A calender material shape color mutation detection method based on image sequence

Abstract

The application discloses a calender material shape and color mutation detection method based on an image sequence, and comprises the following steps: establishing a section corresponding relationship and combining a link state to form a link state level sequence, completing the judgment of a main detection position and an auxiliary detection position, making the front and rear positions of the same material section in the multi-position image connected, and solving the problem that the cross-position fragments are difficult to correspond; extracting shape change characteristics and color change characteristics, generating shape and color mutation candidate sections, and improving the recognition integrity of the shape and color abnormalities of the calender material; performing successive positioning on the candidate sections and generating shape and color mutation continuation sections, reducing the influence of the position switching on the continuity of the abnormal detection; and merging the front and rear sections, improving the continuity and accuracy of the shape and color mutation section judgment.

Description

Technical Field

[0001] This invention relates to the technical field of online detection of materials in calendering mills and industrial image processing, and particularly to a method for detecting abrupt changes in the shape and color of materials in calendering mills based on image sequences. Background Technology

[0002] With the increasing application of calendering processes in continuous manufacturing scenarios such as calendered glass, metal strips, lithium electrode sheets, rubber and plastic sheets, and composite films, online inspection technologies for the surface condition of calendered materials are also gradually developing. Existing technologies typically use industrial cameras to continuously image the surface of materials during operation, combined with image processing methods such as edge recognition, texture analysis, color difference recognition, or defect classification to detect surface anomalies such as indentations, streaks, bulges, loss of gloss, and color differences. While these technologies can improve the online inspection capabilities of calendered materials, in actual production lines, calendered materials are usually in a continuous operating state, with a large inspection area. The image range acquired by a single machine station is limited, and the same material section often needs to pass through multiple machine stations sequentially. Only then can a complete observation be completed. When anomalies on the material surface manifest as both morphological and color changes, and these anomalies span the imaging range of two or more camera positions, existing detection methods tend to fragment the same anomaly into multiple independent segments, resulting in incomplete anomaly boundaries and difficulty in connecting preceding and following segments. Furthermore, in multi-camera detection scenarios, there are also common issues such as varying image transmission dwell times, different degrees of image loss, unstable image arrival intervals, and fluctuating image clarity, leading to differences in the detection status of different camera positions. If detection is still performed according to a fixed camera position sequence, when the status of a certain camera position declines, it is easy to encounter problems such as untimely reception of anomaly segments and difficulty in accurately corresponding subsequent image segments, thereby affecting the continuous judgment results of abrupt changes in the shape and color of the calender material.

[0003] For example, CN105572143A discloses a method for detecting periodic defects on the surface of calendered materials during the calendering process. This method identifies periodic defects by extracting features from surface defect images of calendered materials and combining them with correlation matching. However, this scheme focuses on the periodic analysis of defect images. The detection process revolves around a single image analysis chain and does not involve establishing the correspondence between images of the same material segment at different machine positions, nor does it involve the processing method of switching the detection machine position according to the changes in the state of each machine position. Therefore, it is difficult to form complete and continuous abnormal segments for morphological and color abnormalities that occur across machine positions.

[0004] For example, CN110702699A discloses a rolled glass defect detection device and method. This scheme detects surface defects of rolled glass by setting up a light source component, an imaging component, and an image processing process, which can reduce the influence of rolling texture on defect identification and improve the identification effect of surface defects of rolled glass. However, this scheme mainly focuses on the optimization of imaging conditions of rolled glass and defect identification based on single imaging. The focus is on the suppression of rolling texture and the improvement of defect display effect. It does not limit the segment mapping relationship between adjacent machine positions of continuously running materials, nor does it introduce the difference in machine position status into the connection process of abnormal segments. Therefore, in the scenario of continuous detection at multiple machine positions, there are still problems such as insufficient continuity of abnormal segments and difficulty in merging segments.

[0005] In summary, existing calender material image detection technologies have the following problems: First, the correspondence between the same material segment in images of adjacent machine positions is unclear, making it difficult to accurately connect abnormal segments across machine positions; second, the image transmission status of different machine positions differs, and when the machine position status changes, existing technologies lack corresponding machine position switching methods, which can easily cause interruptions in abnormal segments; third, for anomalies that simultaneously exhibit both morphological and color changes, existing technologies often treat them as local single-image defects, lacking a joint determination of the continuity relationship between preceding and following segments.

[0006] In view of the above-mentioned problems in the existing technology, the present invention is proposed. Therefore, the problem to be solved by the present invention is: how to establish the correspondence between the same material segment under the condition of material image sequence in a multi-station calender, and to continuously detect and merge abnormal segments under the combined effect of shape and color changes by combining the state changes of each station. In response to the above problems, the present invention provides a method for detecting shape and color mutations in calender materials based on image sequence. This method solves the problem of continuous identification of shape and color mutation segments in calender materials under multi-station conditions by establishing segment correspondence, forming a link state level sequence, determining the main detection station and auxiliary detection station, generating candidate segments and continuation segments of shape and color mutations, and merging the segments before and after. Summary of the Invention

[0007] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0008] In view of the aforementioned existing problems, the present invention is proposed.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: As a preferred embodiment of the image sequence-based method for detecting abrupt changes in the shape and color of calender materials according to the present invention, the method involves: establishing a segment correspondence relationship for the image sequence of materials in a multi-station calender according to the material running direction, and forming a link status level sequence by combining the link status of each station, thereby generating the main detection station and auxiliary detection station. Extract morphological and color change features from the image sequence corresponding to the main detection unit, and generate candidate segments for morphological and color abrupt changes according to the continuity relationship of morphological and color changes; According to the segment correspondence, the candidate segments of shape and color mutation are sequentially located in the image sequence corresponding to the auxiliary detection position, and a continuous segment of shape and color mutation is generated when the position sequence is switched. The candidate segments of shape and color change and the continuous segments of shape and color change are merged according to the continuity of material position and the continuity of shape and color change to generate material shape and color change segments.

[0010] The beneficial effects of this invention are as follows: By establishing segment correspondence and combining it with link status to form a link status level sequence, this invention completes the determination of the main detection position and the auxiliary detection position, enabling the connection of the preceding and following positions of the same material segment in multi-position images, thus solving the problem of difficulty in corresponding segments across positions; by extracting morphological change features and color change features, candidate segments for shape and color abrupt changes are generated, improving the completeness of identifying abnormal shapes and colors of materials in the calender; by performing continuous positioning of candidate segments and generating continuous segments for shape and color abrupt changes, the impact of position switching on the continuity of anomaly detection is reduced; and by merging preceding and following segments, the continuity and accuracy of determining segments for shape and color abrupt changes are improved. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments 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 these drawings without creative effort. Wherein: Figure 1 This is a schematic flowchart of the image sequence-based method for detecting abrupt changes in the shape and color of materials in a calender, as shown in this invention. Detailed Implementation

[0012] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0013] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0014] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0015] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a method for detecting abrupt changes in the shape and color of materials in a calender based on image sequences, specifically including the following steps: S1. Establish segment correspondences for the material image sequence of the multi-station calender according to the material running direction, and form a link status level sequence based on the link status of each station, generating the main detection station and auxiliary detection stations. Note that the following should be noted in this step: In this embodiment, the calender is equipped with three camera positions along the material running direction, namely camera position 1, camera position 2, and camera position 3. Each camera position corresponds to one set of industrial camera and one set of supplementary lighting components. The industrial camera has a collection resolution of 2448×2048, a collection frequency of 80 frames / s, an exposure time of 0.8 ms, and a lens optical axis perpendicular to the material surface. The installation spacing between adjacent camera positions is 1.6 m and 1.8 m, respectively. The material is a continuous strip calendered material with a material width of 32 mm and a material running speed of 0.9 m / s. The images from each camera position are formed into a corresponding camera position image sequence according to the order of collection, and are received, cached, and sorted by the on-site industrial control computer.

[0016] S1.1 Arrange the material image sequence of the multi-station calender according to the installation order of each station on the calender, and extract the material edge line, width boundary line and surface reference texture from each station image in combination with the material running direction to form a station section identification group.

[0017] In a preferred embodiment, surface images of the same batch of material at the corresponding machine positions are continuously acquired at the first, second, and third machine positions respectively; an acquisition time mark and machine position mark are added to each frame of image; then, the image sequences of the first, second, and third machine positions are arranged sequentially according to the installation order of the machine positions on the calender to form a multi-machine calender material image sequence; wherein, arranging according to the installation order of each machine position on the calender means first determining the machine position sequence according to the order in which the material passes through the machine positions, and then arranging the corresponding images within each machine position according to the acquisition time, thereby forming a dual arrangement structure of machine position sequence + time sequence.

[0018] In a preferred embodiment, the preset running direction of the material on the equipment is first determined based on the fixed installation positions of the calender's discharge and feed ends. Then, two adjacent frames are continuously captured at any machine position, and the displacement direction of the surface reference texture, edge gaps, or surface micro-color spots in the images is checked. When the displacement direction is consistent with the preset running direction, this direction is determined as the material running direction. When the displacement direction is inconsistent with the preset running direction, the actual forward direction of the feature displacement in the two adjacent frames is taken as the material running direction. That is to say, the material running direction can be obtained according to the equipment installation direction or verified according to the texture displacement direction in the image. After this processing, interference caused by the adjustment of the camera installation direction or changes in image flipping settings after equipment maintenance can be avoided in subsequent position determination.

[0019] Furthermore, the formation method of the machine position segment identification group specifically includes: firstly, separating the material area and the background area in each machine position image; then extracting the material edge lines on the left and right sides of the material area; extracting the width boundary line based on the lateral range between the left and right edges of the material; then extracting the stripes, indentations, roller marks, or particle distributions that appear continuously along the running direction in the middle of the material surface as the surface reference texture; in each frame image, the machine position mark, image acquisition time, left material edge line position, right material edge line position, width center line position, the position of the first obvious feature point of the surface reference texture, and the position of the last obvious feature point of the surface reference texture together constitute the segment identification unit corresponding to that frame image; after multiple segment identification units are arranged in chronological order, they constitute the machine position segment identification group corresponding to that machine position.

[0020] For example, in a frame of image from camera position 2, the left material edge line is located at the horizontal coordinate 112, the right material edge line is located at the horizontal coordinate 2136, the width center line is located at 1124, the first obvious feature point of the surface reference texture is located at the vertical coordinate 368, and the last obvious feature point is located at the vertical coordinate 1496. Then, the segment identification unit corresponding to this frame of image can be recorded according to the above 7 items (i.e., camera position mark, image acquisition time, left material edge line position, right material edge line position, width center line position, first obvious feature point of the surface reference texture position, and last obvious feature point of the surface reference texture position).

[0021] S1.2. Based on the installation spacing between adjacent machine positions and the material running direction, compare the front and back positions of the machine position segment identification group to determine the entry and exit positions of the same material segment in different machine position images, form a segment mapping order, and establish the segment correspondence relationship based on the segment mapping order.

[0022] Specifically, for two adjacent machine positions, such as machine position 1 and machine position 2, the time range corresponding to the transfer of the same material segment from the former machine position to the latter is first estimated based on the installation distance and material running speed between them. Within this time range, candidate images are selected from the image sequence of machine position 2. Then, the material edge line, width boundary line, and surface reference texture corresponding to a certain material segment in machine position 1 are checked against the corresponding contents in the candidate images one by one. When the offset state of the left material edge line relative to the width center line is consistent, the offset state of the right material edge line relative to the width center line is consistent, and the order of at least two obvious texture features in the surface reference texture is consistent, the candidate image is determined. The selected image represents the same material segment as the one in the previous camera position, but the images are displayed from different camera positions. Specifically, the front-to-back position comparison involves sequentially verifying the entry and exit positions of the same material segment in the images from the previous and subsequent camera positions. The entry position correspondence refers to the relationship between the position where the leading edge of the same material segment first enters the detection field of view of the previous camera position and the position where the leading edge of the same material segment first enters the detection field of view of the subsequent camera position. The exit position correspondence refers to the relationship between the position where the trailing edge of the same material segment leaves the detection field of view of the previous camera position and the position where the trailing edge of the same material segment leaves the detection field of view of the subsequent camera position.

[0023] Preferably, the segment mapping order includes: the segment number in the previous camera position, the corresponding segment number in the next camera position, the order of entry positions, and the order of exit positions. For example, when segment A in the first camera position corresponds to segment B in the second camera position, and segment A enters the first camera position before segment C, and segment B enters the second camera position before segment D, while segment A leaves the first camera position before segment C, and segment B leaves the second camera position before segment D, then there is a consistent segment mapping order between segment A and segment B. Based on this order, a segment correspondence relationship between segment A in the first camera position and segment B in the second camera position can be established. After the correspondence relationships of multiple adjacent camera positions are established consecutively, a segment correspondence relationship chain spanning multiple camera positions is obtained.

[0024] Specifically, the segment correspondence includes the entry position correspondence, exit position correspondence, and segment mapping order of the same material segment in adjacent machine images; among which, the entry position correspondence and exit position correspondence are determined by comparing the front and back positions of the material edge line, width boundary line, and surface reference texture in adjacent machine images, and the segment mapping order is determined by the installation spacing between adjacent machine positions and the material running direction.

[0025] S1.3 Extract the number of image missing times, transmission dwell time, image arrival interval variation, and image clarity fluctuation during the image transmission process for each camera position to form a camera position link status group.

[0026] It should be noted that the number of missing images refers to the number of image frames that should have arrived at a certain camera position but did not arrive in the industrial control computer's buffer during the continuous acquisition phase. For example, within a 10-second detection window, if the acquisition frequency is 80 frames / second, the theoretical number of image frames is 800. If 790 frames actually arrive, the number of missing images is 10. The transmission dwell time refers to the cumulative duration of the excess when the arrival time difference between two adjacent image frames at a certain camera position exceeds the normal arrival time difference limit on the industrial control computer side. For example, if the normal arrival time difference is 12.5 ms, consecutive arrival time differences of 18 ms, 21 ms, and 19 ms will result in dwell time increments of 5.5 ms, 8.5 ms, and 6.5 ms respectively, which are accumulated as the transmission dwell time. The variation in image arrival interval refers to the difference between the maximum and minimum values ​​of the arrival time difference between adjacent images within the detection window. For example, if the maximum value of the adjacent arrival time difference at a certain camera position within a 10-second detection window is 23 ms and the minimum value is 11 ms, the transmission dwell time is considered to be 10. If ms, then the image arrival interval variation is 12 ms; the image sharpness fluctuation refers to the difference between the maximum and minimum values ​​obtained after frame-by-frame statistical analysis of the texture sharpness within the material area in a continuous image.

[0027] As an example, the camera position link status group includes: camera position number, number of missing images, transmission dwell time, image arrival interval variation, and image sharpness fluctuation. For example, in a certain detection, the camera position link status group of camera position 1 is: camera position 1, 2 times, 36 ms, 7 ms, 5; the camera position link status group of camera position 2 is: camera position 2, 0 times, 18 ms, 4 ms, 3; and the camera position link status group of camera position 3 is: camera position 3, 0 times, 18 ms, 6 ms, 4.

[0028] S1.4 First, sort the images from smallest to largest according to the number of missing images for each camera position. If two or more camera positions have the same number of missing images, sort them from shortest to longest according to the transmission dwell time for each camera position. If the same camera positions still exist after the second sort, sort them from smallest to largest according to the variation in image arrival interval for the same camera position. If the same camera positions still exist after the third sort, sort them from smallest to largest according to the fluctuation in image clarity for the same camera position. The sorting results form a link status level sequence.

[0029] For example, if the number of image missing times for both the second and third cameras is 0 and the transmission dwell time is 18ms, they are placed in the third order. The image arrival interval variation for the second camera is 4ms, which is less than the 6ms for the third camera, so the second camera is placed before the third camera. Combined with the fact that the first camera has 2 missing times, the link status level sequence in this detection is: second camera - third camera - first camera.

[0030] S1.5 Among the machine positions that have a segment correspondence with the current material segment, the first machine position in the link status level sequence is selected as the main detection machine position; among the machine positions located at subsequent installation positions adjacent to the main detection machine position and that have a segment correspondence with the current material segment, that machine position is selected as the auxiliary detection machine position.

[0031] For example, if the current material segment has already established a segment correspondence in both camera positions 2 and 3, and the link status level sequence is: camera position 2 - camera position 3 - camera position 1, then camera position 2 is the main detection camera position, and camera position 3 is the auxiliary detection camera position. If the current material segment has not yet entered the field of view of camera position 3, then only the main detection camera position is selected from the camera positions where the segment correspondence has been established, and the auxiliary detection camera position is not determined. It will be determined after the subsequent segment enters the adjacent subsequent camera position.

[0032] It should be noted that during the continuous operation of the calender, the same material segment will pass through multiple machine positions in sequence, but the image transmission stability of each machine position is not consistent. If detection is performed based solely on a fixed machine position, when the machine position experiences image loss, transmission interruption, or significant fluctuation in clarity, it is easy to cause interruption or misjudgment in the identification of abrupt segments. Therefore, in this embodiment, the entry position, exit position, and mapping order of the same material segment in different machine positions are first mapped, and then the machine position with the optimal link status is selected from the machine positions with corresponding relationships as the main detection machine position, and adjacent subsequent machine positions are designated as auxiliary detection machine positions. Through the implementation of this step S1, the segment tracking and machine position selection can be synchronously established at the detection starting stage. On the one hand, this enables subsequent candidate segments to have cross-machine position continuation conditions, and on the other hand, it prevents the detection starting machine position from being fixed to a single machine position, reducing the impact of single link fluctuations on the continuity of results.

[0033] S2. Extract morphological and color change features from the image sequence corresponding to the main detection position, and generate candidate segments for morphological and color abrupt changes based on the continuity relationship of morphological and color changes. Note that the following should be noted in this step: S2.1 In the image sequence corresponding to the main detection station, divide the continuous detection zone according to the material width direction, and extract the edge fluctuation, calendering texture direction change and local bulge distribution in each continuous detection zone along the material running direction to form a morphological change feature group.

[0034] In a preferred embodiment, the specific method for dividing the continuous detection bands according to the material width direction is as follows: First, in the current image of the main detection station, the actual width range of the material is determined based on the left and right material edge lines; then, along the material width direction, i.e., the horizontal direction from the left material edge line to the right material edge line, the material area is divided into several adjacent and uniformly wide continuous detection bands; for example, when the effective width of the material is 32 mm, it is divided into 8 continuous detection bands, each with a width of 4 mm, and sequentially labeled as detection band 1 to detection band 8; wherein, detection band 1 and detection band 8 correspond to the edge area, and detection bands 2 to 7 correspond to the middle area; the material width direction is the horizontal direction connecting the left and right material edge lines extracted in step S1.

[0035] In this embodiment, edge fluctuation refers to the undulation of the material edge line along the material running direction, that is, the deviation of the edge line from its straight baseline, such as the edge line protruding outward, shrinking inward, undulating frequently, or the undulation amplitude increasing; calendering texture change refers to the directional deflection of the texture formed by the calendering roller on the material surface in the image, such as the texture that originally extended along the running direction turning to the upper left or the lower right; local bulge distribution refers to the location, dispersion, and continuous distribution of local raised areas on the material surface in the continuous detection zone; for example, if three raised bright spots appear consecutively in the fourth detection zone, and the distance between these three bright spots along the running direction is 6 mm and 8 mm respectively, then they can be classified as local bulge distribution in the continuous detection zone.

[0036] Furthermore, the morphological change feature group includes: the continuous detection zone number, the edge fluctuation change state, the calendering texture direction change state, the local bulge distribution change state, and the corresponding front and rear position range in the image; for example, in a group of continuous images of the main detection position, the edge fluctuation of the second detection zone changes from straight to inward, the calendering texture direction changes from approximately parallel to the running direction to skewed to the right, and the local bulge distribution changes from no bulges to two bulges appearing consecutively, and the corresponding position range is 24 mm to 46 mm from the upper boundary of the image of that position, then a set of morphological change feature records of that continuous detection zone can be formed; the corresponding records of multiple continuous detection zones together form the morphological change feature group.

[0037] S2.2 In the image sequence corresponding to the main detection position, extract the differences in brightness, color shift and surface gloss changes according to the continuous detection band to form a color change feature group.

[0038] In this embodiment, brightness difference refers to the change in grayscale depth between the current image region and the corresponding region of the adjacent preceding image in the same continuous detection band, i.e., the region becomes brighter or darker overall; color shift refers to the change in the overall hue of the material surface in the same continuous detection band relative to the preceding image, such as becoming more reddish, yellowish, bluish, or darker; surface gloss change refers to the change in the area, concentration, and distribution pattern of the high-brightness reflective area in the same continuous detection band relative to the preceding image, such as changing from scattered bright spots to continuous bright bands, or from continuous bright bands to local bright spots.

[0039] Specifically, the color change feature group includes: continuous detection band number, brightness difference change status, color shift change status, surface gloss change status, and the corresponding front and back position range in the image.

[0040] For example, in the continuous images of the 5th detection zone of the main detection unit, the area from 30 mm to 52 mm from the upper boundary of the image changes from a uniform gray-white state to an overall darkening, the overall tone changes from neutral gray to yellowish, and the surface highlight area changes from 3 discrete bright spots to 1 continuous narrow bright band. This area can be recorded as a set of color change characteristics in the 5th detection zone. As another example, in the edge area of ​​the 1st detection zone, the continuous image shows a change from light gray to dark gray, and at the same time the area of ​​the edge bright band shrinks. This can be recorded as a decrease in brightness difference, a color shift towards the dark gray side, and a weakening of surface gloss.

[0041] S2.3 Combine the morphological change feature group and the color change feature group according to the same continuous detection band and the preceding and following positions in adjacent images to form a sequence of morphological and color change segments.

[0042] Specifically, based on the same continuous detection band, the morphological change feature group and the color change feature group are matched within the band; then, segments are spliced ​​according to the front-to-back position relationship in adjacent images; when a certain morphological change feature and a certain color change feature have overlapping position ranges or directly connected position ranges within the same continuous detection band, the two are combined into one morphological and color change segment; multiple morphological and color change segments are arranged according to the sequential position of the material running direction to form a sequence of morphological and color change segments.

[0043] Furthermore, the sequence of shape and color change segments includes continuous detection band number, start position, end position, direction of edge fluctuation change, direction of change of calendering texture direction, direction of change of local bulge distribution, direction of change of brightness difference, direction of change of color shift, and direction of change of surface gloss. Among them, the direction of change is specifically represented by increase or decrease. For example, if the edge fluctuation changes from slight undulation to obvious undulation, it can be recorded as an increase in the amount of edge fluctuation change; if the surface gloss changes from a continuous bright band to scattered bright spots, it can be recorded as a decrease in the amount of surface gloss change.

[0044] S2.4. Sequentially determine two adjacent shape and color change segments in the shape and color change segment sequence; when two shape and color change segments are continuous in material position, compare the shape change direction of the two shape and color change segments; when the shape change direction of the two shape and color change segments is the same, compare the color change direction of the two shape and color change segments; when the color change direction of the two shape and color change segments is the same, merge the two shape and color change segments into the same shape and color change continuation segment.

[0045] As an example, the specific rules for comparing the morphological change directions of two color change segments are as follows: Check the change directions of edge fluctuations, calendering lines, and local bulge distribution in each of the two color change segments. When all three change directions are the same, and there is no intersection where one increases while the other decreases, the morphological change directions of the two color change segments are determined to be the same. For example, if the edge fluctuation, calendering line, and local bulge distribution all increase in the first color change segment, and all three increase in the second color change segment, then the morphological change directions of the two segments are the same. If any one of the change directions is different, then the morphological change directions are not considered to be the same.

[0046] As an example, the specific rules for comparing the color change directions of two shape and color change segments are as follows: Check the direction of change in brightness difference, color shift, and surface gloss in the two shape and color change segments respectively; when all three change directions are the same and there are no conflicting change states, it is determined that the color change directions of the two shape and color change segments are the same; for example, in the first shape and color change segment, the change in brightness difference decreases, the change in color shift increases, and the change in surface gloss decreases; in the second shape and color change segment, the above three items also decrease, increase, and decrease respectively, then the two have the same color change direction; only when the shape change direction is the same and the color change direction is the same, are the two adjacent shape and color change segments merged into the same shape and color change continuation segment.

[0047] Preferably, in this embodiment, the merging order is carried out segment by segment along the material running direction, so that the candidate segments have clear starting and ending boundaries.

[0048] S2.5. Merge the boundaries of the shape and color change continuation segments according to the start and end positions to generate candidate segments for shape and color change.

[0049] Specifically, color change continuation segments belonging to the same continuous detection zone and with consecutive front and rear positions are merged by taking the starting position of the foremost segment as the starting position of the merged segment and the ending position of the last segment as the ending position of the merged segment, thus obtaining candidate segments for color change abrupt changes. If color change continuation segments appear in multiple continuous detection zones within the same front and rear position range, they are then merged according to their parallel positions in the width direction to form candidate segments that cross detection zones.

[0050] It should be noted that abrupt changes on the surface of materials in a calender are usually not only manifested as color abnormalities or morphological abnormalities, but rather as synchronous or continuous correlations between edge fluctuations, texture deflections, bulge distribution, and changes in brightness, color, and gloss. If identification is based solely on single color or texture differences, localized lighting fluctuations, short-term reflections, oil stains, or lens shake can easily be misjudged as abrupt changes. In this embodiment, morphological change feature groups and color change feature groups are first extracted from the continuous detection zone, and then combined and merged according to the relationships of the same zone, the same position, and continuity, so that the identified candidate segments have both morphological and color continuity. Through this step S2, short-term isolated changes and continuous abrupt change segments can be distinguished, improving the stability and verifiability of candidate segments. Compared with the existing technology that only screens the surface color threshold or only locally identifies edge contour fluctuations, this embodiment establishes a joint continuity judgment basis of morphology and color based on the continuous detection zone, making the candidate segment boundaries more complete and reducing sources of false detection.

[0051] S3. Based on the segment correspondence, the candidate segments for shape and color mutation are sequentially located in the image sequence corresponding to the auxiliary detection position, and a continuation segment for shape and color mutation is generated when a position change occurs. It should be noted that in this step: S3.1. Based on the segment correspondence, determine the position of the subsequent material segment of the candidate segment for shape and color change in the image sequence corresponding to the auxiliary detection position, and extract the continuous image corresponding to the position of the subsequent material segment to form the continuous detection segment.

[0052] In this embodiment, the subsequent material segment position refers to the range of subsequent forward positions of the candidate segment with a change in shape and color in the auxiliary detection station image sequence, according to the segment correspondence established in step S1. Specifically, after the termination position of a candidate segment in the main detection station is determined, the image position range corresponding to the candidate segment after it continues to move forward can be determined in the auxiliary detection station image sequence based on the installation distance between the main detection station and the auxiliary detection station, the material running direction, and the correspondence between the entry and exit positions of the same material segment in the two stations. This position range is the subsequent material segment position.

[0053] Specifically, first, based on the location of the subsequent material segment, find the corresponding starting and ending images in the image sequence of the auxiliary detection station; then, starting from the starting image, continuously capture a set of images covering the location of the subsequent material segment along the time sequence; then, reorder them according to the front and back positions of the material in the image's running direction to form a continuous detection segment; for example, if the ending position of a candidate segment in the main detection station is located 58 mm from the upper boundary of the image, after the segment correspondence conversion, the subsequent forward range corresponding to this ending position in the auxiliary detection station is located from 12 mm to 46 mm from the upper boundary of the image, then capture 18 consecutive frames of images covering this range from the image sequence of the auxiliary detection station to form a continuous detection segment.

[0054] S3.2 In the continuous detection segment, according to the termination position of the candidate segment of shape and color change, the edge fluctuation, calendering texture direction change, local bulge distribution, brightness difference, color shift and surface gloss change in adjacent images are compared and connected to form a shape and color continuity segment sequence.

[0055] As an example, the method for performing before-and-after connection comparison specifically includes: using the termination position of the candidate segment of shape and color change as the connection benchmark, in the continuous detection segment, starting from the position closest to the connection benchmark, checking six items frame by frame: edge fluctuation, change of calendering texture direction, distribution of local bulges, brightness difference, color shift, and surface gloss change; when the corresponding position in the later image is directly connected to the end change state in the previous image in terms of position, and at least one of the six items has a corresponding relationship of consistent detection band number, connected position range, unchanged shape change direction, and unchanged color change direction, it is determined that there is a before-and-after connection relationship between the two; that is, the before-and-after connection comparison is not simply comparing pixel similarity, but using the termination position of the candidate segment as the benchmark, checking the position of the same band, the front and back boundaries, and the shape and color change direction item by item.

[0056] Specifically, the shape and color continuity segment sequence includes: segment sequence number, continuous detection zone number, start position, end position, edge fluctuation change direction, calendering texture direction change direction, local bulge distribution change direction, brightness difference change direction, color shift change direction, and surface gloss change direction; each segment is sorted according to the order of the material movement direction.

[0057] For example, in the continuous detection section of the auxiliary detection station, segments P1, P2 and P3 are obtained in sequence. The starting position of P1 is 12 mm and the ending position is 24 mm, the starting position of P2 is 24 mm and the ending position is 37 mm, and the starting position of P3 is 39 mm and the ending position is 46 mm. Then P1 and P2 are continuous in position, and there is a 2 mm gap between P2 and P3, which is not considered as direct continuity.

[0058] S3.3 After the link status level sequence is updated, when the auxiliary detection station is in the first priority and the main detection station is no longer in the first priority, the segment in the shape and color continuation segment sequence that is continuous with the termination position of the shape and color mutation candidate segment is determined as the switching start segment.

[0059] In a preferred embodiment, the arrival of continuous images is used as the basis for updating. After each set of consecutive detection segments is completed, the number of image missing times, transmission dwell time, image arrival interval variation, and image clarity fluctuation of each camera position in the last 5 seconds are re-statistically analyzed. Then, the positions of each camera position are rearranged according to the same sorting order as in step S1.4. The new sorted order is used as the updated link status level sequence. For example, if the initial link status level sequence is camera position 2-camera position 3-camera position 1, and camera position 2 experiences 3 consecutive image pauses during the consecutive detection phase, while camera position 3 has stable image arrival and small clarity fluctuations, then the reordered sequence may become camera position 3-camera position 2-camera position 1. When the position of the auxiliary detection camera position is updated to the first position and the original main detection camera position is no longer the first position, the camera position switching condition is met.

[0060] In this embodiment, the switching start segment refers to the first segment in the shape and color continuation segment sequence that is directly continuous with the termination position of the shape and color change candidate segment. Direct continuity means that the starting position of the segment is after the termination position of the candidate segment, and there are no other shape and color change segments or gaps between them. For example, if the termination position of the candidate segment is 58 mm, and the starting position of segment P1 in the auxiliary detection position is 58 mm and the starting position of segment P2 is 71 mm, then P1 is the switching start segment.

[0061] S3.4. Sequentially determine the adjacent segments after the switching start segment; when the adjacent segments are continuous in the material position and the direction of shape change and the direction of color change are consistent with the switching start segment, merge the adjacent segments into the same shape and color change continuation segment to generate a shape and color change continuation segment.

[0062] In this embodiment, when adjacent segments are continuous in terms of material position, it specifically means that the ending position of the previous segment and the starting position of the next segment are directly connected, or the interval between them does not exceed the material advancement length corresponding to the same continuous detection band in one frame of image, and there are no other segments with shape and color changes within this interval area; for example, if the material running speed is 0.9 m / s and the acquisition frequency is 80 frames / s, then the material advancement length corresponding to one frame of image is approximately 11.25 mm; if the ending position of the previous segment is 24 mm and the starting position of the next segment is 31 mm, and the interval between them is 7 mm, then it can be considered as continuous in terms of material position; if the interval between them is 16 mm, then it is not considered as continuous.

[0063] Furthermore, when judging adjacent segments after switching the starting segment, only when all three conditions are met simultaneously—continuous material position, consistent shape change direction, and consistent color change direction—will the adjacent segments be merged into the same shape and color change continuation segment, and a shape and color change abrupt change continuation segment be generated.

[0064] S4. Merge the candidate segments and the segments showing shape and color change continuation according to the continuity of material location and the continuity of shape and color change to generate material shape and color change segments. Note that the following should be noted in this step: S4.1 Compare the termination position of the candidate segment of shape and color mutation with the starting position of the continuation segment of shape and color mutation; when the termination position and the starting position are continuous in the material movement direction, it is determined that there is a material position continuity relationship between the candidate segment of shape and color mutation and the continuation segment of shape and color mutation.

[0065] As an example, the method for comparing the connection between the preceding and following segments specifically includes: taking the termination position of the candidate segment of shape and color change as the preceding boundary and the starting position of the continuation segment of shape and color change as the following boundary; verifying the sequential relationship between the two in the material running direction; when the starting position of the continuation segment of shape and color change is after the termination position of the candidate segment of shape and color change, and there are no other shape and color change segments between the two, it is determined that there is a material position continuity relationship between the two; if there are other unmerged shape and color change segments between the two, or if there is a gap that exceeds the allowable connection range (i.e., the material forward length corresponding to 1 frame of image, the specific value can be calculated according to the acquisition frequency and material running speed, which will not be elaborated in this embodiment), then it is not considered a material position continuity relationship.

[0066] S4.2 After determining the continuity of material positions, the candidate segments for shape and color change and the continuation segments for shape and color change are judged in sequence. When the shape change direction of the candidate segments for shape and color change and the continuation segments for shape and color change are the same, the color change direction of the candidate segments for shape and color change and the continuation segments for shape and color change are compared. When the color change direction is the same, it is determined that there is a continuity relationship between the candidate segments for shape and color change and the continuation segments for shape and color change.

[0067] Specifically, the method for sequentially judging the candidate segments and the continuation segments of shape and color abrupt changes is as follows: First, check whether the continuous detection zone numbers of the two are the same or directly adjacent in the width direction; then check the direction of shape change; when the three items of edge fluctuation change, calendering line direction change, and local bulge distribution change are all increasing or decreasing in both the candidate segment and the continuation segment, it is determined that the two have the same direction of shape change; then check the direction of color change; when the three items of brightness difference change, color shift change, and surface gloss change are all increasing or decreasing in both the candidate segment and the continuation segment, it is determined that the two have the same direction of color change; only when the material position continuity relationship and the shape and color change continuation relationship are established are they merged into one complete material shape and color abrupt change segment in the order of appearance.

[0068] It should be noted that an increase in edge fluctuation means that the degree of edge undulation in the candidate segment is greater than that in the preceding normal segment, and the degree of edge undulation in the continuing segment also continues to increase compared to its preceding connecting segment; an increase in the direction of calendering lines means that the degree of texture skewness in both the candidate segment and the continuing segment expands towards the same deviation trend; an increase in the distribution of local bumps means that changes such as an increase in the number of bumps, a decrease in the spacing between bumps, or an expansion in the coverage area of ​​bumps occur continuously in both segments; correspondingly, the same direction of color change means that the differences in brightness, color shift, and surface gloss changes in the candidate segment and the continuing segment continue along the same trend; for example, if the candidate segment shows an overall darkening, a yellowish overall tone, and a decrease in the area of ​​high-brightness reflection, and the continuing segment also shows an overall continued darkening, a continued yellowish overall tone, and a continued decrease in the area of ​​high-brightness reflection, then the two are considered to have the same direction of color change; if the candidate segment shows an overall darkening, while the continuing segment turns to an overall brightening, then the color change direction is not considered to be the same.

[0069] S4.3 When both the continuity relationship of material position and the continuity relationship of shape and color change are established, the candidate segments of shape and color change and the continuous segments of shape and color change are merged in order to generate material shape and color change segments.

[0070] Specifically, the continuity of material position includes the connection between the termination position of the candidate segment of shape and color change and the starting position of the continuation segment of shape and color change in the direction of material movement; wherein the connection between the termination and starting positions includes the starting position of the continuation segment of shape and color change being located after the termination position of the candidate segment of shape and color change, and there being no other shape and color change segments between the two. The continuity relationship of shape and color changes includes the same direction of morphological change between the candidate segment of shape and color change and the continuous segment of shape and color change, as well as the same direction of color change between the candidate segment of shape and color change and the continuous segment of shape and color change. Among them, the same direction of morphological change includes the same increase or decrease of the same three items in the candidate segment of shape and color change, such as the change in edge fluctuation, the change in the direction of calendering lines, and the change in the distribution of local bumps, as in the continuous segment of shape and color change. The same direction of color change includes the same increase or decrease of the same three items in the candidate segment of shape and color change, such as the change in brightness difference, the change in color shift, and the change in surface gloss, as in the continuous segment of shape and color change.

[0071] Preferably, in this embodiment, combined with the photovoltaic ribbon rolling scenario, steps S4.2 to S4.3 are illustrated as follows: Assuming the object to be detected is a photovoltaic ribbon rolled material with a width of 32 mm and a thickness of 0.25 mm, the second camera position is determined as the main detection camera position, and the third camera position is determined as the auxiliary detection camera position; in the image sequence of the second camera position, between the third and fifth detection bands, a candidate segment of shape and color abrupt change appears at a position 42 mm to 68 mm from the upper boundary of the image; in this candidate segment, the amount of edge fluctuation change increases, the amount of change in the direction of rolling lines increases, and the amount of change in the distribution of local bulges increases; at the same time, the amount of change in brightness difference decreases, the amount of change in color shift increases, and the amount of change in surface gloss decreases; then, in the continuous detection segment of the third camera position, at a position 69 mm to 96 mm from the upper boundary of the image... A section of shape and color change was identified at the position mm. The three morphological changes corresponding to this section were still increasing, and the three color changes were still decreasing, increasing, and decreasing, respectively. After comparison of the preceding and following sections, the starting position of this section at 69 mm in the third position was located after the ending position of the candidate section at 68 mm in the second position, with a 1 mm interval between them. No other shape and color change segments were identified within this 1 mm range. Therefore, it was determined that there was a continuous material position relationship between the two.

[0072] After determining the continuity of shape and color changes, the candidate segment and the continuous segment have the same direction of three shape changes and the same direction of three color changes. Therefore, the 42 mm to 68 mm segment in the second position and the 69 mm to 96 mm segment in the third position are merged in the order of front and back to obtain the material shape and color change segment, whose total position range is 42 mm to 96 mm.

[0073] It should also be noted that the candidate segments originate from the main detection position, while the continuation segments originate from the auxiliary detection position. Although they belong to images from different positions, they essentially correspond to the continuous abrupt changes in the same material segment during the preceding and following operations. If the two are not uniformly merged, the final detection result can only remain at the level of segment recognition, making it difficult to provide the preceding and following boundaries of the complete abrupt change segment. Through the preceding and following connection comparison and the dual verification of shape and color change direction in this embodiment, the segmented detection results can be integrated into a single material shape and color abrupt change segment in cross-position scenarios. This allows the final output result to no longer be limited to local anomalies in a single position, but to obtain the continuous abrupt change boundary across positions. On the other hand, the merging condition is based on two constraints: positional continuity and shape and color direction continuity, which can distinguish adjacent but different change segments.

[0074] In another preferred embodiment, the acquisition frequency of the industrial camera can be selected from 60 frames / s to 120 frames / s, and the installation spacing between adjacent camera positions can be adjusted from 1.0 m to 3.0 m according to the material running speed; the number of continuous detection strips can be selected as 6, 8 or 10 strips according to the material width; the update window of the link status level sequence can be selected as 3 s, 5 s or 8 s.

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting abrupt changes in shape and color of calender materials based on image sequences, characterized in that, include: Establish segment correspondences for the material image sequence of a multi-station calender according to the material running direction, and form a link status level sequence by combining the link status of each station, thereby generating the main detection station and auxiliary detection station; Extract morphological and color change features from the image sequence corresponding to the main detection unit, and generate candidate segments for morphological and color abrupt changes according to the continuity relationship of morphological and color changes; According to the segment correspondence, the candidate segments of shape and color mutation are sequentially located in the image sequence corresponding to the auxiliary detection position, and a continuous segment of shape and color mutation is generated when the position sequence is switched. The candidate segments of shape and color change and the continuous segments of shape and color change are merged according to the continuity of material position and the continuity of shape and color change to generate material shape and color change segments.

2. The method for detecting abrupt changes in shape and color of calender materials based on image sequences according to claim 1, characterized in that, Establishing the segment correspondence includes: Arrange the material image sequence of the multi-station calender according to the installation order of each station on the calender, and extract the material edge line, width boundary line and surface reference texture from each station image in combination with the material running direction to form a station section identification group; Based on the installation spacing between adjacent machine positions and the material running direction, the machine position segment identification group is compared front and back to determine the entry and exit positions of the same material segment in different machine position images, forming a segment mapping order, and establishing a segment correspondence relationship based on the segment mapping order.

3. The image sequence based color burst detection method for calender material according to claim 2, characterized in that, Forming the link state level sequence includes: Extract the number of image missing times, transmission dwell time, image arrival interval variation, and image clarity fluctuation during the image transmission process for each camera position to form a camera position link status group; First, the images at each camera position are sorted in ascending order of the number of missing images. If two or more camera positions have the same number of missing images, they are sorted in ascending order of the transmission dwell time. If the same camera positions still exist after the second sort, they are sorted in ascending order of the variation in image arrival intervals. If the same camera positions still exist after the third sort, they are sorted in ascending order of the image clarity fluctuation. The link status level sequence is formed based on the sorting results.

4. The image sequence based color burst detection method for calendering machines according to claim 3, characterized in that, Generating the main detection station and the auxiliary detection station includes: Among the machine positions that correspond to the current material segment, the first sequential machine position in the link status level sequence is selected as the main detection machine position; among the subsequent installation positions adjacent to the main detection machine position that correspond to the current material segment, the machine position is selected as the auxiliary detection machine position.

5. The image sequence based color burst detection method for calender material according to claim 1, wherein, Extracting morphological change features and color change features from the image sequence corresponding to the main detection unit, including: In the image sequence corresponding to the main detection station, continuous detection zones are divided according to the material width direction, and edge fluctuations, changes in the direction of calendering lines and the distribution of local bulges in each continuous detection zone are extracted along the material running direction to form a morphological change feature group; In the image sequence corresponding to the main detection position, brightness differences, color shifts, and surface gloss changes are extracted according to the continuous detection bands to form a color change feature group.

6. The image sequence based color burst detection method for calendering machines according to claim 5, characterized in that, Generating the candidate regions for shape and color mutations includes: The morphological change feature group and the color change feature group are combined according to the same continuous detection band and the front and back positions in adjacent images to form a sequence of morphological and color change segments; The process involves sequentially determining two adjacent shape and color change segments in the shape and color change segment sequence; when the two shape and color change segments are continuous in terms of material position, comparing the shape change direction of the two shape and color change segments; when the shape change direction of the two shape and color change segments is the same, comparing the color change direction of the two shape and color change segments; and when the color change direction of the two shape and color change segments is the same, merging the two shape and color change segments into the same shape and color change continuation segment. The boundary of the shape and color change continuation segment is merged according to the start position and the end position to generate the shape and color change candidate segment.

7. The image sequence based color burst detection method for calender material according to claim 1, wherein, Generating the shape and color mutation continuation segment includes: Based on the segment correspondence, the position of the subsequent material segment of the candidate segment of the shape and color change is determined in the image sequence corresponding to the auxiliary detection position, and the continuous image corresponding to the position of the subsequent material segment is captured to form a continuous detection segment. In the continuous detection segment, according to the termination position of the candidate segment for shape and color change, the edge fluctuation, calendering texture direction change, local bulge distribution, brightness difference, color shift and surface gloss change in adjacent images are compared and connected to form a sequence of shape and color continuity segments. After the link status level sequence is updated, when the auxiliary detection station is in the first priority and the main detection station is no longer in the first priority, the segment in the shape and color continuation segment sequence that is continuous with the termination position of the shape and color mutation candidate segment is determined as the switching start segment. The adjacent segments after the switching start segment are judged sequentially; when the adjacent segments are continuous in the material position and the direction of shape change and the direction of color change are consistent with the switching start segment, the adjacent segments are merged into the same shape and color change continuation segment to generate the shape and color change continuation segment.

8. The image sequence based color burst detection method for calendering machines according to claim 7, characterized in that, The segment correspondence includes the entry position correspondence, exit position correspondence, and segment mapping order of the same material segment in adjacent machine position images; wherein, the entry position correspondence and the exit position correspondence are determined by comparing the front and back positions of the material edge line, width boundary line, and surface reference texture in adjacent machine position images, and the segment mapping order is determined by the installation spacing between adjacent machine positions and the material running direction.

9. The image sequence based color burst detection method for calender material according to claim 1, wherein, Generating the material's shape and color abrupt change segment includes: The termination position of the candidate segment of shape and color change is compared with the starting position of the continuation segment of shape and color change; when the termination position and the starting position are continuous in the material movement direction, it is determined that there is a material position continuity relationship between the candidate segment of shape and color change and the continuation segment of shape and color change. After determining the continuity of the material positions, the candidate segments for shape and color abrupt changes and the continuation segments for shape and color abrupt changes are judged sequentially. When the direction of shape change of the candidate segments for shape and color abrupt changes and the continuation segments for shape and color abrupt changes are the same, the direction of color change of the candidate segments for shape and color abrupt changes and the continuation segments for shape and color abrupt changes are compared. When the direction of color change is the same, it is determined that there is a continuity relationship of shape and color change between the candidate segments for shape and color abrupt changes and the continuation segments for shape and color abrupt changes. When the material position continuity relationship and the shape and color change continuity relationship are both true, the shape and color change candidate segment and the shape and color change continuity segment are merged in order to generate the material shape and color change segment.

10. The image sequence based color burst detection method for calendering machines according to claim 9, characterized in that, The continuity of material position includes the connection between the termination position of the candidate segment of shape and color change and the starting position of the continuation segment of shape and color change in the direction of material movement; wherein the connection between the termination and starting positions includes the starting position of the continuation segment of shape and color change being located after the termination position of the candidate segment of shape and color change, and there being no other shape and color change segments between them. The relationship of shape and color change continuity includes the relationship that the shape and color change candidate segment and the shape and color change continuity segment have the same direction of shape change, and the relationship that the shape and color change candidate segment and the shape and color change continuity segment have the same direction of color change; wherein, the relationship that the shape and color change direction is the same includes the relationship that the edge fluctuation change, the calendering line direction change, and the local bulge distribution change in the shape and color change candidate segment are all increased or all decreased in the shape and color change continuity segment; the relationship that the color change direction is the relationship that the brightness difference change, the color shift change, and the surface gloss change in the shape and color change candidate segment are all increased or all decreased in the shape and color change continuity segment.

Images