High-precision thin metal plate detection method

Abstract

The invention discloses a high-precision thin metal plate detection method and a related device. The method comprises the following steps: projecting an optical pattern with a space period to the surface of a metal plate and receiving reflected light; performing parallel separation acquisition on the reflected light at the same acquisition moment to obtain at least two paths of channel signals corresponding to different polarization states and / or different spectrum channels; triggering line-by-line acquisition by using a motion pulse signal of the metal plate, so that the acquisition result of each line corresponds to the fixed length position increment, and keeping the synchronization of optical pattern updating and acquisition; carrying out availability screening on the channel signals, and extracting stripe features representing local distortion of the optical pattern from the channel signals meeting conditions; and determining the position and severity level of the surface defect of the metal plate according to the continuity and abrupt change characteristics of the stripe characteristics along the length direction. According to the scheme, stable and high-precision online detection on the surface defects of the thin metal plate running at a high speed can be realized.

Description

Technical Field

[0001] This invention relates to the field of optical inspection technology for online inspection of metal sheets, and particularly to a high-precision inspection method and related apparatus for thin metal sheets. Background Technology

[0002] Sheet metal typically refers to thin sheets or strips formed from steel, aluminum, stainless steel, and their alloys through plastic processing techniques such as rolling. It is characterized by its thinness, wide width, continuous length, and strong surface reflectivity, and is widely used in automotive body panels, appliance housings, architectural decorative panels, packaging materials, and precision structural components. In actual industrial production, thin sheet metal often runs at high speeds on production lines for rolling, annealing, cleaning, coating, or slitting in continuous strip form. Its surface condition and local morphology directly affect the quality of subsequent forming, coating, and assembly. Therefore, online inspection of thin sheet metal is usually required during production to obtain quality-related information, such as the presence of defects like indentations, creases, scratches, pits, oxide scale, and uneven oil film, as well as the distribution of these defects along the length and width of the strip, thereby enabling process monitoring and quality assessment.

[0003] To address the aforementioned inspection needs, various online inspection methods for metal sheets based on optical principles have been proposed in existing technologies. One typical method involves projecting an optical pattern with a certain spatial period or structural characteristics onto the surface of the metal sheet, acquiring the optical signals after reflection or scattering of the pattern on the sheet surface, and then identifying surface anomalies through image processing or phase analysis. This type of method can reflect surface defects by utilizing the distortion or brightness changes of the optical pattern under experimental conditions or low-speed operation, and has therefore found some application in the field of metal surface inspection. However, this type of method typically relies on stable lighting conditions and clear optical imaging; the inspection results are relatively reliable when the sheet surface has uniform reflectivity and the movement is stable.

[0004] However, in actual continuous production lines, thin metal sheets generally exhibit strong specular reflectivity, and their surface reflectivity fluctuates significantly with changes in local conditions. Furthermore, vibration, tension fluctuations, and slight speed variations are inevitable during high-speed operation. Under these conditions, the optical pattern projected onto the sheet surface is prone to local overexposure, saturation, or abrupt contrast changes, leading to unstable intensity of the acquired optical signal. Simultaneously, due to the continuous movement of the sheet, detection methods relying on multiple sampling or multi-frame analysis are prone to temporal inconsistencies, making it difficult to align optical information at the same location in time. The combined effect of these intensity fluctuations and temporal inconsistencies weakens the identifiability of surface defects in the optical signal, thus affecting the stability and repeatability of the detection results. Therefore, existing online optical inspection methods for metal sheets struggle to stably extract optical information closely related to surface defects in high-speed, high-reflectivity production environments without sacrificing production cycle time. This constitutes a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] This application provides a high-precision method for inspecting thin metal sheets, aiming to solve the technical problems of unstable defect characteristics and poor repeatability of detection results caused by fluctuations in reflection intensity and inconsistencies in acquisition timing in existing online optical inspection methods for thin metal sheets in high-speed, high-reflection production environments.

[0006] To achieve the above objectives, embodiments of this application provide a high-precision method for detecting thin metal sheets, comprising: An optical pattern with a spatial period is projected onto the surface of a metal sheet, and reflected light from the surface of the metal sheet is received. At the same acquisition time, the reflected light is acquired in parallel to obtain at least two independent channel signals, which correspond to reflected light with different polarization states and / or different spectral channels; The motion pulse signal of the metal plate is acquired, and the motion pulse signal is used to trigger the line-by-line acquisition of the at least two independent channel signals, so that the acquisition result of each line corresponds to the fixed position increment of the metal plate along the length direction, and the update of the optical pattern is controlled to keep synchronized with the line-by-line acquisition. The availability of the at least two independent channel signals is screened to obtain the effective channel signal that meets the contrast condition and has not saturated. Then, the fringe features that characterize the local distortion of the optical pattern are extracted from the effective channel signal. When one channel signal is unavailable, the fringe features of the other channel signal are used as a substitute. Based on the continuity and abrupt changes of the stripe features along the length direction, the length position, lateral position, and severity level of the surface defects of the metal sheet are determined.

[0007] To achieve the above objectives, this application also proposes a high-precision thin metal sheet inspection device, comprising: A pattern projection module is used to project an optical pattern with a spatial period onto the surface of a metal sheet and to receive reflected light from the surface of the metal sheet. A multi-channel acquisition module is used to perform parallel separate acquisition of the reflected light at the same acquisition time, and acquire at least two independent channel signals, wherein the at least two independent channel signals correspond to reflected light with different polarization states and / or different spectral channels; The synchronous triggering module is used to acquire the motion pulse signal of the metal plate, and to trigger the line-by-line acquisition of the at least two independent channel signals with the motion pulse signal, so that the acquisition result of each line corresponds to the fixed position increment of the metal plate along the length direction, and to control the update of the optical pattern to keep it synchronized with the line-by-line acquisition. The feature extraction module is used to perform availability screening on the at least two independent channel signals, obtain the effective channel signal that meets the contrast condition and has not saturated, and extract stripe features that characterize the local distortion of the optical pattern on the effective channel signal; wherein, when one channel signal is unavailable, the stripe features of the other channel signal are used as a substitute. The defect determination module is used to determine the length position, lateral position, and severity level of the surface defect of the metal sheet based on the continuity and abrupt change characteristics of the stripe feature along the length direction.

[0008] To achieve the above objectives, this application also proposes a high-precision thin metal sheet inspection device, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit; the at least one processor invokes the instructions in the memory to cause the high-precision thin metal sheet inspection device to execute the steps of the above-described high-precision thin metal sheet inspection method.

[0009] The technical solution provided in this application projects a spatially periodic optical pattern onto the surface of a metal sheet and receives reflected light, allowing surface defects to be revealed through local distortion or modulation of the optical pattern. Defect information is converted into a change that can be measured in the signal domain. The reflected light is separated into at least two independent channel signals in parallel at the same acquisition moment. These channel signals correspond to reflected light of different polarization states and / or different spectral channels, ensuring that optical information at the same location remains consistent in time and does not introduce cross-time alignment errors. Simultaneously, different channels respond differently to specular reflection and scattering components. When localized areas experience enhanced specular reflection, overexposure, or decreased contrast, signal distortion tends to concentrate in some channels rather than simultaneously in all channels, thus providing conditions for subsequent screening and retention of usable information. Motion pulse signals are used to trigger line-by-line acquisition, so that the acquisition result of each line corresponds to a fixed position increment along the length of the metal plate, and control the optical pattern update to keep it synchronized with the line-by-line acquisition. The sampling process is changed from time domain driven to length domain driven, avoiding misalignment of the same position at different sampling times due to slight speed changes. The synchronization of the optical pattern with the acquisition keeps the projection state corresponding to the same line of exposure consistent, reducing the stripe shape drift caused by pattern changes and motion superposition. Therefore, the signal sequence formed along the length direction can stably correspond to the same spatial coordinate system.

[0010] After obtaining the channel signals arranged along the length coordinate, the channel signals are screened for availability, and effective channel signals that meet the contrast condition and have not saturated are selected. This is equivalent to selecting observation results where the information remains linear and the fringe structure is discernible at each position. When one channel signal is unavailable, the fringe features of another channel signal are used as a substitute, so that the information loss caused by local saturation or abrupt changes in reflectivity is no longer directly converted into missed or false defects. Based on the effective channel signals, fringe features characterizing local distortions of the optical pattern are extracted. The relative intensity amplitude of the distortion features more stably reflects the influence of surface geometric disturbances and changes in surface state on the pattern, and can separate the influence of illumination fluctuations and slow changes in reflectivity from the information required for defect determination. The continuity and abrupt changes of the fringe features along the length direction are used to determine the length position, lateral position, and severity level of the defect. Continuity describes the smooth changes of normal texture and operating condition disturbances, while abrupt changes describe the local non-smooth changes caused by defects such as indentations, creases, and scratches. Under the condition of strict alignment of the length coordinate, the abrupt change position corresponds one-to-one with the defect position, and the severity can be obtained by comprehensively considering the abrupt change amplitude, duration, and lateral expansion range. By suppressing the distortion caused by intensity fluctuations through simultaneous multi-channel observation, suppressing the temporal inconsistency caused by high-speed motion through length domain triggering synchronization, and ensuring the stable output of defect information through availability screening and stripe feature extraction, high repeatability defect detection results can still be obtained under high-speed and high-reflection conditions. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of one embodiment of the high-precision thin metal sheet detection method of the present invention; Figure 2 This is a schematic diagram of one embodiment of the high-precision thin metal sheet detection device of the present invention; Figure 3 This is a schematic diagram of one embodiment of the high-precision thin metal sheet inspection equipment of the present invention.

[0013] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0015] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0016] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the term "and / or" throughout the text includes three solutions; taking A and / or B as an example, it includes technical solution A, technical solution B, and a technical solution that simultaneously satisfies A and B. Furthermore, the technical solutions of various embodiments can be combined with each other, provided that they are feasible for those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0017] One embodiment of this application provides a high-precision method for detecting thin metal sheets. Figure 1 This is a flowchart illustrating a high-precision method for inspecting thin metal sheets according to an embodiment of this application. In this embodiment, the method includes: Please see Figure 1 It projects an optical pattern with a spatial period onto the surface of a metal sheet and receives reflected light from the surface of the metal sheet. It should be noted that a spatially periodic optical pattern refers to an optical intensity structure with repeating intervals of light and dark distribution or color distribution formed on the surface of a metal sheet. These repeating intervals are fixed values ​​in the coordinate system of the sheet surface. This is used to map surface defects such as indentations, creases, scratches, pits, oxide scale, or uneven oil film onto the reflected light in the form of local distortion, local displacement, or changes in local contrast. Specifically, projecting a spatially periodic optical pattern onto the surface of a metal sheet can be achieved using projection imaging. That is, a periodic stripe or periodic grid pattern is output above or to the side of the detection area, and the pattern is projected onto the surface of the metal sheet through an imaging optical path, so that the pattern covers the target detection area in the width direction of the metal sheet. The projection angle is set to have an angle with the normal to the surface of the metal sheet, so that the specular reflection main energy deviates from the direction of reflected light reception and reduces the risk of local overexposure, while ensuring that the effective contrast of the pattern is still contained within the field of view of the reflected light reception.

[0018] Reflected light from the surface of a metal sheet includes specular reflection and scattering components. Reflected light reception is achieved through imaging the projection area, forming a stripe structure signal at the receiving end that corresponds to the projected pattern. In high-speed strip applications, reflected light reception can employ a row-by-row acquisition method, ensuring each row covers the transverse range of the metal sheet. This allows the stripe structure to be stitched together along its length in subsequent steps to form a continuous sequence. For example, on rolling or coating production lines, unidirectional, equally spaced stripes can be used for projection. This allows scratches or creases to cause local stripe shifts, manifesting as phase abrupt changes or gradient anomalies in the reflected light. Furthermore, uneven oil film leading to reflectivity variations manifests as localized changes in stripe contrast, providing a distinguishable signal basis for subsequent usability screening and stripe feature extraction. This step, by translating the minute perturbations of defects on the surface into local distortions of a spatially periodic pattern, allows defect information to enter the reflected light signal as structured features. Subsequent contrast determination, saturation determination, and distortion feature calculations can then be performed within a unified feature framework, avoiding misjudgments caused by relying solely on brightness thresholds.

[0019] In other embodiments, the spatial periodic optical pattern can also be achieved using a striped grating imaging method, that is, by forming periodic stripes in the illumination light path through a fixed grating and projecting them onto the surface of the metal plate, or by outputting a periodic pattern on a display screen and receiving it after reflection onto the surface of the metal plate; the above equivalent methods also form an optical structure with repeating intervals on the surface of the metal plate, so that the reflected light carries stripe information that can be used to characterize local distortion, thereby achieving the same or similar basic detection effect.

[0020] Please continue reading. Figure 1 At the same acquisition time, the reflected light is acquired in parallel and separated to obtain at least two independent channel signals, which correspond to reflected light with different polarization states and / or different spectral channels; It should be noted that parallel separation acquisition refers to splitting the reflected light from the surface of the metal plate into at least two non-overlapping optical paths according to their optical properties at the same acquisition time, and obtaining the corresponding channel signals within the same time window. Independent channel signals refer to two signals that are acquired at the same time and have no time difference between them. At the same time, the two signals have different responses to the specular reflection component and the scattering component in the reflected light, thus possessing complementarity when local overexposure or reflectivity fluctuations occur. Different polarization states refer to at least two polarization components formed after polarization separation of the reflected light, such as the two orthogonal direction components of linear polarization. Different spectral channels refer to different wavelength range components formed after the reflected light is split, such as red, green, and blue channels or visible light and near-infrared channels.

[0021] Specifically, parallel separation and acquisition of reflected light at the same acquisition time can be achieved using two methods: one is polarization separation, where the reflected light, after entering the beam splitter, is first split into two polarization components by a polarization separation component, and then enters their respective imaging channels and is sampled synchronously to obtain the first channel signal and the second channel signal; the other is spectral separation, where the reflected light is split into at least two optical paths with different spectral ranges by a color separation or wavelength division component, and then is sampled synchronously to obtain at least two channel signals. To meet the constraint of the same acquisition time, the two channel signals are driven by the same trigger source and share the same exposure time window, with consistent sampling timestamps. In the row-by-row acquisition scenario, the two channel signals are synchronously acquired for the horizontal range of the same row, so that the same length position increment corresponds to the observation results of the two paths at the same moment. For example, in the high-speed strip inspection section of the coating production line, specular reflection will produce strong highlights in some areas, causing one channel signal to saturate and reduce the stripe contrast, while the other channel signal, due to its lower response to the specular component or higher response to the scattering component, can still maintain the distinguishable stripe structure, thus providing usable input for subsequent usability screening and stripe feature extraction. This step, by fixing multiple observations at the same acquisition time, ensures that the differences between channels mainly come from different responses of polarization state and / or spectral channels rather than time misalignment, thus avoiding positional inconsistencies caused by multiple samplings under high-speed operating conditions. At the same time, by making different channels contribute different proportions to specular reflection and scattering, the damage to fringe distortion characteristics caused by local overexposure no longer affects all channels simultaneously, making usability screening and channel replacement a practically feasible premise.

[0022] Please continue reading. Figure 1 The motion pulse signal of the metal plate is acquired, and the motion pulse signal triggers the line-by-line acquisition of the at least two independent channel signals, so that the acquisition result of each line corresponds to the fixed position increment of the metal plate along the length direction, and the update of the optical pattern is controlled to keep synchronized with the line-by-line acquisition. It should be noted that the motion pulse signal is a pulse sequence associated with the actual displacement of the metal sheet along the production line direction. Each pulse in the pulse sequence corresponds to a preset position increment of the metal sheet along the length direction. Line-by-line acquisition refers to synchronously sampling at least two independent channel signals with each line as the acquisition unit. The acquisition result of each line covers the lateral range of the metal sheet and corresponds to a position increment in the length direction. The update of the optical pattern refers to refreshing or switching the spatial periodic optical pattern projected onto the surface of the metal sheet so that the projected pattern remains in a fixed state within the acquisition time window.

[0023] Specifically, after acquiring the motion pulse signal, the motion pulse signal is input into the trigger control link. Upon receiving each motion pulse, the trigger control link generates a line-by-line acquisition trigger. This line-by-line acquisition trigger simultaneously acts on at least two independent channel signals, ensuring that both channel signals obtain the acquisition result for the same line at the same acquisition time. Thus, each line acquisition result corresponds one-to-one with the position increment in the same length direction. When the motion speed fluctuates slightly, the pulse trigger still occurs according to the actual displacement. Therefore, the spacing between adjacent lines in the length direction remains a fixed position increment, preventing compression or stretching in the length direction caused by fixed-time sampling. To ensure synchronization between the optical pattern update and the line-by-line acquisition, the update cycle of the optical pattern is bound to the line-by-line acquisition trigger, ensuring that the optical pattern corresponding to each line-by-line acquisition trigger is in a stable display or stable projection state. When the optical pattern needs to be refreshed, the update is completed between two adjacent line-by-line acquisition triggers and remains unchanged until the next line-by-line acquisition trigger, thus avoiding stripe shape drift caused by pattern changes within the acquisition time window. To ensure the repeatability of the synchronization relationship, a timestamp and pulse number can be recorded each time a line-by-line acquisition is triggered. In the event of trigger loss or jitter, resynchronization can be performed by checking the continuity of the pulse number. For example, when an abnormal pulse interval is detected, the update of the optical pattern can be paused and the pattern state of the previous frame can be maintained until the pulse interval returns to the allowable range before the update continues. This avoids introducing inconsistent pattern states during abnormal periods.

[0024] For example, in the high-speed strip inspection section of a coating production line, motion pulse signals can come from the pulse output of a speed measuring roller synchronized with the strip. When the fixed position increment is set to 0.2 mm, each received motion pulse triggers two independent channel signals to complete one row-by-row acquisition. When the strip speed fluctuates from 80 m / min to 85 m / min, the row-by-row acquisition still strictly corresponds to the 0.2 mm length step. Therefore, the sequence of stripe features along the length direction does not undergo coordinate distortion due to speed fluctuations. At the same time, the update of the optical pattern is completed between two triggers and remains stable within the trigger window, ensuring that the same row acquisition corresponds to a unique projection state, reducing the interference of inconsistent pattern timing on stripe features. This step, because it drives row-by-row acquisition with actual displacement and constrains the optical pattern update outside the acquisition window, synchronously locks the length coordinates and the optical pattern state in the acquisition link. Thus, the time misalignment caused by high-speed motion is transformed into controllable length step sampling. Subsequent analysis of the continuity and abrupt changes of stripe features can directly correspond to spatial position without relying on cross-time alignment. In other embodiments, the motion pulse signal may also be generated by the equivalent displacement pulse output by the metal sheet running device, or by converting the periodic sampling of the running speed of the metal sheet into an equivalent pulse sequence to form a fixed position incremental trigger; the update of the optical pattern may also adopt a fixed refresh frequency and select an update state consistent with the trigger phase when triggering line by line acquisition.

[0025] Please continue reading. Figure 1 The availability of the at least two independent channel signals is screened to obtain the effective channel signal that meets the contrast condition and has not saturated, and the stripe features characterizing the local distortion of the optical pattern are extracted from the effective channel signal; wherein, when one channel signal is unavailable, the stripe features of the other channel signal are used as a substitute. In one embodiment of the present invention, the process of performing availability filtering on the at least two independent channel signals and obtaining valid channel signals includes: When the at least two independent channel signals both meet the contrast condition and do not saturate in the same location area, the channel signals are dynamically weighted and fused according to the reflection characteristics of the location area to obtain an effective channel signal; wherein, in the location area dominated by specular reflection, the fusion weight of the channel signal with a high specular reflection response intensity is reduced; in the location area dominated by scattering, the fusion weight of the channel signal with a high scattering response intensity is increased.

[0026] Specifically, the same location region refers to the area formed by a local horizontal window and its corresponding length position increment selected in the row data acquired line by line. This region is used for statistical analysis and discrimination of the reflected light signal. The contrast condition is a threshold constraint used to characterize the resolvability of spatial periodic optical patterns in the channel signal. It can be determined by indicators such as the difference between the maximum and minimum gray levels within the location region, or the ratio of the dominant fringe frequency amplitude to the DC component. Saturation refers to the phenomenon where the channel signal reaches the upper limit of sampling or a shearing plateau occurs within the location region, resulting in the loss of fringe structure information. The reflection characteristics of the location region describe whether the reflected light within that region is dominated by specular reflection or scattering. Location regions dominated by specular reflection often appear as bright and locally close to saturation in the image, while location regions dominated by scattering appear as having more uniform brightness and more stable fringe texture.

[0027] When at least two independent channel signals satisfy the contrast condition and do not saturate in the same location region, the intensity feature value of the location region is calculated for each channel signal, and the ratio of the intensity feature values ​​between the channels is used as the discrimination metric for the reflection characteristics of the location region. When the intensity ratio exceeds a preset threshold, the reflection type corresponding to the channel signal with the larger intensity feature value is determined to be dominant in the location region, and the location region is marked as specular reflection dominant. At the same time, a smaller fusion weight is assigned to the channel signal with the larger intensity feature value, and a larger fusion weight is assigned to the other channel signal. When the intensity ratio does not exceed the preset threshold, the location region is marked as scattering dominant, and a larger fusion weight is assigned to the channel signal with a higher scattering response intensity. The assignment of the fusion weight can be implemented using a piecewise linear method or a continuous mapping method. For example, the change of the intensity ratio from the threshold to the upper limit can be mapped to the change of the fusion weight from equal distribution to bias, so that the weight adjustment smoothly transitions with the change of reflection state. Then, the two channel signals are weighted and summed or weighted averaged in the location region to obtain the effective channel signal, and the effective channel signal is used for subsequent fringe feature extraction.

[0028] For example, in high-speed inspection of a coating production line, uneven oil film on the surface of a metal sheet can create strong specular highlights, significantly increasing the intensity ratio. In this case, reducing the fusion weight of the channel signal with higher specular reflection response intensity can reduce the dominance of the highlights on the effective channel signal, allowing the effective channel signal to retain the stripe structure and thus enabling continuous extraction of stripe features. In areas with rougher surfaces or oxide scale, the proportion of scattering components increases. Increasing the fusion weight of the channel signal with higher scattering response intensity can improve the stability of the dominant stripe frequency component and reduce baseline drift caused by reflectivity fluctuations. The intensity ratio threshold can be determined based on the sampling dynamic range and saturation margin. A threshold that is too low will cause frequent weight fluctuations, while a threshold that is too high will reduce the ability to suppress highlight areas. Therefore, it is preferable to select a value within the ratio range where the channel signal is still in the linear region and significant reflection differences have already appeared. This step first confirms that both channel signals are available within the location area, and then dynamically allocates weights based on reflection characteristics. This ensures that the effective channel signal is preferentially dominated by the channel with more stable information and a more complete fringe structure. This transforms the local intensity abrupt changes caused by specular highlights into controlled weight changes, preventing highlights from directly entering the fringe features and inducing false detections. In other embodiments, the reflection characteristics of the location area can also be determined by indicators such as the proportion of saturated pixels within the location area, the stability of the fringe dominant frequency amplitude, or the polarization difference amplitude of the channel signals. Fusion weights are then allocated accordingly. The aforementioned equivalent discrimination parameters can also reflect the difference between specular reflection and scattering ratios, thereby achieving similar or identical dynamic weighted fusion effects.

[0029] In one embodiment of the present invention, the step of dynamically weighting and fusing the signals of each channel based on the reflection characteristics of the location region includes: Calculate the intensity characteristic values ​​of the at least two independent channel signals in the location region, and calculate the intensity ratio between the channels based on the intensity characteristic values; Based on the comparison between the intensity ratio and a preset threshold, the reflection characteristics of the location region are determined to be either specular reflection-dominated or scattering-dominated, and a fusion weight coefficient is assigned to each channel signal based on the reflection characteristics; wherein, when the intensity ratio exceeds the preset threshold, the location region is determined to be dominated by the reflection type corresponding to the channel signal with the larger intensity characteristic value, and the fusion weight coefficient of the channel signal with the larger intensity characteristic value is reduced.

[0030] Specifically, intensity feature values ​​are used to quantify the brightness level and effective energy of at least two independent channel signals within the same location area, while intensity ratios are used to quantify the relative degree of intensity differences between channels, thus providing repeatable numerical basis for subsequent determination of the reflection characteristics of the location area. Specifically, in each row of data acquired line by line, the location area for each channel signal is divided by the same horizontal window, and the pixel values ​​within each location area are first subjected to saturation pixel removal and noise suppression processing. Saturation pixel removal can be achieved by excluding pixels that have reached the sampling upper limit from the statistics, and noise suppression can be achieved by smoothing isolated spikes through median filtering or moving average. Subsequently, the intensity feature values ​​of each channel signal in the location area are calculated. The intensity feature values ​​can be selected as the average value within the location area, the truncated average value after removing extreme values, or the root mean square value, to reduce the influence of individual bright spots on the statistics. In metal sheet scenes where strong specular highlights are prone to occur, the truncated average value is more conducive to reflecting the overall energy of the area. After obtaining the two intensity characteristic values, the intensity ratio between the channels is calculated. The intensity ratio can be calculated as the ratio of the larger intensity characteristic value to the smaller one, ensuring that the intensity ratio is always greater than or equal to 1. The intensity ratio is then associated with and stored with the location region index, forming an intensity ratio sequence distributed laterally. For example, in a coating production line, when oil film accumulates in the same location region, it creates a specular highlight. The intensity characteristic value of a certain channel signal will increase significantly, and the intensity ratio sequence will show a peak in that location region, thus providing a stable input for subsequent reflection characteristic discrimination. This step, by performing uniform window statistics on the location region and using intensity characteristic values ​​that are insensitive to extreme values, ensures that the intensity ratio reflects the differences in channel response rather than being dominated by single-point noise, thereby improving the repeatability of reflection characteristic discrimination.

[0031] A preset threshold is used to map the intensity ratio to a discrete determination result of reflection characteristics. The fusion weight coefficient is used to control the contribution ratio of each channel signal in the effective channel signal, so that the effective channel signal is dominated by the channel with more stable information under different reflection states. Specifically, the intensity ratio is compared with the preset threshold. When the intensity ratio exceeds the preset threshold, it indicates that the energy difference between the two channel signals in that location region has reached a significant level. It is determined that the reflection type corresponding to the channel signal with the larger intensity characteristic value dominates in that location region, and the location region is classified as specular reflection dominant. At this time, a smaller fusion weight coefficient is assigned to the channel signal with the larger intensity characteristic value, and a larger fusion weight coefficient is assigned to the other channel signal, thereby reducing the dominance of the specular channel on the effective channel signal.

[0032] The fusion weighting coefficients can be implemented using a piecewise linear allocation method. For example, when the intensity ratio just exceeds a preset threshold, the fusion weighting coefficients are kept nearly evenly distributed to avoid abrupt changes in fringe features caused by sudden weight shifts. As the intensity ratio continues to increase, the fusion weighting coefficients of the channel signals with larger intensity characteristic values ​​are gradually reduced, allowing the weights to transition smoothly with changes in reflection state. When the intensity ratio does not exceed the preset threshold, the region is determined to be dominated by scattering, and a larger fusion weighting coefficient is assigned to the channel signals with higher scattering response intensity. The preset threshold can be determined based on the dynamic range and saturation margin of the linear region of the channel signals, so that the preset threshold corresponds to the state where the two channel signals are not yet saturated but have already shown a significant energy difference, avoiding frequent misjudgments caused by an excessively low threshold or lag in highlight suppression caused by an excessively high threshold. For example, when there are localized high-brightness spots on the surface of a high-speed strip, the fusion weight coefficient of the high-brightness channel is reduced after the intensity ratio exceeds a preset threshold, making the effective channel signal closer to the stripe structure of another channel, thus maintaining the extractability of stripe features. In areas with rough surfaces or oxide scale, the intensity ratio is smaller and does not exceed the preset threshold, so the fusion weight coefficient is biased towards the channel with stronger scattering response, making the stripe dominant frequency component of the effective channel signal more stable and reducing the interference of slow reflectivity changes on subsequent defect abrupt detection. This step transforms the difference in reflection state into a controlled adjustment of the fusion weight coefficient, enabling the effective channel signal in the location region to adaptively switch information sources between specular reflection dominance and scattering dominance, thereby simultaneously suppressing both overexposure distortion and insufficient signal-to-noise ratio problems in high-reflectivity scenes. In other embodiments, the preset threshold can also be determined by the saturated pixel ratio of the location region or the contrast difference between channels, and the fusion weight coefficient can be assigned accordingly.

[0033] In one embodiment of the present invention, extracting fringe features characterizing local distortions of the optical pattern from the effective channel signal includes: Gradient calculations are performed on the effective channel signal along at least two different directions to obtain the fringe gradient distribution in each direction. The dominant direction of the stripe feature is determined based on the stripe gradient distribution in each direction, and the dominant direction is used as the directional characteristic of the surface defect of the metal sheet.

[0034] It should be noted that gradient calculation is used to quantify the rate of brightness change of the effective channel signal in spatial location, fringe gradient distribution is used to describe the edge strength and degree of concentration of change of the spatial periodic optical pattern in different directions, and the dominant direction is used to characterize the direction in which the fringe feature changes most significantly in space. This direction is also used as the directional characteristic of the surface defect of the metal plate to reflect the spatial extension orientation of the defect. Specifically, when performing gradient calculation on the effective channel signal along at least two different directions, the coordinates of the length direction and the lateral direction are first constructed on the two-dimensional signal plane formed by row-by-row acquisition. The length direction is formed by the fixed position increment triggered by row-by-row acquisition, and the lateral direction is formed by the lateral pixel sequence covered by each row of acquisition. In this two-dimensional signal plane, at least two different directions are selected to perform gradient calculation. The directions may include the length direction and the lateral direction, or include the length direction, the lateral direction, and an oblique direction at a preset angle to both. Gradient calculation can be achieved through differential operators. For example, the absolute value of the difference between adjacent sampling points in a selected direction can be taken to form a gradient magnitude map. Then, the distribution of gradient magnitude can be statistically analyzed within the window corresponding to the location region to obtain the fringe gradient distribution in each direction. To suppress the amplification effect of noise on the gradient, the effective channel signal can be lightly smoothed before gradient calculation. The width of the smoothing window is preferably smaller than the period of the spatial periodic optical pattern to avoid the fringe structure being excessively smoothed.

[0035] Subsequently, the dominant direction of the stripe feature is determined based on the stripe gradient distribution in each direction. This can be achieved by summarizing the gradient amplitudes in each direction within the same location area. For example, the mean, root mean square, or high quantile of the gradient amplitudes in each direction can be calculated as directional intensity indicators, and the direction with the largest directional intensity indicator is determined as the dominant direction. When the directional intensity indicators of multiple directions are close, the stripe feature can be judged as having weak directionality, and the dominant direction can be set to a state of no dominant direction or multiple candidate directions can be output according to preset rules. For example, in high-speed inspection of a coating production line, when a crease extends along the length of the strip, the stripe distortion produces a more obvious brightness jump in the transverse direction, making the stripe gradient distribution in the transverse direction significantly higher than that in the length direction. The dominant direction is determined to be the transverse direction, thus the directional characteristic indicates that the defect is a linear defect extending along the length direction. When local pits or dot-like indentations appear, the stripe distortion produces local changes in multiple directions, the difference in directional intensity indicators decreases, and the directional characteristics show weak directionality, providing auxiliary information for subsequent defect type differentiation. This step decomposes the spatial variation of stripe distortion into multiple directions and compares them with a unified directional intensity index, so that the directional characteristics are given by reproducible statistics. This allows us to distinguish between brightness fluctuations in high-reflectivity scenes and structural distortions caused by defects. Structural distortions have spatial directional consistency, which causes the gradient distribution in a certain direction to be consistently high in the location region. In contrast, brightness fluctuations caused by slow changes in reflectivity are mostly low-frequency slow changes. After gradient calculation, the directional differences are not significant, thus reducing directional misjudgments.

[0036] In one embodiment of the present invention, determining the dominant direction of the stripe feature based on the stripe gradient distribution in each direction includes: The gradient amplitude characteristic index is calculated for the fringe gradient distribution in each direction, and the directional strength of the fringe feature is determined based on the degree of difference between the gradient amplitude characteristic index in each direction. When the degree of difference exceeds a preset difference threshold, the fringe feature is determined to have strong directionality, and the direction with the largest gradient amplitude characteristic index is determined as the dominant direction. When the degree of difference does not exceed the preset difference threshold, the fringe feature is determined to be an isotropic feature, and the isotropic feature is used as the basis for determining the surface defect type of the metal sheet.

[0037] In the above embodiments, the gradient magnitude feature index is used to compress the fringe gradient distribution in each direction into a comparable scalar, the degree of difference is used to quantify the difference between gradient magnitude feature indices in different directions, and the preset difference threshold is used to distinguish whether the fringe feature shows a significant directional bias in space; isotropic feature means that the intensity of the fringe gradient distribution is close in multiple directions, and the fringe distortion does not have a significant dominant direction in space.

[0038] Specifically, after obtaining the fringe gradient distribution in at least two directions, gradient amplitude characteristic indices are calculated for the fringe gradient distribution in each direction within the same location region. These indices can be selected from the root mean square (RMS) value, the combination of the mean and high quantile of the gradient amplitude within the location region, or the energy value of the gradient amplitude after filtering in the dominant frequency direction of the fringe. This ensures that the indices reflect the overall intensity of change while avoiding misjudgments caused by individual noise spikes. In high-speed metal strip scenarios, the combination of the RMS value and high quantile is more effective in suppressing the influence of occasional bright spots. Subsequently, the directional strength of the fringe feature is determined based on the degree of difference between the gradient amplitude characteristic indices in each direction. The degree of difference can be characterized by the ratio or difference between the maximum and minimum gradient amplitude characteristic indices, and the degree of difference is compared with a preset difference threshold. When the degree of difference exceeds the preset difference threshold, it indicates that the fringe distortion exhibits a significant directional concentration in space, indicating that the fringe feature has strong directionality. The direction with the largest gradient amplitude characteristic index is determined as the dominant direction, and the dominant direction is output as the directional characteristic of the surface defect of the metal sheet.

[0039] The preset difference threshold can be determined based on the stripe period, sampling noise level, and background texture of the strip surface. This ensures that the preset difference threshold corresponds to a stable difference level exceeding the variation level of the background texture fluctuations caused by defects, thus avoiding misjudging random textures as strong directionality. For example, in the high-speed inspection section of a coating production line, scratches or creases will form continuous boundaries in a certain direction, causing the gradient amplitude characteristic index in that direction to be significantly greater than in other directions. Once the difference exceeds the preset difference threshold, the dominant direction stably indicates the defect extension orientation. Local pits or dot-like indentations will cause local perturbations in multiple directions, reducing the difference in gradient amplitude characteristic index in each direction. When the difference does not exceed the preset difference threshold, the stripe feature is determined to be isotropic. When isotropic features are used as the basis for determining the type of surface defects on metal sheets, weakly directional and isotropic defect features can be categorized into a local defect candidate set. This set can be combined with information such as the defect's duration and lateral expansion range to distinguish between local defects such as pits and indentations and linear defects, thereby providing a reproducible type-aiding signal in addition to defect localization. This step quantifies multi-directional gradient information using a unified gradient magnitude feature index and separates background texture fluctuations from structural defect distortions by using a threshold for the degree of difference. This allows the output of the dominant direction to be determined by statistics rather than subjective judgment, reducing the interference of overall brightness changes caused by reflectivity fluctuations on the judgment of defect directions.

[0040] Please continue reading. Figure 1 Based on the continuity and abrupt changes of the stripe features along the length direction, the length position, lateral position, and severity level of the surface defects of the metal sheet are determined.

[0041] In one embodiment of the present invention, determining the length position, lateral position, and severity level of the surface defect of the metal sheet based on the continuity and abrupt change characteristics of the stripe feature along the length direction includes: The frequency domain transformation is performed on the feature sequence formed by the stripe features along the length direction to obtain the spectrum of the feature sequence, and the energy distribution of the spectrum in different frequency bands is calculated; Based on the proportion of low-frequency energy in the total spectrum energy, the surface defects of the metal sheet are distinguished as geometric defects or surface condition defects; wherein, when the proportion of low-frequency energy exceeds a preset low-frequency energy threshold, the surface defect is determined to be a geometric defect; when the proportion of low-frequency energy does not exceed the preset low-frequency energy threshold, the surface defect is determined to be a surface condition defect.

[0042] It needs to be explained that the feature sequence formed along the length direction of the stripe features refers to a one-dimensional sequence obtained by arranging the stripe features at the same horizontal position or within the same defect candidate region in a fixed position increment order along the length direction after row-by-row acquisition and defect localization. Frequency domain transformation refers to the process of converting this one-dimensional sequence from the length coordinate domain to the spatial frequency domain, used to obtain a spectrum that can reflect the components of change at different scales. The energy distribution of the spectrum in different frequency bands is used to measure the energy ratio of the feature sequence in low-frequency and high-frequency changes. Specifically, when forming the feature sequence, the horizontal window corresponding to the defect candidate region is first selected, and the stripe features within the horizontal window are summarized at each length position increment. The summarization method can be to calculate the mean, the maximum value, or the energy value, thereby obtaining the feature sequence that changes with length. To avoid the DC bias of the feature sequence affecting the spectral energy statistics, the feature sequence can be mean-reduced, and the feature sequence can be windowed to reduce the spectral leakage caused by truncation.

[0043] The feature sequence is then subjected to frequency domain transformation to obtain the spectrum. Frequency domain transformation can be implemented using Discrete Fourier Transform (DFT), which calculates the amplitude of each spatial frequency component for a finite-length feature sequence to obtain the spectral amplitude distribution. When online real-time processing is required, Fast Fourier Transform (FFT) can be used to achieve efficient computation of DFT. After obtaining the spectrum, it is divided into low-frequency and high-frequency bands according to preset boundary frequencies, and the energy distribution of each band is calculated. The energy can be obtained by summing the squares of the spectral amplitudes. The boundary frequencies can be set according to the resolution requirements of fixed position increments and defect scales, so that the low-frequency band corresponds to changes in longer spatial scales, and the high-frequency band corresponds to changes in shorter spatial scales.

[0044] For example, in coating production line inspection, creases or localized surface undulations can create long, continuous stripe distortions along the length direction. Their characteristic sequence changes slowly and continuously, with spectral energy concentrated in the low-frequency band. Oil film textures or subtle scattering variations cause the characteristic sequence to exhibit faster, more random fluctuations, with spectral energy more pronounced in the high-frequency and broadband ranges. This step transforms the scale of characteristic sequence changes into the distribution of spectral energy across different frequency bands, allowing subsequent defect category determination to be based on reproducible scale statistics rather than single-point thresholds, thus reducing the interference of slow reflectivity variations and random noise on the determination. In other embodiments, frequency domain transformation can also be achieved using discrete cosine transform or wavelet transform, still obtaining spectral or equivalent energy distributions reflecting different scale components, thereby achieving the same or similar energy statistical effects.

[0045] Low-frequency energy ratio refers to the proportion of low-frequency energy in the total spectrum energy, used to measure whether the characteristic sequence is mainly dominated by long-scale changes or short-scale changes; geometric defects refer to defects mainly characterized by morphological disturbances such as indentations, creases, and pits, while surface state defects refer to defects mainly characterized by reflection state disturbances such as uneven oil film, oxide scale, and changes in surface texture. The two have distinguishable differences in scale characteristics in the length direction.

[0046] After obtaining the low-frequency energy and total spectral energy, the proportion of low-frequency energy is calculated and compared with a preset low-frequency energy threshold. When the proportion of low-frequency energy exceeds the preset low-frequency energy threshold, the surface defect is determined to be a geometric defect. This is because geometric defects often cause more continuous and smoother changes in stripe distortion along the length direction, and the main energy of the characteristic sequence is concentrated in the low-frequency band. When the proportion of low-frequency energy does not exceed the preset low-frequency energy threshold, the surface defect is determined to be a surface state defect. This is because surface state defects are more likely to exhibit local scattering and fluctuations in reflectivity on a short scale, and the characteristic sequence contains richer high-frequency components, resulting in a relatively low proportion of low-frequency energy. The preset low-frequency energy threshold can be determined based on fixed position increments, production line speed ranges, and the scale distribution of typical defect samples. This ensures that the low-frequency energy proportion corresponding to the threshold for geometric defects is consistently higher than that for surface state defects, while maintaining a low false positive probability in defect-free background areas. To adapt to different steel grades or different surface treatment processes, the preset low-frequency energy threshold can be configured in stages or applied according to the process formula during production switching.

[0047] For example, when creases extending along the length of the strip surface appear, the feature sequence continuously deviates from the baseline within a range of tens to hundreds of positional increments. Once the low-frequency energy proportion exceeds a preset low-frequency energy threshold, it is classified as a geometric defect. When uneven oil film causes mottled textures on the surface, the feature sequence fluctuates rapidly within a shorter length range. Since the low-frequency energy proportion does not exceed the preset low-frequency energy threshold, it is classified as a surface condition defect. This step maps the scale difference of defects along the length direction to a threshold for low-frequency energy proportion, enabling defect category differentiation to be completed with a small amount of statistics in high-speed online scenarios, reducing the impact of overall amplitude drift introduced by illumination fluctuations on category determination. In other embodiments, the division between low-frequency and high-frequency bands can also employ multi-band division and be determined by the joint proportion of low-frequency and mid-frequency bands, or by using the cumulative percentile of spectral energy distribution as the proportion indicator.

[0048] In one embodiment of the present invention, when the surface defect is determined to be a geometric defect, the method further includes: Calculate the characteristic frequencies of energy distribution in the low-frequency band of the spectrum; Based on the comparison result between the characteristic frequency and the preset characteristic frequency threshold, the geometric defect is distinguished as either a local geometric defect or an extended geometric defect; wherein, when the characteristic frequency is higher than the preset characteristic frequency threshold, the geometric defect is determined to be a local geometric defect; when the characteristic frequency is lower than or equal to the preset characteristic frequency threshold, the geometric defect is determined to be an extended geometric defect.

[0049] It should be noted that the characteristic frequency of energy distribution in the low-frequency band is used to characterize the dominant spatial scale of geometric defects in the length direction. The higher the characteristic frequency, the more the characteristic sequence is dominated by relatively short-scale changes in the low-frequency band, while the lower the characteristic frequency, the more the characteristic sequence is dominated by longer-scale gradual changes.

[0050] After obtaining the spectrum of the characteristic sequence and completing the geometric defect determination, the low-frequency band of the spectrum is selected as the analysis interval. The energy value of each spatial frequency point within the low-frequency band is normalized to make the energy distribution under different defect intensities comparable. Subsequently, the characteristic frequency of the energy distribution within the low-frequency band is calculated. The characteristic frequency can be realized by using the weighted center frequency of the energy in the low-frequency band, that is, calculating the weighted average with each spatial frequency point as the independent variable and the corresponding energy value as the weight to obtain the frequency representing the location of energy concentration. To avoid noise biasing the weighted result at the boundary of the low-frequency band, frequency points with energy below the preset noise floor in the low-frequency band can be removed before calculation, or the characteristic frequency can be calculated only within the effective frequency range where the energy accumulation in the low-frequency band reaches a preset proportion.

[0051] For example, when local indentations appear on rolled or annealed strip, the stripe feature exhibits undulations over a short length. Although the low-frequency energy accounts for a high proportion, its distribution is closer to the high end of the low-frequency range, resulting in a relatively high calculated characteristic frequency. Conversely, when extended creases or slow undulations occur, the stripe feature continuously deviates from the baseline over a longer length, with the low-frequency energy concentrated in a region closer to zero frequency, resulting in a relatively low characteristic frequency. This step, by compressing the energy concentration location within the low-frequency range into a single characteristic frequency, allows subsequent subdivision and determination of geometric defects to be completed with lower computational load in online scenarios, while avoiding scale misjudgments caused by relying solely on defect amplitude.

[0052] A preset feature frequency threshold is used to map feature frequencies to discrete classification results of local and extended geometric defects. Local geometric defects correspond to geometric perturbations with shorter length scales, while extended geometric defects correspond to geometric perturbations with longer length scales. Specifically, the calculated feature frequencies are compared with the preset feature frequency threshold. When the feature frequency is higher than the preset feature frequency threshold, the geometric defect is determined to be a local geometric defect. This is because a higher feature frequency means that the low-frequency energy distribution is more biased towards the higher frequency side, and the main change scale of the feature sequence is relatively shorter, which is consistent with the behavior of defects such as local indentations and local pits that appear concentrated in the length direction. When the feature frequency is lower than or equal to the preset feature frequency threshold, the geometric defect is determined to be an extended geometric defect. This is because a lower feature frequency means that the low-frequency energy is closer to zero frequency, and the feature sequence is dominated by continuous deviations at longer scales, which is consistent with the behavior of defects such as creases and extended undulations that persist along the length direction. The preset characteristic frequency threshold can be determined by combining the fixed position increment and the target classification scale: the fixed position increment determines the frequency resolution of the spectrum and the shortest distinguishable scale. The preset characteristic frequency threshold preferably corresponds to the boundary scale that distinguishes the typical length of local defects from the typical length of extended defects, so that the characteristic frequency of local geometric defects is statistically stable above the threshold, and the characteristic frequency of extended geometric defects is stable below or equal to the threshold. When switching between different production line speeds or different surface treatment processes, the preset characteristic frequency threshold can be updated according to the typical defect scale of the process segment.

[0053] For example, when the fixed position increment is 0.2 mm and the typical length of a local defect is less than 20 mm, the characteristic frequency threshold can be selected to correspond to the spatial frequency boundary at the 20 mm scale. This classifies local indentations as local geometric defects, while creases extending for hundreds of millimeters along the length direction are classified as extended geometric defects. This step establishes an executable discriminative relationship between the concentration location of low-frequency energy and the defect scale using the characteristic frequency threshold, allowing for further subdivision of geometric defects within the same inspection process, facilitating subsequent quality assessment and handling strategy formulation. In other embodiments, the preset characteristic frequency threshold can also be obtained from historical production line data, or set separately for different transverse regions to accommodate scale differences caused by tension variations between the edges and the center.

[0054] In one embodiment of the present invention, when the surface defect is determined to be a surface condition type defect, the method further includes: Calculate the proportion of high-frequency energy in the total energy of the spectrum; Based on the comparison between the high-frequency energy ratio and the preset high-frequency energy threshold, the surface condition defects are distinguished as random surface condition defects or distributed surface condition defects; wherein, when the high-frequency energy ratio exceeds the preset high-frequency energy threshold, the surface condition defects are determined to be random surface condition defects; when the high-frequency energy ratio does not exceed the preset high-frequency energy threshold, the surface condition defects are determined to be distributed surface condition defects.

[0055] The high-frequency energy proportion is used to quantify the proportion of the rapidly fluctuating components of the stripe feature along its length in the overall energy. The high-frequency band corresponds to variations at shorter spatial scales and can reflect the texturing, granular, or mottled characteristics of surface condition defects. After obtaining the spectrum of the feature sequence and determining that the surface defect is a surface condition defect, the spectrum is divided into high-frequency bands according to preset boundary frequencies. The energy values ​​of each spatial frequency point within the high-frequency band are accumulated to obtain the high-frequency energy. Simultaneously, the energy values ​​of the entire frequency band are accumulated to obtain the total spectral energy. The energy value can be obtained by squared the spectral amplitude to ensure that the energy accumulation remains monotonically consistent with the signal amplitude change. To avoid the noise floor at both ends of the spectrum dominating the high-frequency energy, noise floor suppression processing can be performed on the spectral amplitude before accumulation. For example, the energy at frequencies below the noise floor threshold can be set to zero, or energy can only be calculated within the frequency range where the spectral amplitude exceeds the noise floor threshold. Subsequently, the proportion of high-frequency energy in the total spectral energy is calculated, and the high-frequency energy proportion is associated with and stored with the defect candidate region index to form a statistic that can be used for classification.

[0056] For example, on a coating production line, uneven oil film may manifest as fine spots or localized texture fluctuations, with stripe features exhibiting rapid random undulations along the length direction. The spectrum shows a wider energy distribution in the high-frequency band, increasing the proportion of high-frequency energy. Conversely, when there is a relatively uniform hazy contamination or a slower coating thickness drift, the stripe features change more smoothly along the length direction, and the proportion of high-frequency energy is relatively low. This step characterizes the short-scale fluctuations of surface condition defects by using the proportion of high-frequency energy, allowing subsequent classification to be based on the statistical regularity of spectral energy, rather than relying on single-point brightness thresholds or human experience. This reduces the impact of lighting fluctuations and overall reflectivity changes on classification.

[0057] A preset high-frequency energy threshold is used to map the proportion of high-frequency energy to discrete classification results of random and distributed surface state defects. Random surface state defects correspond to irregular, short-scale, and rapidly fluctuating surface states along the length direction, while distributed surface state defects correspond to relatively continuous, uniform, or slowly changing banded distributions along the length direction. Specifically, the calculated proportion of high-frequency energy is compared with the preset high-frequency energy threshold. When the proportion of high-frequency energy exceeds the preset high-frequency energy threshold, the surface state defect is determined to be a random surface state defect. This is because a high proportion of high-frequency energy indicates a large proportion of short-scale, rapidly changing components in the feature sequence, consistent with defects such as random spots, particulate contamination, splatter points, or localized oil film patches. When the proportion of high-frequency energy does not exceed the preset high-frequency energy threshold, the surface state defect is determined to be a distributed surface state defect. This is because the proportion of short-scale fluctuation components is limited, and the feature sequence is more likely to be dominated by low- and mid-frequency components, consistent with defects such as slow coating thickness drift, banded oil film distribution, or relatively uniform oxide scale coverage. The preset high-frequency energy threshold can be determined by combining the fixed position increment with the typical texture scale of surface defects: the smaller the fixed position increment, the wider the high-frequency range that the spectrum can cover, and the more likely random defects are to form a significant energy proportion in the high-frequency band. Therefore, the threshold should be set in conjunction with the fixed position increment so that the high-frequency energy proportion of the defect-free background texture is mostly lower than the threshold, while the high-frequency energy proportion of random defects is mostly higher than the threshold. When switching between different steel grades or different surface treatment processes, the corresponding preset high-frequency energy threshold can be loaded according to the process formula.

[0058] For example, when random spatter points appear on the strip surface causing rapid fluctuations in the stripe pattern, the high-frequency energy percentage exceeding a preset high-frequency energy threshold is classified as a random surface condition defect. Conversely, when the coating thickness slowly drifts along its length, forming a wide-band distribution, the high-frequency energy percentage not exceeding a preset high-frequency energy threshold is classified as a distributed surface condition defect. This step further subdivides surface condition defects based on the high-frequency energy percentage threshold, enabling the production line to adopt different handling strategies for different types of surface condition anomalies while maintaining reproducible classification rules within the same inspection process.

[0059] The high-precision thin metal sheet detection method of the present invention has been described above. The high-precision thin metal sheet detection device of the present invention is described below. Please refer to [link / reference]. Figure 2 One embodiment of the high-precision thin metal sheet inspection device of the present invention includes: The pattern projection module 101 is used to project an optical pattern with a spatial period onto the surface of a metal plate and to receive reflected light from the surface of the metal plate. The multi-channel acquisition module 102 is used to perform parallel separate acquisition of the reflected light at the same acquisition time, and acquire at least two independent channel signals, wherein the at least two independent channel signals correspond to reflected light with different polarization states and / or different spectral channels; The synchronous trigger module 103 is used to acquire the motion pulse signal of the metal plate, trigger the line-by-line acquisition of the at least two independent channel signals with the motion pulse signal, so that the acquisition result of each line corresponds to the fixed position increment of the metal plate along the length direction, and control the update of the optical pattern to keep it synchronized with the line-by-line acquisition. The feature extraction module 104 is used to perform availability screening on the at least two independent channel signals, obtain effective channel signals that meet the contrast conditions and have not saturated, and extract stripe features that characterize the local distortion of the optical pattern on the effective channel signals; wherein, when one channel signal is unavailable, the stripe features of the other channel signal are used as a substitute. The defect determination module 105 is used to determine the length position, lateral position and severity level of the surface defect of the metal plate based on the continuity and abrupt change characteristics of the stripe feature along the length direction.

[0060] above Figure 2 The high-precision thin metal sheet detection device in the embodiments of the present invention will be described in detail from the perspective of modular functional entities. The high-precision thin metal sheet detection equipment in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0061] Figure 3This is a schematic diagram of a high-precision thin metal sheet inspection device 200 provided in an embodiment of the present invention. The high-precision thin metal sheet inspection device 200 can vary significantly due to different configurations or performance characteristics. It may include one or more processors 210 (e.g., one or more processors) and a memory 220, and one or more storage media 230 (e.g., one or more mass storage devices) for storing application programs 233 or data 232. The memory 220 and storage media 230 can be temporary or persistent storage. The program stored in the storage media 230 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the high-precision thin metal sheet inspection device 200. Furthermore, the processor 210 may be configured to communicate with the storage media 230 and execute the series of instruction operations in the storage media 230 on the high-precision thin metal sheet inspection device 200 to implement the steps of the aforementioned high-precision thin metal sheet inspection method.

[0062] The high-precision thin metal sheet inspection equipment 200 may also include one or more power supplies 240, one or more wired or wireless network interfaces 250, one or more input / output interfaces 260, and / or one or more operating systems 231, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The structure of the high-precision thin metal sheet inspection equipment shown does not constitute a limitation on the high-precision thin metal sheet inspection equipment provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0063] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A high-precision method for detecting thin metal sheets, characterized in that, include: An optical pattern with a spatial period is projected onto the surface of a metal sheet, and reflected light from the surface of the metal sheet is received. At the same acquisition time, the reflected light is acquired in parallel to obtain at least two independent channel signals, which correspond to reflected light with different polarization states and / or different spectral channels; The motion pulse signal of the metal plate is acquired, and the motion pulse signal is used to trigger the line-by-line acquisition of the at least two independent channel signals, so that the acquisition result of each line corresponds to the fixed position increment of the metal plate along the length direction, and the update of the optical pattern is controlled to keep synchronized with the line-by-line acquisition. The availability of the at least two independent channel signals is screened to obtain the effective channel signal that meets the contrast condition and has not saturated. Then, the fringe features that characterize the local distortion of the optical pattern are extracted from the effective channel signal. When one channel signal is unavailable, the fringe features of the other channel signal are used as a substitute. Based on the continuity and abrupt changes of the stripe features along the length direction, the length position, lateral position, and severity level of the surface defects of the metal sheet are determined.

2. The high-precision thin metal sheet inspection method according to claim 1, characterized in that, The process of performing availability filtering on the at least two independent channel signals and obtaining valid channel signals includes: When the at least two independent channel signals both meet the contrast condition and do not saturate in the same location area, the channel signals are dynamically weighted and fused according to the reflection characteristics of the location area to obtain an effective channel signal. Specifically, in regions dominated by specular reflection, the fusion weight of channel signals with high specular reflection response intensity is reduced; in regions dominated by scattering, the fusion weight of channel signals with high scattering response intensity is increased.

3. The high-precision thin metal sheet inspection method according to claim 2, characterized in that, The dynamic weighted fusion of signals from each channel based on the reflection characteristics of the location region includes: Calculate the intensity characteristic values ​​of the at least two independent channel signals in the location region, and calculate the intensity ratio between the channels based on the intensity characteristic values; Based on the comparison result between the intensity ratio and the preset threshold, it is determined whether the reflection characteristics of the location region are dominated by specular reflection or scattering, and a fusion weighting coefficient is assigned to each channel signal according to the reflection characteristics. When the intensity ratio exceeds the preset threshold, the location region is determined to be dominated by the reflection type corresponding to the channel signal with a larger intensity feature value, and the fusion weight coefficient of the channel signal with a larger intensity feature value is reduced.

4. The high-precision thin metal sheet inspection method according to claim 1, characterized in that, Extracting fringe features characterizing local distortions of the optical pattern from the effective channel signal includes: Gradient calculations are performed on the effective channel signal along at least two different directions to obtain the fringe gradient distribution in each direction. The dominant direction of the stripe feature is determined based on the stripe gradient distribution in each direction, and the dominant direction is used as the directional characteristic of the surface defect of the metal sheet.

5. The high-precision thin metal sheet inspection method according to claim 4, characterized in that, The process of determining the dominant direction of the stripe features based on the stripe gradient distribution in each direction includes: The gradient amplitude characteristic index is calculated for the fringe gradient distribution in each direction, and the directional strength of the fringe feature is determined based on the degree of difference of the gradient amplitude characteristic index in each direction. When the degree of difference exceeds a preset difference threshold, the stripe feature is determined to have strong directionality, and the direction with the largest gradient amplitude feature index is determined as the dominant direction. When the degree of difference does not exceed the preset difference threshold, the stripe feature is determined to be an isotropic feature, and the isotropic feature is used as the basis for determining the surface defect type of the metal sheet.

6. The high-precision thin metal sheet inspection method according to claim 1, characterized in that, The step of determining the length position, lateral position, and severity level of the surface defect of the metal sheet based on the continuity and abrupt changes of the stripe features along the length direction includes: The frequency domain transformation is performed on the feature sequence formed by the stripe features along the length direction to obtain the spectrum of the feature sequence, and the energy distribution of the spectrum in different frequency bands is calculated; Based on the proportion of low-frequency energy in the total energy of the spectrum, the surface defects of the metal sheet can be distinguished as geometric defects or surface condition defects. Specifically, when the proportion of low-frequency energy exceeds a preset low-frequency energy threshold, the surface defect is determined to be a geometric defect; when the proportion of low-frequency energy does not exceed the preset low-frequency energy threshold, the surface defect is determined to be a surface state defect.

7. The high-precision thin metal sheet inspection method according to claim 6, characterized in that, When the surface defect is determined to be a geometric defect, the method further includes: Calculate the characteristic frequencies of energy distribution in the low-frequency band of the spectrum; Based on the comparison result between the characteristic frequency and the preset characteristic frequency threshold, the geometric defect is distinguished as a local geometric defect or an extended geometric defect. Specifically, when the characteristic frequency is higher than the preset characteristic frequency threshold, the geometric defect is determined to be a local geometric defect; when the characteristic frequency is lower than or equal to the preset characteristic frequency threshold, the geometric defect is determined to be an extended geometric defect.

8. The high-precision thin metal sheet inspection method according to claim 6, characterized in that, When the surface defect is determined to be a surface condition type defect, the following is also included: Calculate the proportion of high-frequency energy in the total energy of the spectrum; Based on the comparison between the high-frequency energy ratio and the preset high-frequency energy threshold, the surface state defects are distinguished as random surface state defects or distributed surface state defects. Specifically, when the proportion of high-frequency energy exceeds the preset high-frequency energy threshold, the surface condition defect is determined to be a random surface condition defect; when the proportion of high-frequency energy does not exceed the preset high-frequency energy threshold, the surface condition defect is determined to be a distributed surface condition defect.

9. A high-precision thin metal sheet inspection device, characterized in that, include: A pattern projection module is used to project an optical pattern with a spatial period onto the surface of a metal sheet and to receive reflected light from the surface of the metal sheet. A multi-channel acquisition module is used to perform parallel separate acquisition of the reflected light at the same acquisition time, and acquire at least two independent channel signals, wherein the at least two independent channel signals correspond to reflected light with different polarization states and / or different spectral channels; The synchronous triggering module is used to acquire the motion pulse signal of the metal plate, and to trigger the line-by-line acquisition of the at least two independent channel signals with the motion pulse signal, so that the acquisition result of each line corresponds to the fixed position increment of the metal plate along the length direction, and to control the update of the optical pattern to keep it synchronized with the line-by-line acquisition. The feature extraction module is used to perform availability screening on the at least two independent channel signals, obtain the effective channel signal that meets the contrast condition and has not saturated, and extract stripe features that characterize the local distortion of the optical pattern on the effective channel signal; wherein, when one channel signal is unavailable, the stripe features of the other channel signal are used as a substitute. The defect determination module is used to determine the length position, lateral position, and severity level of the surface defect of the metal sheet based on the continuity and abrupt change characteristics of the stripe feature along the length direction.

10. A high-precision testing device for thin metal sheets, characterized in that, The high-precision thin metal sheet inspection equipment includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the high-precision thin metal sheet inspection device to perform the steps of the high-precision thin metal sheet inspection method as described in any one of claims 1 to 8.

Images