Data management methods for smart warehousing

Abstract

This invention discloses a data management method for intelligent warehousing, relating to the field of intelligent warehousing information technology. The method includes: acquiring a grayscale image of the vehicle's imprinted code area and a recognition string; performing robust illumination normalization on the image, extracting positive and negative polarity skeletons, and generating bistable reliability weights and structural signatures; calculating the weighted similarity between the structural signature and the vehicle's main object based on the weights, determining whether to merge or create a new main object through a competitive decision-making process, and using the recognition string as an observation alias; updating the template signature using the weights and associating business data with the unique identifier of the vehicle's main object. This invention solves the problem of primary key splitting and dual identity coexistence for the same vehicle caused by the bistable polarity of the imprinted code, ensuring the accuracy of warehousing data.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing information technology, and in particular to a data management method for intelligent warehousing. Background Technology

[0002] In smart warehousing scenarios, reusable containers such as crates, bins, and pallets are widely used in various stages of goods storage, sorting, and transfer. To achieve precise management of these containers, traceability of business records, and improved warehousing efficiency, embossed codes are typically placed on the surface of the containers as their unique identifiers. Identifying these embossed codes confirms the container's identity and associates various warehousing business records with their corresponding containers. However, in actual warehousing operations, embossed codes are affected by changes in lighting angle and intensity, exhibiting two stable forms: shadow groove development and highlight ridge development. This polarity reversal phenomenon in visual presentation is an inherent optical property of such non-printed markings. Furthermore, long-term cyclical use of containers can lead to wear and tear on the embossed codes, as well as surface stains obscuring them, resulting in differences in the identification results of the embossed codes for the same container under different observation scenarios. This, in turn, affects the uniformity and accuracy of warehousing data management and reduces the level of automation and intelligence in warehousing operations.

[0003] Existing technologies typically utilize optical character recognition (OCR) algorithms to analyze captured images and directly use the output strings as primary key indexes for asset ledgers and business transaction logs. However, conventional recognition logic struggles to effectively address the appearance differences caused by polarity reversal, leading to different character recognition results for the same physical vehicle under varying lighting conditions. Directly equating the recognized string with a unique identity results in two or more sets of encoding for the same vehicle at the system data level, causing errors such as primary key splitting and dual identities. This not only severs the business data association chain of the vehicle but also leads to inconsistencies between asset ledgers and actual inventory status, severely impacting the accuracy of warehouse data management and end-to-end traceability capabilities. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies that directly use identification strings as vehicle identification, which lead to the splitting of the primary key of the same vehicle, the coexistence of multiple identities, and the breakage of business data association due to the fluctuation of identification results caused by the bistable phenomenon of imprinted code polarity. Therefore, this invention proposes a data management method for intelligent warehousing.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: Data management methods for smart warehousing include: S1. Obtain the grayscale image of the vehicle imprint code area and the recognition string corresponding to this visual code reading; S2. Perform robust illumination normalization on the grayscale image, decompose the normalized image and extract the positive and negative polarity skeletons, and generate bistable reliability weights and structural signatures based on the positive and negative polarity skeletons. S3. Based on the bistable reliability weight, calculate the weighted similarity between the structural signature and the template signatures of each vehicle master object in the pre-stored set of vehicle master objects; S4. Based on weighted similarity, make a competitive decision on the attribution of this observation, determine whether to merge this observation into an existing vehicle master object or a newly created vehicle master object, and write the identification string as the observation alias into the vehicle master object selected by the competitive decision on attribution; S5. Update the template signature of the selected vehicle master object according to the bistable reliability weight, and associate the warehousing business data with the unique identifier of the vehicle master object.

[0006] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs a polarity-independent structural signature and bistable reliability weight by performing robust illumination normalization processing on grayscale images and extracting positive and negative polarity skeletons. It utilizes the projection contour features of the joint skeleton to eliminate visual difference interference caused by the polarity reversal of the development of shadow grooves and highlight ridges in imprinted codes, thus achieving stable identity anchoring across illumination geometry conditions. At the same time, based on the reliability weight calculated by the spatial fit of the positive and negative skeletons, the credibility of structural features is quantified. Therefore, without relying on a specific illumination steady state, the consistency of structural representation of the same physical vehicle under different work positions and postures is ensured, avoiding misjudgment of identity due to changes in appearance from the source of identification.

[0007] 2. This invention establishes a thresholdless competitive decision-making mechanism based on weighted similarity and the Softmax function. It downgrades visually recognized strings to observation aliases and converges them to a unique vehicle master object. By maintaining an alias mapping table, it achieves a unified index of the same master object ID for multiple recognition results, completely eliminating data pollution caused by primary key splitting and dual identities. It uses bistable reliability weights to perform weighted average updates on the master object template signature, enabling the template to dynamically integrate high-reliability observation features and resist low-quality data interference, ensuring long-term accurate association and full-process traceability of asset ledgers, business flow, and actual vehicle physical status. Attached Figure Description

[0008] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart illustrating the data management method for intelligent warehousing provided by the present invention; Figure 2 A functional block diagram of the data management system for intelligent warehousing provided by the present invention; Figure 3 This is a schematic diagram of the process for generating a polarity-independent imprinted signature provided by the present invention. Detailed Implementation

[0009] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0010] Example: This example provides a data management method for intelligent warehousing. See [link to example]. Figure 1 Specifically, including: In an embodiment of the present invention, acquiring the grayscale image of the vehicle imprint code area and the recognition string corresponding to this visual code reading includes: In the intelligent warehouse, industrial area array cameras are fixedly deployed on the sides and above the conveyor paths of the vehicles at the inbound conveyor lines, sorting ports, and shelving ports. Each workstation is equipped with at least two cameras at different installation angles to cover the tilting and flipping postures that may occur during the movement of the imprinted codes. The cameras are CMOS industrial cameras with a resolution of no less than 1920×1080 pixels and a frame rate of no less than 30fps, equipped with fixed-focus lenses with a focal length of 12mm. The optical axis of the lens is at an angle of 30 to 60 degrees to the conveyor plane of the vehicle to optimize the imaging effect of the imprinted codes. At the same time, a ring shadowless light source and a strip light source are installed next to each camera. The ring shadowless light source is used to weaken the specular reflection caused by the material characteristics of the vehicle surface, and the strip light source is used to specifically enhance the contrast between the imprinted codes and the vehicle substrate. The brightness of both light sources can be adaptively adjusted according to the material of the vehicle, the degree of surface wear, and the ambient lighting environment, with the adjustment range controlled between 500 and 3000 lux.

[0011] As the vehicle passes through each workstation, the photoelectric sensor beside the conveyor line detects the vehicle signal and triggers the camera to start acquiring data. The trigger signal is synchronously transmitted to the image acquisition card, which controls the camera to accurately focus on the imprinted code area on the surface of the vehicle. During the focusing process, the Laplacian gradient method, which is well-known in the field, is used as the image sharpness evaluation algorithm to detect the sharpness value of the image in the imprinted code area in real time. The preset sharpness threshold is determined through multiple sets of comparative experiments and is set in combination with camera parameters, vehicle material, and minimum recognizable sharpness requirements of the imprinted code characters. When the detected sharpness value is higher than the sharpness threshold, focusing is completed and the shooting operation is performed. During the shooting process, the exposure time is automatically adjusted to control the exposure time between 10 and 50 microseconds to avoid motion blur caused by the movement of the vehicle.

[0012] After the camera acquires a complete image of the vehicle surface, an image cropping algorithm is used to perform dual positioning based on the vehicle's outline and the preset installation position of the imprinted code. The area where the imprinted code is located is automatically selected and cropped to obtain a grayscale image of the imprinted code area. After cropping, the imprinted code and a small amount of surrounding background area are retained to reduce irrelevant interference. At the same time, a median filtering algorithm known in the field is used to remove salt-and-pepper noise and Gaussian noise in the grayscale image to improve image quality and ensure the accuracy of subsequent character recognition.

[0013] After grayscale image preprocessing, the embedded visual code reading module performs character recognition. This module integrates a deep learning-based character recognition algorithm adapted to the existing AI recognition scheme for the bistable morphology of embossed code development (shadow groove development and highlight ridge development). Specifically, the grayscale image is first stretched to further amplify the grayscale difference between the embossed code characters and the carrier substrate. Then, a contour extraction algorithm separates the embossed code character contours. The extracted character contours are normalized to unify the character size to a preset standard size. Finally, character matching logic compares the image with a preset character template library, outputting the recognized string corresponding to this visual code reading. If the recognition process is hindered by wear or stains on the embossed code... If characters are blurred, missing, or misidentified due to changes in lighting, resulting in the inability to output a valid recognition string, the reading result is recorded as invalid. Simultaneously, the grayscale image of the imprinted code area, acquisition time, workstation number, and carrier conveyor speed are retained and transmitted along with the invalid recognition result to the data processing unit for subsequent processes. If recognition is successful, the recognition string is directly bound to the corresponding grayscale image and associated information, and synchronously transmitted to the data processing unit. This completes the acquisition of the grayscale image of the carrier imprinted code area and the recognition string corresponding to this visual reading, providing basic data support for subsequent robust lighting normalization, structural signature construction, and main object merging steps.

[0014] It should be noted that the embossed code refers to a permanent mark made by embossing, raised lettering, or molding processes on the surface of recyclable carriers (turnover boxes, bins, pallets, etc.). It is made of the same material as the carrier's substrate and serves as the core identifier for the carrier. Its visibility depends on shadows and specular reflection, and it easily exhibits two stable forms: shadow groove development and highlight ridge development. The identification string refers to the character sequence output after performing visual code reading processing on the grayscale image of the embossed code area of ​​the carrier. It is used to express the embossed code reading value under this image observation in string form. The identification string corresponds one-to-one with the acquired image and allows for stable misreadings due to the difference between the two stable development states dominated by highlights and shadows, or differences in output caused by character polarity reversal. In the data management process of this invention, the identification string is written as an observation alias into the carrier's main object and used for alias mapping index, rather than as the unique identifier of the carrier's main object, thereby avoiding data pollution caused by the coexistence of dual identities for the same physical carrier under bistable conditions.

[0015] It should be noted that the ring-shaped shadowless light source is an auxiliary lighting component used with industrial cameras. Its ring-shaped structure provides uniform, shadowless illumination. Its core function is to reduce specular reflections caused by the material properties of the carrier surface (such as metal and plastic), minimizing the interference of reflected light on the clarity of the embossed code image, thus meeting the imaging requirements of areas with low contrast between the embossed code and the substrate. The strip-shaped supplementary light source is an auxiliary lighting component used in conjunction with the ring-shaped shadowless light source. Its strip-shaped structure allows for targeted projection of light onto the embossed code area. Its core function is to enhance the brightness difference between the embossed code characters and the carrier substrate, compensating for the insufficient targeted illumination intensity of the ring-shaped shadowless light source, and assisting in subsequent image preprocessing and character recognition. The bistable morphology of the embossed code refers to the two stable appearance morphologies of the same embossed code area under different lighting geometry conditions. One is development with shadow grooves, where the grooves formed by the embossing create the character outline through shadow; the other is development with highlight ridges, where the protrusions formed by the embossing create the character outline through light reflection. These two morphologies are visually polar opposites and can both potentially trigger stable character recognition output.

[0016] It should be noted that the image sharpness evaluation algorithm is used to detect the sharpness of the embossed code area image and determine the image sharpness. This embodiment uses the well-known Laplacian gradient method, whose core principle is to extract image edge information by calculating the second derivative of the pixel grayscale value in the image. The sharper the edge, the larger the magnitude of the second derivative, and the higher the corresponding sharpness value, thereby realizing the quantitative detection of the sharpness of the embossed code image and providing a basis for focusing operations. The sharpness threshold is the critical sharpness value for judging whether the embossed code area image meets the requirements of subsequent processing and recognition. It is determined by calibration through multiple sets of control experiments. The calibration process takes into account the parameters of the industrial camera, the material of the carrier, the size of the embossed code characters, and the minimum recognizable sharpness requirement. When the sharpness value detected by the image sharpness evaluation algorithm is higher than this threshold, it indicates that the embossed code image is clear, and focusing and shooting can be completed to ensure the accuracy of subsequent character recognition.

[0017] It should be noted that the deep learning character recognition algorithm adapted to the bistable form of the embossed code in this embodiment adopts a lightweight convolutional neural network architecture, balancing real-time performance and recognition accuracy in embedded deployment. It is optimized to address issues such as polarity reversal of the embossed code, carrier wear, dirt occlusion, and uneven lighting, and can stably output the recognized string. The specific implementation is as follows: The algorithm is based on an improved MobileNetV2 architecture that constructs a lightweight three-stage architecture of feature extraction, feature fusion, and character prediction. Redundant convolution and pooling layers are removed to adapt to embedded computing power, and a new dual-branch feature fusion and attention mechanism module is added. It includes a seven-layer structure including an input layer and a preprocessing adaptation layer, and is compatible with two steady-state forms of imprinted codes to achieve accurate recognition.

[0018] Input layer: The input is a grayscale image of the imprint code that has been denoised in the early stage. It is uniformly normalized to 32×128 pixels and the grayscale value is mapped to the 0-1 range. No color conversion is required. It takes into account both computing power optimization and grayscale difference preservation, and is adapted to the characteristics of bistable imaging.

[0019] Preprocessing adaptation layer: Perform grayscale stretching and adaptive threshold segmentation: expand the contrast between characters and the substrate through linear transformation, automatically determine the segmentation threshold using the Otsu method, convert the grayscale image to a binary image, preserve the character outline and filter out slight noise, and adapt to the problem of blurred character edges caused by wear.

[0020] Basic feature extraction layer: An improved MobileNetV2 is adopted, which replaces the traditional convolution with depthwise separable convolution to reduce the number of parameters and improve speed. It contains 6 convolutional blocks (each containing depthwise separable convolution, batch normalization and ReLU6 activation function) to extract low-level edge texture and mid-to-high-level overall structural features, while taking into account the commonalities of bistable character structure.

[0021] Dual-branch feature enhancement layer: The dual-branch layer adapts to the bistable morphology: the 3×3 convolutional kernel branch enhances the shadow contour features of the shadow groove development, and the 1×3 convolutional kernel branch captures the fine texture of the highlight ridge development. After parallel processing, the targeted enhanced feature map is output to ensure that the bistable features are fully extracted.

[0022] Feature fusion layer: Combining channel attention mechanism with feature concatenation to achieve fusion, adaptively adjusting feature channel weights, strengthening character features and suppressing background interference, and obtaining a global feature map after concatenation, ensuring that complete character features can be extracted under bistable morphology, avoiding misidentification.

[0023] Character prediction layer: It adopts a temporal classification structure, eliminating the need for manual character segmentation. It captures character contextual relationships through a bidirectional long short-term memory network and maps them to the character category space (including 0-9 and AZ) through a fully connected layer, outputting the character prediction probability distribution, which is suitable for scenarios where characters are closely connected.

[0024] Post-processing layer: The predicted probability distribution is converted into a character sequence through CTC decoding. A confidence threshold of ≥0.7 is set to filter low-confidence characters. Invalid sequences are removed by combining the preset character length, and the accurately recognized string is output.

[0025] Algorithm training process: Supervised learning is used to optimize model parameters and adapt them to the intelligent warehousing scenario. The specific training steps are as follows: Training dataset construction: Construct an embossed code dataset containing bistable morphology, multiple materials, multiple wear levels, and multiple lighting scenarios, label real character sequences, and divide it into training set, validation set, and test set in a 7:2:1 ratio to ensure data diversity and representativeness and support model generalization.

[0026] Data augmentation: Enhancement operations such as illumination perturbation, rotation, scaling, noise addition, horizontal flipping, and polarity flipping are performed on the training set to simulate complex scenes inside the warehouse, avoid model overfitting, and improve generalization ability.

[0027] Loss function selection: We adopt a weighted fusion of CTC loss and contrastive loss (weight ratio 7:3). CTC loss optimizes the character prediction accuracy, while contrastive loss enhances the common features of the same character under bistable conditions, thus accurately optimizing the model parameters.

[0028] Training parameters and strategy settings: The Adam optimizer is used (initial learning rate 0.001, cosine annealing decays to 1e-6), batch size 32, 50 epochs of iteration with an early stopping mechanism (stop if there is no improvement in the accuracy of the validation set after 5 epochs), and the optimal model is saved for deployment.

[0029] The model is embedded with an embedded visual code reading module. The inference process is connected to the previous image acquisition and preprocessing. It sequentially completes image normalization, feature extraction, fusion, prediction and post-processing, outputs the recognition string and binds the associated information to the data processing unit. The inference time is ≤50 milliseconds, which meets the requirements of high-speed transmission.

[0030] Optimizations targeting core pain points include: first, dual-branch extraction and contrast loss training to achieve bistable unified recognition; second, expanding the coverage of wear and tear scenarios and adding an attention mechanism to improve the ability to recognize damaged characters; and third, using scenario-based data augmentation to reduce the impact of lighting and posture changes and ensure stable recognition across multiple workstations.

[0031] In an embodiment of the present invention, a robust illumination normalization process is performed on the grayscale image, the normalized image is decomposed and a positive and negative polarity skeleton is extracted, and a bistable reliability weight and structural signature are generated based on the positive and negative polarity skeletons, including: Calculate the median gray level and the median absolute gray level deviation of a grayscale image; Calculate the difference between the grayscale image and the median grayscale value, and obtain the normalized image based on the ratio of the difference to the median of the absolute grayscale deviation. In detail, the grayscale image obtains the grayscale values ​​of all pixels. It iterates through each pixel, collecting the grayscale value for each pixel to form a grayscale value set. This set is then sorted in ascending order. The median grayscale value is determined based on this sorted set. If the number of pixels in the set is odd, the middle grayscale value is selected as the median. If the number of pixels is even, the arithmetic mean of the two middle grayscale values ​​is selected as the median. This median grayscale value is used to characterize the overall brightness baseline of the grayscale image, reducing the impact of extremely bright or dark pixels. Interference in overall brightness assessment; then the median of grayscale absolute deviation is calculated. For each collected pixel grayscale value, the absolute difference between it and the grayscale median obtained above is calculated. All absolute differences constitute an absolute deviation set. The absolute deviation set is sorted in ascending order. The median of the absolute deviation set is selected as the median of grayscale absolute deviation using the same rules as for determining the grayscale median. This value is used to quantify the dispersion of pixel grayscale values ​​in the grayscale image relative to the overall brightness benchmark, adapting to the grayscale distribution fluctuation scene caused by uneven illumination and material reflection in the embossed code grayscale image.

[0032] The difference between the grayscale image and the median grayscale value is calculated. This is done by iterating through each pixel of the grayscale image and subtracting the median grayscale value from the grayscale value of each individual pixel. This difference highlights the grayscale difference between the embossed code character and the carrier substrate while preserving the structural features of the embossed code. Finally, a normalized image is obtained based on the ratio of the above difference to the median of the absolute grayscale deviation. A very small positive number is introduced during the calculation to avoid a zero denominator. This very small positive number is selected as a value no greater than 10 to the power of -6 to ensure the stability of the calculation process. For each pixel... The corresponding difference is divided by the sum of the median of the absolute grayscale deviation and the smallest positive number to obtain the normalized grayscale value of each pixel. The normalized grayscale values ​​of all pixels constitute the normalized image. This normalized image can effectively reduce the influence of brightness drift and local highlights on pixel amplitude, unify the illumination distribution of the imprinted code area, highlight the character outline and structural details of the imprinted code, and provide high-quality image data support for subsequent steps such as imprinted code polarity decomposition, structural signature construction and vehicle identification. It is suitable for processing imprinted codes of different steady-state forms under multiple illumination conditions in the warehouse.

[0033] The portion of the normalized image that is greater than zero is taken as the positive polarity image, and the absolute value of the portion that is less than zero is taken as the negative polarity image. The positive polarity map and the negative polarity map are binarized and the skeleton is refined respectively to obtain the positive polarity skeleton and the negative polarity skeleton. Calculate the distance transformation diagrams for the positive and negative polarity frameworks respectively; In detail, the process begins with positive and negative polarity image splitting. This involves traversing every pixel in the normalized image, extracting all pixels with normalized grayscale values ​​greater than zero, retaining these values, and setting the grayscale values ​​of the remaining pixels to zero, thus obtaining a positive polarity image. This positive polarity image corresponds to the characteristic region of the highlight ridge development pattern of the embossed code, highlighting the highlight structure formed by the raised characters. Simultaneously, all pixels with normalized grayscale values ​​less than zero in the normalized image are extracted. The absolute value of these normalized grayscale values ​​is taken and retained, while the grayscale values ​​of the remaining pixels are set to zero, resulting in a negative polarity image. This negative polarity image corresponds to the characteristic region of the shadow groove development pattern of the embossed code, highlighting the shadow structure formed by the grooves in the characters. This positive and negative polarity splitting allows for the capture of the core structural features of the two stable states of the embossed code, providing a foundation for subsequent bistable consistency recognition.

[0034] Binarization was performed on both the positive and negative polarity images. The Otsu method, known in the art, was used to automatically determine the binarization threshold, eliminating the need for manual parameter setting. This method adapts to the differences in grayscale distribution of polarity images under different lighting conditions, converting both images into binary images containing only foreground and background pixels. The foreground pixels correspond to the character structure region of the imprinted code, while the background pixels correspond to the irrelevant background region, effectively filtering out minor noise interference and enhancing the character structure outline. After binarization, skeleton thinning was performed on both the positive and negative polarity binary images using the Zhang-Suen thinning algorithm, known in the art. This algorithm iteratively removes redundant edge pixels of the foreground pixels in the binary image while retaining the central skeleton of the character structure. The iteration process strictly adheres to the principle of not destroying character connectivity and not losing key character structures until redundant pixels can no longer be removed. Finally, positive and negative polarity skeletons are obtained. These two skeletons correspond to the core structure of the characters in the two steady-state forms of the imprinted code, accurately representing the structural features of the imprinted code without being affected by polarity reversal.

[0035] The distance transformation maps of the positive and negative skeletons are calculated separately. Using the Euclidean distance transformation algorithm, the binary image corresponding to the positive skeleton is traversed, and the Euclidean distance from each foreground pixel to the nearest background pixel is calculated. This distance value is used as the gray value of the corresponding pixel. The distance values ​​of all pixels constitute the distance transformation map of the positive skeleton. The same algorithm and process are used to process the binary image corresponding to the negative skeleton to obtain the distance transformation map of the negative skeleton. The distance transformation map can quantify the distance from each pixel in the skeleton to the background, which is used for subsequent calculation of bistable reliability weights. This provides quantitative data support for vehicle identity anchoring and main object attribution, and adapts to the structural feature extraction requirements of the bistable morphology of the imprinted code under multi-light conditions in the warehouse.

[0036] It should be noted that a positive polarity image refers to a positive component image generated from a normalized image, used to characterize the developed areas in the normalized image dominated by highlight ridges or bright side slopes. Its pixel values ​​are obtained by directly retaining the pixel values ​​greater than zero in the normalized image. Pixels less than or equal to zero in the normalized image are set to zero in the positive polarity image, thus the positive polarity image only includes the portion whose brightness is relatively higher than the local robust center. A negative polarity image refers to a negative component image generated from a normalized image, used to characterize the developed areas in the normalized image dominated by shadow grooves or dark side slopes. Its pixel values ​​are obtained by taking the absolute value of the pixel values ​​less than zero in the normalized image. Pixels greater than or equal to zero in the normalized image are set to zero in the negative polarity image, thus the negative polarity image only includes the portion whose brightness is relatively lower than the local robust center.

[0037] It should be noted that the positive and negative polarity skeletons refer to the sets of single-pixel wide center lines obtained by refining the binary images generated from the positive and negative polarity images, respectively. The positive polarity skeleton is used to depict the stroke center direction of the bright and dark regions in the positive polarity image, while the negative polarity skeleton is used to depict the stroke center direction of the dark and dark regions in the negative polarity image. Both aim to maintain the connectivity and topological structure of the original binary image and are formed by gradually deleting boundary pixels while retaining the central axis, thereby compressing the thick stroke region into a center line representation that can be used for geometric alignment and distance calculation.

[0038] It should be noted that the distance transformation map refers to the distance field image formed by calculating the shortest Euclidean distance from each pixel in the image to the reference set, using the skeleton or binary foreground set as a reference. The distance transformation map of the positive polar skeleton represents the distance from any point to the nearest point of the positive polar skeleton, and the distance transformation map of the negative polar skeleton represents the distance from any point to the nearest point of the negative polar skeleton. The smaller the distance value, the closer the pixel is to the corresponding skeleton, and the larger the distance value, the more obvious the deviation of the pixel from the corresponding skeleton. Therefore, it can be used to calculate the degree of fit between the two polar skeletons and the generation of bistable reliability weights.

[0039] In detail, the first step is to calculate the mean distance, which is the mean distance of pixels in the positive polarity skeleton to the corresponding distance transformation map of the negative polarity skeleton. The positive polarity skeleton is the core structure of the character with a highlight ridge development pattern obtained by binarizing the positive polarity image and processing it with the Zhang-Suen thinning algorithm. The distance transformation map corresponding to the negative polarity skeleton is a feature map obtained by performing an Euclidean distance transformation on the binary image of the negative polarity skeleton. The gray value of each pixel in the map represents the Euclidean distance from that pixel to the nearest background pixel. During the calculation, each foreground pixel of the positive polarity skeleton is traversed, and the two-dimensional coordinate information of each pixel is recorded one by one. Based on the coordinates, the pixel is accurately matched to the corresponding pixel in the distance transformation map of the negative polarity skeleton. For each pixel location, the corresponding distance value is extracted. If the pixel coordinates of the positive skeleton exceed the effective pixel range of the distance transformation map corresponding to the negative skeleton, the distance value corresponding to that pixel is recorded as a preset maximum value. The preset maximum value is set to ten times the maximum value of all effective distance values ​​in the distance transformation map of the negative skeleton. This avoids interference from pixels with abnormal coordinate matching in the overall mean calculation, ensuring the rationality of the calculation results. After collecting the distance values ​​corresponding to all foreground pixels of the positive skeleton, all distance values ​​are summed, and then the sum is divided by the total number of foreground pixels of the positive skeleton to obtain the first mean distance value. This mean value is used to quantify the spatial proximity between the positive skeleton and the background of the negative skeleton.

[0040] The second distance mean is calculated, which is the average distance of the pixels in the negative polarity skeleton in the distance transformation map corresponding to the positive polarity skeleton. The calculation process is exactly the same as that of the first distance mean. The negative polarity skeleton is the core structure of the character in the shadow groove development form obtained by binarizing the negative polarity image and processing it with the Zhang-Suen thinning algorithm. The distance transformation map corresponding to the positive polarity skeleton is the feature map obtained by performing Euclidean distance transformation on the binary image of the positive polarity skeleton. Each foreground pixel of the negative polarity skeleton is traversed, the two-dimensional coordinates of each pixel are recorded and matched to the same position in the distance transformation map corresponding to the positive polarity skeleton, and the corresponding distance value is extracted. Pixels that are outside the effective range are also recorded as the above-mentioned preset maximum value. After collecting all the distance values, they are summed and then divided by the total number of foreground pixels of the negative polarity skeleton to obtain the second distance mean. This mean quantifies the spatial proximity between the negative polarity skeleton and the background of the positive polarity skeleton, and together with the first distance mean, it reflects the structural consistency of the bistable form of the imprinted code.

[0041] After calculating the two distance means, the bistable reliability weight is calculated based on both, using the following formula: The bistable reliability weight is used This indicates that it is the core parameter for quantifying the reliability of the structural feature extraction of the imprinted code. Its value ranges from zero to one and can be directly used for weighted calculations in subsequent structural signature merging decisions and template updates, thus meeting the quantitative requirements of reliability assessment. The natural constant is chosen as the exponent base because it has the characteristics of being continuously differentiable and having a reasonable value range distribution. It can stably convert the sum of distances to the mean into weight values ​​in the range of zero to one, thus avoiding abnormal weight value ranges. This represents the first distance mean, which is the average distance of positive skeleton pixels in the negative skeleton distance transform map. The second distance mean represents the average distance of the negative skeleton pixels in the distance transformation map of the positive skeleton. The sum of the two directly reflects the overall fit of the positive and negative skeletons and is the core foundation for reliability quantification. A coefficient of -1 / 2 is used to scale the sum of the two distance means to prevent the weight value from approaching zero due to an excessively large sum, ensuring that the weight values ​​corresponding to different fits have clear distinction. The smaller the sum of the two distance means, the tighter the fit of the positive and negative skeleton structures, and the more reliable the extraction of imprinted code structural features. After negative exponential operation, the weight value approaches one. Conversely, when one skeleton mismatches with another due to stray highlights, stains, or vehicle damage, both distance means will increase, their sum will rise, and the weight value will decrease accordingly. This accurately characterizes the reliability of structural feature extraction and effectively ensures the accuracy of vehicle identity anchoring.

[0042] It should be noted that the bistable reliability weight refers to the weight value calculated based on the consistency of the stroke structure of the carrier imprint code under two polarity steady states: highlight-dominated and shadow-dominated. It is used to quantify whether the positive and negative polarity skeletons originate from complementary development of the same set of imprinted strokes rather than stray highlights or dirt artifacts. The calculation idea is to first obtain the positive and negative polarity skeletons separately and generate corresponding distance transformation maps. Then, the mean distance of the positive polarity skeleton pixels in the distance transformation map of the negative polarity skeleton and the mean distance of the negative polarity skeleton pixels in the distance transformation map of the positive polarity skeleton are calculated separately. The two values ​​are then combined... The mean cross-skeletal distance is obtained, and a negative exponential function value is applied to the mean cross-skeletal distance as the bistable reliability weight. Thus, when the two polar skeletons are spatially close to each other and their shapes correspond to each other, the mean cross-skeletal distance is small, and the bistable reliability weight approaches one, indicating that the bistable structure observed in this study is highly reliable and suitable for vehicle master object merging and template updating. When the two polar skeletons are far apart from each other or there is a significant skeleton on only one side, the mean cross-skeletal distance is large, and the bistable reliability weight approaches zero, indicating that this study is more likely to be affected by stray specular highlights or local contamination, and its contribution to merging decisions and template updating should be reduced.

[0043] In embodiments of the present invention, based on bistable reliability weights, the weighted similarity between the structural signature and the template signatures of each vehicle master object in the pre-stored set of vehicle master objects is calculated, including: The positive and negative polarity skeletons are combined to obtain a polarity-independent joint skeleton. The positive polarity skeleton corresponds to the character core structure of the embossed code's highlight ridge development pattern, while the negative polarity skeleton corresponds to the character core structure of the embossed code's shadow groove development pattern. Both are foreground pixel sets obtained after binarization and Zhang-Suen thinning algorithm processing. During processing, each pixel position of the two skeleton images is traversed to determine whether the position is a foreground pixel of the positive or negative polarity skeleton. If either condition is met, the pixel at that position is set as a foreground pixel; otherwise, it is set as a background pixel. The foreground pixel set formed after traversal is the polarity-independent joint skeleton. This method integrates all core structural features of the bistable morphology of the embossed code, eliminates background interference, and removes the influence of brightness polarity, ensuring that the joint skeleton only represents the inherent structural geometric distribution of the embossed code, providing a basis for the polarity independence of subsequent structural signatures.

[0044] The polarity-independent joint skeleton's projected contours in the horizontal and vertical directions are calculated separately. The horizontal projection contour characterizes the pixel density of the joint skeleton in each row, and the vertical projection contour characterizes the pixel density of the joint skeleton in each column. Together, they reflect the overall structural morphology of the embossed character. The formula for calculating the horizontal projection contour is as follows: ,in, represents the projection value corresponding to the i-th row in the horizontal projection contour, where i is the horizontal row index. The value range is the total number of rows in the joint skeleton image, increasing sequentially from the first row to the last row. Each row corresponds to a projection value, forming a complete horizontal projection contour sequence. The pixel state at the position of the polarity-independent joint skeleton in the i-th row and y-th column is represented by a value of 1 if the position is a foreground pixel and 0 if it is a background pixel. By summing the pixel states of all columns in each row, the total number of foreground pixels in that row can be obtained, which is the projection value of that row. This calculation method can intuitively quantify the structural distribution density of each row and preserve the structural features of the character in the horizontal direction.

[0045] The formula for calculating the projected profile in the vertical direction is: , This represents the projection value corresponding to the j-th column in the vertical projection contour, where j is the vertical column index. The value range is the total number of columns in the joint skeleton image, increasing sequentially from the first column to the last column. The pixel state at the x-th row and j-th column of the polarity-independent joint skeleton is represented by a value of 1 for the foreground pixel and a value of 0 for the background pixel. The pixel states of all rows in each column are summed to obtain the total number of foreground pixels in that column, which is the projection value. This forms a vertical projection contour sequence to capture the structural features of the character in the vertical direction.

[0046] After the projection contour calculation is completed, the projection contours in the horizontal and vertical directions are normalized separately. This is to eliminate the influence of differences in the size and pixel count of the joint skeleton image, ensuring the comparability of the structural signatures of the imprinted codes under different acquisition scenarios and postures, and adapting to the requirements of subsequent similarity calculations and vehicle identity matching. The normalization process uses the following formula: and ,in, Let be the projection value of the i-th row of the normalized horizontal projection contour. The projection value of the j-th column of the normalized vertical projection contour; molecule part and These are the unnormalized horizontal and vertical projection values, respectively, preserving the distribution characteristics of the original projected contours; in the denominator, This is the sum of the projection values ​​of all rows of the horizontal projection contour, i.e., the total number of foreground pixels in the polarity-independent joint skeleton. The sum of the projection values ​​of all columns of the vertical projection contour is consistent with the total number of foreground pixels of the horizontal projection contour, since both are the total number of foreground pixels of the joint skeleton. The introduction of a very small positive number ε is to avoid the case where the denominator is zero. The value of ε is no greater than 10 to the power of negative 6 to ensure the stability of the calculation process and avoid division by zero error.

[0047] After normalization, the normalized horizontal projection contour sequence and vertical projection contour sequence are combined to form a one-dimensional vector structural signature. This structural signature depends only on the inherent structural geometric distribution of the imprint code and is independent of brightness polarity. It can maintain consistency across the two steady-state forms of imprint code shadow groove development and highlight ridge development, providing core structural feature basis for subsequent vehicle main object merging decisions and identity anchoring, ensuring the stability and accuracy of vehicle identity recognition.

[0048] It should be noted that the polarity-independent joint skeleton refers to the skeleton set obtained by merging the positive polarity skeleton and the negative polarity skeleton. It is used to uniformly represent the geometric center line distribution of the embossed code strokes without relying on the brightness polarity determination. The positive polarity skeleton depicts the center direction of the bright development stroke dominated by the highlight, while the negative polarity skeleton depicts the center direction of the dark development stroke dominated by the shadow. The two reflect the complementary visible parts of the same embossed stroke under different lighting geometries. The polarity-independent joint skeleton formed by the union fusion can cover the effective stroke center lines under the two steady states, thereby reducing the structural loss caused by polarity reversal and providing a basis for the subsequent generation of polarity-independent identity anchors. A structural signature is a vectorized description of the geometric structure distribution of an imprinted code, constructed based on a polarity-independent joint skeleton. It is independent of pixel brightness polarity and can remain relatively stable between a highlight-dominated steady state and a shadow-dominated steady state. The preferred generation method is to calculate the projection contours of the polarity-independent joint skeleton in the horizontal direction and the projection contours in the vertical direction, respectively. The projection contours are used to describe the cumulative number distribution of skeleton pixels in each row or column. Then, the projection contours are normalized to eliminate the influence of changes in the total skeleton quantity on the amplitude. Finally, the normalized horizontal projection contours and the vertical projection contours are combined to form a structural signature. This allows the structural signature to be used as an identity anchor field of the vehicle's main object for calculating similarity with the template signature and supporting the merging and creation decision of the vehicle's main object.

[0049] In embodiments of the present invention, based on bistable reliability weights, the weighted similarity between the structural signature and the template signatures of each vehicle master object in the pre-stored set of vehicle master objects is calculated, including: Detailed Examples of Cosine Similarity and Weighted Similarity Calculation The cosine similarity between the structural signature and the template signatures of each vehicle master object in the pre-stored set of vehicle master objects is calculated. The structural signature is a one-dimensional vector formed by combining the horizontal and vertical normalized projection contours of the polarity-independent joint skeleton, representing only the inherent structural geometric distribution of the imprinted code and is independent of brightness polarity. The set of vehicle master objects is a pre-maintained set of all vehicle identity cores. Each vehicle master object stores a corresponding template signature, which is a structural signature benchmark obtained after multiple reliability-weighted updates. This template signature can stably represent the core structural features of the corresponding vehicle's imprinted code, ensuring consistency in identity matching. When calculating the cosine similarity, each vehicle master object in the set of vehicle master objects is traversed, and the template signature of each master object is extracted one by one. This template signature is then compared with the structural signature generated in this operation. The formula for calculating the cosine similarity is: ,in, represents the cosine similarity between the generated structural signature and the template signature of the k-th vehicle main object, where k is the main object index and its value range covers all main objects in the set of vehicle main objects, with each main object corresponding to a cosine similarity value; v represents the generated structural signature vector, which is formed by combining the normalized horizontal projection contour sequence and the vertical projection contour sequence, and is the core vector characterizing the structural features of the imprinted code. The template signature vector representing the k-th vehicle master object is the structural signature benchmark of the master object after multiple observations and updates, which can reflect the stable structural characteristics of the corresponding vehicle imprint code. The introduction of a very small positive number ε is to avoid the case where the denominator is zero. The value of ε is no greater than 10 to the power of negative 6, which can effectively avoid the calculation anomaly caused by the vector magnitude being zero and ensure the stability of the calculation process.

[0050] The cosine similarity value ranges from zero to one. A value closer to one indicates a higher similarity between the current imprinted code structure and the imprinted code structure corresponding to the k-th main object, suggesting they likely belong to the same physical vehicle. A value closer to zero indicates a greater structural difference, suggesting they likely belong to different physical vehicles. After calculating the cosine similarity for all main objects, each cosine similarity is multiplied by a bistable reliability weight to obtain a weighted similarity. The bistable reliability weight is used to weight and calibrate the cosine similarity. The core purpose of this weighted calculation is to integrate structural matching degree and feature extraction reliability, avoiding the inclusion of low-reliability structural matching results in the decision-making process. When the current structural feature extraction reliability is high, the weighted similarity is close to the original cosine similarity, allowing for full reliance on structural matching degree for main object matching. When the current extraction reliability is low, the weighted similarity is weakened, reducing the interference of low-reliability matching results on subsequent merging or creation decisions, ensuring decision accuracy, and providing more precise quantitative support for subsequent vehicle main object merging or creation decisions.

[0051] It should be noted that weighted similarity refers to the similarity metric used to characterize the degree of matching between the structural signature of the current observation and the template signature of an existing vehicle owner during the competition for ownership of the vehicle owner object. It is determined by the cosine similarity between the structural signature and the template signature and the bistable reliability weight, so that the similarity not only reflects the consistency of the geometric structure, but also reflects the credibility of the structure corresponding to the current observation under the two stable states of highlight dominance and shadow dominance.

[0052] In an embodiment of the present invention, a classification competition decision is made for the current observation based on weighted similarity, determining whether to merge the current observation into an existing vehicle master object or a newly created vehicle master object, and the identification string is written as an observation alias into the vehicle master object selected by the classification competition decision, including: A new candidate node is defined to represent a new physical vehicle that is not covered by the pre-stored vehicle master object set. To ensure that the node has no initial matching priority and does not interfere with the matching decision of the existing vehicle master objects, the fixed score of the new candidate node is set to zero. This fixed score is only used to participate in the calculation of the attribution weight and does not change with the results of the current imprint code structural feature extraction or the weighted similarity value. Its core function is to determine whether the currently observed imprint code belongs to the existing vehicle master object set through weight quantization. If the new candidate node has the highest attribution weight, it means that the vehicle corresponding to the currently observed imprint code is a new vehicle, and a new vehicle master object needs to be created and the master object set updated.

[0053] After the new candidate nodes are set, the weighted similarity of each vehicle owner and the attribution weight corresponding to the fixed score of the new candidate node are calculated using the Softmax function. The core function of the Softmax function is to normalize all the scores involved in the calculation into weight values ​​between zero and one, and the sum of all weight values ​​is one, which meets the quantitative requirements of attribution probability and can accurately characterize the probability that the current observation result belongs to each vehicle owner and the new candidate node. During the calculation, the weighted similarity corresponding to all vehicle owners is collected first. The weighted similarity of the kth vehicle owner has been calculated above. k is the index of the main object, and its value range covers all main objects in the vehicle main object set. It also extracts the fixed score of newly created candidate nodes. , A value of zero represents all weighted similarities. With fixed score Both are used as input parameters to the Softmax function to calculate the ownership weight corresponding to each vehicle master object and the ownership weight corresponding to the newly created candidate node.

[0054] The formula for calculating the attribution weight corresponding to the kth vehicle master object is as follows: The formula for calculating the attribution weight of a newly created candidate node is as follows: ,in, This represents the ownership weight corresponding to the k-th vehicle owner object. This represents the affiliation weight corresponding to the newly created candidate node. Both values ​​range from zero to one, and all... and The sum of all values ​​is one. The higher the weight value, the greater the likelihood that the observation result belongs to the corresponding object. e is the natural constant. Using the natural constant as the exponent base can stably convert the score difference into a significant weight difference, while avoiding weight distortion caused by extreme scores. The weighted similarity corresponding to the kth vehicle main object represents the matching degree between the imprinted code structure and the main object structure, as well as the reliability of feature extraction. The fixed score for newly created candidate nodes is zero, ensuring that the new vehicle initially has no matching advantage. The weight of the newly created candidate node will only increase when the weighted similarity of all existing main objects is extremely low; N represents the total number of main objects in the pre-stored set of vehicle main objects. The sum of the weighted similarity indices representing the similarity of all vehicle main objects. The index value representing the fixed score of newly created candidate nodes, and the sum of the two constitute the core denominator basis for weight calculation. The introduction of a very small positive number ε avoids the case where the denominator is zero. ε takes a value no greater than 10 to the power of -6, effectively avoiding calculation anomalies caused by all main objects having weighted similarities of negative infinity and newly created candidate node scores of zero, thus ensuring the stability of the calculation process. Through the normalization property of the Softmax function, matching degree and reliability are transformed into assignment probabilities that can be directly used for decision-making. This not only highlights the assignment priority of high-matching main objects but also quantifies the probability of new vehicles through the weight of newly created candidate nodes, avoiding the omission of new vehicles or the misallocation of new vehicles to existing main objects. Simultaneously, it weakens the interference of low-reliability and low-matching main objects on decision-making, providing accurate quantitative basis for subsequent vehicle main object merging or creation decisions, adapting to the dynamic management needs of vehicle recycling and new vehicle entry in intelligent warehousing.

[0055] The attribution weights corresponding to all vehicle owner objects are traversed and filtered to extract the maximum attribution weight. This maximum attribution weight corresponds to a specific vehicle owner object, indicating that the observed imprint code structure feature has the highest probability of belonging to that vehicle owner object. Then, this maximum attribution weight is compared with the attribution weights corresponding to newly created candidate nodes. If the maximum attribution weight in the vehicle owner object is not less than the attribution weight of the newly created candidate node, it is determined that the current observation result belongs to the vehicle owner object corresponding to the maximum attribution weight, and a merging operation is performed. During merging, the recognition string obtained from the current visual code reading is used as the observation of that vehicle owner object. The aliases are recorded and merged. The observation aliases are used to supplement the identity information of the vehicle's main object. They can adapt to the situation where the identification string of the same vehicle imprint code has slight differences under different observation scenarios due to wear and light changes. At the same time, the structural signature, bistable reliability weight and attribution weight corresponding to the observation alias are retained. They are associated and integrated with the template signature, historical observation aliases and related data already stored for the vehicle's main object to ensure that the identity information of the vehicle's main object is complete and traceable. Subsequently, the template signature and cumulative weight can be updated and optimized based on the observation data merged multiple times to improve the accuracy of subsequent identity matching.

[0056] If, after traversing the ownership weights of all existing vehicle master objects, it is found that the ownership weights of all existing vehicle master objects are less than the ownership weight of the newly created candidate node, then the vehicle corresponding to the imprint code observed this time is determined to be a new physical vehicle not covered by the pre-stored set of vehicle master objects. The operation of creating a new vehicle master object is then executed. During creation, the structural signature generated this time is directly initialized as the template signature of the new vehicle master object. The template signature serves as the core structural identifier of the new vehicle master object, providing a benchmark for identity matching when the vehicle is observed again. Simultaneously, the bistable reliability weight calculated this time is initialized as the cumulative weight of the new vehicle master object. The cumulative weight is used for subsequent weighted updates based on multiple observation data, recording the initial identity information of the new vehicle master object. The recognition string obtained from this visual code reading is used as the initial observation alias of the new vehicle master object. After the creation of the new vehicle master object is completed, it is added to the pre-stored set of vehicle master objects, realizing dynamic updates of the set of vehicle master objects. This ensures that the new vehicle can be accurately identified and merged during subsequent cyclical use, guaranteeing the comprehensiveness and dynamic adaptability of intelligent warehousing vehicle identity management.

[0057] It should be noted that the "vehicle master object" refers to a unique data entity established for each physical vehicle in the data management of intelligent warehousing. This entity carries the unique identifier of the physical vehicle, along with its associated historical observation aliases, structural characterization information, and records related to warehousing business data. The vehicle master object serves as a unified primary key, used for recording business activities such as inbound, outbound, transfer, picking, inventory, cleaning, and maintenance. This avoids primary key splitting and dual identities caused by fluctuations in the identification string due to the bistable nature of the imprinted code. The "new candidate node" is a virtual competing object introduced in the vehicle master object attribution competition decision. It has a fixed score of zero and participates in the attribution weight calculation. It represents the possibility that the current observation may not belong to any existing vehicle master object and a new vehicle master object should be created. The attribution weight is obtained by inputting the weighted similarity of each existing vehicle master object and the fixed score of the new candidate node into the Softmax function. The maximum attribution weight is compared with the attribution weight of the new candidate node to determine whether to merge or create a new one. This achieves adaptive creation decisions without relying on preset thresholds.

[0058] It should be noted that the template signature refers to the structural signature template vector stored in the vehicle master object, used to characterize the stable reference form of the imprinted code geometry of the physical vehicle corresponding to the vehicle master object. The template signature is formed by weighted updating of structural signatures obtained from multiple historical observations, and is used as a comparison object in subsequent attribution competition to calculate the similarity with the currently observed structural signature, thereby realizing identity anchoring and merging judgment across illumination geometry. The cumulative weight refers to the cumulative weight value stored in the vehicle master object, used to quantify the total amount of effective observation contributions adopted by the template signature of the vehicle master object in the historical update process. The cumulative weight is preferably obtained by accumulating the bistable reliability weight of each observation, and is used together with the bistable reliability weight of the current observation as a weighting coefficient when updating the template signature, so that the template signature is more biased towards the stable structure formed by high-reliability observations, thereby reducing the impact of low-reliability observations on template drift.

[0059] In an embodiment of the present invention, the template signature of the selected vehicle owner is updated according to the bistable reliability weight, and the warehousing business data is associated with the unique identifier of the vehicle owner, including: The system retrieves the existing cumulative weight of the vehicle master object. This cumulative weight is the accumulated value obtained by gradually adding bistable reliability weights during the creation and merging processes of the vehicle master object. When the vehicle master object is created, its cumulative weight is initialized to the bistable reliability weight calculated at that time. After each merging operation is completed, the bistable reliability weight corresponding to this merging is added to the cumulative weight. This cumulative weight is stored in the vehicle master object information pre-stored in the system and is saved in association with information such as template signature and observation alias. When retrieving, the system directly retrieves the cumulative weight W of the vehicle master object pointed to by the maximum attribution weight corresponding to this merging. The value of the cumulative weight W increases with the number of mergings and is used to characterize the cumulative reliability of the historical observation data of the vehicle master object.

[0060] Using the bistable reliability weight and cumulative weight as weighting coefficients, a weighted average update is performed on the current structural signature and the original template signature of the vehicle's main object. The current structural signature is the polarity-independent structural signature generated during this observation process, denoted as... The original template signature of the vehicle's main object is the base structure signature currently stored in that main object, denoted as... The formula for weighted average update is: ,in, The new template signature representing the updated vehicle master object serves as the benchmark for subsequent identity matching; (Molecular part) This represents the weighted contribution value of this structural signature. For bistable reliability weights, For structural signature, This represents the weighted contribution value of the original template signature. For cumulative weighting, The template signature is calculated using a weighted approach because a higher bistable reliability weight indicates a more reliable structural feature extraction and a greater impact on the new template signature. A higher cumulative weight also indicates a stronger historical cumulative reliability of the original template signature, resulting in a higher contribution. This avoids interference from a single low-reliability observation and gradually integrates structural features from multiple high-reliability observations, optimizing the template signature's accuracy. The denominator is the sum of the two weighting coefficients, used to normalize the numerator and ensure that the numerical scale of the new template signature is consistent with the original template signature and the current structural signature. A very small positive number ε is introduced to avoid a zero denominator; ε is no greater than 10 to the power of -6 to ensure the stability of the calculation process and prevent division by zero errors.

[0061] After the template signature is updated, the bistable reliability weight calculated in this instance is added to the current cumulative weight to complete the update of the cumulative weight. This accumulation operation can accumulate the reliability of historical observations and the reliability of the current observation, so that the cumulative weight can truly reflect the overall reliability of all observation data of the vehicle's main object. This provides reasonable weighting coefficient support for subsequent template signature weighting updates, ensuring that the template signature is gradually optimized as the number of observations increases, further improving the accuracy and stability of subsequent vehicle identity matching. After the update is completed, the new template signature and the new cumulative weight are synchronously stored in the system, overwriting the original template signature and cumulative weight. At the same time, the observation aliases, structural signatures and attribution weights merged in this instance are also associated and saved to ensure that the identity information of the vehicle's main object is complete and traceable.

[0062] A dedicated alias mapping table is maintained in the warehouse data storage database. This alias mapping table serves as the core association table of the warehouse data management system, establishing a bidirectional association with the pre-stored vehicle owner information table to ensure the consistency and traceability of the mapping relationship. The alias mapping table adopts a relational data table structure. The core fields of the table include an identification string field and a vehicle owner unique identifier field. The identification string field stores all valid identification strings obtained during the visual code reading process, covering the initial identification string when a vehicle owner is created and the observation aliases added during each merging process. The vehicle owner unique identifier field stores a globally unique identifier automatically assigned by the system when each vehicle owner is created. This identifier is generated by the system according to preset rules, is non-repeatable, and is used to uniquely distinguish different vehicle owner objects, ensuring the uniqueness of the mapping relationship.

[0063] The alias mapping table establishes a strict mapping relationship between the identification strings and the unique identifiers of the vehicle's main object. One identification string corresponds to only one unique identifier of the vehicle's main object, avoiding indexing errors caused by mapping ambiguity. One vehicle's main object can correspond to multiple different identification strings, adapting to the actual needs of different identification strings for the same physical vehicle in different observation scenarios. The maintenance of the alias mapping table is performed synchronously with the merging and creation operations of vehicle's main objects. When the vehicle's main object merging operation is completed, the identification string corresponding to this merging is inserted into the alias mapping table as an observation alias and associated with the unique identifier of the vehicle's main object corresponding to the merging. When the vehicle's main object creation operation is completed, the identification string obtained from this visual code reading is inserted into the alias mapping table as the initial observation alias and bound to the unique identifier of the newly created vehicle's main object. At the same time, the system periodically verifies the alias mapping table, removing mapping records corresponding to invalid identification strings to ensure the accuracy of the mapping relationship and the cleanliness of the data table.

[0064] When the system receives a new warehousing operation request, this request includes, but is not limited to, various operation requests that require vehicle identification, such as vehicle inbound, outbound, sorting, and inventory counting. Each request carries a corresponding identification string as a vehicle identifier. First, the corresponding identification string is extracted from the request. The extraction process strictly follows the format specifications consistent with the visual code identification string described earlier to avoid query failures due to format differences. Then, the system automatically triggers a query operation in the alias mapping table, using the extracted identification string as the query condition to search the identification string field of the alias mapping table. It determines whether the identification string already exists in the alias mapping table. If the query result shows that the identification string exists in the alias mapping table, then the system uses the preset identification string in the table. The system establishes a mapping relationship between the identification string and the unique identifier of the vehicle owner object. This allows for rapid indexing to the unique identifier of the vehicle owner object corresponding to the identification string. Based on this unique identifier, the system accurately locates the corresponding vehicle owner object from the pre-stored set of vehicle owner objects. The system then completely mounts the business record corresponding to this business request to the vehicle owner object. The business record includes the business request type, request execution time, execution workstation, operation parameters, and related verification information. After mounting, a unique association is established between the business record and the vehicle owner object, enabling the linkage management of business data and vehicle identity information. This ensures the traceability of various warehousing business operations, improves the processing efficiency of business requests, avoids errors in business record mounting due to differences in identification strings, and guarantees the uniformity and accuracy of warehousing data management.

[0065] like Figure 2 The diagram shown is a functional block diagram of a smart warehouse data management system provided in an embodiment of the present invention.

[0066] In this embodiment, the functions of each module / unit are as follows: The code reading module is used to acquire the grayscale image of the vehicle imprint code area and the recognition string corresponding to this visual code reading; The structural signature module is used to perform robust illumination normalization processing on grayscale images, decompose the normalized image and extract the positive and negative polarity skeletons, and generate bistable reliability weights and structural signatures based on the positive and negative polarity skeletons. The weighted similarity module is used to calculate the weighted similarity between the structural signature and the template signatures of each vehicle master object in the pre-stored set of vehicle master objects, based on the bistable reliability weight. The attribution competition module is used to make attribution competition decisions for this observation based on weighted similarity, determine whether to merge this observation into an existing vehicle master object or a newly created vehicle master object, and write the identification string as an observation alias into the vehicle master object selected by the attribution competition decision; The template update and association module is used to update the template signature of the selected vehicle owner object according to the bistable reliability weight, and associate the warehousing business data with the unique identifier of the vehicle owner object.

[0067] like Figure 3 As shown, using a normalized image as input, positive and negative polarity maps are generated for the normalized image of the imprinted code area on the carrier. The positive polarity map is used to retain the bright development components greater than zero in the normalized image, while the negative polarity map is used to retain the dark development components less than zero in the normalized image and take their absolute values, thus separating the effective stroke information under the two steady states of highlight dominance and shadow dominance. Subsequently, skeleton extraction is performed on the positive and negative polarity maps respectively to obtain positive and negative polarity skeletons, which are used to depict the centerline direction of the imprinted strokes under the two polarity steady states. Based on this, the positive polarity map is further processed... The polar skeleton and the negative polar skeleton are fused together to generate a polarity-independent joint skeleton, forming a geometric centerline representation that is independent of light and dark polarities and covers two steady-state complementary strokes. Further, the polarity-independent joint skeleton is projected in the horizontal and vertical directions and normalized to obtain the horizontal projection and normalization results and the vertical projection and normalization results. The projected contours are used to characterize the cumulative distribution of the joint skeleton in each row and column. Finally, the horizontal projection and normalization results and the vertical projection and normalization results are combined to generate a structural signature.

[0068] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A data management method for intelligent warehousing, characterized in that, Includes the following steps: S1. Obtain the grayscale image of the vehicle imprint code area and the recognition string corresponding to this visual code reading; S2. Perform robust illumination normalization on the grayscale image, decompose the normalized image and extract the positive and negative polarity skeletons, and generate bistable reliability weights and structural signatures based on the positive and negative polarity skeletons. S3. Based on the bistable reliability weight, calculate the weighted similarity between the structural signature and the template signatures of each vehicle master object in the pre-stored set of vehicle master objects; S4. Based on weighted similarity, make a competitive decision on the attribution of this observation, determine whether to merge this observation into an existing vehicle master object or a newly created vehicle master object, and write the identification string as the observation alias into the vehicle master object selected by the attribution competitive decision. S5. Update the template signature of the selected vehicle master object according to the bistable reliability weight, and associate the warehousing business data with the unique identifier of the vehicle master object. 2.The data management method for intelligent warehousing according to claim 1, wherein, Robust illumination normalization processing for grayscale images includes: Calculate the median gray level and the median absolute gray level deviation of a grayscale image; Calculate the difference between the grayscale image and the median grayscale value, and obtain the normalized image based on the ratio of the difference to the median of the absolute grayscale deviation. 3.The data management method for intelligent warehousing of claim 2, wherein, Generate bistable reliability weights, including: The portion of the normalized image that is greater than zero is taken as the positive polarity image, and the absolute value of the portion that is less than zero is taken as the negative polarity image. The positive polarity map and the negative polarity map are binarized and the skeleton is refined respectively to obtain the positive polarity skeleton and the negative polarity skeleton. Calculate the distance transformation diagrams for the positive and negative polarity frameworks respectively; Calculate the mean distance of pixels in the positive polarity skeleton in the distance transformation map corresponding to the negative polarity skeleton, and the mean distance of pixels in the negative polarity skeleton in the distance transformation map corresponding to the positive polarity skeleton; calculate the bistable reliability weight based on the negative exponential function value of the sum of the two mean distances. 4.The data management method for intelligent warehousing of claim 3, wherein, Generate a structure signature, including: By performing a union operation on the positive and negative polarity frameworks, a polarity-independent joint framework is obtained. Calculate the projected profiles of the polarity-independent joint skeleton in the horizontal and vertical directions, respectively; The projected contours are normalized and combined to form a structural signature.

5. The data management method for intelligent warehousing according to claim 1, characterized in that, Calculate the weighted similarity between the structural signature and the template signatures of each vehicle master object in the pre-stored set of vehicle master objects, including: Calculate the cosine similarity between the structural signature and the template signature of the vehicle master object; Multiplying the cosine similarity by the bistable reliability weight yields the weighted similarity.

6. The data management method for intelligent warehousing according to claim 1, characterized in that, The attribution competition decision for this observation is based on weighted similarity, including: Set a new candidate node with a fixed score of zero; The weighted similarity of each vehicle's main object and the attribution weight corresponding to the fixed score of the newly created candidate node are calculated using the Softmax function. If the maximum attribution weight in the vehicle master object is not less than the attribution weight of the newly created candidate node, then merging is determined, and the identification string is merged into the vehicle master object corresponding to the maximum attribution weight as the observation alias. If the ownership weights of all existing vehicle master objects are less than the ownership weights of the newly created candidate node, then a new vehicle master object is created, and the structural signature is initialized to the template signature of the new object, and the bistable reliability weight is initialized to the cumulative weight of the new object.

7. The data management method for intelligent warehousing according to claim 6, characterized in that, The template signature of the selected vehicle master object is updated according to the bistable reliability weight, including: Get the existing cumulative weight of the vehicle's main object; The bistable reliability weight and cumulative weight are used as weighting coefficients to perform a weighted average update on the current structural signature and the original template signature of the vehicle master object. The bistable reliability weight is added to the current cumulative weight.

8. The data management method for intelligent warehousing according to claim 1, characterized in that, The unique identifier that links warehousing business data to the vehicle owner includes: Maintain an alias mapping table in the database for storing warehouse data, and establish a mapping relationship between identification strings and the unique identifier of the vehicle master object; When a new business request is received, if the identification string in the request exists in the alias mapping table, the corresponding vehicle master object is indexed through the mapping relationship, and the business record is mounted to the vehicle master object.

9. A data management system for intelligent warehousing, applied in the data management method for intelligent warehousing as described in any one of claims 1-8, characterized in that, The system includes: The code reading module is used to acquire the grayscale image of the vehicle imprint code area and the recognition string corresponding to this visual code reading; The structural signature module is used to perform robust illumination normalization processing on grayscale images, decompose the normalized image and extract the positive and negative polarity skeletons, and generate bistable reliability weights and structural signatures based on the positive and negative polarity skeletons. The weighted similarity module is used to calculate the weighted similarity between the structural signature and the template signatures of each vehicle master object in the pre-stored set of vehicle master objects, based on the bistable reliability weight. The attribution competition module is used to make attribution competition decisions for this observation based on weighted similarity, determine whether to merge this observation into an existing vehicle master object or a newly created vehicle master object, and write the identification string as an observation alias into the vehicle master object selected by the attribution competition decision; The template update and association module is used to update the template signature of the selected vehicle master object according to the bistable reliability weight, and associate the warehousing business data with the unique identifier of the vehicle master object.

Images