CN-121981998-A - Cable sheath surface defect real-time detection and classification system based on machine vision

CN121981998ACN 121981998 ACN121981998 ACN 121981998ACN-121981998-A

Abstract

The invention relates to the technical field of machine image detection, and particularly discloses a system for detecting and classifying surface defects of a cable sheath in real time based on machine vision. The system comprises a high-speed synchronous image acquisition module, an illumination self-adaptive compensation module, an image preprocessing and enhancing module, a multi-stage defect detection and feature extraction module, a real-time classification decision module and a system control and feedback execution module. Through the collaborative work of the modules, the high-quality image acquisition, the high-efficiency and accurate detection classification and the real-time decision feedback of the surface defects of the cable sheath are realized, and the detection automation efficiency and the intelligent level of the production line are improved.

Inventors

GUO YAPENG
SHEN FANG
WANG QI
HAN YINGHAO
HE RUIQI
CHANG ZHIWEI
LI ZONGFANG
SUN SHIHAO

Assignees

河南国网电缆集团有限公司

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. The utility model provides a cable sheath surface defect real-time detection and classification system based on machine vision which characterized in that includes: The linear array image acquisition module is used for acquiring complete image data of the surface of the sheath of the cable in the continuous movement process of the cable; the illumination self-adaptive compensation module is used for eliminating interference of uneven ambient illumination and reflection of light on the surface of the cable sheath on imaging quality; The image preprocessing and enhancing module is used for receiving the original image data stream from the linear array image acquisition module and executing a series of operations aiming at improving the signal-to-noise ratio of the defect area; The multistage defect detection and feature extraction module adopts a cascade architecture to perform coarse-to-fine defect positioning and feature quantization on the preprocessed image; the real-time classification decision module is used for receiving the high-dimensional feature vector output by the multi-stage defect detection and feature extraction module and executing final defect type accurate classification and confidence assessment; and the system control and feedback execution module is used as a dispatching center and an execution terminal of the whole system.
2. The machine vision-based cable sheath surface defect real-time detection and classification system according to claim 1, wherein the linear array image acquisition module comprises a linear array camera, a pulse trigger unit synchronous with a production line spindle encoder and an auxiliary positioning laser ranging unit; The linear array camera is vertically arranged along the axial direction of the cable, and the acquisition line frequency is strictly synchronous with the pulse signal output by the encoder through the pulse triggering unit; The laser ranging unit measures the working distance between the surface of the cable and the camera lens in real time, and feeds back working distance data to an automatic focusing servo mechanism of the camera lens to dynamically compensate imaging focal length change caused by radial runout of the cable.
3. The machine vision-based real-time detection and classification system for cable jacket surface defects according to claim 2, wherein the illumination self-adaptive compensation module comprises an annular multi-partition programmable light source and an embedded light intensity sensor array, the annular light source is arranged around a camera lens of a linear array, and the light emitting surface is divided into a plurality of independently controlled light emitting partitions; the embedded light intensity sensor array is integrated in the light source, and the illumination intensity of each direction projected to the surface of the cable is monitored in real time in an annular arrangement; the illumination self-adaptive compensation module is internally provided with an illumination balance control algorithm, the illumination balance control algorithm aims at obtaining uniform illumination gray level distribution, reads the data of each sensor in real time, calculates uniformity deviation of a current illumination field, and dynamically adjusts driving current of each luminous partition until readings of all the sensors reach a preset balance threshold range.
4. The machine vision based cable sheath surface defect real-time detection and classification system according to claim 3, wherein the image preprocessing and enhancement module firstly performs background estimation and subtraction based on a gaussian model to eliminate periodic interference of the cable sheath base texture; Then, an adaptive contrast stretching algorithm based on local gray statistics is adopted, the adaptive contrast stretching algorithm divides an image into a plurality of overlapped subareas, gray average value and standard deviation of each subarea are calculated, and contrast gain of the subarea is dynamically adjusted according to deviation degree of pixel gray values in the subarea relative to regional statistical characteristics; And finally, applying an anisotropic diffusion filter with adjustable direction, wherein the diffusion coefficient of the anisotropic diffusion filter is inversely related to the local gradient amplitude of the image, and the diffusion direction of the anisotropic diffusion filter is calibrated according to the main direction of the surface texture of the cable.
5. The system for detecting and classifying surface defects of a cable sheath in real time based on machine vision according to claim 4, wherein the multi-stage defect detection and feature extraction module comprises a first-stage fast suspected region detection unit, a second-stage defect confirmation and rough classification unit and a third-stage high-dimensional feature vector extraction unit; the first-stage rapid suspected region detection unit calculates an absolute difference image of a current frame image and a dynamically updated background template image, adopts a double threshold value adaptive to local gray variance of the image to carry out binarization segmentation, and initially extracts all gray abnormal regions as suspected defect candidate regions; The second-stage defect confirmation and rough classification unit builds a multi-scale gradient amplitude pyramid at the corresponding position of the original image of each candidate region, calculates a statistical histogram of the gradient direction in the region under each scale and the spatial distribution entropy of the gradient amplitude, and distinguishes the candidate region into edge defects, planar defects or noise interference by analyzing the consistency degree of the gradient direction and the numerical value of the distribution entropy; The third-stage high-dimensional feature vector extraction unit extracts a group of quantized feature vectors containing geometric, texture and frequency domain information for the defect region confirmed by the second stage.
6. The machine vision-based cable sheath surface defect real-time detection and classification system according to claim 5, wherein the core of the real-time classification decision module is a hybrid classification model which is subjected to offline training and online lightweight deployment; The mixed classification model is formed by connecting a support vector machine classifier and a shallow convolutional neural network classifier in parallel; the support vector machine classifier adopts a radial basis function as a kernel function and is responsible for classifying the feature vectors and outputting decision function values; the shallow convolutional neural network classifier directly takes a normalized image block of a defect area as an input and is used for learning local pixel mode characteristics and outputting classification probability distribution; A decision fusion unit is arranged in the real-time classification decision module; The decision fusion unit converts the decision function value of the support vector machine classifier into a probability value through a sigmoid function, carries out weighted average fusion with probability distribution output by a shallow convolutional neural network, dynamically sets a weighting coefficient according to the historical accuracy of the two classes of classifiers on a verification set, determines the class corresponding to the highest probability value in the fused probability distribution as a final classification result, and records the highest probability value as the confidence coefficient of the classification result.
7. The machine vision-based cable jacket surface defect real-time detection and classification system of claim 6, wherein said system control and feedback execution module comprises a real-time industrial controller; The real-time industrial controller runs a multi-task real-time operation system, the system control and feedback execution module receives a defect classification result, position information and confidence coefficient from the real-time classification decision module, and when the classification confidence coefficient is larger than a preset confidence coefficient threshold value, the system control and feedback execution module judges that the system control and feedback operation system is effective defect detection and effective defect detection; The system control and feedback execution module precisely calculates the absolute length position of the defect on the whole cable according to the position information in the linear array image stream by combining with the pulse count of the encoder, and then sends a structured report containing the defect type, position, size and confidence to an upper monitoring system through an industrial Ethernet interface, and simultaneously generates a synchronous trigger signal; the synchronous trigger signal is accurately triggered when a defect point moves along with the cable to reach a marking station of a production line, and the inkjet marking device is driven to mark at a position corresponding to the surface defect of the cable sheath; The system control and feedback execution module is also responsible for coordinating the triggering time sequence of the linear array image acquisition module, the control period of the illumination self-adaptive compensation module and the priority and resource allocation of each software processing thread.
8. The system for detecting and classifying surface defects of a cable sheath based on machine vision according to claim 7, wherein the multi-scale gradient field analysis process adopted by the second-stage defect confirmation and rough classification unit is as follows, namely, for each suspected defect candidate region, intercepting and expanding a region-of-interest image block of 10 pixels at an original image corresponding position; On each scale image, calculating Sobel gradient in x direction and y direction, and synthesizing gradient amplitude diagram and gradient direction diagram; on the gradient direction diagram, counting the gradient directions of all pixel points in the candidate region, quantizing the gradient directions into 8 direction intervals, forming a direction histogram, and calculating the entropy value of the direction histogram as a gradient direction consistency measure; on the gradient amplitude diagram, calculating the mean value and standard deviation of gradient amplitudes in the candidate region, and calculating the local entropy of the spatial distribution of the gradient amplitudes; Setting a group of judgment rules, namely judging the edge type defect if the entropy value of the direction histogram is smaller than a first threshold value and the average value of the gradient amplitude is larger than a second threshold value; If the entropy of the direction histogram is high and the spatial distribution entropy of the gradient amplitude is greater than a third threshold, judging that the defect is a planar defect; if the gradient amplitude mean value is smaller than the fourth threshold value, the noise interference is determined to be filtered.
9. The machine vision-based cable sheath surface defect real-time detection and classification system according to claim 8, wherein the specific execution steps of the illumination balance control algorithm are as follows: After each image acquisition period is finished, reading the numerical values of 8 light intensity sensors integrated in the annular light source; calculating the average value of the 8 values as the target illumination intensity; Calculating absolute differences of each sensor reading and the target intensity; If the difference value of a certain sensor is larger than a preset tolerance range, generating a current adjustment instruction of the sensor corresponding to the luminous subarea, wherein the current adjustment quantity is in direct proportion to the difference value, the proportionality coefficient is determined through pre-calibration, the adjustment instruction is sent to a constant current driving circuit of each subarea through a digital-analog converter, after adjustment is completed, the short stabilization time is waited, the sensor value is read again for verification until the difference value of all the sensor readings and the target intensity is within the tolerance range, or the maximum adjustment iteration times are reached for 5 times.
10. The system for detecting and classifying surface defects of a cable sheath in real time based on machine vision according to claim 9, wherein the method for calculating the absolute length position in the system control and feedback execution module is as follows: When the system is initialized, the actual physical length corresponding to each line of images of the linear array camera is marked as the resolution of each line by the laser ranging unit; When the real-time classification decision module reports the defect, the line number of the defect in the image frame is reported at the same time; Calculating the length offset of the defect relative to the initial position of the current image frame according to the defect line number and the resolution of each line; Calculating the absolute length of the initial position of the current image frame on the whole cable according to the total number of the encoder pulses and the preset corresponding length of each pulse; and finally, adding the absolute length and the length offset to obtain the accurate absolute length position of the defect.

Description

Cable sheath surface defect real-time detection and classification system based on machine vision Technical Field The invention belongs to the technical field of machine image detection, and particularly relates to a cable sheath surface defect real-time detection and classification system based on machine vision. Background Machine vision and artificial intelligence technology play an increasingly important role in the field of industrial automation and quality detection, and by simulating the human visual system and combining with powerful data processing capability, the quick and non-contact assessment of the appearance, size and surface quality of products on a production line is realized. Automatic detection technology of surface defects based on image processing has become a key means for improving the intelligent level of manufacturing industry and the quality control precision of products. The automatic detection of the surface defects of the cable sheath is an important link for guaranteeing the safety of power transmission and the reliability of cable products. The technology aims at acquiring visual information of the surface of the cable sheath through an image acquisition device, and automatically identifying and classifying various defects such as scratches, pits, bulges, impurity embedding and the like by utilizing an image processing and pattern recognition algorithm so as to replace the traditional low-efficiency, subjective and fatigued operation mode depending on manual visual inspection. The prior art generally adopts a fixed industrial camera for image acquisition and combines a traditional image processing algorithm or a deep learning model based on a convolutional neural network for defect identification. However, under the high-speed running scene of an actual continuous production line, the existing scheme has the following problems that the quality of an original image is obviously reduced due to the problems of image blurring and incomplete acquisition caused by high-speed movement, the robustness of feature extraction is insufficient due to the complex illumination environment and the reflection characteristic of the cable surface, the capability of distinguishing micro defects from similar texture backgrounds is limited by a traditional classification algorithm, and the deep learning method is limited by contradiction between real-time requirements and scarcity of marked data. The existing system is often used for loosely coupling the image acquisition, processing and classification modules, and the end-to-end optimization aiming at the beat of the production line is lacking, so that the detection delay is high, and real-time feedback and control are difficult to realize. Therefore, how to realize real-time detection and classification of cable sheath surface defects with high accuracy, high robustness and low delay in high-speed industrial sites becomes a technical problem to be solved urgently. Disclosure of Invention The invention aims to provide a machine vision-based real-time detection and classification system for surface defects of a cable sheath, which is used for solving the technical contradiction that the automatic detection efficiency of the surface defects of the cable sheath is low due to the fact that the image acquisition quality is affected by movement and illumination, the defect feature extraction robustness is insufficient, and the detection instantaneity and the classification precision are difficult to consider in a high-speed continuous production line scene. In order to achieve the above object, the present invention provides a system for detecting and classifying surface defects of a cable sheath in real time based on machine vision, comprising: The linear array image acquisition module is used for acquiring complete image data of the surface of the sheath of the cable in the continuous movement process of the cable; the illumination self-adaptive compensation module is used for eliminating interference of uneven ambient illumination and reflection of light on the surface of the cable sheath on imaging quality; The image preprocessing and enhancing module is used for receiving the original image data stream from the linear array image acquisition module and executing a series of operations aiming at improving the signal-to-noise ratio of the defect area; The multistage defect detection and feature extraction module adopts a cascade architecture to perform coarse-to-fine defect positioning and feature quantization on the preprocessed image; the real-time classification decision module is used for receiving the high-dimensional feature vector output by the multi-stage defect detection and feature extraction module and executing final defect type accurate classification and confidence assessment; and the system control and feedback execution module is used as a dispatching center and an execution terminal of the whole system. Preferably, the linear array i