A method for optimizing edge searching precision of a table searching system

Abstract

The present application relates to the technical field of thermal process, and especially relates to a method for optimizing edge searching precision of a surface inspection system, which divides a gray scale distribution into two regions of light and dark based on a gray scale reference value; sets a fixed edge threshold parameter or a variable edge threshold parameter for the surface inspection system; resets a periodic parameter; collects defect data in a way of training, modifying and adding; and optimizes a detection region. The present application has the following advantages: the gray scale range is expanded from 0-255 to 0-280, the reference value is increased from 128 to 140, and the light and dark regions are redivided, so that the defect pixel distribution is more balanced; for the working condition that the dark strip steel is close to the worn roller surface, the edge searching accuracy is improved, the positioning error is reduced, and the edge searching drift caused by the change of steel type and the wear of roller surface is solved; and the fixed and adaptive double-mode edge threshold strategy is adopted, the threshold value in the region within 30mm from the edge is narrowed by 20%-30%, and the fixed mode reduces the false positive rate.

Description

Technical Field

[0001] This invention relates to the field of thermal process technology, and in particular to a method for optimizing the edge-finding accuracy of a surface inspection system. Background Technology

[0002] In the production of sheet and strip metal, surface quality inspection is a crucial step in controlling product quality and meeting customer needs. Different application areas have significantly different requirements for the surface quality of strip steel. For example, automotive exterior panels have extremely stringent surface quality requirements, typically requiring a surface grade of FD, detecting micron-level defects, and allowing no flaws that could affect painting or stamping. In contrast, structural materials such as truck body panels have relatively relaxed surface quality requirements, with FB grade being sufficient. These differences in surface quality grades directly determine the types of defects detected, the detection rate, and the judgment criteria.

[0003] Currently, most strip inspection systems employ an edge-finding principle based on grayscale value changes: utilizing the difference between the grayscale values ​​on the strip surface and the roll surface, the rate of change of grayscale values ​​is calculated over a fixed distance along the strip's running width. When the grayscale value change reaches a set threshold, the area is identified as the edge of the strip. However, the following technical problems exist in actual production: 1. Variations in steel grades result in significant differences in the color, brightness, and texture of the strip surface. For example, different chemical compositions, hot rolling processes, and cooling regimes can cause the strip surface to exhibit various shades such as grayish-white, grayish-black, and dark black. The grayscale characteristics of the strip are close to or even overlap with the grayscale values ​​of the background roll surface. This makes it difficult for a single, fixed edge-finding parameter to adapt to various steel grade conditions, often resulting in misjudgments of the edge or drift in the edge-finding position, which in turn affects the effective detection of edge defects.

[0004] 2. Long-term use of the roller conveyor results in uneven wear and surface contamination, altering the reflective properties of the roller surface. This reduces the difference between the grayscale value of the roller surface and the grayscale value of the strip edge, further exacerbating the problem of inaccurate edge detection. This is especially pronounced when the strip is untensioned at the head and tail, leading to more noticeable unsteady conditions such as strip deviation and vibration.

[0005] 3. Different customers have different tolerances for product surface defects. If a single detection scheme is used, the following two extreme situations are likely to occur: First, some defects that affect the performance of the product may be missed due to the detection threshold being set too high; second, non-critical defects may be over-detected, causing quality inspectors to spend a lot of time identifying invalid alarms, which affects production efficiency and the accuracy of quality judgment.

[0006] 4. Existing surface inspection systems have limitations in terms of light source arrangement, edge lighting, and algorithm processing. Single-sided light sources cause uneven incident light in the edge areas, which can easily produce shadows or glare; conventional lighting intensity settings cannot effectively improve the contrast between the edge of dark strip steel and the roll surface; and the algorithm lacks optimized design for reserving buffer zones for physical edges and glare shielding.

[0007] Therefore, there is an urgent need for a method that can automatically switch detection schemes and dynamically optimize edge-finding parameters according to the characteristics of different steel grades, surface quality requirements and production conditions, in order to solve the problems of low edge-finding accuracy, poor defect detection adaptability and low system resource utilization in the existing technology, and achieve accurate and efficient judgment of strip steel surface quality. Summary of the Invention

[0008] The purpose of this invention is to provide a method for optimizing the edge-finding accuracy of a surface inspection system, solving the problems of edge drift and misjudgment caused by the similarity of grayscale between the strip and the roll surface due to changes in steel grade and roll surface wear. This method improves the edge-finding adaptability of strips with different surface characteristics by expanding the grayscale range and optimizing the grayscale reference value and interval division. It also enhances the accuracy of edge defect detection by setting fixed and adaptive dual-mode edge thresholds. Furthermore, it suppresses physical edge interference by optimizing periodic parameters and detection area, improving light source configuration and glare shielding. Finally, it achieves dynamic optimization of the defect judgment model through continuous classifier training and data accumulation. Ultimately, this method synergistically improves the edge-finding accuracy and defect detection rate of the surface inspection system, meeting the requirements of multi-level surface quality inspection.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A method for optimizing the edge-finding accuracy of a table inspection system includes: S1. Optimization of grayscale parameters: Adjust the grayscale parameter value range from 0~255 to 0~280, set the grayscale reference value, and divide the grayscale distribution into two regions: bright area and dark area based on the grayscale reference value; S2. Adjustment and optimization of edge threshold parameters: Based on the chemical composition of the steel, a fixed or variable edge threshold parameter is set for the surface inspection system to detect defects that are less than 30mm away from the edge of the strip. S3, Optimization of resetting periodic parameters: Set the parameters for detecting periodically recurring defects; S4. Design Optimization: Using the quality data provided by the inspection system, combined with secondary process control and tertiary distribution strategies, quality control is implemented on a per-volume basis; defect data is collected through training, modification, and addition. S5. Optimization of the detection area: Reduce the effective detection area, avoid the physical edge of the strip, and reserve a 1-2mm buffer detection area; A dual-sided symmetrical light source is used, and the incident angle is adjusted to 15°~30°; Adjust the edge illumination intensity of the dual-side light source from 70%~80% to 90%~95%; Add glare shielding to the defect identification algorithm of the surface inspection system; S6. Running trajectory and speed control: The material trajectory is centered to prevent edge deviation; The speed of the strip passing through the pinch rolls is controlled to a set low speed; The position at which the tail of the strip begins to decelerate on the K roller is calculated, and when it reaches the deceleration point, it decelerates at the set deceleration rate.

[0010] In S1, the grayscale reference value is 140; The bright and dark areas are each divided into at least 5 smaller intervals to balance the distribution of defective image pixels within the grayscale range.

[0011] The five sub-intervals of the dark zone are: 0~20, 21~40, 41~70, 71~100, and 101~139. The five intervals of the bright area are: 141~180, 181~210, 211~240, 241~260, and 261~280.

[0012] In S2, the fixed edge threshold parameter is a set upper or lower limit value, which is used for detection scenarios where the gray uniformity of the strip surface is high. The variable edge threshold parameter is an upper or lower limit that changes with the brightness and grayscale of the strip surface, and is used for detection scenarios where the grayscale of the strip surface fluctuates due to changes in steel type or process.

[0013] In S3, the parameters for periodic repetitive defects include the lower limit of the period, the upper limit of the period, the number of repetitive defects, the number of ignored defects, the lateral deviation, and the longitudinal deviation. in: The lower limit of the cycle is set to 1000~1400mm; The upper limit of the cycle is set to 2200~2600mm; The lateral deviation is 5~10mm; The longitudinal deviation is 10~15mm.

[0014] In S3, the periodicity parameter is used to adjust the detection of periodically repeating defects.

[0015] In S4, training is the number of defects collected for training the current reference classifier; The modification is the number of defects that the current reference classifier was used for training but was subsequently modified. The newly added number is the number of defects collected since the current reference classifier was established.

[0016] In S5, the defect identification algorithms include grayscale threshold segmentation algorithms, edge detection algorithms, or deep learning-based defect classification algorithms.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. By expanding the grayscale parameter range from 0-255 to 0-280 and increasing the grayscale reference value from 128 to 140, the grayscale distribution intervals of the bright and dark areas are redefined, making the distribution of defect image pixels more balanced within each small interval. Actual testing shows that for working conditions where the grayscale values ​​of dark strip steel (grayscale value 90-110) and worn roll surface (grayscale value 85-105) are similar, the edge finding accuracy is improved, the edge positioning error is reduced, and the edge finding drift problem caused by changes in steel grade and roll surface wear is effectively solved. 2. A dual-mode edge threshold strategy with fixed and variable modes is adopted. A narrowing threshold (20%–30%) is set for the area within 30mm of the strip edge. For strips with uniform surface grayscale, the fixed threshold mode reduces the false alarm rate by 35%. For strips with fluctuating surface grayscale, the adaptive threshold mode dynamically adjusts the threshold range according to the grayscale standard deviation, improving the edge defect detection rate and reducing the missed detection rate. At the same time, the dual-sided symmetrical light source, combined with an incident angle of 15°–30° and an increase in supplementary light intensity to 90%–95%, expands the grayscale difference between the edge and the roll surface from 5–10 to 15–25, improving the edge imaging contrast by 1.8 times. 3. By optimizing the periodic parameter settings, the lower limit of the period is set to 1000-1400mm, the upper limit of the period is set to 2200-2600mm, the lateral deviation is 5-10mm, and the longitudinal deviation is 10-15mm. This matches the actual working conditions and correction capability range of the working roller diameter of 400-800mm. Field application shows that this improves the accuracy of identifying periodic repetitive defects, reduces repeated alarms caused by unreasonable parameter settings, reduces the CPU usage of the inspection system, and improves the inspection speed. 4. Through a three-level data accumulation mechanism of classifier training, modification, and addition, the defect identification model can be continuously iterated and optimized. Taking the data of a production line optimized for four consecutive weeks as an example: the number of training samples increased from 12,500 to 14,600, the number of modified samples decreased from 340 to 120, and the number of newly added samples totaled 2,550. This improved the accuracy of defect identification after optimization, reduced the false alarm rate, reduced the workload of manual review, and enabled real-time quality control and automatic judgment on a steel coil basis. 5. By reserving a 1-2mm buffer detection area inside the physical edge of the strip, interference from physical edges such as burrs and folds is eliminated; a glare shielding module is integrated into the defect identification algorithm to identify and shield local overexposed areas with gray values ​​>250. For black and gray strips with a final rolling temperature ≤880℃ and a coiling temperature ≤580℃, the number of abnormal alarms in program learning is reduced, and the system stability is improved. 6. The strip running trajectory is controlled within ±2mm of the camera's field of view center line by the correction system. It is combined with a fixed low speed (0.8~1.2m / s) through the pinch rollers, and the timing of the tail deceleration is controlled according to the deceleration point position calculation formula. Field tests show that the imaging tilt angle of the strip head and tail under the untension state is reduced from an average of 3.5° to 0.8°, and the edge positioning deviation is reduced from ±10mm to ±2.5mm, thereby improving the detection rate of head and tail defects. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the strip steel running trajectory and speed control.

[0019] Figure 2 This is a screenshot of the threshold parameter setting window interface.

[0020] Figure 3 This is a statistical chart of data from the CBE classifier optimization process. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings, but it should be noted that the implementation of the present invention is not limited to the following embodiments.

[0022] The following embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments. Unless otherwise specified, the methods used in the following embodiments are conventional methods.

[0023] Example 1: A method for optimizing the edge-finding accuracy of a table inspection system includes: S1. Optimization of grayscale parameters: The basic principle of the edge detection system is based on the difference in gray values ​​between the strip surface and the roll surface. The edge position is determined by calculating the rate of change of gray values ​​within a set distance in the running width direction of the strip.

[0024] The rate of change of grayscale values ​​is calculated using the following formula: ; in: For the first The grayscale value of each pixel; To set the number of pixels within a given distance; It is the grayscale value of the adjacent preceding pixel; When the calculated mean grayscale change When the value exceeds the preset threshold T, the location is determined to be the edge of the strip.

[0025] For example, a detection point is taken every 1 mm along the width of the strip, with a detection distance of 10 mm, for a total of 10 pixels. When the grayscale value of the strip surface is 140-145 and the grayscale value of the roll surface is 110-115, the grayscale change value at the edge position is... The grayscale value is approximately 30-35; however, when the strip has a dark surface (grayscale value 115-120) and the grayscale value rises to 105-110 after the roll surface is worn, the grayscale change value at the edge position is... The threshold value is only 10-15, which is close to the fluctuation of the roller surface itself, resulting in inaccurate edge detection. In this case, the threshold value T needs to be lowered from 25 to 12, and the detection distance needs to be extended from 10mm to 15mm to enhance the ability to identify weak edge signals.

[0026] To address the aforementioned issues, the grayscale parameter range was adjusted from 0-255 to 0-280, and the grayscale baseline value was increased from 128 to 140. The area was divided into bright and dark zones, with 140 as the boundary. Each zone was further divided into five sub-zones. The endpoint grayscale values ​​were set based on the following: 1) Balance: By conducting statistical analysis on 100,000 defective images and approximately 230 million pixels, it was ensured that the probability of defective pixels appearing in each cell was equal. 2) Consistency: The grayscale column parameters for surface brightness and darkness are kept consistent to facilitate defect library maintenance.

[0027] The grayscale column distribution is shown in Table 1.

[0028] Table 1 shows the grayscale column distribution.

[0029] S2. Adjustment and optimization of edge threshold parameters: Threshold parameters refer to the upper and lower limits of grayscale values ​​at which the surface inspection system identifies detected images as defects, as well as parameters such as defect recognition sensitivity and the adjustment method of the upper and lower limits. The surface inspection system has a total of 7 detection ports, including horizontal threshold, baseline threshold, noise reduction threshold, stripe threshold, and edge threshold, and the threshold parameters of each port are independently adjustable.

[0030] The edge threshold parameter is specifically designed for defect detection in areas less than 30mm from the edge of the strip. Due to edge-finding errors, this method narrows the edge threshold by 20%–30% to improve the accuracy of edge defect identification.

[0031] Based on the chemical composition and surface characteristics of the steel, two edge threshold modes are set: 1) Fixed mode: Suitable for strip steel with high surface gray uniformity, such as Q235B carbon structural steel.

[0032] For example, the threshold is set to ±15 gray values, the gray value of the strip surface is stable between 135 and 145, and the detection threshold for the edge area is fixed at 120 to 160.

[0033] 2) Adaptive Mode: Suitable for strip steel whose surface grayscale varies due to steel grade or process variations, such as DC04 deep-drawing steel, whose cold-rolled grayscale value is 140-150, decreasing to 125-135 after annealing. The adaptive threshold algorithm is as follows: ; in: The baseline threshold is set at 30. The standard deviation of grayscale values ​​for the most recent 100 frames; This is the adjustment coefficient (taken as 1.5 to 2.5).

[0034] When the surface grayscale fluctuates greatly ( The threshold is automatically relaxed to ±45. When the surface grayscale is stable ( The threshold was narrowed to ±20.

[0035] S3, Optimization of resetting periodic parameters: Periodic parameters are used to detect periodic repetitive defects caused by rotating components such as work rolls and roller conveyors. These parameters include the lower limit of the period, the upper limit of the period, the number of repetitive defects, the number of defects to be ignored, lateral deviation, and longitudinal deviation.

[0036] Set periodic parameters: On the one hand, it should be based on the amount of deviation that the strip steel will produce near the detection position of the surface inspection system during the actual production process, within the correction capability of the CPC (correction system); On the other hand, it is necessary to consider the parts on site that are prone to periodic recurring defects, such as the actual diameter of the work roll and the roller conveyor, and optimize reasonable periodic parameters. This will not only ensure the accurate detection of periodic defects, but also avoid the over-utilization of the surface inspection system resources, and improve the utilization rate and detection speed of the surface inspection system resources.

[0037] Basis for setting: The working roll diameter ranges from 400 to 800 mm, corresponding to a defect repetition cycle of 1256 to 2512 mm; Within the CPC correction capability range, the lateral deviation of the strip is ≤5~10mm, and the longitudinal deviation is ≤10~15mm.

[0038] The optimized periodic parameter settings are shown in Table 2.

[0039] Table 2 shows the optimized periodic parameter settings.

[0040] The above parameters apply to the surface quality level testing schemes of FB, FC, and FD: in: FB: Testing solutions for the surface quality of commercial-grade automotive exterior panels; FC: Testing solution for high-grade automotive exterior panel surface quality; FD: Testing solution for ultra-high-level automotive exterior panel surface quality.

[0041] S4. Design Optimization: Based on the defect data provided by the surface inspection system, combined with the two-level process control system and the three-level distribution strategy, the surface quality of steel coils can be judged and sealed in real time.

[0042] Defect data is continuously optimized through three methods: training, modification, and addition. CBE (ClassifierBaselineEngine) is a reference classifier. Training: The number of defects used for training by the current CBE reference classifier; Modification: The number of defects that were adopted by the current CBE reference classifier but subsequently manually corrected; Added: The number of newly recorded defects since the current CBE reference classifier was established.

[0043] See Figure 3 Taking the optimization process in a certain month as an example: Initial state (Week 1): 12,500 training samples, 340 modified samples, and 0 new samples; Week 2: 12,500 training samples, 280 modified samples, and 1,240 new samples; Week 3: 13,740 training samples, 190 modified samples, and 860 new samples; Week 4: 14,600 training samples, 120 modified samples, and 450 new samples.

[0044] After each retraining, the new classifier is automatically activated as the CBE test classifier. The initial classifier is compared with the optimized classifier from week 4 to improve defect identification accuracy and reduce false positive rate.

[0045] S5. Optimization of the detection area: For hot-rolled or pickled strips with a final rolling temperature ≤880℃ and a coiling temperature ≤580℃, the strip turns blackish-gray (gray value 90~110) after laminar flow cooling, which highly overlaps with the gray value (85~105) of the worn roll surface, easily leading to edge misjudgment and abnormal learning of the automatic judgment program.

[0046] The following optimization measures will be adopted: To reduce the effective detection area, a 1-2mm buffer detection area is reserved inside the physical edge of the strip to eliminate interference from edge burrs and minor edge turning. A dual-sided symmetrical LED light source is used, with the incident angle adjusted to 15°~30° to ensure that the illuminance uniformity in the edge area is ≥90%; The edge lighting intensity is increased from 70%–80% to 90%–95%, which increases the gray difference between the edge and the roller surface from 5–10 to 15–25. A glare shielding module is added to the defect identification algorithm to identify and shield local overexposed areas (grayscale value > 250) caused by improper light source angle.

[0047] Through the above measures, the number of abnormal alarms during program learning decreased from an average of 12.5 times per day before optimization to an average of 1.8 times per day after optimization, a reduction of 85.6%.

[0048] Defect identification algorithms include: Segmentation algorithms based on grayscale thresholds (Otsu method, maximum entropy method). Edge detection algorithms (Canny, Sobel); Defect classification algorithms based on deep learning (CNN, ResNet, YOLO).

[0049] S6. Running trajectory and speed control: See Figure 1 To ensure stable imaging under tension loss at the head and tail of the strip, the following control strategies are adopted: 1) Trajectory control: The CPC correction system controls the strip's trajectory within ±2mm of the camera's field of view center line to prevent edge offset from causing image tilting. 2) Speed ​​control: The strip should be kept at a constant low speed (0.8-1.2 m / s) as it passes through the pinch rolls to ensure that the head and tail sections pass smoothly through the inspection area while in a state of tension loss. 3) Calculation of deceleration position: The formula for calculating the position of the deceleration point at the tail of the strip is as follows: ; in: This indicates the distance from the mill exit to the point where the tail begins to decelerate, in meters. This indicates the total length of the strip, in meters (m). This indicates the stable operating speed, expressed in m / s. This indicates the set deceleration rate, in m / s², with a value range of 0.3 to 0.5. This represents the safety buffer distance, in meters (m), and its value ranges from 3 to 5.

[0050] For example, strip length running speed Deceleration rate Safety buffer ,but: ; When the tail of the strip reaches 764.75m, it begins to decelerate at a rate of 0.4m / s² until it comes to a smooth stop.

[0051] This invention expands the grayscale parameter range from 0-255 to 0-280 and increases the grayscale reference value from 128 to 140, redefining the grayscale distribution intervals of bright and dark areas to achieve a more balanced distribution of defective image pixels within each interval. Actual testing shows that for conditions where the grayscale values ​​of dark strip steel (grayscale value 90-110) and worn roll surfaces (grayscale value 85-105) are similar, it improves edge-finding accuracy, reduces edge positioning errors, and effectively solves the edge-finding drift problem caused by steel grade changes and roll surface wear. A dual-mode edge threshold strategy, combining fixed and variable thresholds, is used. A narrowing threshold (20%-30%) is set for areas within 30mm of the strip edge. For strip steel with uniform surface grayscale, a fixed threshold mode... This method reduces the false alarm rate by 35%. For strip steel with fluctuating surface grayscale, the adaptive threshold mode dynamically adjusts the threshold range based on the grayscale standard deviation, improving the detection rate of edge defects and reducing the missed detection rate. Simultaneously, the dual-sided symmetrical light source, combined with an incident angle of 15°–30° and an increase in supplementary light intensity to 90%–95%, expands the grayscale difference between the edge and the roll surface from 5–10 to 15–25, improving the edge imaging contrast by 1.8 times. By optimizing the periodic parameter settings, the lower limit of the period is set to 1000–1400 mm, the upper limit to 2200–2600 mm, the lateral deviation to 5–10 mm, and the longitudinal deviation to 10–15 mm, matching the actual working conditions and correction capability range of the work roll diameter of 400–800 mm, and applicable in the field. This demonstrates that the system improves the accuracy of identifying recurring defects, reduces repeated alarms caused by unreasonable parameter settings, lowers the CPU usage of the surface inspection system, and increases detection speed. Through a three-level data accumulation mechanism of classifier training, modification, and addition, the defect identification model achieves continuous iterative optimization. Taking four consecutive weeks of data optimization on a production line as an example: the number of training samples increased from 12,500 to 14,600, the number of modified samples decreased from 340 to 120, and 2,550 new samples were added. This improves the accuracy of defect identification after optimization, reduces the false alarm rate, reduces the workload of manual review, and enables real-time quality control and automatic judgment at the steel coil level. Furthermore, by reserving a 1-2mm buffer detection area inside the physical edge of the strip, burrs and flaking are eliminated. Physical edge interference is eliminated; a glare shielding module is integrated into the defect recognition algorithm to identify and shield local overexposed areas with grayscale values ​​>250. For black and gray strip steel with a final rolling temperature ≤880℃ and a coiling temperature ≤580℃, the number of abnormal alarms in program learning is reduced, and the system stability is improved; the strip steel running trajectory is controlled within ±2mm of the camera's field of view center line through the correction system, and the strip steel passes through the pinch roll at a fixed low speed (0.8~1.2m / s). The timing of the tail deceleration is controlled according to the deceleration point position calculation formula. Field tests show that the imaging tilt angle of the strip steel head and tail under untensioned state is reduced from an average of 3.5° to 0.8°, and the edge positioning deviation is reduced from ±10mm to ±2.5mm, improving the detection rate of head and tail defects.

Claims

1. A method for optimizing the edge-finding accuracy of a table inspection system, characterized in that, include: S1. Optimization of grayscale parameters: Adjust the grayscale parameter value range from 0~255 to 0~280, set the grayscale reference value, and divide the grayscale distribution into two regions: bright area and dark area based on the grayscale reference value; S2. Adjustment and optimization of edge threshold parameters: Based on the chemical composition of the steel, a fixed or variable edge threshold parameter is set for the surface inspection system to detect defects that are less than 30mm away from the edge of the strip. S3, Optimization of resetting periodic parameters: Set the parameters for detecting periodically recurring defects; S4. Design Optimization: Using the quality data provided by the inspection system, combined with secondary process control and tertiary distribution strategies, quality control is implemented on a per-volume basis; defect data is collected through training, modification, and addition. S5. Optimization of the detection area: Reduce the effective detection area, avoid the physical edge of the strip, and reserve a 1-2mm buffer detection area; A dual-sided symmetrical light source is used, and the incident angle is adjusted to 15°~30°; Adjust the edge illumination intensity of the dual-side light source from 70%~80% to 90%~95%; Add glare shielding to the defect identification algorithm of the surface inspection system; S6. Running trajectory and speed control: The material trajectory is centered to prevent edge deviation; The speed of the strip passing through the pinch rolls is controlled to a set low speed; The position at which the tail of the strip begins to decelerate on the K roller is calculated, and when it reaches the deceleration point, it decelerates at the set deceleration rate.

2. The method for optimizing the edge-finding accuracy of a table inspection system according to claim 1, characterized in that, In S1, the grayscale reference value is 140; The bright and dark areas are each divided into at least 5 smaller intervals to balance the distribution of defective image pixels within the grayscale range.

3. The method for optimizing the edge-finding accuracy of a table inspection system according to claim 2, characterized in that, The five sub-intervals of the dark zone are: 0~20, 21~40, 41~70, 71~100, and 101~139. The five intervals of the bright area are: 141~180, 181~210, 211~240, 241~260, and 261~280.

4. The method for optimizing the edge-finding accuracy of a table inspection system according to claim 1, characterized in that, In S2, the fixed edge threshold parameter is a set upper or lower limit value, which is used for detection scenarios where the gray uniformity of the strip surface is high. The variable edge threshold parameter is an upper or lower limit value that changes with the brightness and grayscale of the strip surface, and is used for detection scenarios where the grayscale of the strip surface fluctuates due to changes in steel type or process.

5. The method for optimizing the edge-finding accuracy of a table inspection system according to claim 1, characterized in that, In S3, the parameters of the periodic repetitive defects include the lower limit of the period, the upper limit of the period, the number of repetitive defects, the number of ignored defects, the lateral deviation, and the longitudinal deviation. in: The lower limit of the cycle is set to 1000~1400mm; The upper limit of the cycle is set to 2200~2600mm; The lateral deviation is 5~10mm; The longitudinal deviation is 10~15mm.

6. The method for optimizing the edge-finding accuracy of a table inspection system according to claim 1, characterized in that, In S3, the periodic parameter is used to adjust the detection of periodically repeating defects.

7. The method for optimizing the edge-finding accuracy of a table inspection system according to claim 1, characterized in that, In S4, training is the number of defects collected for training the current reference classifier; The modification is the number of defects that the current reference classifier was used for training but was subsequently modified. The newly added number is the number of defects collected since the current reference classifier was established.

8. The method for optimizing the edge-finding accuracy of a table inspection system according to claim 1, characterized in that, In S5, the defect identification algorithms include grayscale threshold segmentation algorithms, edge detection algorithms, or deep learning-based defect classification algorithms.

Images