the corresponding feature map. The method for segmenting the comprehensive defect index feature map comprises the following steps: A dynamic threshold of the comprehensive defect index feature map is calculated, and the method for calculating the dynamic threshold comprises the following steps: a preset multiple of a mean value and a standard deviation of the comprehensive defect index feature map is added; Pixel points in the comprehensive defect index feature map greater than the dynamic threshold are taken as candidate defect pixel points. The candidate defect pixel points are subjected to connected domain analysis to obtain the candidate defect region.

6. The cable production process real-time monitoring method of claim 1, wherein The calculation method of the cross-frame stability is: An area sequence of the candidate defect region in continuous multiple frames of images is obtained, a variance of the area sequence is calculated, and the variance is subjected to exponential decay to obtain the cross-frame stability of the candidate defect region.

7. The cable production process real-time monitoring method of claim 1, wherein The low-pass filtering adopts Gaussian filtering.

8. A real-time monitoring system for a cable production process, comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to realize the cable production process real-time monitoring method according to any one of claims 1-7.

Images