Laundry detergent packaging defect detection method and system based on image recognition

Abstract

The invention provides a laundry detergent package defect detection method and system based on image recognition, and relates to the technical field of defect detection.The method comprises the steps that type defects are determined and extracted according to different types of laundry detergent packages, and surface image information is extracted to obtain package defect data; and classifying the packaging defect data to obtain type defect data, and detecting the type defect data by adopting a mixed imaging method to obtain defect detection data. And analyzing the defect detection data to obtain a defect analysis result, comparing the defect analysis result with a preset defect judgment threshold value, and judging whether the current laundry bag package is a defective product or not. And recording defect types, positions, parameters and production batch information of the defective products to generate a defect statistical report, and generating a defect adjustment scheme according to the defect statistical report. According to the method, the detection precision and efficiency are improved, the adaptability and robustness of the system are enhanced, and powerful support is provided for optimization of the production process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of defect detection, in particular to a laundry liquid packaging defect detection method and system based on image recognition. BACKGROUND

[0002] In today's laundry liquid production industry, packaging quality is crucial to product market competitiveness and brand image. However, laundry liquid packaging is prone to various defects during production, such as surface scratches, stains, printing errors, and poor sealing, etc. These defects not only affect the appearance of the product, but also may cause leakage during transportation or storage, thereby affecting product quality and consumer experience.

[0003] Currently, traditional laundry liquid packaging defect detection methods mainly rely on manual detection. Manual detection is a method in which an inspector evaluates laundry liquid packaging samples under specific light conditions and determines whether they have defects based on visual inspection. However, this method has many drawbacks, such as being easily affected by the inspector's experience, psychology, and physiology, and being prone to visual fatigue after long hours of work, leading to frequent missed and false detections. It is also difficult to establish a unified detection standard and quantify the work, and manual detection is inefficient and costly.

[0004] In addition, some enterprises have tried to use some machine learning-based detection methods, but these methods rely on manual feature design in industrial defect detection, and face great challenges in the case of complex and diverse defect features of laundry liquid packaging, leading to frequent missed and false detections. With the continuous updating of laundry liquid packaging materials, such as the application of new composite materials, and the increasing diversification of packaging design, which incorporates more technology and fashion elements, packaging defects are more difficult to separate from the background, further increasing the difficulty of defect identification.

[0005] At the same time, existing technologies also have deficiencies in laundry liquid packaging defect cause tracing, which cannot accurately determine the root cause of the defect, leading to a lack of clear direction for production optimization, making it difficult for enterprises to take targeted measures to improve production processes and improve product quality.

[0006] Therefore, in order to solve the problems of difficult laundry liquid packaging defect identification and difficult defect cause tracing leading to no direction for production optimization in the prior art, there is an urgent need for a laundry liquid packaging defect detection method and system based on image recognition. SUMMARY

[0007] The present application provides a laundry liquid packaging defect detection method and system based on image recognition, which solves the problem of difficult laundry liquid packaging defect identification and difficult defect cause tracing leading to no direction for production optimization in the prior art.

[0008] In one aspect, the present application provides a laundry liquid packaging defect detection method based on image recognition, comprising: Collecting surface image information of the laundry liquid packaging bag, determining extraction type defects according to different types of laundry liquid packaging, and extracting surface image information to obtain packaging defect data.

[0009] Classifying the packaging defect data according to the package defect number classification method to obtain type defect data, and detecting the type defect data using a hybrid imaging method to obtain defect detection data.

[0010] Analyzing the defect detection data to obtain defect analysis results, and comparing the defect analysis results with a preset defect determination threshold to determine whether the current laundry bag packaging is a defective product.

[0011] Recording the defect type, position, parameter and production batch information of the defective product to generate a defect statistical report, and generating a defect adjustment scheme according to the defect statistical report.

[0012] The present application provides a laundry liquid packaging defect detection method based on image recognition, and the step of obtaining surface image information comprises: Isolating and cleaning the detection area of the production line of the laundry liquid packaging, and matching the conveying speed of the conveyor belt with the beat of the production line.

[0013] According to the detection focus of the laundry liquid packaging, a combination of top, front and side industrial cameras is used, and corresponding light sources are matched according to the detection target.

[0014] According to the transmission speed of the laundry liquid packaging, a trigger signal is set, and the surface image information is obtained by synchronously collecting the images of the laundry liquid packaging through the camera combination.

[0015] The present application provides a laundry liquid packaging defect detection method based on image recognition, and the step of determining extraction type defects comprises: According to the form of the laundry liquid packaging, the laundry liquid packaging is classified according to the material characteristics and structural differences, and the core structure and detection focus area of each type of packaging are determined to obtain multiple types of packaging.

[0016] For each type of packaging, the corresponding extraction type defects are matched and analyzed from the structural weak points and functional failure scenarios.

[0017] The present application provides a laundry liquid packaging defect detection method based on image recognition, and the step of extracting the packaging defect data comprises: The surface image information is processed to remove noise, optimize brightness and contrast, correct geometry and crop regions to obtain surface image processing data.

[0018] Based on the distribution area and visual features of the identified defects on the packaging, the corresponding algorithm is selected to locate the defect area and obtain the location of the defect to be extracted.

[0019] Based on the judgment indicators for different defect types, corresponding defect parameters are extracted from heat-sealing defects, surface defects, printing defects, and structural and functional defects, and then integrated to obtain packaging defect data.

[0020] This invention provides a method for detecting defects in laundry detergent packaging based on image recognition. The steps for obtaining type defect data include: Packaging defect data is organized into standardized fields, duplicate data is removed, and parameters from different units are converted into industry-standard units to obtain processed defect data.

[0021] The primary classification rules are obtained by classifying the impact of defect data on packaging functionality, and the secondary classification rules are obtained by further subdividing the primary classification rules according to parameter characteristics and defect manifestations.

[0022] Based on primary and secondary classification rules, the defect data is categorized one by one to obtain type defect data.

[0023] This invention provides a method for detecting defects in laundry detergent packaging based on image recognition. The steps for obtaining defect detection data include: To address the detection needs of different types of defect data, hybrid imaging combination schemes are developed based on the characteristics of hybrid imaging technology.

[0024] The equipment is deployed according to the hybrid imaging scheme and the detection area, and resolution, synchronization, brightness and color calibration are performed.

[0025] Multimodal images are obtained by acquiring feature images of heat-sealing defects, printing defects, surface appearance defects, and structural and functional defects according to the hybrid imaging combination scheme.

[0026] Defect detection data is obtained by fusing multimodal images captured by different hybrid imaging schemes with feature parameters and classifying defects into defect levels according to preset standards.

[0027] This invention provides a method for detecting defects in laundry detergent packaging based on image recognition. The steps for obtaining defect analysis results include: The defect detection data is analyzed to output a grade statistics table based on the total number, proportion and grade distribution of defect types, and a spatiotemporal distribution heat map is drawn from the time and space dimensions.

[0028] For multimodal images of the same defect, the correlation between defect detection data is analyzed, and defect control directions are obtained by combining grade statistics tables and spatiotemporal distribution heat maps.

[0029] By analyzing process parameter matching, equipment status checks, and personnel operation traceability, the root causes of defects are identified and addressed in the defect control strategy.

[0030] Based on the degree of impact on production efficiency from defect levels, level statistics tables, and root cause assessments, risk levels for product quality and customer experience are classified.

[0031] This invention provides a method for detecting defects in laundry detergent packaging based on image recognition, and the steps for generating a defect statistical report include: Recording dimensions are set according to the defect itself, product attributes, and production scenario, and stored in a hierarchical directory of production batch-packaging type-defect type.

[0032] The records are categorized and summarized according to their dimensions, transforming scattered individual records into a dimensional dataset.

[0033] Based on dimensional datasets, we analyze time, parameters, and correlations to uncover potential patterns in defects and generate defect statistical reports.

[0034] This invention provides a method for detecting defects in laundry detergent packaging based on image recognition, and the steps for generating a defect adjustment plan include: A list of high-risk defects is compiled by extracting defects with high impact and risk level from the defect statistics report.

[0035] Design corresponding solutions for defects in the high-risk defect list by focusing on equipment maintenance, process optimization, and personnel training.

[0036] Based on technical feasibility, cost controllability, and production compatibility, the corresponding solutions are adjusted to generate defect adjustment plans.

[0037] On the other hand, the present invention provides an image recognition-based laundry detergent packaging defect detection system, comprising: The image inspection defect extraction module is used to collect surface image information of laundry detergent packaging bags, determine the type of defect to be extracted according to the different types of laundry detergent packaging, and extract packaging defect data from the surface image information.

[0038] The hybrid defect detection module is used to classify packaging defect data according to the package defect number regularization method to obtain type defect data, and to detect the type defect data using a hybrid imaging method to obtain defect detection data.

[0039] The defect threshold determination module is used to analyze defect detection data to obtain defect analysis results, and compare them with preset defect determination thresholds to determine whether the current laundry bag packaging is a defective product.

[0040] The defect statistics and analysis module is used to record the defect type, location, parameters, and production batch information of defective products, generate defect statistics reports, and generate defect adjustment plans based on the defect statistics reports.

[0041] This invention provides a method and system for detecting defects in laundry detergent packaging based on image recognition. Through the logic of "classifying by packaging form and material → focusing on structural weaknesses → matching with a dedicated extraction algorithm," it focuses on extracting heat-sealing defects for composite plastic bags and bottom forming defects for stand-up pouches, achieving "one extraction scheme per type of packaging," adapting to multiple packaging types and improving the accuracy of defect extraction. It employs a hybrid imaging scheme to cover all types of defects, combining multimodal parameter fusion and preset standards to classify defect levels, realizing a shift from "qualitative" to "quantitative" defect detection and improving detection reliability. Furthermore, it locates the root cause of defects through "process parameter matching + equipment status investigation + personnel operation traceability" and generates a statistical report containing "defect distribution, risk level, and improvement suggestions," thereby reducing production losses. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0043] Figure 1 This is one of the flowcharts of the image recognition-based method for detecting defects in laundry detergent packaging provided in this embodiment of the invention; Figure 2 This is the second schematic flowchart of the image recognition-based laundry detergent packaging defect detection method provided in this embodiment of the invention; Figure 3 This is a flowchart illustrating the image recognition-based laundry detergent packaging defect detection system provided in this embodiment of the invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0045] The following is combined Figures 1-3 This invention describes a method and system for detecting defects in laundry detergent packaging based on image recognition.

[0046] like Figure 1 As shown, the image recognition-based method for detecting defects in laundry detergent packaging provided in this embodiment of the invention includes: Collect surface image information of laundry detergent packaging bags, determine the type of defect to be extracted according to the different types of laundry detergent packaging, and extract packaging defect data from the surface image information.

[0047] The steps for obtaining surface image information include: The inspection area of ​​the laundry detergent packaging production line is isolated and cleaned, and the conveyor belt speed is matched with the production line cycle time.

[0048] Based on the key points of inspection for laundry detergent packaging, a combination of top, front, and side industrial cameras is used, with corresponding light sources matched to the inspection targets.

[0049] The top camera can be a 5-megapixel CMOS industrial camera (resolution 2592×1944) with an 8mm focal length low distortion lens (distortion rate <1%). The installation height is 30cm from the surface of the conveyor belt, and the lens is vertically downward aimed at the bag opening area to capture the integrity of the heat seal line of the bag opening and the alignment of the top of the label (such as whether the label is offset).

[0050] The front camera can be an 8-megapixel CMOS industrial camera (resolution 3264×2448) with a 12mm focal length lens. The installation height is 40cm above the surface of the conveyor belt, and the lens is at a 45° angle to the front of the packaging bag, focusing on capturing the printed patterns, surface stains or scratches on the bag.

[0051] The side camera can be a 5-megapixel CMOS industrial camera (resolution 2592×1944) with a 6mm focal length lens. The installation height is 20cm above the surface of the conveyor belt, and the lens is horizontally aligned with the side edge of the packaging bag to detect perforations (such as tiny holes) on the edge of the bag and whether the side seal is broken.

[0052] A trigger signal is set according to the conveying speed of the laundry detergent packaging, and surface image information is obtained by synchronously acquiring images of the laundry detergent packaging through a camera combination.

[0053] A diffuse reflection photoelectric sensor is installed at the conveyor belt inlet. When the laundry detergent bag enters the sensor's detection range (coverage ≥50%), the sensor outputs a high-level trigger signal to inform the system that "the package to be detected has arrived." A linkage logic between the sensor and cameras is established through an industrial PLC controller: after the sensor sends the trigger signal, the controller delays for 200ms and simultaneously sends shooting commands to three cameras, achieving synchronous acquisition of "top + front + side" images. This avoids packaging bag position shifts due to acquisition time differences, which could affect subsequent defect location.

[0054] The steps to determine the extraction type of defect include: Based on the packaging form of laundry detergent, combined with material characteristics and structural differences, laundry detergent packaging is classified, and the core structure and key testing areas of each type of packaging are determined, resulting in multiple packaging types.

[0055] Laundry bag packaging may include: ordinary composite plastic bags, stand-up pouches, tote bags, transparent PET bottles (extended bag types), etc.

[0056] For each type of packaging, we analyze and match the corresponding extracted type defects based on structural weaknesses and functional failure scenarios.

[0057] Ordinary composite plastic bags have no complex structure; the risks are concentrated in "heat-sealing performance (key to leak prevention), surface integrity (appearance), and printing compliance (information transmission)." The specific defect types to be identified are as follows: Heat sealing defects: Heat sealing is the core of leak prevention, and it is necessary to identify heat sealing defects such as leaks, incomplete seals, and wrinkles.

[0058] Surface defects: Since the bag body is in direct contact with the outside world, it is necessary to remove surface scratches, stains, and perforations.

[0059] Printing defects: The bag body printing contains key information such as product name and shelf life, and it is necessary to extract "missing / blurred text" and "color registration deviation".

[0060] The core value of stand-up pouches lies in their upright posture and convenient access. Their shortcomings should be identified by addressing issues such as bottom shaping (the foundation of upright posture), spout sealing (the key to leak prevention), and functionality (whether they can be poured out properly). Bottom forming defects: The bottom is crucial for uprightness, and it is necessary to identify "bottom deformation" and "bottom sealing defects".

[0061] Nozzle sealing defects: The nozzle is the retrieval channel, and it is necessary to remove "leaky weld seams in the nozzle" and "loose nozzle cover".

[0062] Functional defects: To ensure user experience, issues such as "mouthpiece blockage" and "leaking due to bag creases" need to be addressed.

[0063] The core function of a tote bag is "portability," and its shortcomings should be addressed by focusing on "the strength of the handle connections (whether it can bear weight), the integrity of the bag's structure (preventing damage), and the neatness of its appearance (user experience)." Handle strap connection defects: The handle strap is the key to load-bearing, and it is necessary to investigate "handle strap welding failure" and "handle strap sewing breakage".

[0064] Structural defects: The bag body must be sealed and intact, and "missing side seal" and "torn bag body" must be identified.

[0065] Appearance defects: To ensure user experience, issues such as "stains on the carrying strap" and "misaligned printing" need to be addressed.

[0066] The characteristics of transparent PET bottles are "transparency + rigidity". Defects to watch out for include "material flaws (affecting transparency), component assembly (compliance of labels and caps), and sealing performance (leakage prevention)". Material defects: The transparency of the bottle is a selling point, so it is necessary to extract "air bubbles" and "scratches" from the bottle.

[0067] Assembly defects: Labels and bottle caps are key components, and "label offset" and "label wrinkling" need to be identified.

[0068] Sealing defects: The bottle cap is the core of leak prevention, and it is necessary to identify the "bottle cap sealing gap" and "missing bottle cap sealing ring".

[0069] The steps for extracting packaging defect data include: The surface image information is processed by noise removal, brightness and contrast optimization, geometric correction and region cropping to obtain surface image processing data.

[0070] Noise Removal: For noise in the image caused by dust and camera sensor interference, a combination of "Gaussian filtering + median filtering" is used for processing - first, a 3×3 core Gaussian filter is used to smooth the image (reduce high-frequency noise), and then a 5×5 core median filter is used to remove salt and pepper noise (such as white dust spots on the surface of the bag) while preserving the details of defect edges (such as the edges of scratches and gaps in the seal).

[0071] Brightness and contrast optimization: Adjust image brightness according to packaging material—for matte composite bags, use the "grayscale stretching algorithm" to map the image grayscale range to 0-255 (enhancing the visibility of dark defects such as shallow scratches). For transparent PET bottles, use "histogram equalization" to balance differences in light transmission across the bottle.

[0072] Geometric correction: If the packaging bag tilts due to conveyor belt vibration, it is corrected by "edge detection + perspective transformation" - first, the Canny edge detection algorithm (threshold range 50-150) is used to extract the packaging outline and determine the tilt angle, and then perspective transformation is used to correct the tilted image to a horizontal state to ensure accurate measurement of the defect location.

[0073] Region cropping: Based on the detection area corresponding to the defect type, irrelevant parts of the image are cropped—for example, when extracting heat-sealing defects, only the heat-sealed area at the bag opening / bottom is retained. When extracting nozzle defects, only the nozzle area at the top of the stand-up pouch is retained, reducing interference from irrelevant areas in subsequent extractions.

[0074] Based on the distribution area and visual features of the identified defects on the packaging, the corresponding algorithm is selected to locate the defect area and obtain the location of the defect to be extracted.

[0075] Region localization based on grayscale differences (applicable to defects such as heat-sealed leaks and perforations): Heat-sealed leak areas (blank areas where the material is not fused) and bag perforation areas (dark areas under backlight) have significant grayscale differences from normal areas. By automatically calculating the segmentation threshold through an "adaptive threshold segmentation algorithm" (such as the Otsu thresholding method), the image is divided into "defect candidate areas" and "normal areas". For example, when extracting heat-sealed leaks in composite bags, the grayscale value of the normal heat-sealed area is low (high material density after fusion), while the grayscale value of the leaked area is high (blank area). The Otsu algorithm can accurately segment the leaked area candidate region.

[0076] Template-match-based region localization (applicable to defects such as label misalignment and nozzle blockage): The positions of components like labels and nozzles are relatively fixed. First, a "standard template for qualified components" is established (e.g., the position of a qualified label, the outline of a qualified nozzle). Then, a "normalized cross-correlation template matching algorithm" is used to search for the region in the preprocessed image that is most similar to the standard template. For example, when extracting label misalignment, the actual label position is located through template matching and compared with the standard template position to lock the misaligned label region. When extracting nozzle blockage, after matching the nozzle outline, the candidate region for foreign objects inside the nozzle is locked.

[0077] Edge-feature-based region localization (applicable to defects such as scratches and printing color misregistration): The linear edges of scratches and the color layer edges of printing color misregistration differ significantly from normal areas—using a combination of "Canny edge detection + Hough transform" for localization: First, use the Canny algorithm to extract edges, then use the Hough linear transform to detect the linear edges of scratches (locating the scratch area), or use the Hough circular transform to detect the circular features of printed patterns (such as the circular border of a logo) to locate the color layer edge area of ​​the color misregistration.

[0078] Based on the judgment indicators of different defect types, corresponding defect parameters are extracted from heat-sealing defects, surface defects, printing defects, and structural and functional defects, and then integrated to obtain packaging defect data.

[0079] Heat-sealing defects include heat-sealing omissions: within the located omission candidate area, the outline of the omission area is extracted using a "sub-pixel edge detection algorithm", and the length and width of the omission are calculated by converting "pixel coordinates to actual size".

[0080] Heat sealing with false sealing: Parameters are extracted through "grayscale value analysis" - multiple sampling points are selected in the heat-sealed area, the grayscale value of each point is measured, the grayscale difference between the false sealing area and the normal heat-sealed area is calculated, and the heat sealing strength value is estimated.

[0081] Heat-sealing wrinkles: Use the "morphological dilation algorithm" (3×3 kernel) to fill the tiny gaps in the wrinkle area, and then extract the actual area of the wrinkles through "contour area calculation" (the number of pixels in the wrinkle area × the actual area of a single pixel). At the same time, calculate the proportion of the wrinkle area in the total heat-sealing area.

[0082] Surface defects can include surface scratches: In the located scratch area, use the "skeleton extraction algorithm" to simplify the scratch edge into a center line, calculate the actual length of the center line, measure the scratch width at the same time, and extract the gray-scale difference value between the scratch area and the background area (to judge the depth of the scratch, the greater the gray-scale difference, the deeper the scratch).

[0083] Surface stains: Extract the stain area through "HSV color space segmentation" - set the color range of the stains in the HSV space, segment the stain candidate area, and calculate the actual area and the center coordinates of the stains (to locate the position of the stains on the bag body, such as whether it is in the front printing area).

[0084] Printing defects can include missing printed characters: Identify the characters in the printing area, compare the recognition results with the standard character library, locate the position coordinates of the missing characters, and calculate the missing area of the missing characters.

[0085] Printing blurriness: Select typical characters (such as "衣" "液") in the printed character area, and use the "edge sharpness calculation" to extract parameters - calculate the gray-scale gradient value of the character edge, and measure the actual thickness deviation of the character strokes at the same time.

[0086] Color registration deviation: Separate the printed image into RGB three channels (such as a red logo, a blue border), use "template matching" to locate the same feature point (such as the upper right corner of the logo) in each channel, record the pixel coordinates of the feature points in each channel, and calculate the offset of the feature points in different channels after converting to actual coordinates.

[0087] Structure and function defects can include label offset: Locate the center coordinates of the actual label through template matching, compare with the center coordinates of the standard label, calculate the horizontal offset and vertical offset, and calculate the distance deviation between the label edge and the packaging edge at the same time.

[0088] Nozzle blockage: In the located nozzle area, use "threshold segmentation" to extract the contour of the blocking foreign object, calculate the maximum diameter and area of the foreign object, and measure the distance between the foreign object and the nozzle liquid outlet at the same time (to judge whether the liquid outlet channel is completely blocked).

[0089] Bottom deformation: For the image of the bottom of the stand-up bag, use the "ellipse fitting algorithm" to fit the normal contour of the bottom (the standard bottom is an ellipse), then fit the actual bottom contour, calculate the long-axis deviation and short-axis deviation between the actual contour and the standard contour, and measure the actual upright angle of the bottom at the same time.

[0090] The extracted defect parameters are logically integrated according to "defect type - packaging type - quantitative parameters - image association information" to form standardized packaging defect data.

[0091] Packaging defect data is classified into types based on the package defect number regularization method to obtain type defect data. A hybrid imaging method is then used to detect these type defect data to obtain defect detection data. Type defect data can include appearance type defect data, structural type defect data, printing type defect data, and functional type defect data.

[0092] The steps to obtain type defect data include: Packaging defect data is organized into standardized fields, duplicate data is removed, and parameters from different units are converted into industry-standard units to obtain processed defect data.

[0093] The primary classification rules are obtained by classifying the impact of defect data on packaging functionality, and the secondary classification rules are obtained by further subdividing the primary classification rules according to parameter characteristics and defect manifestations.

[0094] Based on the impact of defects on packaging functionality, all defects are divided into four categories, each with clearly defined parameter characteristics, resulting in a primary classification rule: Heat-sealing defects: These affect the leak-proof performance of packaging, such as the length of the leaked seal, the gap between leaked seals, and the heat seal strength.

[0095] Surface appearance defects: These affect the visual experience of the packaging, such as the area of ​​surface stains, the length of scratches, and the color difference in printing.

[0096] Defects related to printing information: These affect the transmission of product information, such as missing characters, color misregistration, and deviations in stroke thickness.

[0097] Structural and functional defects: These affect the stability or functionality of the packaging structure, and the parameters are mostly "offset, angle, and blockage size". Under the primary category, subcategories are further subdivided according to "parameter characteristics + defect manifestations," with each subcategory having a clearly defined parameter judgment range, resulting in secondary category rules: Heat-sealing products are further categorized into "heat-sealed leaks" (leakage length ≥ 1 mm), "heat-sealed incomplete seals" (heat seal strength < 5 N / 15 mm), and "heat-sealed wrinkles" (wrinkle area ≥ 0.5 cm²). 2 ).

[0098] The surface appearance category is further divided into "surface scratches" (scratch length ≥ 2mm) and "surface stains" (stain area ≥ 0.3cm²). 2 ), "Pore through the bag" (perforation diameter ≥ 0.5mm).

[0099] The printed information category is further divided into "missing characters" (missing key characters), "blurry printing" (stroke thickness deviation > 10%), and "color registration deviation" (color layer offset ≥ 0.2mm).

[0100] The structural function category is further divided into "label offset" (offset > 2mm), "nozzle blockage" (foreign object diameter ≥ 0.3mm), and "bottom deformation" (tilt angle > 15°).

[0101] Based on primary and secondary classification rules, the defect data is categorized one by one to obtain type defect data.

[0102] like Figure 2 As shown, the steps for obtaining defect detection data include: To address the detection needs of different types of defect data, hybrid imaging combination schemes are developed based on the characteristics of hybrid imaging technology.

[0103] For heat-sealing defects: the core requirement is to "clearly present the gaps and fusion state of the heat-sealed edges, and eliminate interference from plastic reflections"—matching a hybrid solution of "polarization imaging + 2D visible light imaging": polarization imaging filters plastic reflections in the heat-sealed area, highlighting gaps in incomplete sealing and uneven fusion areas in partial sealing. 2D visible light imaging restores the color and texture of the heat-sealed edges, assisting in determining the shape of wrinkles.

[0104] For printing defects: the core requirement is "accurate reproduction of printed colors and capture of character details and color layer misalignment"—matching a hybrid solution of "standard color temperature visible light imaging + UV imaging": Standard color temperature (D65) visible light imaging ensures accurate reproduction of printed colors, facilitating the detection of color registration deviations and character blurring. UV imaging identifies invisible anti-counterfeiting printing (such as UV anti-counterfeiting codes on some laundry detergent packaging), supplementing the detection dimension of missing printing defects.

[0105] For surface appearance defects: the core requirement is to "highlight subtle surface differences and identify the light transmission characteristics of perforations"—matching a hybrid solution of "side-light imaging + backlight imaging": side-light imaging illuminates from the side of the packaging, creating a light-dark boundary line on the scratches and magnifying minor scratches (such as 0.5mm shallow scratches). Backlight imaging illuminates from the back of the packaging; the perforated areas appear as dark spots due to light transmission, creating a strong contrast with normal areas and preventing stains from being confused with perforations.

[0106] Structural and functional defects: The core requirement is to "capture three-dimensional deviations, internal foreign objects and structural morphology" - matching a hybrid solution of "3D laser imaging + 2D backlight imaging": 3D laser imaging scans the label, nozzle and bottom structure to obtain three-dimensional dimensions (such as the tilt angle of the label and the height difference of the bottom deformation).

[0107] The equipment is deployed according to the hybrid imaging scheme and the detection area, and resolution, synchronization, brightness and color calibration are performed.

[0108] Resolution calibration: The detection resolution of all imaging devices must match the defect accuracy requirements. Calibration is performed using a standard calibration board to ensure that the pixel-to-actual-size conversion factor is consistent across different devices.

[0109] Brightness and color calibration: The brightness of visible light, UV, and polarized imaging must be consistent. Standard color temperature visible light imaging requires white balance calibration using a color chart to ensure that the printing color reproduction error ΔE < 2, avoiding misjudgment due to color misregistration.

[0110] Synchronization calibration: Multiple imaging devices need to be synchronously triggered through an industrial PLC controller (trigger delay ≤ 1ms) to ensure that the same defect is captured at the same time in different images, and to avoid misalignment of the detection area due to time difference.

[0111] Multimodal images are obtained by acquiring feature images of heat-sealing defects, printing defects, surface appearance defects, and structural and functional defects according to the hybrid imaging combination scheme.

[0112] Defect detection data is obtained by fusing multimodal images captured by different hybrid imaging schemes with feature parameters and classifying defects into defect levels according to preset standards.

[0113] Using "defect ID + detection area coordinates" as the association key, detection data from different imaging methods are bound together. For each type of defect, multimodal image annotation information is added to the detection data, classifying defects into mild, moderate, and severe levels. For example, a heat seal leakage length of 1-2mm is considered mild, 2-5mm moderate, and >5mm severe. Similarly, a label offset of 1-2mm is considered mild, 2-3mm moderate, and >3mm severe.

[0114] The defect detection data is analyzed to obtain defect analysis results, which are then compared with preset defect judgment thresholds to determine whether the current laundry bag packaging is a defective product.

[0115] The steps to obtain defect analysis results include: The defect detection data is analyzed to output a grade statistics table based on the total number, proportion and grade distribution of defect types, and a spatiotemporal distribution heat map is drawn from the time and space dimensions.

[0116] For multimodal images of the same defect, the correlation between defect detection data is analyzed, and defect control directions are obtained by combining grade statistics tables and spatiotemporal distribution heat maps.

[0117] By analyzing process parameter matching, equipment status checks, and personnel operation traceability, the root causes of defects are identified and addressed in the defect control strategy.

[0118] The root cause may include: if heat sealing defects occur frequently in production line A between 8:00 and 10:00, the corresponding log shows "the temperature of the heat sealing machine rose from 155℃ to 165℃ during this period (preheating process)", verifying that "unstable temperature leads to uneven heat sealing fusion" as the root cause.

[0119] If there are many label offset defects associated with the labeling machine, the equipment calibration record shows that "the positioning sensor of this labeling machine has not been calibrated for more than 7 days (the standard calibration cycle is 3 days)," verifying that "the sensor misalignment caused the label positioning deviation" as the root cause.

[0120] Investigate the material batches associated with the defects (such as composite film, ink, and labels). For example, if the thickness deviation of the heat-sealing layer of a certain batch of composite film reaches 15%, the heat-sealing defect rate of the packaging using the corresponding batch of material increases by 3 times, verifying that "material quality fluctuation" is the root cause.

[0121] Based on the degree of impact on production efficiency from defect levels, level statistics tables, and root cause assessments, risk levels for product quality and customer experience are classified.

[0122] The impact on production efficiency can include: calculating the "rework / scrap cost" caused by defects. For example, a moderate heat-sealing defect requires manual rework (each piece takes 2 minutes and the labor cost is 0.5 yuan / piece). If 100 pieces are reworked in a certain batch, the direct cost is 50 yuan, which indirectly affects the efficiency of the production line (rework leads to a 5% decrease in production capacity).

[0123] Severe perforation defects require direct scrapping (scrapping rate 1.2%, cost per bag 2 yuan). In one batch, 24 bags were scrapped, resulting in a direct loss of 48 yuan. Defects are classified into "high impact (>500 yuan / batch), medium impact (100-500 yuan / batch), and low impact (<100 yuan / batch)" according to the "cost loss amount".

[0124] Risk level classification can include: determining whether the defect violates industry standards or company regulations. For example, "missing production date in printing" violates the "Product Quality Law of the People's Republic of China" and is classified as "high compliance risk." "Minor surface scratches" do not affect use and are classified as "low compliance risk."

[0125] Based on historical data and customer feedback, for example, "heat sealing failure" once led to 30% of customer complaints (leakage issues), which is classified as "high complaint risk". "Slight color misalignment" has no customer complaint records and is classified as "low complaint risk". Taking into account "compliance risk + complaint risk", the defects are divided into "high risk (requires immediate handling), medium risk (requires handling within a time limit), and low risk (continuous monitoring)".

[0126] Record the defect type, location, parameters, and production batch information of defective products to generate a defect statistics report, and generate a defect adjustment plan based on the defect statistics report.

[0127] The steps to generate a defect statistics report include: Recording dimensions are set according to the defect itself, product attributes, and production scenario, and stored in a hierarchical directory of production batch-packaging type-defect type.

[0128] Recording dimensions may include: unique identifier of defective products: including "production serial number" and "packaging type".

[0129] Defect type: Fill in according to the previously classified "first-level category + second-level category" (such as "heat sealing defect - heat sealing omission" or "printing defect - color registration deviation").

[0130] Defect location: Precise to the specific area of ​​the packaging, using coordinates + area name.

[0131] Defect Quantification Parameters: Fill in the core parameters of the hybrid imaging detection, in the format of "parameter name + value + unit" (e.g., "missing seal length 2.5mm, missing seal width 0.3mm", "color offset 0.25mm, character blur gradient value 30").

[0132] Defect level: Labeled as “minor / moderate / severe” according to the previous assessment (e.g., “moderate” or “severe”).

[0133] Production batch information: including "production batch number", "production line number", and "production date".

[0134] Testing information includes "testing equipment number", "testing time", and "testing personnel".

[0135] The records are categorized and summarized according to their dimensions, transforming scattered individual records into a dimensional dataset.

[0136] For example: Using the "production batch number" as the core, summarize the overall defect situation of the batch: basic indicators: total output of the batch, number of defective products, and total defect rate (number of defects / total output × 100%).

[0137] Defect distribution: The number and percentage of various defects (heat sealing, printing, surface, structure) (e.g., "The total defect rate of batch 20240520-B03 is 4.2%, of which 18 pieces have heat sealing defects (accounting for 42.9%) and 12 pieces have printing defects (accounting for 28.6%)").

[0138] Grade distribution: the number and percentage of minor / moderate / severe defects (e.g., "15 minor defects, 13 moderate defects, and 2 severe defects"). Create a "Production Batch Defect Summary Table," marking each batch as an "abnormal batch."

[0139] Locate the problematic links in the production process: calculate the defect rate of each production line (e.g., "Line A defect rate 3.5%, Line B defect rate 5.2%").

[0140] Compile statistics on the main defect types and high-incidence workstations for each production line. Create a "Production Line-Workstation Defect Summary Table" to identify "high-risk production lines / workstations".

[0141] Based on dimensional datasets, we analyze time, parameters, and correlations to uncover potential patterns in defects and generate defect statistical reports.

[0142] Defect rate changes are statistically analyzed by "hour / day / week": for example, "the defect rate reaches 6% from 8:00 to 10:00 every day, and is stable at 2%-3% at other times" and "the defect rate is higher on Mondays than on other workdays", suggesting that "insufficient equipment warm-up in the morning and unstable parameters after restarting after shutdown on weekends" are potential causes.

[0143] Plot a frequency distribution histogram for key parameters (such as heat seal missing length and color offset) to determine whether the parameters conform to a normal distribution. Analyze the correspondence between parameter values ​​and defect levels. For example, “missing seal length <2mm is mostly a minor defect (accounting for 90%), 2-5mm is mostly a moderate defect (accounting for 85%), and >5mm is a severe defect (100%)” to verify the rationality of the level classification.

[0144] By correlating defect data with production processes and equipment status, we can pinpoint root cause clues: process parameter correlation: for example, "when the heat sealing temperature at station 3 of line B is <155℃, the incomplete heat sealing rate reaches 30%. When the temperature is 155-165℃, the incomplete sealing rate drops to 5%", indicating that "insufficient temperature" is directly related to incomplete sealing.

[0145] Equipment status correlation: For example, "When the calibration interval of the labeling machine's positioning sensor exceeds 7 days, the label offset defect rate rises to 8%. After calibration, it drops to 2%", verifying the impact of "equipment calibration cycle".

[0146] like Figure 3 As shown, based on the same general inventive concept, this invention also protects an image recognition-based laundry detergent packaging defect detection system, the detection system comprising: The image inspection defect extraction module is used to collect surface image information of laundry detergent packaging bags, determine the type of defect to be extracted according to the different types of laundry detergent packaging, and extract packaging defect data from the surface image information.

[0147] The hybrid defect detection module is used to classify packaging defect data according to the package defect number regularization method to obtain type defect data, and to detect the type defect data using a hybrid imaging method to obtain defect detection data.

[0148] The defect threshold determination module is used to analyze defect detection data to obtain defect analysis results, and compare them with preset defect determination thresholds to determine whether the current laundry bag packaging is a defective product.

[0149] The defect statistics and analysis module is used to record the defect type, location, parameters, and production batch information of defective products, generate defect statistics reports, and generate defect adjustment plans based on the defect statistics reports.

[0150] This embodiment provides a method and system for detecting defects in laundry detergent packaging based on image recognition. By employing hybrid imaging technology and deep learning algorithms, it accurately detects various defects in laundry detergent packaging, including irregular defects, thus improving detection accuracy and reliability. The automated image acquisition and processing workflow, combined with a highly efficient deep learning model, significantly increases detection speed, meeting the demands of high-speed production lines and enhancing overall production efficiency. Furthermore, by generating defect statistical reports and defect adjustment plans, it helps production managers quickly pinpoint the causes of defects, adjust production parameters in a timely manner, optimize production processes, reduce the production of defective products, and improve product quality and customer satisfaction. Through multi-dimensional data recording and analysis, it generates detailed defect statistical reports, providing strong data support for production management and facilitating refined management of the production process. It significantly improves detection accuracy and efficiency, enhances the system's adaptability and robustness, and provides strong support for optimizing production processes.

[0151] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting defects in laundry detergent packaging based on image recognition, characterized in that, include: Collect surface image information of laundry detergent packaging bags, determine the type of defect to be extracted according to different types of laundry detergent packaging, and extract packaging defect data from the surface image information; The packaging defect data is classified according to the package defect number regularization method to obtain type defect data, and the type defect data is detected by the hybrid imaging method to obtain defect detection data; The defect detection data is analyzed to obtain defect analysis results, which are then compared with a preset defect judgment threshold to determine whether the current laundry bag packaging is a defective product. Record the defect type, location, parameters, and production batch information of defective products to generate a defect statistics report, and generate a defect adjustment plan based on the defect statistics report.

2. The method for detecting defects in laundry detergent packaging based on image recognition according to claim 1, characterized in that, The steps for obtaining the surface image information include: The inspection area of ​​the laundry detergent packaging production line is isolated and cleaned, and the conveyor belt speed is matched with the production line cycle time. Based on the key points of inspection for laundry detergent packaging, a combination of top, front, and side industrial cameras is used, with corresponding light sources matched to the inspection targets; A trigger signal is set according to the conveying speed of the laundry detergent packaging, and the surface image information is obtained by synchronously acquiring images of the laundry detergent packaging through the camera combination.

3. The method for detecting defects in laundry detergent packaging based on image recognition according to claim 1, characterized in that, The steps for determining the extraction type of defect include: Based on the packaging form of laundry detergent, combined with material characteristics and structural differences, laundry detergent packaging is classified, and the core structure and key testing areas of each type of packaging are determined to obtain multiple packaging types; For each type of packaging, we analyze and match the corresponding extracted type defects based on structural weaknesses and functional failure scenarios.

4. The method for detecting defects in laundry detergent packaging based on image recognition according to claim 1, characterized in that, The steps for extracting the packaging defect data include: The surface image information is subjected to noise removal, brightness and contrast optimization, geometric correction and region cropping to obtain surface image processing data; Based on the distribution area and visual features of the identified defects on the packaging, the corresponding algorithm is selected to locate the defect area and obtain the location of the defect to be extracted. Based on the judgment indicators for different defect types, corresponding defect parameters are extracted from heat-sealing defects, surface defects, printing defects, and structural and functional defects, and then integrated to obtain the packaging defect data.

5. The method for detecting defects in laundry detergent packaging based on image recognition according to claim 1, characterized in that, The steps for obtaining the defect data of the aforementioned type include: The packaging defect data is organized according to a unified field, duplicate data is deleted, and parameters of different units are converted into industry-standard units to obtain processed defect data. The impact of the processed defect data on packaging function is classified to obtain a primary classification rule, and the primary classification rule is further subdivided according to parameter characteristics and defect manifestations to obtain a secondary classification rule; Based on the primary classification rules and the secondary classification rules, the processed defect data are classified one by one to obtain the type of defect data.

6. The method for detecting defects in laundry detergent packaging based on image recognition according to claim 1, characterized in that, The steps for obtaining the defect detection data include: To address the detection requirements of the aforementioned types of defect data, hybrid imaging combination schemes corresponding to different types of defect data are formulated based on the characteristics of hybrid imaging technology. The device is deployed according to the hybrid imaging combination scheme and the detection area, and resolution, synchronization, brightness and color calibration are performed; Multimodal images are obtained by acquiring feature images of heat-sealing defects, printing defects, surface appearance defects, and structural and functional defects according to the hybrid imaging combination scheme. The defect detection data is obtained by fusing multimodal images captured by different hybrid imaging schemes with feature parameters and classifying defects into defect levels according to preset standards.

7. The method for detecting defects in laundry detergent packaging based on image recognition according to claim 6, characterized in that, The steps to obtain the defect analysis results include: The defect detection data is analyzed from the total number, proportion and grade distribution of defect types, and a grade statistics table is output. A spatiotemporal distribution heat map is drawn from the time and space dimensions. For multimodal images of the same defect, the correlation between defect detection data is analyzed, and the defect control direction is obtained by combining the grade statistics table and the spatiotemporal distribution heat map. By analyzing the root causes of defects in the defect control direction through process parameter matching, equipment status investigation, and personnel operation tracing, the following methods are used: Based on the defect level, the level statistics table, and the impact of the root cause assessment on production efficiency, risk levels for product quality and customer experience are classified.

8. The method for detecting defects in laundry detergent packaging based on image recognition according to claim 7, characterized in that, The steps for generating the defect statistics report include: Recording dimensions are set according to the defect itself, product attributes, and production scenario, and stored in a hierarchical directory of production batch-packaging type-defect type; Based on the recorded dimensions, the scattered individual records are categorized and summarized, transforming them into a dimensional dataset; Based on the dimensional dataset, the potential patterns of defects are analyzed from the perspectives of time, parameters, and correlations to obtain the defect statistical report.

9. The method for detecting defects in laundry detergent packaging based on image recognition according to claim 1, characterized in that, The steps for generating the defect adjustment plan include: A high-risk defect list is constructed by extracting defects with high impact and risk level from the defect statistics report. Design corresponding solutions for the defects in the high-risk defect list, focusing on equipment maintenance, process optimization, and personnel training; The defect adjustment plan is generated by adjusting the corresponding solution based on technical feasibility, cost controllability, and production compatibility.

10. A laundry detergent packaging defect detection system based on image recognition, comprising the laundry detergent packaging defect detection method based on image recognition as described in any one of claims 1 to 9, characterized in that, The detection system includes: The image inspection defect extraction module is used to collect surface image information of laundry detergent packaging bags, determine the type of defect to be extracted according to different types of laundry detergent packaging, and extract packaging defect data from the surface image information. The hybrid defect detection module is used to classify the packaging defect data according to the package defect number regularization method to obtain type defect data, and to detect the type defect data using a hybrid imaging method to obtain defect detection data. The defect threshold determination module is used to analyze the defect detection data to obtain defect analysis results, and compare them with the preset defect determination threshold to determine whether the current laundry bag packaging is a defective product. The defect statistics and analysis module is used to record the defect type, location, parameters, and production batch information of defective products, generate a defect statistics report, and generate a defect adjustment plan based on the defect statistics report.

Images