Intelligent visual recognition food material acceptance automatic checking system

Abstract

This invention discloses an automated verification system for food ingredient acceptance based on intelligent visual recognition, relating to the fields of food ingredient acceptance and image recognition technology, and addressing the front-end screening scenario of suspicious local anomalies near the sealing edge of pre-packaged easily separable liquid food trays. The system first locates the suspected anomaly area and determines the corresponding sealing edge segment in the image of the pre-packaged easily separable liquid food tray. A local sealing edge geometric benchmark is established around the corresponding sealing edge segment. On the same normal sampling line, inner source candidates on the inner side of the sealing edge and outer release candidates on the outer side of the sealing edge are identified first. Then, valid cross-sealing edge channels are verified from the middle sealing edge area, and the closed valid cross-sealing edge release lines are processed and bound together to form a cross-sealing edge anomaly band. The system further completes regional category sorting and piece-level acceptance result output based on the cross-sealing edge anomaly band and its inner and outer support states, thereby improving the stability of front-end sorting and the consistency of piece-level output.

Description

Technical Field

[0001] This invention relates to the field of food ingredient inspection and image recognition technology, and more specifically, to an automated verification system for food ingredient inspection based on intelligent visual recognition. Background Technology

[0002] Food ingredient inspection and image recognition technology are commonly used for the acceptance of incoming food ingredients. During the food ingredient inspection process, inspectors typically combine images of trays, label information, packaging condition, and overall appearance to perform basic checks on the arriving food ingredients, outputting item-level inspection results such as pass, review, or rejection. For routine food ingredients with intact packaging, stable appearance, and no local anomalies, existing inspection processes can usually complete the basic judgment. This type of technology is also commonly used for observing anomalies near the sealing edges of pre-packaged food ingredient trays. Current practices usually involve first identifying local anomalies near the sealing edges in the whole-box image, then cropping the local area and directly judging based on the overall appearance of that local area. For obvious objects, a preliminary conclusion is given; for less obvious objects, manual review or more complex subsequent recognition processes are initiated.

[0003] The existing technology has the following shortcomings: On the one hand, in pre-packaged food containers prone to leakage, phenomena such as food leakage, membrane reflection, localized darkening, blunted boundaries, and external adhesion can easily create a mixed appearance near the sealing edge. While existing methods can detect local anomalies near the sealing edge, the entry point is usually still the local anomaly block itself. There is a lack of a processing mechanism to establish a stable local spatial reference around the corresponding sealing edge segment most relevant to the anomaly, and there is also a lack of a unified reading framework that simultaneously constrains the inner side, sealing area, and outer side of the sealing edge under the same normal sampling path. Therefore, it is difficult to reliably distinguish obvious illusory objects, non-closed anomaly objects, objects to be verified, and truly high-risk objects at the front end.

[0004] On the other hand, even when abnormal appearances are observed on both the inner and outer sides of the sealing layer, existing practices typically lack a closure verification mechanism for the sealing area between the two. This makes it impossible to confirm whether the source of the abnormality on the inner side and the release phenomenon on the outer side are truly connected by the same cross-sealing path, easily leading to the conflation of reflections, localized adhesions, external residues, and actual leaks. Consequently, the front-end diversion caliber becomes unstable, obvious artifacts may still enter the review process, non-closed objects are difficult to effectively distinguish from truly high-risk objects, and regional-level identification results lack a stable, structured chain leading to component-level acceptance results.

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an automated verification system for food ingredient acceptance based on intelligent visual recognition, so as to solve the problems mentioned in the background art. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides the following technical solution: An automated food ingredient verification system based on intelligent visual recognition includes: The suspected anomaly location unit is used to locate suspected anomaly areas within the sealing edge neighborhood of the image of the pre-packaged easily separable liquid food tray, and to determine the corresponding sealing edge segment for each suspected anomaly area. The inner and outer observation zone construction unit is used to establish the local edge sealing center line and normal around the corresponding edge sealing segment, and generate the inner observation zone, the outer observation zone, and the normal sampling line for subsequent identification. The cross-edge anomaly zone identification unit is used to first read the inner edge coverage length, inner texture attenuation intensity, and inner edge continuity in the inner observation zone on the same normal sampling line to identify whether there are inner source candidates inside the normal sampling line; then read the outer extension length and outer edge divergence in the outer observation zone to identify whether there are outer release candidates outside the normal sampling line. When both inner source candidates and outer release candidates exist, the cross-border penetration width, cross-border grayscale continuity, and border boundary weakening degree in the border area between them are read to verify whether there is an effective cross-border channel between the inner source candidates and the outer release candidates. Only when an effective cross-border channel is established is it determined that the normal sampling line forms an effective cross-border release line. Adjacent normal sampling lines that both form effective cross-border release lines are merged along the local border centerline to form a cross-border anomaly zone, and the support status of the cross-border anomaly zone on the inner and outer sides of the border is recorded. The result determination unit is used to determine the area category of the suspected abnormal area based on the cross-sealing abnormal strip corresponding to the suspected abnormal area and its inner and outer support states, and output the acceptance result of the current tray according to the combination relationship of the area categories.

[0007] In a preferred embodiment, the suspected anomaly location unit extracts the amount of bright spot aggregation, dark wet infiltration, and edge relaxation within the edge sealing neighborhood; when any abnormal response amount reaches the corresponding abnormal condition, the corresponding area is marked as a candidate abnormal area; the candidate abnormal area is subjected to minimum connected area screening, maximum allowable distance screening, and minimum edge-fitting ratio screening to determine the suspected abnormal area.

[0008] In a preferred embodiment, the suspected anomaly location unit extracts the projection range of the suspected anomaly area along the sealing direction and extracts candidate sealing segments near the projection range; the candidate sealing segment with the smallest normal distance to the suspected anomaly area or the largest projection overlap ratio is determined as the corresponding sealing segment.

[0009] In a preferred embodiment, the inner and outer observation zone construction unit establishes a local edge sealing center line based on the boundary positions on both sides of the corresponding edge sealing section, and determines the normal direction based on the local direction of the local edge sealing center line; multiple sampling points are selected in sequence along the local edge sealing center line, and normal sampling lines are arranged along the normal direction at each sampling point; the normal lines are extended towards the inner side and the outer side of the edge sealing respectively along the normal direction at each position to form an inner observation zone and an outer observation zone.

[0010] In a preferred embodiment, the cross-edge anomaly identification unit reads the abnormal coverage segment, texture response changes, and continuous connection along the line within a local area where the current normal sampling line passes through the inside of the edge, so as to form the inner edge coverage length, inner texture attenuation intensity, and inner edge continuity. When the inner edge coverage length is not less than the minimum edge extension length, the inner texture attenuation intensity reaches a preset attenuation threshold, and the inner edge continuity meets the continuous connection condition, it is determined that there is an inner source candidate for the current normal sampling line.

[0011] In a preferred embodiment, the cross-sealing anomaly identification unit reads the outer anomaly coverage segment and its outer edge expansion shape within a local area where the current normal sampling line passes through the outer side of the sealing edge, so as to form the outer expansion length and the outer edge divergence; when the outer expansion length is not less than the minimum outer expansion length and the outer edge divergence meets the divergence judgment condition, it is determined that there is an outer release candidate for the current normal sampling line.

[0012] In a preferred embodiment, the cross-edge anomaly identification unit extracts the abnormal coverage segment, grayscale sequence along the line, and boundary response sequence across the edge only for normal sampling lines that simultaneously have inner source candidates and outer release candidates. This is done within the local area of ​​the edge zone where the current normal sampling line passes through the corresponding edge segment, to form the cross-edge penetration width, cross-edge grayscale continuity, and edge boundary weakening degree. When the continuous crossing length corresponding to the cross-edge penetration width is not less than the minimum crossing length and not less than a preset ratio of the nominal edge width, the cross-edge grayscale continuity falls within the allowable range of continuous change, and the edge boundary weakening degree meets the boundary instability condition, it is determined that there is a valid cross-edge channel on the current normal sampling line.

[0013] In a preferred embodiment, when the current normal sampling line has both an inner source candidate and an outer release candidate and a valid cross-edge channel, the cross-edge anomaly identification unit determines that the current normal sampling line forms a valid cross-edge release line; when the current normal sampling line lacks any of the aforementioned conditions, the cross-edge anomaly identification unit registers that the current normal sampling line has not formed a valid cross-edge release line and does not include the current normal sampling line in subsequent processing.

[0014] In a preferred embodiment, the cross-edge sealing anomaly zone identification unit merges adjacent normal sampling lines that form effective cross-edge sealing release lines along the local edge sealing centerline into the same cross-edge sealing anomaly zone; when the number of normal sampling lines that do not form effective cross-edge sealing release lines in the middle does not exceed the allowable number of gaps, the normal sampling lines on both sides of the gap are merged into the same cross-edge sealing anomaly zone; when the number of effective normal sampling lines merged into the same cross-edge sealing anomaly zone and having inner source candidates reaches the minimum number of inner support lines, or their continuous projection length reaches the minimum inner support length, the inner support state is determined to be established; when the number of effective normal sampling lines merged into the same cross-edge sealing anomaly zone and having outer release candidates reaches the minimum number of outer support lines, or their continuous projection length reaches the minimum outer support length, the outer support state is determined to be established.

[0015] In a preferred embodiment, the result determination unit identifies suspected abnormal areas that do not form an abnormal band across the sealing edge as obvious artifact areas; identifies suspected abnormal areas that form an abnormal band across the sealing edge but whose inner and outer support states are not established as non-closed abnormal areas; identifies suspected abnormal areas that form an abnormal band across the sealing edge and whose support state is established on only one side as areas to be reviewed; identifies suspected abnormal areas that form an abnormal band across the sealing edge and whose inner and outer support states are both established as leakage candidate areas; when all suspected abnormal areas are obvious artifact areas, a pass result is output; when there are no leakage candidate areas and at least one non-closed abnormal area or area to be reviewed exists, a review result is output; when there is at least one leakage candidate area, a rejection result is output.

[0016] The effects and advantages of the intelligent visual recognition-based automated food inspection system of this invention: This invention provides an automated food ingredient inspection and verification system based on intelligent visual recognition. By establishing a local edge-sealing geometric benchmark around the corresponding edge-sealing section and using the normal sampling line as the smallest identification carrier, the system first identifies inner source candidates and outer release candidates on the same normal sampling line. Then, it verifies valid cross-edge-sealing channels from the middle edge-sealing area. This prevents similar appearances near the edge-sealing from being directly considered as the same abnormal object; instead, it only performs subsequent consolidation and aggregation on the line-by-line results of closed, valid cross-edge-sealing release lines. Thus, the system can distinguish between reflections, localized adhesions, external residues, and genuine cross-edge-sealing release phenomena. It stably converges objects that have not formed a valid closed chain into obvious artifacts or non-closed abnormal objects, objects that have formed partial closed chains into objects awaiting verification, and objects that have formed complete closed chains and have bilateral support into leakage candidates. Furthermore, the system completes regional-level identification through cross-edge-sealing abnormal zones and their inner and outer support states, and then outputs item-level inspection results based on the regional category combination relationship. This improves the stability, verifiability, and item-level output consistency of the front-end triage process and reduces reliance on human experience. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the system structure of the present invention; Figure 2 This is a schematic diagram showing the partial identification results of the sealing edge of the chicken breast tray. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides an automated verification system for food ingredient acceptance based on intelligent visual recognition. It is applicable to scenarios where suspicious local anomalies appear near the sealing edge of pre-packaged food ingredients prone to leakage during food ingredient arrival acceptance. The system performs front-end screening on such objects and supports item-level acceptance result output. During food ingredient arrival acceptance, inspectors typically need to check the label information, packaging status, and appearance of the arriving objects. For routine food ingredients with intact packaging, stable appearance, and no local anomalies, existing acceptance processes are usually sufficient for basic judgment.

[0020] However, for pre-packaged food containers prone to leakage, phenomena such as food leakage, film reflection, localized darkening, and blunted boundaries can easily create a mixture of similar abnormal appearances near the sealing edge, making this area difficult to use as a stable background to directly support front-end judgment. Current practices, when encountering such suspicious local anomalies near the sealing edge, typically involve directly judging the overall appearance of the local anomaly block, or transferring to manual review and a more complex subsequent identification process. There is a lack of a front-end judgment chain that establishes a unified geometric benchmark around the same corresponding sealing edge segment and verifies the relationship between the inner and outer ends under the same normal sampling path. Therefore, it is difficult to reliably distinguish between obvious artifacts, non-closed anomalies, objects awaiting review, and truly high-risk objects.

[0021] Therefore, the technical problem to be solved by this invention is how to establish a unified local sealing geometric benchmark around the corresponding sealing segment most relevant to the suspected local area near the sealing edge of the pre-packaged easily separable food tray, identify the source of abnormality on the inside of the sealing edge and the release phenomenon on the outside of the sealing edge on the same normal sampling line, and verify whether there is a valid cross-sealing channel through the sealing area located between the two, thereby completing the front-end diversion of obvious illusory objects, non-closed abnormal objects, objects to be verified and leakage candidates, and supporting the output of component-level acceptance results.

[0022] Based on the above design, this invention constructs a complete process for an automated food inspection system using intelligent visual recognition, consisting of a suspected anomaly location unit, an inner and outer observation zone construction unit, a cross-sealing anomaly zone identification unit, and a result judgment unit. (Refer to...) Figure 1 , Figure 1 This is a schematic diagram of the system structure of the present invention. The system includes: The suspected anomaly localization unit is used to locate suspected anomaly areas and determine the corresponding edge segments within the edge-sealed neighborhood of the tray image, forming suspected anomaly localization result information R101. This unit reads the tray input information X101, which includes at least the tray image, and completes the location of suspected anomaly areas and determination of corresponding edge segments within the edge-sealed neighborhood based on the tray image. R101 includes at least the suspected anomaly area and the corresponding edge segment. The edge-sealed neighborhood refers to the local image region formed by expanding the tray's edge-sealing strip to the inside and outside of the edge according to the nominal width of the edge or the image resolution. The suspected anomaly area refers to an abnormal connected region located within the edge-sealed neighborhood that meets at least one type of anomaly response condition and is retained after screening. The corresponding edge segment refers to the local edge segment most relevant to the suspected anomaly area in local spatial relationship.

[0023] The inner and outer observation zone construction unit is used to establish the local edge sealing centerline and normal around the corresponding edge sealing segment, and generate the inner observation zone, outer observation zone, and normal sampling line to form the basic information R102 for local edge sealing observation. This unit reads R101 and, if necessary, combines the tray specification information, image resolution, or acquisition parameter information to reinforce the establishment range and sampling density of the local geometric reference. R102 includes at least the local edge sealing centerline, normal, inner observation zone, outer observation zone, and normal sampling line. Among them, the local edge sealing centerline refers to the local edge sealing centerline determined based on the corresponding edge sealing segment; the normal refers to the direction perpendicular to the local direction of the local edge sealing centerline and pointing to the inside or outside of the edge sealing; the inner and outer observation zones refer to the local strip areas formed by extending along the normal to the inside and outside of the edge sealing, respectively; and the normal sampling line refers to the sampling line arranged in order of the position of the local edge sealing centerline and arranged along the normal at each position.

[0024] The cross-edge sealing anomaly identification unit is used to read the basic information R102 of local edge sealing observation. Using the normal sampling line as the smallest identification carrier, it first identifies the inner source candidate on the inner side of the edge sealing, and then identifies the outer release candidate on the outer side of the edge sealing. When both inner source and outer release candidates exist simultaneously, it verifies whether a valid cross-edge sealing channel exists within the local area of ​​the edge sealing section through which the current normal sampling line passes. Only when a valid cross-edge sealing channel is established is the current normal sampling line determined to form a valid cross-edge sealing release line. This unit then merges adjacent valid cross-edge sealing release lines along the local edge sealing centerline to form a cross-edge sealing anomaly zone, and statistically analyzes the support status of the cross-edge sealing anomaly zone on the inner and outer sides of the edge sealing, generating cross-edge sealing anomaly zone identification result information R103. R103 includes at least the cross-edge sealing anomaly zone, the inner support status, and the outer support status.

[0025] The result determination unit is used to read R103, perform area category determination and part-level closing on the cross-sealing anomaly identification results, and form acceptance determination result information R104. R104 includes at least the area category and the part-level acceptance result. Among them, the area category preferably includes obvious false appearance area, non-closed anomaly area, area to be reviewed, and leakage candidate area. The part-level acceptance result preferably includes pass result, review result, and rejection result.

[0026] The system of this invention no longer directly uses the entire local appearance near the edge banding as the judgment object. Instead, it establishes a local edge banding geometric benchmark around the corresponding edge banding segment and uses the normal sampling line as the smallest identification carrier. On the same normal sampling line, it first identifies the inner source candidates on the inner side of the edge banding and the outer release candidates on the outer side of the edge banding. Then, it verifies the valid cross-edge banding channel by the edge banding area located between the two. It only performs line-by-line verification on the results of closing and forming a valid cross-edge banding release line, forming a structured intermediate basis of the cross-edge banding anomaly zone. On this basis, the system completes the regional category classification based on the cross-edge banding anomaly zone and its inner and outer support states, and collects the results of the regional category combination into the part-level acceptance result. Thus, the front-end screening is closed into a line-by-line closed verification and part-level output chain around the corresponding edge banding segment.

[0027] The implementation process and operational effects of the system of the present invention will be described in detail below with reference to specific embodiments. It should be understood that the embodiments are only used to illustrate the technical solution of the present invention, and not to limit it. Without changing the essence of the invention, the relevant unit division, steps and parameters can be appropriately adjusted.

[0028] In an optional embodiment, the suspected anomaly location unit reads the tray input information X101 and performs edge sealing neighborhood limitation, anomaly response extraction, region filtering and corresponding edge sealing segment determination within the edge sealing neighborhood to form suspected anomaly location result information R101.

[0029] In defining the sealing neighborhood, this unit identifies the position of the sealing strip in the tray image and uses the pixel width of the sealing strip as a reference, expanding it to the inside and outside of the sealing strip according to an expansion coefficient to form the sealing neighborhood. The expansion width is preferably determined by multiplying the nominal width of the sealing strip by the expansion coefficient, which can be selected based on the tray specifications and image resolution conversion results. When tray specification information is provided, the actual width of the sealing strip corresponding to the specification is preferably used as the reference; when specification information is not provided, the pixel width of the sealing strip in the image is preferably used as a substitute reference.

[0030] In the anomaly response extraction, this unit performs local response extraction on the image within the edge-sealing neighborhood, forming three types of anomaly response quantities: bright spot aggregation, dark wetting, and edge relaxation. Bright spot aggregation is preferably obtained by extracting the set of pixels with brightness above a brightness threshold and performing connected component segmentation, then statistically analyzing the proportion of bright spot connected component area or the number density of bright spots per unit area. Dark wetting is preferably obtained by extracting regions with brightness below a darkness threshold and analyzing their diffusion range in the normal direction in conjunction with gray-level gradient changes. Edge relaxation is preferably obtained by calculating the gradient peak at the edge of the edge-sealing and comparing it with the normal edge-sealing area; when the edge gradient peak significantly decreases or the boundary response changes from a single peak to broadening, the region is marked as a candidate anomaly region.

[0031] In the region screening process, this unit further performs connectivity and spatial relationship screening on candidate abnormal regions. First, this unit calculates the area of ​​the candidate regions and retains only those regions with an area not less than the minimum connectivity area threshold. Next, this unit calculates the minimum distance between the candidate region and the edge banding, retaining only those regions whose distance does not exceed the maximum allowable distance; the maximum allowable distance is preferably determined as a multiple of the nominal width of the edge banding. Then, this unit calculates the overlap between the candidate region and the edge banding boundary to obtain the edge coverage ratio, retaining only those regions with a coverage ratio not less than the minimum edge coverage ratio; the minimum edge coverage ratio is preferably determined based on statistics of effective edge coverage areas in historical abnormal samples. After the above screening, suspected abnormal regions that simultaneously possess abnormal response characteristics and maintain spatial correlation with the edge banding are retained.

[0032] In determining the corresponding edge sealing segment, this unit projects each suspected abnormal area along the edge sealing direction to obtain its corresponding position range on the edge sealing strip, and then extracts candidate edge sealing segments near this position range along the edge sealing direction. The length of the candidate edge sealing segment is preferably extended to both sides by a predetermined proportion based on the projected length of the suspected abnormal area in the edge sealing direction, or a fixed length window is extracted as a multiple of the nominal width of the edge sealing. Among multiple candidate edge sealing segments, the edge sealing segment with the smallest distance from the suspected abnormal area in the normal direction is preferably selected as the corresponding edge sealing segment; when multiple candidate edge sealing segments are close in distance, the edge sealing segment with the largest overlap ratio with the projected suspected abnormal area is preferably selected, or the optimal selection is made based on the continuity of the distribution of the abnormal response amount in the edge sealing direction.

[0033] Through the above processing, this unit generates suspected anomaly location result information R101 and provides R101 to the inner and outer observation zone construction unit for reading. To improve the stability of suspected anomaly area location, this unit can also perform brightness normalization, local illumination correction, and noise suppression on the tray image before the above processing to reduce the impact of ambient light changes on anomaly response extraction and region selection, but without changing the core processing logic of this unit to locate suspected anomaly areas and bind corresponding edge segments within the edge-sealing neighborhood.

[0034] In an optional embodiment, the inner and outer observation zone construction unit reads the suspected anomaly location result information R101, and performs local edge sealing centerline establishment, normal determination, inner and outer observation zone construction and normal sampling line arrangement around the corresponding edge sealing segment in R101, thereby forming local edge sealing observation basic information R102.

[0035] In establishing the local edge sealing centerline, this unit first extracts the local edge distribution of the corresponding edge sealing segment based on its spatial position in the tray image, and then determines the center direction of the edge sealing area within this local edge distribution. Preferably, the local edge sealing centerline can be established by reading the positions of the two side boundaries of the corresponding edge sealing segment and calculating its center trajectory. When there are slight bends or image disturbances in the local edge sealing edge, a moving average or local linear fitting can be performed on the adjacent boundary positions to obtain a continuous and stable local edge sealing centerline.

[0036] In determining the normal direction, this unit calculates the local tangential direction at each position along the extension direction of the local edge sealing centerline, and determines the normal direction at the corresponding position using the direction perpendicular to the local tangential direction. Preferably, when the local edge sealing centerline maintains an approximately straight trend near the current position, the perpendicular tangential line can be directly taken as the normal direction. When the local edge sealing centerline exhibits a slow curvature, the normal direction can be continuously updated according to the directional changes of adjacent positions of the centerline to ensure that the distribution of the normal direction remains continuous along the entire local edge sealing centerline.

[0037] In constructing the inner and outer observation zones, this unit uses the local edge sealing centerline as a reference and expands along the normal direction at each position towards the inner and outer sides of the edge sealing, respectively, forming an inner observation zone and an outer observation zone. Preferably, the width of the inner observation zone can be determined based on the safe observation distance on the inner side of the edge sealing corresponding to the tray specification, the image resolution conversion result, or the inner extension range in historical valid abnormal samples. The width of the outer observation zone can be determined based on the range of the liquid-readable area on the outer side of the edge sealing corresponding to the tray specification, the image resolution conversion result, or the outer extension range in historical valid abnormal samples. When multiple width sources exist, it is preferable to use the width corresponding to the tray specification as the main aperture, the image resolution conversion result as the scale correction basis, and the range of historical valid abnormal samples as the boundary correction basis. When tray specification information is missing, the widths of the inner and outer observation zones can also be converted by multiples of the nominal width of the edge sealing.

[0038] In the arrangement of normal sampling lines, this unit selects multiple sampling points sequentially along the local edge centerline, and arranges normal sampling lines along the corresponding normal at each sampling point, forming a set of normal sampling lines for subsequent line-by-line identification. Preferably, the layout density of normal sampling lines can be determined jointly based on the length of the local edge centerline, the image resolution, and the minimum number of sampling lines required to form a stable banded object. When the local edge centerline is short, the sampling density per unit length can be increased; when the local edge centerline is long, a uniform arrangement can be maintained while ensuring that the spacing between adjacent sampling lines does not exceed a certain proportion of the nominal width of the edge, and the total number of normal sampling lines is not less than the minimum number of lines required for subsequent stability determination. Each normal sampling line is preferably numbered according to its position on the local edge centerline so that the subsequent third unit can perform adjacent relationship checks and parallel processing along the direction of the local edge centerline.

[0039] Through the above processing, this unit establishes a local edge sealing centerline and normal around the corresponding edge sealing segment, forming an inner observation zone, an outer observation zone, and a normal sampling line, and writes these objects together into the local edge sealing observation basic information R102. R102 is provided to the cross-edge sealing anomaly zone identification unit for reading.

[0040] To improve the stability of establishing the local edge sealing geometric reference, this unit can also perform edge enhancement, local smoothing, and geometric correction on the local image near the corresponding edge sealing segment before the above processing, so as to reduce the impact of local noise, slight deformation, or changes in acquisition angle on the determination of the local edge sealing center line and normal. It can also adaptively adjust the width of the observation strip and the density of the sampling line according to different tray specifications, different image resolutions, or different acquisition device parameters, but without changing the core processing logic of this unit to establish the local edge sealing geometric reference around the corresponding edge sealing segment and form R102.

[0041] In one optional embodiment, the cross-edge sealing anomaly identification unit reads the basic information R102 of the local edge sealing observation and performs line-by-line identification around the normal sampling line in R102. The implementation process of this unit includes line-by-line preparation and sorting, identification of inner source candidates, identification of outer release candidates, verification of valid cross-edge sealing channels, confirmation of valid cross-edge sealing release lines, and cross-edge sealing anomaly zone and result output.

[0042] In the line-by-line preparation and sorting process, this unit reads the local edge sealing centerline, normal, inner observation zone, outer observation zone, and normal sampling line from R102. It then sorts the positions of each normal sampling line along the extension direction of the local edge sealing centerline, ensuring that the arrangement order of each normal sampling line on the local edge sealing centerline is consistent with the subsequent parallel zone order. The sorted normal sampling lines pass through the inner edge sealing, the edge sealing area, and the outer edge sealing, respectively. Therefore, the identification of inner source candidates, the identification of outer release candidates, and the verification of valid cross-edge sealing channels are all constrained to be completed under the unified geometric relationship of the same corresponding edge sealing segment.

[0043] In the identification of inner source candidates, this unit first reads the abnormal coverage segment, texture response changes, and continuous connection along the line within a local area where the current normal sampling line passes through the inner side of the edge, and determines whether the current normal sampling line forms an inner source candidate based on this. An inner source candidate refers to an inner abnormal line segment on the current normal sampling line that is located inside the edge, continuously attached to the inner boundary of the edge, and meets the requirements of inner edge coverage length, inner texture attenuation intensity, and inner edge continuity. It is used to indicate that there is an abnormal source inside the edge that can be continuously transmitted to the edge area.

[0044] When forming candidates for inner sources, this unit preferably extracts the inner edge coverage length, inner texture attenuation intensity, and inner edge continuity. The inner edge coverage length is obtained by the length of the coverage segment that continuously attaches to the current abnormal coverage range from the inner boundary of the sealing edge in the inward direction, and is used to characterize the edge extension range of the abnormal source along the inner side of the sealing edge. The inner texture attenuation intensity is preferably obtained by the difference between the local edge density at the abnormal coverage position and the local edge density at the adjacent normal position. The local edge density is preferably defined as the proportion of pixels whose gradient magnitude exceeds the edge determination threshold within the local window, and is used to characterize the degree of weakening of the original texture of the flesh surface under the effect of abnormal coverage. The inner edge continuity is obtained by the continuous connection of the inner abnormal coverage segment near the sealing edge position, and is used to characterize whether the abnormal source can continuously attach to the sealing edge position along the edge direction. When the inner edge coverage length is not less than the minimum edge extension length, and the local edge density at the abnormal coverage location is lower than the local edge density at the adjacent normal location to the preset attenuation threshold, and the number of interruptions of the inner abnormal coverage segment does not exceed the interruption allowable value and remains continuously connected, the current normal sampling line is registered to have an inner source candidate; otherwise, the current normal sampling line is registered to have no inner source candidate.

[0045] In the outer release candidate identification, this unit further reads the outer abnormal coverage segment and its outer edge expansion shape within the local area where the current normal sampling line passes through the outer edge of the seal, and determines whether the current normal sampling line forms an outer release candidate. An outer release candidate refers to an outer abnormal line segment on the current normal sampling line that is located outside the seal, continuously attached to the outer edge boundary, and meets the requirements of outer extension length and outer edge divergence. It is used to indicate that there is a release phenomenon outside the seal that needs further verification.

[0046] When forming candidates for outward release, this unit preferably extracts the outward extension length and the outward edge divergence. The outward extension length is obtained by the length of the coverage segment that continuously attaches to the current anomaly coverage area from the outer boundary of the sealing edge in the outward direction, and is used to characterize the expansion range of the anomaly release outside the sealing edge. The outward edge divergence is preferably obtained by the difference in the outer edge expansion width between the position near the sealing edge and the position far from the sealing edge, and is used to characterize whether the outward anomaly release shows a continuous spreading trend during the outward expansion process. When the outward extension length is not less than the minimum outward extension length and the outward edge divergence meets the divergence judgment condition, the existence of an outward release candidate is registered for the current normal sampling line; otherwise, the existence of an outward release candidate is registered for the current normal sampling line.

[0047] In the verification of effective cross-border channels, only for normal sampling lines where both inner source candidates and outer release candidates exist simultaneously, this unit continues to extract the grayscale sequence, boundary response sequence, and abnormal coverage segments crossing the border within the local area of ​​the corresponding border segment crossed by the current normal sampling line, and determines whether a valid cross-border channel exists on the current normal sampling line. An effective cross-border channel refers to a continuous crossing path located within the local area of ​​the border zone on the current normal sampling line, satisfying the requirements for cross-border width, cross-border grayscale continuity, and border boundary weakening. This path is used to verify whether a genuine cross-border connection exists between the current inner source candidate and the current outer release candidate. The purpose of this verification is not to re-identify an independent anomaly, but to determine whether the inner source candidate and outer release candidate already identified on the current normal sampling line can be closed into a continuous release chain by the same cross-border path.

[0048] When forming an effective cross-border channel, this unit preferably extracts the cross-border width, cross-border grayscale continuity, and border boundary weakening degree. The cross-border width is obtained by the length of the abnormal coverage segment that maintains continuous coverage within the border area and when crossing both sides of the border, and is used to characterize whether the current abnormal coverage truly and continuously crosses the border area. The cross-border grayscale continuity is preferably obtained by the difference between the maximum and minimum values ​​of the absolute value sequence of grayscale differences between adjacent sampling points during the process of the abnormal coverage segment crossing the border area, and is used to characterize whether the abnormal coverage that crosses the border area maintains continuous transmission in optical response. The border boundary weakening degree is preferably obtained by comparing the peak value of the border edge gradient at the abnormal coverage location with the peak value of the border edge gradient at the corresponding non-abnormal location. The corresponding non-abnormal location is preferably the position of the normal sampling line adjacent to the current normal sampling line and not intersecting with the suspected abnormal area, and is used to characterize whether the original border boundary response is weakened, widened, or locally missing at the abnormal coverage location. When the continuous span of the abnormal coverage segment is not less than the minimum span length and not less than the preset proportion of the nominal width of the sealing edge, and the difference between the maximum and minimum values ​​of the absolute value sequence of gray difference between adjacent sampling points falls within the range of continuous change, and the peak value of the sealing edge gradient at the abnormal coverage location is lower than the peak value at the corresponding non-abnormal location by a preset proportion threshold, or the peak width of the sealing edge response is greater than the peak width at the corresponding non-abnormal location by a preset proportion threshold, or the sealing edge changes from a clear single-peak response to weakening, widening, or local absence, a valid cross-sealing edge channel is identified on the current normal sampling line; otherwise, a valid cross-sealing edge channel is not identified.

[0049] In the confirmation of valid cross-edge release lines, when the current normal sampling line simultaneously has an inner source candidate, an outer release candidate, and a valid cross-edge channel, this unit determines that the current normal sampling line forms a valid cross-edge release line. A valid cross-edge release line refers to the line-by-line establishment result where an inner source candidate, an outer release candidate, and a valid cross-edge channel are simultaneously formed on the current normal sampling line. This is used as the smallest valid line unit for the subsequent parallel band formation of cross-edge anomaly bands. If the current normal sampling line lacks any of the aforementioned establishment conditions, it is registered that the current normal sampling line has not formed a valid cross-edge release line, and the current normal sampling line is not included in the subsequent parallel band processing.

[0050] In the cross-edge anomaly zone merging and result output, this unit retains three types of line-by-line results for each normal sampling line: whether there are inner source candidates, whether there are outer release candidates, and whether a valid cross-edge release line is formed. Subsequently, this unit takes the formation of a valid cross-edge release line by adjacent normal sampling lines as the premise for merging, checks the continuous positional relationship of each normal sampling line along the direction of the local edge centerline, and merges the valid line-by-line results with adjacent connection relationships into a cross-edge anomaly zone. The cross-edge anomaly zone is composed of continuously distributed valid cross-edge release lines, used to receive the continuous results of each valid cross-edge release line, and used for subsequent statistical analysis of the support status of the inner and outer sides of the edge. The adjacent connection relationship can preferably be determined based on the numbering continuity of the normal sampling lines on the local edge centerline; when the local sampling lines are not strictly evenly spaced, it can also be determined based on whether the projection distance of adjacent normal sampling lines on the local edge centerline is not greater than the upper limit of the allowable distance. For continuous normal sampling lines that all form effective cross-edge release lines, this unit will group them into the same cross-edge anomaly zone. For cases where there are only a few missing normal sampling lines in the middle but the number of gaps does not exceed the allowable number of gaps, this unit can still group the effective normal sampling lines on both sides of the gap into the same cross-edge anomaly zone. When the distance between adjacent effective normal sampling lines exceeds the upper limit of the allowable distance, or the number of gaps in the middle exceeds the allowable number of gaps, this unit will divide them into different cross-edge anomaly zones.

[0051] After forming the cross-edge anomaly zone, this unit further calculates the number of valid normal sampling lines satisfying the inner source candidate identification results and their continuous projection lengths based on the inner source candidate identification results and outer release candidate identification results corresponding to each normal sampling line incorporated into the cross-edge anomaly zone. It also calculates the number of valid normal sampling lines satisfying the outer release candidate identification results and their continuous projection lengths. Since all normal sampling lines included in the zone statistics have already formed valid cross-edge release lines, the inner source candidates and outer release candidates calculated here have all passed the valid cross-edge channel verification and are no longer isolated candidate results. When the number of valid normal sampling lines satisfying the inner source candidate reaches the minimum number of inner support lines, or the continuous projection length formed by them along the local edge sealing center line reaches the minimum inner support length, the inner support state of the cross-edge sealing anomaly zone is recorded as established; when the number of valid normal sampling lines satisfying the outer release candidate reaches the minimum number of outer support lines, or the continuous projection length formed by them along the local edge sealing center line reaches the minimum outer support length, the outer support state of the cross-edge sealing anomaly zone is recorded as established; at the same time, the length of the cross-edge sealing anomaly zone and the number of valid normal sampling lines are written together into the cross-edge sealing anomaly zone identification result information R103, and R103 is provided to the result judgment unit for reading.

[0052] To improve the stability and anti-interference capability of the cross-sealing abnormal zone identification results, this unit can further adopt the following engineering implementation method.

[0053] When multiple criteria exist, it is preferable to first calculate the basic threshold based on the nominal scale corresponding to the current tray specification and the image resolution, and then perform boundary correction using the upper bound of historical normal samples or the lower bound of historical valid abnormal samples. When tray specification information is missing, it is preferable to determine the basic threshold as a multiple of the width of the edge pixels in the image.

[0054] For the minimum edge extension length, it can preferably be calculated based on the effective width of the inner observation zone of the seal at the current image resolution and the safe observation distance of the inner edge corresponding to the tray specification, and then corrected by combining the maximum extension range of short-term edge disturbances in historical normal samples, for the identification of inner source candidates. For the minimum outward expansion length, it can preferably be calculated based on the effective width of the outer observation zone of the seal at the current image resolution and the nominal expansion range of the liquid-extractable area on the outer side of the tray, and then corrected by combining the maximum extension range of scattered attachments on the outer side in historical normal samples, for the identification of outer release candidates. For the minimum crossing length, it can preferably be calculated based on the nominal width of the seal edge of the current tray and the image resolution to obtain the seal edge pixel width, and then corrected by combining the continuous crossing lower limit of the complete seal edge area in historical normal samples, for the verification of effective crossing of the seal edge channel. For the upper limit of allowable spacing and the number of allowable gaps, it can preferably be determined by combining the local seal edge centerline length, sampling line layout density, centerline pixel interval corresponding to the image resolution, and the continuous coverage distribution of historical effective abnormal bands in the centerline direction. The minimum strip length and minimum number of effective normal sampling lines can be preferably determined by combining the length of the local edge sealing centerline, the projected length of the centerline, the sampling line layout density, and the minimum number of continuous effective lines required to form a stable strip-shaped object. Here, the requirement for a stable strip-shaped object refers to the condition for the same cross-edge sealing anomaly strip to simultaneously meet the minimum strip length requirement and the minimum number of effective normal sampling lines requirement.

[0055] Furthermore, the minimum number of inner and outer support lines can be preferably determined based on the local edge centerline length range, the density of normal sampling lines, and the minimum number of consecutive normal sampling lines in historical valid anomaly zones. When the local edge centerline is short, it is preferable to use two or three consecutive normal sampling lines as the starting point for the basic number of support lines, and to make corrections based on the lower limit of short-band continuity in historical valid anomaly samples. The minimum inner and outer support lengths can be preferably determined based on the centerline projection length in the direction of the local edge centerline, the pixel scale corresponding to the nominal width of the edge, and the continuous projection range of the support segment in historical valid anomaly zones. The preset ratio threshold in the valid cross-edge channel verification can be preferably determined based on the historical statistical results of the boundary gradient peak ratio distribution between the normal edge area and the valid anomaly area, the boundary response peak width ratio distribution, or the proportion of the continuous crossing length to the nominal width of the edge. The preset attenuation threshold in the inner source candidate identification can be preferably determined based on the distribution of the local edge density difference between the anomaly coverage position and the adjacent normal position in the valid anomaly samples. Preferably, the above-mentioned ratio threshold and attenuation threshold can be determined based on the separation interval between the upper limit of historical normal samples and the lower limit of historical effective abnormal samples, and then boundary correction can be performed in combination with the current tray specifications, image resolution and acquisition device parameters; when historical samples are insufficient, conservative thresholds can also be taken based on the pixel scale corresponding to the nominal width of the sealing edge and the local strip statistical stability requirements.

[0056] Before generating the aforementioned readings, brightness normalization, local illumination correction, and noise suppression can be performed on the current local area to reduce the impact of changes in the acquisition environment on the continuity of grayscale across the edge, the degree of edge weakening, the intensity of inner texture attenuation, and the divergence of outer edge. However, this does not change the main processing order of this unit, which uses the normal sampling line as the smallest identification carrier, first identifies inner source candidates, then identifies outer release candidates, then performs valid cross-edge channel verification, and then outputs the data.

[0057] In an optional embodiment, the result determination unit reads the cross-sealing abnormal band identification result information R103, and performs area category determination, combination relationship merging and part-level acceptance result output, thereby forming the acceptance determination result information R104.

[0058] In determining the area category, this unit performs formal classification based on the cross-edge anomaly zone corresponding to each suspected anomaly zone and its inner and outer support states. When a suspected anomaly zone does not form a corresponding cross-edge anomaly zone, this unit identifies the suspected anomaly zone as a significant false alarm zone; when a suspected anomaly zone has formed a cross-edge anomaly zone, but neither the inner nor outer support states are established, this unit identifies the suspected anomaly zone as a non-closed anomaly zone; when a suspected anomaly zone has formed a cross-edge anomaly zone, and only one side of the support state is established, this unit identifies the suspected anomaly zone as a zone awaiting verification; when a suspected anomaly zone has formed a cross-edge anomaly zone, and both the inner and outer support states are established, this unit identifies the suspected anomaly zone as a leak candidate zone.

[0059] In the merge operation, this unit summarizes the region categories of all suspected abnormal areas within the current tray and performs item-level merge according to region category priority. When all suspected abnormal areas within the current tray are identified as obvious false positives, this unit merges the current tray into a pass result. When there are no leakage candidate areas within the current tray, but at least one non-closed abnormal area or area awaiting review exists, this unit merges the current tray into a review result. When there is at least one leakage candidate area within the current tray, this unit merges the current tray into a rejection result. When non-closed abnormal areas, areas awaiting review, and leakage candidate areas coexist, this unit prioritizes outputting the item-level result corresponding to the leakage candidate area.

[0060] In the output of the part-level acceptance results, this unit writes the area category corresponding to each suspected abnormal area and the part-level acceptance result corresponding to the current tray into the acceptance judgment result information R104, and uses R104 as the final output of the system main chain.

[0061] To improve the stability of region category determination and component-level finishing, this unit can further adopt the following engineering implementation methods.

[0062] For the boundary determination of obvious false areas, non-closed abnormal areas, areas to be verified, and leakage candidate areas, it can be refined by combining the length of the cross-sealing abnormal zone, the number of effective normal sampling lines, the integrity of the inner support, and the integrity of the outer support. Among them, the integrity of the inner support refers to the proportion of continuous coverage or continuous projection length of the effective normal sampling lines forming inner source candidates in the cross-sealing abnormal zone under the premise that the inner support state has been established. The integrity of the outer support refers to the proportion of continuous coverage or continuous projection length of the effective normal sampling lines forming outer release candidates in the cross-sealing abnormal zone under the premise that the outer support state has been established.

[0063] For merging the piece-level acceptance results, stability reinforcement can be performed by combining the number of candidate leakage areas, the number of non-closed abnormal areas, the number of areas to be reviewed, and the distribution of each area in the sealing direction. Furthermore, when the current tray image simultaneously provides label information, batch information, or tray specification information, this unit can also combine the above-mentioned auxiliary input information to verify the results of the area category combination relationship. However, the auxiliary input information is only used for scale conversion, specification matching, result verification, or review prompts, and does not directly replace the area category determination and piece-level acceptance result output based on the cross-sealing abnormal zone and its inner and outer support states.

[0064] In one specific embodiment, taking the arrival and acceptance scenario of a single refrigerated pre-packaged chicken breast tray as an example, the operation process of the aforementioned system is described. The system receives a top-view image of the chicken breast tray as the current tray image and locates a suspected abnormal area within the sealing edge neighborhood of the tray image. Subsequently, the system establishes a local identification range around the sealing edge neighborhood where the suspected abnormal area is located, and determines the corresponding sealing edge segment based on the spatial correspondence between the local identification range and the sealing edge strip. The local identification range refers to the local observation area intercepted around the suspected abnormal area and its corresponding sealing edge segment. Based on this, the system establishes a local sealing edge centerline around the corresponding sealing edge segment and arranges normal sampling lines that penetrate the inner observation strip, the sealing edge area, and the outer observation strip, thereby forming a unified local sealing edge geometric reference.

[0065] like Figure 2 As shown, under the local edge sealing geometric reference, the system first identifies the inner source candidate on the inner side of the edge sealing along the same normal sampling line, and then identifies the outer release candidate on the outer side of the edge sealing. When both the inner source candidate and the outer release candidate exist simultaneously, the system verifies whether a valid cross-edge sealing channel exists in the edge sealing area between them. Only when a valid cross-edge sealing channel is established is the current normal sampling line determined to form a valid cross-edge sealing release line. Subsequently, the system merges adjacent normal sampling lines along the local edge sealing centerline direction that all form valid cross-edge sealing release lines to form a cross-edge sealing anomaly zone.

[0066] Within the inner observation zone, the system reads the inner edge coverage length, inner texture attenuation intensity, and inner edge continuity to determine whether the current normal sampling line forms an inner source candidate within the edge. The inner edge coverage length characterizes the extent to which the anomalous coverage extends inward from the inner boundary of the edge; the inner texture attenuation intensity characterizes whether the texture in the inner region is weakened compared to the normal position; and the inner edge continuity characterizes whether the anomalous coverage remains continuously attached along the edge direction. When the above readings meet the conditions for establishing an inner source candidate, the system registers the existence of an inner source candidate for the current normal sampling line.

[0067] Within the outer observation zone, the system reads the outer extension length and outer edge divergence to determine whether the current normal sampling line forms an outer release candidate on the outer side of the sealing edge. The outer extension length characterizes the extent to which the anomalous coverage extends outward from the outer boundary of the sealing edge, while the outer edge divergence characterizes whether the outer edge exhibits a continuous outward expansion of the boundary trend. When the above reading results meet the conditions for the establishment of an outer release candidate, the system registers the existence of an outer release candidate for the current normal sampling line.

[0068] For normal sampling lines that simultaneously possess both inner source candidates and outer release candidates, the system continues to read the cross-border width, cross-border grayscale continuity, and border boundary weakening degree in the border area located between them to verify whether a valid cross-border channel exists on the current normal sampling line. Specifically, the cross-border width characterizes whether the anomalous coverage continuously crosses the border area; the cross-border grayscale continuity characterizes whether the anomalous coverage maintains continuous transmission within the border area; and the border boundary weakening degree characterizes whether the original border boundary response weakens, widens, or is partially missing. Only when the above border area reading results meet the conditions for a valid cross-border channel are the system determined that a genuine cross-border connection exists between the inner source candidate and the outer release candidate on the current normal sampling line, and further confirms that the normal sampling line forms a valid cross-border release line.

[0069] In this embodiment, the system confirms the formation of valid cross-edge release lines on multiple adjacent normal sampling lines within the local identification range. That is, near the same corresponding edge segment, the system not only identifies source candidates on the inner side of the edge and release candidates on the outer side, but also verifies valid cross-edge channels in the edge area between them. Based on this, the system merges the continuously established valid cross-edge release lines along the local edge centerline direction to form... Figure 2 The cross-edge anomaly band shown is not an isolated judgment result of a single normal sampling line, but a continuous band-shaped recognition result formed around the same corresponding edge segment. It is used to characterize that a stable cross-edge anomaly recognition structure has been formed in the vicinity of the edge in the current local recognition range.

[0070] During the result determination phase, the system performs region category determination on the area corresponding to the current suspected anomaly zone based on the cross-sealing anomaly band and its inner and outer support states. Since the cross-sealing anomaly band corresponding to the suspected anomaly zone has already formed, and both its inner and outer support states are valid, the result determination unit identifies the area corresponding to the suspected anomaly zone as a leakage candidate area. Furthermore, since the current tray only involves the region category result corresponding to this one suspected anomaly zone, the region category directly participates in the determination of the region category combination relationship of the current tray, and the result determination unit outputs the piece-level acceptance result of the current tray as a rejection result accordingly.

[0071] Through the above methods Figure 2 The candidate sources from the inner side, the candidate release from the outer side, the effective cross-sealing channel, the effective cross-sealing release line, and the cross-sealing anomaly zone shown in the diagram are expanded into an identification process that involves line-by-line identification, closure verification, convergence, and result output around the same corresponding seaming segment. As a result, local anomalies near the seam that are easily confused with reflections, wrinkles, or local attachments are unified under the same local seaming geometric benchmark for identification, and further converged into region categories and piece-level acceptance results. This improves the stability of identifying genuine high-risk trays and reduces reliance on human experience.

[0072] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0073] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0074] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0075] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0076] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent visual recognition food material acceptance automated checking system, characterized in that, include: The suspected anomaly location unit is used to locate suspected anomaly areas within the sealing edge neighborhood of the image of the pre-packaged easily separable liquid food tray, and to determine the corresponding sealing edge segment for each suspected anomaly area. The inner and outer observation zone construction unit is used to establish the local edge sealing center line and normal around the corresponding edge sealing segment, and generate the inner observation zone, the outer observation zone, and the normal sampling line for subsequent identification. The cross-edge anomaly zone identification unit is used to first read the inner edge coverage length, inner texture attenuation intensity, and inner edge continuity in the inner observation zone on the same normal sampling line to identify whether there are inner source candidates inside the normal sampling line; then read the outer extension length and outer edge divergence in the outer observation zone to identify whether there are outer release candidates outside the normal sampling line. When both inner source candidates and outer release candidates exist, the cross-border penetration width, cross-border grayscale continuity, and border boundary weakening degree in the border area between them are read to verify whether there is an effective cross-border channel between the inner source candidates and the outer release candidates. Only when an effective cross-border channel is established is it determined that the normal sampling line forms an effective cross-border release line. Adjacent normal sampling lines that both form effective cross-border release lines are merged along the local border centerline to form a cross-border anomaly zone, and the support status of the cross-border anomaly zone on the inner and outer sides of the border is recorded. The result determination unit is used to determine the area category of the suspected abnormal area based on the cross-sealing abnormal strip corresponding to the suspected abnormal area and its inner and outer support states, and output the acceptance result of the current tray according to the combination relationship of the area categories.

2. The smart visual recognition food ingredient receiving and automated checking system according to claim 1, wherein, The suspected anomaly localization unit extracts the amount of bright spot aggregation, dark wet infiltration, and edge relaxation within the edge sealing neighborhood; when any abnormal response amount reaches the corresponding abnormal condition, the corresponding area is marked as a candidate abnormal area. The candidate abnormal regions are filtered by minimum connected area, maximum allowable distance, and minimum edge ratio to identify suspected abnormal regions.

3. The intelligent visual recognition automated food inspection and verification system according to claim 2, characterized in that, The suspected anomaly location unit extracts the projection range of the suspected anomaly area along the sealing direction and extracts candidate sealing segments near the projection range; the candidate sealing segment with the smallest normal distance to the suspected anomaly area or the largest projection overlap ratio is determined as the corresponding sealing segment.

4. The intelligent visual recognition automated food inspection and verification system according to claim 3, characterized in that, The inner and outer observation zone construction unit establishes a local edge sealing center line based on the boundary positions on both sides of the corresponding edge sealing section, and determines the normal direction based on the local direction of the local edge sealing center line; multiple sampling points are selected in sequence along the local edge sealing center line, and normal sampling lines are arranged along the normal direction at each sampling point; the normal direction at each position is extended to the inner side and the outer side of the edge sealing respectively to form the inner observation zone and the outer observation zone.

5. The intelligent visual recognition automated food inspection and verification system according to claim 4, characterized in that, The cross-edge anomaly identification unit reads the abnormal coverage segment, texture response changes, and continuous connection along the line within a local area where the current normal sampling line passes through the inside of the edge, in order to form the inner edge coverage length, inner texture attenuation intensity, and inner edge continuity. When the inner edge coverage length is not less than the minimum edge extension length, the inner texture attenuation intensity reaches the preset attenuation threshold, and the inner edge continuity meets the continuous connection condition, it is determined that there is an inner source candidate for the current normal sampling line.

6. The intelligent visual recognition automated food inspection and verification system according to claim 5, characterized in that, The cross-sealing anomaly identification unit reads the outer anomaly coverage segment and its outer edge expansion shape within a local area where the current normal sampling line passes through the outer side of the sealing edge, so as to form the outer expansion length and the outer edge divergence. When the outer expansion length is not less than the minimum outer expansion length and the outer edge divergence meets the divergence judgment condition, it is determined that there is an outer release candidate for the current normal sampling line.

7. The intelligent visual recognition automated food inspection and verification system according to claim 6, characterized in that, The cross-edge anomaly identification unit only extracts the abnormal coverage segment, grayscale sequence along the line, and boundary response sequence across the edge within the local area of ​​the edge zone where the current normal sampling line passes through the corresponding edge segment, for normal sampling lines that simultaneously have inner source candidates and outer release candidates. This forms the cross-edge penetration width, cross-edge grayscale continuity, and edge boundary weakening degree. When the continuous crossing length corresponding to the cross-edge penetration width is not less than the minimum crossing length and not less than the preset ratio of the nominal edge width, the cross-edge grayscale continuity falls within the allowable range of continuous change, and the edge boundary weakening degree meets the boundary instability condition, it is determined that there is a valid cross-edge channel on the current normal sampling line.

8. The intelligent visual recognition automated food inspection and verification system according to claim 7, characterized in that, When the current normal sampling line has both an inner source candidate and an outer release candidate and a valid cross-edge channel, the cross-edge anomaly identification unit determines that the current normal sampling line forms a valid cross-edge release line; when the current normal sampling line lacks any of the aforementioned conditions, the cross-edge anomaly identification unit registers that the current normal sampling line has not formed a valid cross-edge release line and does not include the current normal sampling line in subsequent processing.

9. The intelligent visual recognition automated food inspection and verification system according to claim 8, characterized in that, The cross-edge sealing anomaly zone identification unit merges adjacent normal sampling lines that form effective cross-edge sealing release lines along the local edge sealing centerline into the same cross-edge sealing anomaly zone. When the number of normal sampling lines that do not form effective cross-edge sealing release lines in the middle does not exceed the allowable number of gaps, the normal sampling lines on both sides of the gap are merged into the same cross-edge sealing anomaly zone. When the number of effective normal sampling lines merged into the same cross-edge sealing anomaly zone and having inner source candidates reaches the minimum number of inner support lines, or their continuous projection length reaches the minimum inner support length, the inner support state is determined to be established. When the number of effective normal sampling lines merged into the same cross-edge sealing anomaly zone and having outer release candidates reaches the minimum number of outer support lines, or their continuous projection length reaches the minimum outer support length, the outer support state is determined to be established.

10. The intelligent visual recognition automated food inspection and verification system according to claim 9, characterized in that, The result determination unit identifies suspected abnormal areas that do not form cross-border abnormal zones as obvious false areas. Suspected anomalous areas that cross the edge sealing anomaly zone but whose inner and outer support states are not established are identified as non-closed anomalous areas; suspected anomalous areas that cross the edge sealing anomaly zone and whose support state is established on only one side are identified as areas to be reviewed; suspected anomalous areas that cross the edge sealing anomaly zone and whose inner and outer support states are established are identified as leakage candidate areas; when all suspected anomalous areas are obvious false positives, the pass result is output; when there are no leakage candidate areas and there is at least one non-closed anomalous area or area to be reviewed, the review result is output; when there is at least one leakage candidate area, the rejection result is output.

Images