Multi-process parameter collaborative control method and system for hot compress and warm patch

Abstract

The application relates to the technical field of process parameter control, in particular to a multi-process parameter cooperative control method and system for hot compress and warm patch, and the following scheme is provided: in the transmission belt conveying process, an image sensor is used to collect a patch original image, a connected domain formed by glare and motion trailing is screened and processed, and a restored image is generated; in the restored image, hole domain segmentation is performed to obtain a real hole domain distribution; based on the segmentation result, a hole index is calculated, and in combination with a preset index mapping standard, corresponding process compensation parameters are determined; through closed-loop feedback of image processing and process parameter mapping, false hole interference is effectively inhibited, the accuracy of the hole detection result is ensured, and therefore the stability and performance consistency of the hot compress and warm patch in the production process are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process parameter control, in particular to a multi-process parameter collaborative control method and system for hot compress and warm patch. BACKGROUND

[0002] The existing quality detection method of hot compress and warm patch generally relies on sampling inspection or low-precision surface observation, and the detection result often cannot fully reflect the overall pore distribution of the product, and there is a problem of insufficient coverage. Even if image acquisition means is used, the detection result may still have problems such as statistical distortion of pore size, large deviation of porosity, or unstable pore domain morphology, resulting in inconsistency between the detection result and the actual performance. Further, there is a lack of effective linkage between detection and process in the production process, and the detection data is often only used for recording, and cannot be fed back to the production link in time, resulting in insufficient consistency between product batches and problems such as uneven heating or reduced use comfort.

[0003] To solve the above problems, the present application designs a multi-process parameter collaborative control method and system for hot compress and warm patch. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a multi-process parameter collaborative control method and system for hot compress and warm patch to solve the problems of the prior art. In the transmission belt conveying process, an image sensor is used to collect a patch original image, and connected domains formed by glare and motion smearing are screened and processed to generate a restored image. In the restored image, pore domain segmentation is performed to obtain the real pore domain distribution. Based on the segmentation result, pore index is calculated, and combined with a preset index mapping standard, corresponding process compensation parameters are determined to correct the heating reaction formula proportion, the amount of moisture regulator, the thickness distribution of the coating layer, or the process link of the packaging film air permeability.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] A multi-process parameter collaborative control method for hot compress and warm patch is applied to a finished product detection device for hot compress and warm patch, the finished product detection device comprising an image sensor arranged directly above a transmission belt, and the method comprising:

[0007] When the hot compress and warm patch is transmitted on the transmission belt, the image sensor is used to acquire a patch original image;

[0008] A plurality of connected domains to be processed are screened out in the original image, and the connected domains are processed to obtain a restored image;

[0009] Pore domain segmentation is performed in the restored image to obtain a pore domain segmentation result;

[0010] According to the hole domain segmentation result, a hole index of the hot compress warm patch is calculated, a corresponding process compensation parameter is calculated combined with a preset index mapping standard, and a process parameter is controlled to be updated.

[0011] A plurality of to-be-processed connected domains are screened out in the original image, including:

[0012] Brightness detection and saturation detection are performed on the original image to obtain a first candidate mask, wherein a threshold of the brightness detection is set according to an upper limit of a gray scale of the image sensor, and a threshold of the saturation detection is set according to a statistical distribution of historical product samples.

[0013] Multi-directional fine line response analysis is performed on the first candidate mask to screen out an area continuously extending in a single direction, having a consistent edge orientation and being in an elongated form, and generate a second candidate mask.

[0014] A candidate set is initialized, and brightness time series analysis in a row scanning direction is performed on an area covered by the second candidate mask, and when a periodic strip structure is detected, a region corresponding to the strip structure is integrated into the candidate set.

[0015] The regions corresponding to the strip structures in the candidate set are subjected to connectedness clustering to obtain a plurality of to-be-processed connected domains.

[0016] The connected domains are processed to obtain a restored image, including:

[0017] For each to-be-processed connected domain, a main extension direction is calculated according to an elongated form and an edge orientation corresponding to the connected domain, and semi-morphological thinning is performed on the connected domain according to the main extension direction to obtain a target connected domain, wherein the semi-morphological thinning includes a semi-dilation matrix and an erosion matrix.

[0018] In the target connected domain, brightness saturation features, overexposure halo distribution and strip undulation features parallel to the row scanning direction of all pixel points are calculated to obtain a plurality of composite masks, wherein the composite masks are used to locate stripe-shaped highlight structures.

[0019] An iteration operation is performed to calculate a motion smear initial value of a current composite mask, to perform a smear removal process on the current composite mask according to the motion smear initial value, and to calculate a stripe intensity of the current composite mask after the smear removal process, and if the stripe intensity is greater than a preset prior condition, the current composite mask is updated according to the stripe intensity and replaced by the current composite mask, until the stripe intensity is less than or equal to the preset prior condition.

[0020] Each composite mask is processed one by one to obtain a local restoration result of a target connected domain, the local restoration results of the target connected domains are fused with the original image in a boundary smoothing manner to generate a restored image.

[0021] performing semi-morphological thinning on the connected domain according to the main extension direction, comprising:

[0022] determining a positive half-space according to the luminance gradient and energy symmetry ratio of the connected domain in the main extension direction;

[0023] constructing a first structure matrix as a semi-inflating matrix according to the main extension direction and the positive half-space, wherein the semi-inflating matrix takes the main extension direction as the long axis and takes the elements in the positive half-space as the one-way element set;

[0024] constructing a second structure matrix as an eroding matrix according to the width distribution of the connected domain in the direction perpendicular to the main extension direction, wherein the eroding matrix is an all-directional element set;

[0025] performing aggregation processing on the connected domain by using the semi-inflating matrix to obtain an intermediate connected domain, and performing shaping processing on the intermediate connected domain by using the eroding matrix to obtain a target connected domain.

[0026] the calculation of the motion smearing initial value of the current composite mask, comprising:

[0027] performing frequency domain transformation on the coverage area of the current composite mask, and counting the energy distribution and energy gap bandwidth of the power spectrum on the concentric ring band to obtain a frequency domain index, wherein the frequency domain index is used to represent the frequency band compression degree corresponding to the smearing;

[0028] performing line scanning on the luminance distribution of the coverage area of the current composite mask, and extracting the luminance fluctuation curve along the scanning direction;

[0029] detecting the periodic strip features in the luminance fluctuation curve, calculating the average interval of adjacent peak values and the peak amplitude difference to obtain a spatial domain index;

[0030] performing weighted fusion on the frequency domain index and the spatial domain index to obtain the motion smearing initial value.

[0031] the de-smearing processing of the current composite mask according to the motion smearing initial value, comprising:

[0032] extracting the skeleton in the coverage area of the current composite mask along the strip direction, and dividing the strip into a plurality of micro-segments according to the direction of the skeleton, wherein the strip is a stripe-shaped highlight structure in the coverage area of the current composite mask;

[0033] determining the processing parameters of each micro-segment according to the motion smearing initial value, wherein the processing parameters include effective length, processing intensity and start and end positions;

[0034] According to the processing parameter, a directional microkernel corresponding to the microsegment is calculated, the microsegment is processed according to the directional microkernel, and a restored segment is obtained;

[0035] The restored segment of the microsegment is spliced and energy-consistent along the skeleton direction to obtain a candidate restored segment;

[0036] The candidate restored segment is subjected to topological conservation checking, and topological indexes at least include the number of holes, the connectivity, the roundness distribution and the ring structure count; when any topological index does not satisfy a preset interval, the processing intensity is updated and spliced again according to the deviation between the topological index not satisfying the preset interval and the candidate restored segment until the topological index satisfies the preset interval, and the current composite mask after the ghosting processing is output.

[0037] According to the stripe intensity, the current composite mask is updated and replaced by the current composite mask, including:

[0038] The difference between the stripe intensity and the prior condition is calculated;

[0039] The mask parameters of the current composite mask are adjusted according to the difference, and the adjustment at least includes one of shortening the effective length, reducing the processing intensity and shrinking the coverage range;

[0040] The current composite mask is updated according to the adjusted mask parameters and replaced by the current composite mask.

[0041] The hole domain segmentation is performed in the restored image, including:

[0042] The threshold segmentation and edge detection are performed on the restored image, and the closed region is extracted as the hole domain;

[0043] The roundness screening is performed on the hole domain to obtain the hole domain segmentation result.

[0044] The process compensation parameters at least include one of the ratio of iron powder to activated carbon in the heating reaction formula, the amount of water regulator, the uniformity parameter of raw material stirring, the thickness distribution of the coating layer and the air permeability of the packaging film.

[0045] A multi-process parameter collaborative control system of a hot compress and warm patch, applied to a finished product detection device of a hot compress and warm patch, the finished product detection device includes an image sensor arranged directly above a transmission belt, and the system includes:

[0046] An image acquisition unit, configured to acquire an original image of the hot compress and warm patch during the transmission belt conveying process;

[0047] A feature analysis unit, configured to perform brightness, saturation and fine line response analysis on the original image, screen a to-be-processed connected domain and generate a composite mask;

[0048] The defect repairing unit performs de-smearing processing and topological conservation checking on the composite mask based on the motion smearing initial value, to generate a recovered image;

[0049] The parameter calculation unit performs hole domain segmentation in the recovered image, calculates a pore index, and outputs corresponding process compensation parameters to control the update of process parameters.

[0050] Compared with the prior art, the beneficial effects of the present application are:

[0051] The present application introduces a connected domain processing and recovery processing logic for pseudo-hole interference in the finished product detection stage, so that the hole domain segmentation result can accurately reflect the real pore distribution of the hot compress and warm patch, and avoid the problem of virtual high porosity caused by glare and smearing. At the same time, combined with the pore index and the preset index mapping standard, the detection result is directly converted into process compensation parameters to realize dynamic control in the production link. BRIEF DESCRIPTION OF DRAWINGS

[0052] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0053] Figure 1 An exemplary application scenario provided for the embodiments of the present application is shown in the following figure:

[0054] Figure 2 A module schematic diagram of the finished product detection device provided for the embodiments of the present application is shown in the following figure:

[0055] Figure 3 A problem principle schematic diagram provided for the embodiments of the present application is shown in the following figure:

[0056] Figure 4 A flowchart of the hot compress and warm patch multi-process parameter collaborative control method provided for the embodiments of the present application is shown in the following figure:

[0057] Figure 5 A flowchart of the connected domain screening provided for the embodiments of the present application is shown in the following figure:

[0058] Figure 6 An inflation comparison schematic diagram provided for the embodiments of the present application is shown in the following figure.

[0059] Reference signs: 100, transmission belt; 101, hot compress and warm patch; 102, finished product detection device; 1022, processor; 1021, image sensor. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments.

[0061] Reference herein to an "embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. As will be apparent to those of ordinary skill in the art, embodiments described herein can be combined with other embodiments.

[0062] The hot compress and heating patch in the present application specifically refers to a patch-shaped product that provides a sustained warm effect for the human body through exothermic reaction or physical heat preservation means, which is usually composed of a multi-layer structure, including:

[0063] The heating layer contains iron powder, activated carbon, inorganic salt, and moisture-regulating agent, etc., which can release heat through exothermic reaction in the air;

[0064] The carrier layer, usually non-woven fabric or fiber substrate, is used to support the heating layer and maintain its stable form;

[0065] The control layer, including air-permeable film, back adhesive or functional coating layer, is used to adjust the oxygen permeation rate, control the heating duration and temperature peak;

[0066] The packaging layer, mostly polymer film or composite film, is used to isolate water vapor, fix the patch structure and achieve adhesion to the skin or clothing.

[0067] The hot compress and heating patch can be a common medical physiotherapy type heating patch, or a daily heating patch for warming or cold prevention during exercise, and the commonality is that it realizes a stable and controllable exothermic process through a patch-shaped laminated structure, and relies on process parameters such as porosity, air permeability and coating uniformity to determine its thermal performance.

[0068] It can be understood that the process parameter optimization control in the present application does not optimize all processes of the hot compress and heating patch, but focuses on the key processes that directly affect the pore structure of the finished product. The present application can realize dynamic compensation and iterative optimization of local key parameters while keeping the overall production rhythm unchanged, thereby ensuring the consistency of the heat release rate and the temperature distribution in the plane of the heating patch during use.

[0069] In a mass production heating patch factory, the porosity-related qualification test is usually carried out in a stable laboratory or offline:

[0070] After the sample is cut, the oxygen permeability or gas permeability is tested, combined with the conventional measurement of moisture content and water activity, and if necessary, destructive pore size characterization is used as a sampling review.

[0071] Such a procedure is reliable in static conditions, but slow-paced and narrow in coverage, and often only gives the average value of the whole piece or the representative value of the local section, lacking intuitive mapping of the in-plane differences between the edge seal and the center, the printing area and the blank area. More importantly, it is difficult to rely on these offline data to achieve timely closed-loop of the process fine-tuning of the production line.

[0072] In the example technology, more and more production lines attempt to move the visible light / near-infrared camera forward to the finished product section, hoping to use non-contact, full field of view, and beat-by-beat imaging to complement the detection information. With the easily established index mapping model, stable aperture index can be output on each product, and then drive the process compensation.

[0073] It can be understood that the difficulty of detection by image lies in the fact that the surface state of the patch has two distinct features:

[0074] First, the transparent or semi-transparent barrier film, the back adhesive and the decorative coating superimposed into a high-gloss interface, the mirror surface reflection is obvious.

[0075] Second, the patch passes through the overlying camera at high speed with the transmission belt, and the imaging window is limited to a single external view angle and a rolling scanning timing.

[0076] Any visual assessment based on texture and pore group has a congenital condition. In the real pore domain, mirror surface reflection will generate elongated highlight strips and halos. Rolling collection will stretch these highlights into band-like undulations in the row direction. Common threshold or edge segmentation will treat these strips as pores, thereby statistically magnifying the porosity and misleading subsequent process correction.

[0077] The embodiment is aimed at the imaging scene of superimposing real pore texture and mirror surface strips, and does not require the patch to be detected to have a pre-defined area template or fixed configuration partition. Whether it is a high-barrier film for long-time preservation, a strong back adhesive optimized for sports scenes, or a durable version using thickened heat sealing in the edge sealing area, as long as it meets the typical mirror surface reflection path constraints and view angle shielding restrictions, the embodiment maintains the same processing strategy and judgment caliber.

[0078] Without changing the hardware, the embodiment regards the abnormal bright spots as strip-shaped external components that are jointly affected by the collection timing and surface reflection geometry, and regards the pore group related to the patch material-process as a structurally stable component.

[0079] After imaging, first integrate the three types of signals of brightness saturation, halo expansion and row band undulation in a single frame to generate a composite mask for positioning the strips; then perform directional local repair on each elongated suspected area, and maintain the number of pores, connectivity and circularity distribution from changing unexplainably through topological conservation constraints.

[0080] The restored image thus obtained does not pursue the reconstruction of a new image, but eliminates the misleading of segmentation and statistics by the strip with minimal intervention, and retains the geometric details of the real pore edge. The final pore segmentation and in-plane pore index thus obtained have stable statistical consistency, which can be directly used to drive the joint correction of coating edge compensation, micro-pore density gradient adjustment and sealing energy distribution, to form a closed loop of the production section.

[0081] Reference Figure 1 , Figure 1 An exemplary application scenario provided by an embodiment of the present application is shown in the following figure.

[0082] Figure 1 The application scenario includes a transmission belt 100, a hot compress patch 101, and a finished product detection device 102. The arrow in the figure indicates the conveying direction of the transmission belt 100, wherein:

[0083] The transmission belt 100 is used to carry and convey the hot compress patch to be detected;

[0084] The hot compress patch 101 is the detection object, which passes through the detection area during the transmission process;

[0085] The finished product detection device 102 is arranged above the transmission belt 100, and is used to collect the surface image of the hot compress patch 101 during the transmission process, so as to subsequently extract the pore index and optimize the process parameters.

[0086] It can be understood that the arrangement position and arrangement manner of the finished product detection device can be flexibly adjusted according to the space structure of the production line, the direction of the transmission belt, and the lighting conditions, for example, it can be arranged above, beside, or at a certain angle with the transmission direction of the transmission belt, to adapt to different detection needs. The present application does not limit this.

[0087] In some optional specific embodiments, the finished product detection device is also used to process the surface image, and output the porosity analysis result, referring to Figure 2 , Figure 2 A module schematic diagram of the finished product detection device provided by an embodiment of the present application is shown in the following figure.

[0088] Figure 2 The finished product detection device 102 includes an image sensor 1021 and a processor 1022, wherein:

[0089] The image sensor 1021 is used to collect image data of the surface of the hot compress patch in real time during the movement of the hot compress patch along with the transmission belt 100;

[0090] The processor 1022 is electrically connected with the image sensor 1021, and is used to pre-process the collected image data, filter connected domains, generate a restored image, and perform pore segmentation, and calculate a pore index based on the pore segmentation result, so as to obtain a compensation parameter used for process adjustment.

[0091] It can be understood that the processor 1022 can be an independent industrial control unit or integrated in an image sensor module, which is not limited herein.

[0092] Reference Figure 3 , Figure 3 The problem principle schematic diagram provided by the embodiment of the application is shown.

[0093] Figure 3 It is shown that there are regularly distributed real pores on the surface of the hot compress and warming patch. In the finished product detection process, due to the mirror reflection characteristics of the surface film or back adhesive, when the imaging light forms a high reflection angle with the film layer, local areas will appear overexposure or saturation, thereby producing a strip-shaped bright spot area in the image. These bright spots are misjudged as pores after high threshold segmentation, forming false pores, resulting in systematic overestimation of the porosity calculation result.

[0094] It can be understood that although the pores of the hot compress and warming patch are usually regularly arrayed in design, they cannot be simply detected by presetting coordinates and point-by-point comparison in actual production process:

[0095] Firstly, due to process disturbances such as coating thickness fluctuation, uneven drying speed or edge force difference, the position, shape and gray scale of the pores will appear slight deviation or irregular change.

[0096] Secondly, the strong reflection characteristics of the film and back adhesive layer will introduce strip-shaped artifacts in the image, which will cover or distort part of the real pores, so that the template comparison cannot distinguish the real pores covered by optical interference from the false pores formed by reflection.

[0097] Thirdly, the industrial standard more stipulates the statistical range of porosity and distribution uniformity, rather than the fixed geometric position one by one, so the detection target is to extract the global reliable pore structure and pore index to support the subsequent process compensation, rather than only judging whether a coordinate point exists pores.

[0098] It should be noted that in the embodiment not shown in the figure, due to the high-speed operation of the production line, the real pores may also appear elongated due to motion blur in the shooting process, thereby further affecting the confusion of the false pores and the real pores in shape.

[0099] Next, the hot compress and warming patch multi-process parameter collaborative control method provided by the embodiment of the application is further introduced in combination with the drawings, Figure 4 The method shown is applied to a finished product detection device of a hot compress and warming patch, the finished product detection device comprising an image sensor arranged directly above a transmission belt, and the method comprising:

[0100] S1: acquiring a raw image of the patch by the image sensor when the hot compress patch is conveyed on the transmission belt;

[0101] In the embodiment, the raw image not only contains the real texture of the aperture region, but also may superimpose strip-shaped bright spots caused by film reflection and high-speed movement. By directly capturing such images at the imaging end, on the one hand, full-field detection data of each product can be ensured, and on the other hand, the lag of relying on sampling inspection is avoided, so that the subsequent processing has a data basis.

[0102] As can be appreciated by those skilled in the art, the type of image sensor can be a surface array CCD, a linear array CMOS, or an industrial camera with a near-infrared channel, as long as it can obtain sufficient resolution of the aperture texture at the production rhythm, and the present application does not make more limitations.

[0103] S2: screening a plurality of connected domains to be processed in the raw image;

[0104] S3: processing the connected domains to obtain a restored image;

[0105] In the embodiment, the screening of the connected domains not only depends on the brightness and saturation characteristics, but also combines multi-directional fine line response and time sequence scanning results, so that the strip region caused by the superposition of glare and motion blur can be effectively located. Then, in these regions, a composite morphological method combining semi-inflation and erosion matrix is used to repair the local image, so that the strip region is restored to an approximate real aperture background, which does not change the statistical characteristics of the overall image, and can maximize the preservation of real aperture edge details, thereby avoiding the interference of false bright bands on the subsequent aperture segmentation.

[0106] S4: performing aperture segmentation in the restored image to obtain an aperture segmentation result;

[0107] In the embodiment, the segmentation adopts a strategy combining threshold value and edge detection, which quickly identifies the closed region, and then performs roundness and connectivity screening on the closed region to eliminate false regions that do not meet the aperture characteristics.

[0108] S5: calculating the aperture index of the hot compress patch according to the aperture segmentation result, combining a preset index mapping standard, calculating the corresponding process compensation parameter, and controlling the process parameter to update;

[0109] In the embodiment, the aperture index can include aperture rate, distribution uniformity, edge aperture loss rate and other statistical quantities; the index mapping standard is established by comparing the laboratory oxygen permeability test with the imaging data in advance, which can convert the detection result into an actual process adjustment amount, such as coating thickness correction, drying air speed curve fine tuning, sealing energy distribution adjustment, etc.

[0110] The skilled in the art can understand that the specific application mode of the compensation parameter can be flexibly selected according to the configuration of the production line, as long as the dynamic correction of the key process variable in the subsequent process is ensured, and the present application does not make more limitations.

[0111] Before the specific technical content corresponding to the unfolding step, in order to facilitate the understanding of the method logic of the present application, the embodiments of the present application need to be emphasized again.

[0112] The finished product surface presents strong mirror reflection under the joint action of the film and the back glue, and the external imaging is limited to a single viewing angle and line scanning timing, so that a highlight band extending along the line or obliquely appears on the image. Unlike the isotropic dark structure of the real hole domain, such highlight band has stable directionality and energy bunching characteristics in space, and is often accompanied by a halo expanding on one side. The present embodiment does not simply rely on bright or dark threshold, but uses the directionality as independent evidence:

[0113] First, observe whether the energy is gathered in a certain direction and evolves in a band in a single frame, and then decide whether to enter the subsequent restoration.

[0114] The purpose of this is to statistically separate the extraneous form caused by optics and timing from the hole domain form generated by materials and processes, so that subsequent segmentation and index calculation are based on interpretable structural differences; without increasing the viewing angle or modifying the illumination, a high recognition rate of strip pseudo-hole can still be obtained.

[0115] The positioning of the region to be processed follows the order of first suppressing false and then protecting true. First, the continuous highlight domain close to saturation in the original image is weakly constrained and screened, and then the periodic fluctuation of line scanning brightness and the consistency of edge direction are introduced for secondary discrimination, and only when two or more types of evidence are established at the same time, the candidate mask is generated.

[0116] In some optional specific embodiments, in order to avoid false touch of the real hole edge, a guard band can be set on the periphery of the mask, and the positive half space is determined according to the energy symmetry of the candidate domain.

[0117] Further, directional aggregation is performed in the positive half space to fill the strip breakage and halo gap, and then the excess width and lateral overflow are shaped and withdrawn uniformly in all directions, so as to obtain a geometrically narrow and coherent target connected domain, with the smallest geometric change to obtain the most clear strip profile.

[0118] The skilled in the art can understand that the positive half space is determined by the combined index of energy symmetry ratio and brightness gradient, and the size and shape of the morphological matrix can be set in combination with the lower limit of aperture and the desired resolution.

[0119] In the present application, the specific determination condition of the elongated morphology can be understood as that in the candidate connected domain, the ratio of the major axis to the minor axis is greater than a preset length threshold, and the continuity of the major axis direction of the region is good, without large bending or local expansion, and the boundary curvature of the region remains smooth transition in the main direction, so that it can be classified as a connected domain of elongated morphology.

[0120] The specific determination condition of the stripe highlight can be understood as that the pixel brightness in the connected domain is greater than a preset brightness threshold, and presents a band-shaped extension in a certain direction, the band width is maintained within a limited range and the edge trend is highly consistent, the internal brightness distribution shows the characteristic morphology of high center and gradually decaying edge, and a periodic or quasi-periodic brightness fluctuation curve can be detected in the row scanning direction.

[0121] Next, the technical content of the connected domain screening method of the present application embodiment is further expanded.

[0122] Reference Figure 5 , Figure 5 The flowchart of the connected domain screening provided by the present application embodiment is shown.

[0123] In one example, a plurality of connected domains to be processed are screened out in the original image, comprising:

[0124] S2.1: performing brightness detection and saturation detection on the original image to obtain a first candidate mask, wherein the threshold of the brightness detection is set according to the upper limit of the gray scale of the image sensor, and the threshold of the saturation detection is set according to the statistical distribution of historical product samples;

[0125] Specifically, in order to avoid the interference of large-area background information in the overall image, in the first step, the potential highlight region and high saturation region are first distinguished through the joint determination of brightness and saturation. The highlight region often corresponds to a local overexposure or a specular reflection pseudo-hole, while the high saturation region is prone to appear at the position of local reflection of the film layer or the bonding surface. If these regions are not removed in advance, they will directly enter the subsequent hole domain segmentation link, resulting in misjudgment. The setting of the brightness threshold is related to the upper limit of the gray scale of the sensor, so that the threshold can match the dynamic range of the device itself, avoiding missing detection due to too high threshold, or causing large area redundancy due to too low threshold. The saturation threshold depends on the statistical distribution of historical samples, and the characteristics of typical reflection points in the samples can be used to adaptively extract the strong saturation region prone to artifacts in the current batch, so as to ensure that the coverage range of the first candidate mask is more accurate.

[0126] In the embodiment, the first candidate mask is generated by converting the image into a lightness-saturation space, detecting the gray scale value and saturation value of each pixel, and then synthesizing the mask through logical OR relationship, that is, as long as any one of the high light or high saturation conditions is met, the pixel will be marked as a candidate region.

[0127] It can be understood that the first candidate mask often presents as scattered points or blocks in spatial distribution, covering the potential glare false hole positions. The first candidate mask provides an input basis for subsequent fine line direction analysis, which is equivalent to preliminarily extracting the region that may contain artifacts in the entire image, thereby reducing the computational complexity.

[0128] S2.2: performing multi-directional fine line response analysis in the first candidate mask, screening out regions that are continuously extended in a single direction, have consistent edge orientation and are in an elongated form, and generating a second candidate mask;

[0129] Specifically, the most prominent form of the false hole is that the bright spot is elongated into a strip under the action of motion blur, so it is necessary to further identify whether there is an elongated region with consistent direction in the first candidate mask. Through multi-directional fine line response analysis, it can be detected whether the candidate region has directional texture, that is, it is continuously extended in a certain direction and has weak response in other directions. Such a region is most likely to be a ghost false hole, rather than a real hole domain with isotropic distribution.

[0130] In the embodiment, the fine line response analysis uses a directional gradient operator to detect the candidate mask region. Specifically, the gradient cumulative value is calculated in multiple directions, and when the cumulative value in a certain direction is significantly higher than that in other directions, it indicates that the region has consistent direction. Then, a further judgment is made in combination with the aspect ratio of the region. Only when the region presents an elongated form and the edge orientation is consistent, it will be marked as the second candidate mask. The second candidate mask selected in this way has actually distinguished the original high-light noise from the real hole domain, and only the candidate strip region that is most likely to be a ghost false hole is retained.

[0131] S2.3: initializing a candidate set, performing lightness time sequence analysis in the row scanning direction on the region covered by the second candidate mask, and when a periodic strip structure is detected, the region corresponding to the strip structure is incorporated into the candidate set;

[0132] Specifically, the distribution of the real hole domain is often regular, while the strip false hole forms a periodic light-dark alternation in the row scanning direction. Therefore, on the basis of the second candidate mask, it is necessary to combine the lightness time sequence analysis in the scanning direction to determine whether there is a periodic strip structure.

[0133] In the embodiment, the initialization of the candidate set is to record all the detected periodic strip regions and establish the corresponding index.

[0134] It can be understood that the strip pseudo-hole can cross multiple candidate mask regions, and only through the set processing, it can be uniformly marked as a pseudo-hole candidate. In the timing analysis process, the periodicity of brightness is detected through Fourier transform or autocorrelation function, so as to obtain the periodicity characteristics. When the obvious periodicity strip is detected, the corresponding region is added to the candidate set, so that the pseudo-hole is not missed due to the region segmentation.

[0135] S2.4: performing connectivity clustering on the region corresponding to the strip structure in the candidate set to obtain a plurality of to-be-processed connected domains;

[0136] In the embodiment, the clustering method adopts the connectivity determination of four-neighborhood or eight-neighborhood, and combines the direction vector consistency for secondary screening. When the direction deviation of two regions is within a set threshold, and the spatial distance does not exceed a set range, the two regions can be merged into the same connected domain. In this way, the isolated bright spots can be prevented from being mistakenly entered into the clustering, and the integrity of the strip pseudo-hole can be ensured.

[0137] Next, the technical content of the method of the application on the connected domain processing is further expanded.

[0138] It can be understood that the connected domain processing aims to the strip type foreign form introduced by the superposition of glare and high-speed imaging, and under the premise of not changing the real hole rim geometric information, the suspected region is directionally repaired and constrainedly withdrawn, so that it is restored to a stable background which can be used for subsequent hole domain segmentation determination. The processing logic of the application does not take the hole-by-hole template alignment as the premise, but completes the local correction and global consistency checking according to the directionality, elongation and neighborhood energy distribution of the candidate connected domain.

[0139] In one example, the specific steps of S3 are as follows:

[0140] S3.1: For each to-be-processed connected domain, a main extension direction is calculated according to the elongated form and the edge direction corresponding to the connected domain, and a semi-morphological thinning is performed on the connected domain according to the main extension direction to obtain a target connected domain, wherein the semi-morphological thinning includes a semi-dilation matrix and an erosion matrix;

[0141] Specifically, in order to make the de-smearing occur only in the actual direction of strip formation and expansion, it is necessary to first robustly estimate the main extension direction of the to-be-processed connected domain. The direction is not obtained from a single feature, but is given by three types of evidence: the elongated form, the edge direction consistency and the row-wise brightness fluctuation. The elongated form describes the main axis orientation of the connected domain, the edge direction reflects the collective direction of the local boundary, and the row-wise brightness fluctuation reveals the strip evolution trend under rolling acquisition. When the three are consistent in space, the main extension direction has interpretability and stability, which facilitates subsequent processing only on the trailing side of the strip, and avoids touching the real hole rim side.

[0142] In the embodiment, the main extension direction is first obtained as an initial value through the long axis direction of the minimum circumscribed rectangle, and then a plurality of edge micro-sections are extracted at a fixed step length in the connected domain, the tangential direction distribution thereof is counted, and the initial value is offset and corrected according to the tangential direction distribution, and then the phase extension direction of the row direction brightness fluctuation is combined for final correction.

[0143] It can be understood that the morphological processing of the present application is used for local constraint repair of the strip-shaped pseudo-hole generated by superposition of glare and motion blur, so that it is suppressed or shrunk in the restoration process. Due to the particularity of the hot compress warm patch in the process scene, that is, the apertures of the hot compress warm patch in the process scene are mostly regularly distributed circular or quasi-circular hole positions, the boundary is stable and the directionality is weak, and the pseudo-hole formed by the glare trail has obvious direction extension. If omnidirectional dilation operation is used, it will be uniformly expanded around the aperture, which is easy to cover or even erode the real hole rim, resulting in error in the aperture calculation result. Therefore, it is necessary to perform half-space constraint on the dilation operation, that is, only allow dilation in the direction of the trail extension, and keep the hole rim unchanged on the other side.

[0144] Reference Figure 6 , Figure 6 The expansion contrast schematic diagram provided by the embodiment of the present application is shown.

[0145] Figure 6 The morphological changes of the connected domain in different cases are shown, wherein:

[0146] The left graph is the original connected domain, which shows the regular circular aperture and the pseudo-hole strip structure extended below it; the middle graph is the full dilation connected domain, and it can be seen that the dilation is uniformly expanded around the aperture, resulting in that the real hole rim is covered and the aperture boundary is distorted; the right graph is the half-dilation connected domain, which expands only in the direction of the strip tail while keeping the real hole rim unchanged, so that the pseudo-hole and the real aperture are effectively distinguished, providing a basis for subsequent erosion and correction operations.

[0147] In one example, the connected domain is half-morphologically refined according to the main extension direction, comprising:

[0148] S3.1.1: determining a positive half-space according to the brightness gradient and energy symmetry ratio of the connected domain in the main extension direction;

[0149] Specifically, the strip false hole of the connected domain has the characteristics of one-sided trailing and energy expansion, and there is a directional difference with the isotropic dark structure of the real hole edge. In order to limit the subsequent processing to the trailing side, it is necessary to first determine which side is the positive half space in the main extension direction. The basis for the determination comes from the brightness gradient sign distribution along the normal of the main axis and the bias of the energy symmetry ratio: one side of the strip usually shows a gradual change from high brightness to medium brightness, and the other side tends to be dark or dark area of the real hole edge, which forms an asymmetric energy distribution when sampling at equal distance.

[0150] In this embodiment, with the main extension direction as the reference, a plurality of normal sampling lines are constructed on both sides of the connected domain skeleton, and the brightness mean sequence from the skeleton to the equidistant position on both sides and the gradient direction consistency are counted line by line. The energy accumulation difference of both sides at the same distance is compared and a majority vote is made, and the side with the majority vote result is determined as the positive half space; if there are local hole edge neighbors, the voting rule also includes the processing logic of giving priority to avoiding the hole edge:

[0151] When the hole edge marker is detected, the side vote is halved to ensure that the half space does not point to the real hole edge. The positive half space obtained in this way is saved in the form of a pixel-level mask, and a guard band is generated outside its boundary. The guard band width is adaptively calculated according to the minimum distance from the suspected hole edge, the boundary gradient amplitude and the historical error tolerance.

[0152] S3.1.2: Construct a first structure matrix as a semi-inflating matrix according to the main extension direction and the positive half space, wherein the semi-inflating matrix has a long axis in the main extension direction and a single-element set in the positive half space;

[0153] Specifically, the semi-inflating matrix is used to perform one-way aggregation and connection of the strip breakage in the positive half space along the main extension direction, so that the false hole presents a continuous and retractable shape, while the real hole edge on the other side is not changed. The geometric properties of the matrix need to match the long and narrow and clear direction of the strip itself:

[0154] The long axis is aligned with the main direction, the short axis is limited to the actual width of the strip, and only takes values in the positive half space, so as to ensure the one-sidedness of the processing. The length is set according to the connected gap statistics and the continuous length of the main ridge of the segment, so that the one-time aggregation can cross the typical breakpoint without excessive extension; the width takes the median and the safe interval between the lower quantile of the strip width distribution, avoiding the halo boundary together; the anchor point is located on the skeleton center line, only the pixels in the positive half space are reserved in the matrix, and the other half is empty, thereby forming a single-element set.

[0155] S3.1.3: Construct a second structure matrix as an erosion matrix according to the width distribution of the connected domain in the vertical direction of the main extension direction, wherein the erosion matrix is an all-directional element set;

[0156] Specifically, after one-way aggregation, the strip may produce slight overflow in the transverse direction, which needs to be shaped in width with isotropic retreat operation while keeping the continuity of the strip unbroken. The erosion matrix is therefore set as a set of isotropic elements, the scale of which should be smaller than the common safety interval of the real aperture and the strip width to retreat the aggregation redundancy without eroding the aperture edge.

[0157] In this embodiment, the strip width distribution is first counted in the vertical direction of the main direction of the target connected domain, and the median and lower quantile are selected to form a shaping radius candidate, and then the smaller value of the fixed safety proportion of the lower limit of the aperture is taken as the matrix radius; when the strip width gradually changes along the section, the matrix radius slowly changes with the position to ensure smooth transition of the shape.

[0158] S3.1.4: performing aggregation processing on the connected domain by using the semi-inflating matrix to obtain an intermediate connected domain, and performing shaping processing on the intermediate connected domain by using the erosion matrix to obtain a target connected domain;

[0159] S3.2: In the target connected domain, the brightness saturation feature, the overexposure halo distribution, and the strip-shaped undulation feature parallel to the row scanning direction of all pixel points are calculated to obtain a plurality of composite masks, wherein the composite mask is used to locate the stripe-shaped highlight structure.

[0160] In this application, the stripe-shaped highlight structure specifically refers to a strip formed by superimposing glare reflection and motion trail in the connected domain to be processed, which shows that the brightness is close to the saturation value, there is a strong halo effect at the edge, and there is a periodic or quasi-periodic light and dark undulation along the row scanning direction. The difference between this kind of structure and the real pore is that the real pore generally shows a low-brightness closed dark area, the edge profile is stable and approximately circular or quasi-circular, while the stripe-shaped highlight structure lacks a closed boundary and shows stretching or trailing characteristics in the row scanning direction, which is easy to be misidentified as a pore when segmented at a high threshold.

[0161] Specifically, the strip-shaped highlight structure is not a single representation, so it is necessary to extract and fuse the pixel-level evidence in the target connected domain. The brightness saturation feature is used to locate the overexposure core area, the halo distribution describes the ring belt of the overexposure energy decaying outward, and the row strip-shaped undulation reveals the periodic light and dark alternation caused by rolling acquisition in the row direction.

[0162] In this embodiment, the composite mask is generated layer by layer according to the principle of core priority and boundary conservation.

[0163] The first layer mask mainly uses brightness and saturation to mark the continuous area close to the upper limit of the gray scale and the high gradient narrow band adjacent thereto.

[0164] The second layer mask is based on the radial or normal brightness decay to mark the halo outer edge and its gradual change area, allowing low-intensity processing.

[0165] The third layer mask is confirmed by the peak-to-valley stability and peak spacing consistency of the row-wise brightness curve, and a periodic undulating strip paragraph is marked in the row direction of the target connected domain.

[0166] It can be understood that the three-layer mask is superimposed on the pixel coordinates to form a processing parameter map: the core layer pixels are assigned a higher processing intensity and a longer effective length, the ring belt layer pixels are assigned a low intensity and a short effective length, and the row direction layer pixels are used to adjust the start and end positions of the processing and the row-by-row intensity curve.

[0167] S3.3: Perform an iterative operation to calculate the motion smearing initial value of the current composite mask, de-smear the current composite mask according to the motion smearing initial value, and calculate the stripe intensity of the current composite mask after de-smearing processing. If the stripe intensity is greater than the preset prior condition, update the current composite mask according to the stripe intensity and replace the current composite mask, until the stripe intensity is less than or equal to the preset prior condition;

[0168] Specifically, the de-smearing under the single-frame condition needs to gradually eliminate the band residue within a bounded range, and whether to continue is determined by the observable index. The stripe intensity evaluation is not determined by a single quantitative measure, but takes into account three types of quantitative measures: band energy, elongation, and row fluctuation amplitude, with the purpose of corresponding to the morphology, contrast, and time sequence traces of the band. The mask update is performed in the order of shortening the effective length, reducing the processing intensity, shrinking the coverage range, and removing low-confidence pixels, to avoid over-repair.

[0169] It should be noted that the motion smearing initial value in the present application can be understood as a quantitative estimate of the degree of blur of the stripe-shaped highlight structure caused by the production line motion in the imaging process. The quantitative estimate is based on the comprehensive analysis of multi-dimensional features such as frequency energy distribution, brightness fluctuation curve, and local structure extension, and is used to represent the amplitude and direction consistency of the spatial stretching of the smearing.

[0170] The stripe intensity can be understood as a measure of the saliency of the band structure in the composite mask. Its value not only reflects the amplitude of the brightness difference, but also comprehensively considers the continuity, contrast, and periodic stability of the band. When the stripe intensity is high, it usually means that the pseudo-hole band still exists, and its morphology is sufficient to interfere with the accuracy of the hole segmentation.

[0171] The prior condition is specifically a preset threshold value, which is used to limit the convergence target of the fringe intensity within a reasonable range. The prior condition can be obtained by statistical analysis of historical production samples, and can be set in combination with the dynamic range of the imaging device and the allowable error range of the porosity detection. The prior condition described in the present application is neither a single fixed value nor completely dependent on experience, but is obtained by statistical modeling of the differences between the pseudo-hole interference and the real pore boundary in different batches of products, so that it has both universality and adaptability to different process scenarios in engineering practice.

[0172] In one example, the calculation of the motion smearing initial value of the current composite mask includes: performing frequency domain transformation on the coverage area of the current composite mask, and statistically analyzing the energy distribution and energy gap bandwidth of the power spectrum on the concentric ring band to obtain a frequency domain index, wherein the frequency domain index is used to represent the frequency band compression degree corresponding to the smearing; performing line scanning on the brightness distribution of the coverage area of the current composite mask, and extracting the brightness fluctuation curve along the scanning direction; detecting the periodic strip features in the brightness fluctuation curve, and calculating the average interval and amplitude difference of adjacent peaks to obtain a spatial domain index; and weighting and fusing the frequency domain index and the spatial domain index to obtain the motion smearing initial value.

[0173] It can be understood that the coverage area refers to the local candidate fringe area corresponding to the current composite mask. Since the composite mask is a binary or semi-binary marking of the original image, the coverage area is not the entire image, but is limited to the local area circled by the mask. Frequency domain transformation is performed on the local area to avoid interference from global background noise, making the smearing features more prominent. The frequency domain transformation can use fast Fourier transform, discrete cosine transform or other time-frequency analysis methods commonly used in the art, and the present application does not limit the specific implementation.

[0174] Further, the power spectrum obtained after transformation is divided into several concentric ring bands in the radial direction, and the concentric ring bands are used to distinguish the energy distribution in different spatial frequency ranges. When smearing exists, the spectral energy is often concentrated in the low frequency or narrow band region, so that the energy distribution is unbalanced and the gap bandwidth is expanded in the power spectrum. This distribution feature can be used as a frequency domain index to represent the frequency band compression degree corresponding to the smearing.

[0175] Further, in the spatial domain processing, the brightness distribution in the coverage area is unfolded row by row along the row scanning direction of the image to form a one-dimensional brightness curve. Since smearing is often accompanied by periodic or quasi-periodic strip features, the one-dimensional brightness curve will show a fluctuation pattern with alternating peaks and valleys. By detecting the peaks and calculating the interval of adjacent peaks in the one-dimensional brightness curve, the average interval of adjacent peaks and the amplitude difference of adjacent peaks and valleys can be calculated. These features together serve as a spatial domain index, which can reflect the extension degree and period stability of the strip.

[0176] Further, the frequency domain index and the spatial domain index can be normalized in the same dimension and then fused, the fusion relationship can be any one of linear weight, segmented weight or adaptive weight, the weight value can be determined according to the batch calibration data and the target error tolerance, or can be given by a mapping table or a lightweight model trained based on historical samples; in order to suppress the instability caused by extreme values, truncation and robust scaling can be introduced before fusion, and monotonicity and boundary consistency constraints can be set after fusion to ensure that the ghosting intensity increases with the enhancement of the two types of indexes. The above weight and mapping form are all conventional choices in the field, and their implementation can refer to common signal fusion and index aggregation methods, which are not limited by the present application.

[0177] In one example, the de-ghosting processing of the current composite mask according to the motion ghosting initial value includes: extracting a skeleton in a strip direction of a coverage area of the current composite mask, and dividing the strip into a plurality of micro segments according to the direction of the skeleton, wherein the strip is a stripe type highlight structure in the coverage area of the current composite mask; determining a processing parameter of each micro segment according to the motion ghosting initial value, wherein the processing parameter includes an effective length, a processing intensity and a start and end position; calculating a directional micro kernel corresponding to the micro segment according to the processing parameter, processing the micro segment according to the directional micro kernel to obtain a restored segment; performing transition splicing and energy consistency of the restored segment of the micro segment in the skeleton direction to obtain a candidate restored segment; performing topological conservation checking on the candidate restored segment, and the topological index at least includes the number of holes, the connectivity, the roundness distribution and the ring structure count; when any topological index does not satisfy a preset interval, updating the processing intensity according to the deviation between the topological index that does not satisfy the preset interval and the candidate restored segment and re-splicing until the topological index satisfies the preset interval, and outputting the current composite mask after de-ghosting processing.

[0178] In the embodiment, the skeleton in the strip direction is obtained by extracting the geometric center line or the connected domain thinning result of the pixels in the region, and is used to depict the dominant direction of the stripe. The micro segment refers to a plurality of local segments divided on the skeleton according to a fixed length or an adaptive segmentation principle, and the local characteristics of the ghosting can be estimated for each micro segment. The motion ghosting initial value provides a reference for the processing parameters, which include the effective length, the processing intensity and the start and end position of the micro segment.

[0179] Further, the directional micro kernel is a local operator generated according to the skeleton direction and the parameter configuration, and the shape and directionality thereof can be dynamically adjusted according to the micro segment, which is commonly implemented by a directional convolution template or an adaptive filtering window.

[0180] It can be understood that the micro segment region is processed by the micro kernel, the strip ghosting effect can be weakened segment by segment, and the restored segment is generated.

[0181] Further, to avoid breaks or uneven brightness between local corrections, the restored fragments need to be transitioned along the skeleton direction and energy uniformity correction is applied to ensure overall brightness and texture continuity.

[0182] Further, after the candidate restored fragments are formed, they need to be verified through a topological conservation checking link. The topological conservation checking link takes the number of holes, connectivity, circularity distribution, and ring structure count as core indicators, and uses a method based on geometric feature matching and region connectivity analysis to determine whether the correction result destroys the structural characteristics of real pores. When a certain indicator deviates from the preset interval, the processing intensity is adjusted according to the degree of deviation, and the restored fragments are re-spliced until the topological characteristics return to a reasonable range, thereby ensuring that the final output of the de-smearing composite mask eliminates smearing interference and maintains the geometric properties of real pores.

[0183] In one example, updating and replacing the current composite mask according to the stripe intensity includes: calculating the difference between the stripe intensity and the prior condition; adjusting the mask parameters of the current composite mask according to the difference, wherein the adjustment at least includes one of shortening the effective length, reducing the processing intensity, and shrinking the coverage range; and updating and replacing the current composite mask according to the adjusted mask parameters.

[0184] S3.4: Process each composite mask one by one to obtain the local restoration result of the target connected domain, and perform boundary smoothing fusion of the local restoration result of each target connected domain and the original image to generate a restored image.

[0185] Specifically, after de-smearing is completed, the local restoration result of the target connected domain needs to be fused with the original image for consistency, so that the processing area and the non-processing area form a natural transition in geometry and brightness, avoiding new splicing marks or edge distortion. Fusion is not simply weighted superposition, but the transition band width is adaptively determined according to the boundary normal gradient. A narrow transition band is used to protect the edge where the gradient is large, and a wide transition band is used to eliminate brightness jump where the gradient is small. At the same time, the brightness statistics before and after processing need to be matched, so that the gray mean and variance of the local restoration result fall within the interval consistent with the surrounding background.

[0186] In the embodiment, the target connected domain outer edge is first calculated to generate a transition zone, and the transition zone is smoothly transitioned from 0 to 1 according to the weight of the normal coordinates; then the local restoration result is taken as an inner layer, the original image is taken as an outer layer, and the two are fused in the transition zone according to the weight, and the local restoration result is directly replaced in the non-transition zone. The change ratio and boundary displacement are monitored before and after the fusion, and when the change ratio exceeds the limited range or the boundary displacement exceeds the threshold, the intensity coefficient of the local restoration result is automatically reduced and fused again until the index returns to the interval or the fusion times reach the upper limit. After the fusion of all target connected domains is completed, a global brightness consistency correction is performed on the entire restored image, so that the local adjustment of different connected domains maintains the consistency of tone and contrast on the whole image scale, avoiding local repair oversaturation and local grayness.

[0187] In one example, a hole domain segmentation is performed in the restored image, including:

[0188] S4.1: Threshold segmentation and edge detection are performed on the restored image, and a closed region is extracted as a hole domain;

[0189] Specifically, although the restored image has removed the strip-shaped highlight interference, there may still be slow luminance drift, local contrast difference, and slight tone inconsistency between the restored area and the original background. To ensure that the hole domain boundary is dominated by the true geometry, a local adaptive threshold is needed to cooperate with gradient-based edge extraction to determine the closed region:

[0190] The threshold is used to quickly give a dark domain candidate, and the edge is used to limit the boundary trend of the candidate region and close small cracks. The two are superimposed to form a strong constraint closed contour to avoid adhesion or breakage caused by single threshold bias.

[0191] S4.2: Roundness screening is performed on the hole domain to obtain a hole domain segmentation result;

[0192] In the embodiment, a group of shape and stability indicators is calculated for each candidate closed region in the hole domain:

[0193] With the area centroid as a reference, the radial distance sequence of the contour to the centroid is measured along the equi-angle sampling, and the range and uniformity of the radial fluctuation are counted; the eccentricity is read as the eccentricity by fitting the contour to the best approximation ellipse; the inside-out adhesion of the boundary is reflected by the diameter difference between the circumscribed and inscribed circles of the region; and the boundary noise stability is evaluated by combining the centroid drift before and after multi-scale smoothing of the contour.

[0194] In one example, the process compensation parameters at least include one of the proportion of iron powder and activated carbon in the heating reaction formula, the amount of moisture regulator, the uniformity parameter of raw material stirring, the thickness distribution of the coating layer, and the air permeability of the packaging film.

[0195] It can be understood that after obtaining the pore domain segmentation result, the calculation of the pore index can refer to the conventional image metrology method, for example, a set of index sets capable of representing the diffusion performance of the finished product gas and the uniformity of the thermal reaction are obtained according to the segmented pore domain quantity, pore size distribution, porosity (the proportion of the total area of the pore domain to the effective area of the patch), pore domain morphological parameters (roundness, eccentricity, uniformity), etc. These indexes are used as pore indexes in the embodiment, not limited to a single value, but can be composed of multiple sub-indexes, which are used to reflect the overall quality state of the patch from different angles.

[0196] The index mapping standard can be understood as a corresponding relationship established based on batch experiments and process experience, for example, by calibrating the porosity, pore size distribution, and heat duration, peak temperature, and safety performance of the thermal patch during use through a large number of trial samples, a mapping rule between the input index and the process compensation parameter is formed. The mapping standard can be in a static mode such as a linear threshold table, an interval mapping table, or in a dynamic updating mode formed by statistical learning or model regression, which is not limited in the present application.

[0197] As can be appreciated by those skilled in the art, when the pore index and the index mapping standard are known, the corresponding process compensation parameter can be calculated by conventional means such as table lookup, interpolation or fitting calculation. For example, when the porosity is high, the air permeability of the coating layer can be directly reduced or the amount of moisture regulator can be increased; when the pore size distribution is biased towards large pores, the proportion of iron powder and activated carbon can be adjusted to change the heat release reaction rate; when the pore domain morphology is uneven, the stirring uniformity parameter can be improved to improve the material distribution. In this way, a closed-loop correction from the detection end to the production end can be realized, so that the process of the thermal patch is always in a stable and controllable state.

[0198] In one example, the embodiment of the present application provides a thermal patch multi-process parameter collaborative control system applied to a finished product detection device of a thermal patch, the finished product detection device comprising an image sensor arranged directly above a transmission belt, and the system comprising:

[0199] An image acquisition unit for acquiring an original image of the thermal patch during the transmission belt conveying process;

[0200] A feature analysis unit for performing brightness, saturation and fine line response analysis on the original image, screening connected domains to be processed and generating a composite mask;

[0201] A defect repair unit for performing de-smearing processing and topological conservation checking on the composite mask based on the motion smearing initial value, and generating a restored image;

[0202] A parameter calculation unit is configured to perform hole region segmentation in the recovered image, calculate a pore index, and output corresponding process compensation parameters to control the update of process parameters.

[0203] Although the embodiments of the present application have been shown and described above, it should be understood by those skilled in the art that the above embodiments are exemplary and cannot be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for coordinated control of multiple process parameters of a heat therapy patch, applied to a finished product testing device for heat therapy patches, characterized in that, The finished product inspection device includes an image sensor arranged directly above the transmission belt, and the method includes: When the heat therapy patch is transported on the conveyor belt, the image sensor acquires the original image of the patch. Multiple connected components to be processed are selected from the original image, and the connected components are processed to obtain the restored image; Aperture domain segmentation is performed on the restored image to obtain aperture domain segmentation results; The porosity index of the heat therapy patch is calculated based on the pore domain segmentation results. Combined with the preset index mapping standard, the corresponding process compensation parameters are calculated, and the process parameters are updated accordingly. Processing the connected components to obtain the restored image includes: For each connected region to be processed, the main extension direction is calculated based on the elongated shape and edge direction of the connected region. The connected region is then semi-morphologically refined according to the main extension direction to obtain the target connected region. The semi-morphological refinement includes a semi-dilation matrix and an erosion matrix. Within the target connected region, the brightness saturation features, overexposure halo distribution, and banded undulation features parallel to the line scanning direction of all pixels are calculated to obtain multiple composite masks, wherein the composite masks are used to locate striped high-brightness structures. Perform iterative operations, calculate the initial value of motion blur of the current composite mask, perform de-blurring on the current composite mask according to the initial value of motion blur, and calculate the fringe intensity of the current composite mask after de-blurring. If the fringe intensity is greater than a preset prior condition, update the current composite mask according to the fringe intensity and replace the current composite mask until the fringe intensity is less than or equal to the preset prior condition. Each composite mask is processed one by one to obtain the local restoration result of the target connected region. The local restoration result of each target connected region is then fused with the original image through boundary smoothing to generate the restored image.

2. The method for coordinated control of multiple process parameters of a heat therapy patch according to claim 1, characterized in that, Multiple connected components to be processed are selected from the original image, including: Brightness and saturation are detected on the original image to obtain a first candidate mask, wherein the threshold for brightness detection is set according to the upper limit of grayscale of the image sensor, and the threshold for saturation detection is set according to the statistical distribution of historical product samples. Multi-directional fine texture response analysis is performed on the first candidate mask to screen out regions that extend continuously in a single direction, have consistent edge direction, and are elongated in shape, thereby generating the second candidate mask. A candidate set is initialized, and the brightness timing analysis in the line scanning direction is performed on the area covered by the second candidate mask. When a periodic band structure is detected, the area corresponding to the band structure is incorporated into the candidate set. Connectivity clustering is performed on the regions corresponding to the band structures in the candidate set to obtain multiple connected domains to be processed.

3. The method for coordinated control of multiple process parameters of a heat therapy patch according to claim 1, characterized in that, Semi-morphological thinning of the connected components according to the main extension direction includes: The positive half-space is determined based on the brightness gradient and energy symmetry ratio of the connected domain in the main extension direction; Based on the main extension direction and the positive half space, a first structure matrix is ​​constructed as a semi-expansion matrix, wherein the semi-expansion matrix has the main extension direction as the major axis and the elements in the positive half space as a unidirectional element set. Based on the width distribution of the connected components in the direction perpendicular to the main extension direction, a second structure matrix is ​​constructed as the erosion matrix, wherein the erosion matrix is ​​an omnidirectional set of elements; The connected components are aggregated using the semi-dilation matrix to obtain intermediate connected components, and the intermediate connected components are shaped using the erosion matrix to obtain the target connected component.

4. The method for coordinated control of multiple process parameters of a heat therapy patch according to claim 1, characterized in that, The calculation of the initial value of the motion trail of the current composite mask includes: The frequency domain transformation is performed on the coverage area of ​​the current composite mask, and the power spectrum energy distribution and energy gap bandwidth on the concentric ring are statistically analyzed to obtain the frequency domain index, wherein the frequency domain index is used to characterize the frequency band compression degree corresponding to the trailing image. The brightness distribution of the current composite mask's coverage area is scanned in the row direction, and the brightness fluctuation curve along the scanning direction is extracted. Periodic strip features are detected in the brightness fluctuation curve, and the average interval and peak amplitude difference between adjacent peaks are calculated to obtain spatial indexes. The frequency domain index and the spatial domain index are weighted and fused to obtain the initial value of motion blur.

5. The method for coordinated control of multiple process parameters of a heat therapy patch according to claim 1, characterized in that, Based on the initial value of the motion blur, the current composite mask is subjected to motion blur removal processing, including: The skeleton is extracted along the strip direction in the coverage area of ​​the current composite mask, and the strip is divided into multiple micro segments according to the direction of the skeleton. The strip is a striped high-brightness structure in the coverage area of ​​the current composite mask, and the micro segments are determined by local segments divided into fixed lengths on the skeleton. The processing parameters for each micro-segment are determined based on the initial value of the motion blur, wherein the processing parameters include the effective length, processing intensity, and start and end positions; The directional microkernel corresponding to the micro-segment is calculated according to the processing parameters, and the micro-segment is processed according to the directional microkernel to obtain the restored fragment. The directional microkernel represents a local operator, and the local operator is calculated according to the skeleton direction and parameter configuration. The restored fragments of the micro-segments are spliced ​​together along the skeleton direction and their energy is made uniform to obtain candidate restored fragments; The candidate restored fragments are subjected to topological conservation verification. The topological indicators include at least the number of holes, connectivity, roundness distribution and ring structure count. When any topological indicator does not meet the preset interval, the processing intensity is updated and the fragments are re-stitched according to the deviation between the topological indicator that does not meet the preset interval and the candidate restored fragments, until the topological indicator meets the preset interval, and the current composite mask after the removal of ghosting is output.

6. The method for coordinated control of multiple process parameters of a heat therapy patch according to claim 1, characterized in that, Updating and replacing the current composite mask based on the stripe intensity includes: Calculate the difference between the fringe intensity and the prior condition; The mask parameters of the current composite mask are adjusted based on the difference, wherein the adjustment includes at least one of shortening the effective length, reducing the processing intensity, and shrinking the coverage area; The current composite mask is updated and replaced based on the adjusted mask parameters.

7. The method for coordinated control of multiple process parameters of a heat therapy patch according to claim 1, characterized in that, The process of segmenting the aperture region in the restored image includes: The restored image is subjected to threshold segmentation and edge detection, and the closed region is extracted as the aperture region; The pore regions are then subjected to roundness screening to obtain the pore region segmentation results.

8. The method for coordinated control of multiple process parameters of a heat therapy patch according to claim 1, characterized in that, The process compensation parameters include at least one of the following: the ratio of iron powder to activated carbon in the heating reaction formula, the amount of moisture regulator, the uniformity parameter of raw material stirring, the thickness distribution of the coating layer, and the air permeability of the encapsulation film.

9. A multi-process parameter collaborative control system for heat therapy patches, used to implement the multi-process parameter collaborative control method for heat therapy patches as described in any one of claims 1-8, applied to a finished product inspection device for heat therapy patches, the finished product inspection device comprising an image sensor arranged directly above a transmission belt, characterized in that, The system includes: The image acquisition unit is used to acquire raw images of the heat therapy patch during the conveyor belt transport process; The feature parsing unit is used to perform brightness, saturation and fine texture response analysis on the original image, filter the connected components to be processed and generate a composite mask; The defect repair unit performs de-moiring processing and topology conservation verification on the composite mask based on the initial value of motion trailing, and generates a restored image. The parameter calculation unit is used to segment the pore region in the restored image, calculate the porosity index and output the corresponding process compensation parameters, and control the process parameters to be updated.

Images