An Automatic Loading Quantity Identification Method and System for Intelligent Factories

Abstract

The present invention discloses an automatic recognition method and system for loading quantity in an intelligent factory, which relates to the technical field of goods recognition. The method includes: connecting to a task platform to obtain loading task information and obtaining the composition of order goods; using the goods numbers of the order goods composition as indexes to search a goods management feature library to obtain goods recognition features, performing interference feature recognition within the order on the goods recognition features to determine interference recognition features; obtaining interference adjacent goods and their adjacent interference features in the intelligent factory; performing feature selection processing on the goods recognition features to obtain robust features; using the robust features as recognition features and performing loading task recognition processing according to the quantity of goods. It solves the technical problem that existing loading quantity recognition cannot efficiently and accurately recognize goods combinations in the face of diversified and customized production requirements, resulting in low loading accuracy and efficiency, and achieves the technical effect of improving the efficiency and accuracy of factory loading.

Description

Technical Field

[0001] This application relates to the technical field of goods identification, and specifically relates to an automatic loading quantity identification method and system for intelligent factories. Background Art

[0002] In the current wave of intelligent manufacturing, intelligent factories have gradually become the core driving force of modern manufacturing, and are gradually realizing the automation, intelligence, and high efficiency of the production process. With the booming development of the e-commerce industry and the increasing demand of consumers for logistics distribution efficiency, the loading operations in intelligent factories are facing unprecedented challenges. Loading in intelligent factories usually needs to complete the assembly and transportation of specific goods according to production orders and customer requirements. With the expansion of the factory production scale and the increase in the number of orders, the loading tasks have become more complex and diverse. The traditional loading process is difficult to handle various different types of goods combinations, especially customized products. In the factory's logistics warehouse, the goods are stacked closely, and it is easy for adjacent goods to interfere with each other. There are often various interference factors at the loading site, such as the confusion of similar goods and the damage of packaging labels. The traditional manual checking of the loading list and the quantity of goods is not only inefficient, but also difficult to accurately identify the target goods during the actual loading process, and is prone to errors, thus affecting the loading accuracy and efficiency, and seriously restricting the overall efficiency of the logistics supply chain.

[0003] Therefore, in the current related technologies for loading quantity identification in intelligent factories, there are technical problems that in the face of diverse and customized production requirements, it is impossible to efficiently and accurately identify the goods combination, resulting in low loading accuracy and efficiency. Summary of the Invention

[0004] This application provides an automatic loading quantity identification method and system for intelligent factories, solves the technical problem that in the existing loading quantity identification, in the face of diverse and customized production requirements, it is impossible to efficiently and accurately identify the goods combination, resulting in low loading accuracy and efficiency, and achieves the technical effect of improving the efficiency and accuracy of factory loading.

[0005] The present application provides an automatic loading quantity recognition method for an intelligent factory. The method includes: connecting to a task platform to obtain loading task information, decomposing the loading task information to obtain the order goods composition, where the order goods composition is the information of all goods that need to be assembled in the loading task, including the goods number and the quantity of goods; using the goods number in the order goods composition as an index to search the goods management feature library to obtain goods recognition features, and performing interference feature recognition within the order on the goods recognition features to determine interference recognition features; using the goods recognition features as an index to search the goods management feature library to obtain interference adjacent goods and their adjacent interference features in the intelligent factory; using the interference recognition features and the adjacent interference features as constraint conditions to perform feature selection processing on the goods recognition features to obtain robust features, where the robust features are core recognition features or combinations of core recognition features that are not affected by interference features; using the robust features as recognition features to perform loading task recognition processing according to the quantity of goods.

[0006] In a possible implementation manner, safety triage is performed through the abnormal risk mark, and the following processing is executed: based on a machine learning algorithm, a vulnerability scanning detection model is established, where the vulnerability scanning detection model includes a quantity feature recognition scanning layer of event types and a trigger frequency feature recognition scanning layer; based on the vulnerability scanning detection model, safety triage is performed through the abnormal risk mark to determine high-confidence attack behaviors.

[0007] In a possible implementation manner, before searching the goods management feature library, the following processing is also executed: collecting the physical features, image features, and kit composition features of all incoming goods in the intelligent factory, where the physical features include weight, density, and material, the image features include size, shape, color, texture, and light reflection features, and the kit composition features include the number of kit components and the installation and assembly positions; encoding the physical features, image features, and kit composition features, converting them into a storable format, establishing a mapping relationship between the goods number of the incoming goods and the feature encoding, and constructing the goods management feature library, where a multi-level mapping relationship from components to kits and from kits to goods numbers is established for the kit composition features.

[0008] In a possible implementation, interference feature recognition within an order is performed on the goods identification features to determine interference recognition features, and the following processing is further performed: partitioning is carried out according to the goods number, physical features, image features, and kit composition features; using the goods number as the only index, it is integrated into the goods number partition of the goods management feature library; using the physical features, image features, and kit composition features as the main indexes respectively, and the specific features under the features as the auxiliary indexes, combined indexes are established respectively; the combined indexes corresponding to each feature category are respectively integrated into the physical feature, image feature, and kit composition feature partitions of the goods management feature library.

[0009] In a possible implementation, when obtaining the comprehensive similarity value, the following processing is further performed: obtaining a set of loading abnormal cases, clustering the set of loading abnormal cases according to the factory operation process and goods number to construct an abnormal case cluster; through the abnormal case cluster, fitting the abnormal features and abnormal probabilities of the factory operation process and the goods number respectively to obtain the abnormal features and abnormal probabilities of the factory operation process and the abnormal features and abnormal probabilities of the goods number; extracting the factory operation process of the loading task according to the loading task information, and matching the abnormal features and abnormal probabilities of the process; according to the goods number of the goods composition in the order, matching the abnormal features and abnormal probabilities of the goods number; according to the abnormal features and abnormal probabilities of the matching process and the abnormal features and abnormal probabilities of the goods number, performing abnormal probability matching on the features of each pair of goods combinations and their feature category similarities, calculating the average abnormal probability, and the average abnormal probability is the average of the abnormal probabilities of each good in the goods combination in this feature category and the abnormal probability of the matching process; using the average abnormal probability to weight the comprehensive similarity, and using the weighting to perform similarity weighted calculation on each pair of goods combinations to obtain the comprehensive similarity value.

[0010] In a possible implementation, using the interference recognition features and the adjacent interference features as constraint conditions, feature selection processing is performed on the goods identification features to obtain robust features, and the following processing is further performed: using the interference recognition features and the adjacent interference features as the features to be excluded, excluding them from the goods identification features to construct a pure identification feature set;

[0011] Randomly sample from the set of pure recognition features to construct a number of feature subsets. The subsets include different numbers of features or feature combinations. Use the recognition evaluation model to evaluate the recognition probabilities of each random feature or feature combination, obtain the recognition evaluation probabilities, and select the random sampling feature with the highest recognition evaluation probability as the candidate robust feature. Determine whether the recognition evaluation probability reaches the preset recognition target. If it does, use the candidate robust feature as the robust feature. When it does not reach, use the interference recognition feature and the adjacent interference feature as additional features, randomly select from the additional features, and add the randomly added features to the feature subset to obtain a newly constructed feature combination. Use the recognition evaluation model to evaluate the recognition probability of the newly constructed feature combination until the recognition evaluation probability reaches the preset recognition target, and determine the robust feature.

[0012] In a possible implementation, when using the recognition evaluation model to evaluate the recognition probability of the newly constructed feature combination until the recognition evaluation probability reaches the preset recognition target, the following processing is also performed: adjust the recognition weight of the newly constructed feature combination, where the interference feature weight of the randomly added feature is less than the recognition feature weight of the original feature combination; use the recognition evaluation model to evaluate the recognition probability of the newly constructed feature combination. If the recognition evaluation probability does not reach the preset recognition target, continue to add interference features to update the recognition feature combination; among them, perform recognition weight adjustment before each iteration to ensure that the interference feature weight of the randomly added feature is less than the recognition feature weight of the original feature combination, and repeat the iteration until the recognition evaluation probability reaches the preset recognition target.

[0013] This application also provides an automatic loading quantity recognition system for an intelligent factory, including: an order goods composition acquisition module, used to connect to the task platform to obtain loading task information, decompose the loading task information to obtain the order goods composition, and the order goods composition is all the goods information that needs to be assembled in the loading task, including the goods number and the quantity of goods; an interference recognition feature determination module, used to use the goods number of the order goods composition as an index to search the goods management feature library to obtain goods recognition features, and perform interference feature recognition within the order on the goods recognition features to determine interference recognition features; a goods management feature library search module, used to use the goods recognition features as an index to search the goods management feature library to obtain interference adjacent goods and their adjacent interference features in the intelligent factory; a goods recognition feature processing module, used to use the interference recognition features and the adjacent interference features as constraint conditions to perform feature selection processing on the goods recognition features to obtain robust features, and the robust features are core recognition features or core recognition feature combinations that are not affected by interference features; a loading task recognition processing module, used to use the robust features as recognition features to perform loading task recognition processing according to the quantity of goods.

[0014] An automatic loading quantity recognition method and system for an intelligent factory proposed in this application connect to a task platform to obtain loading task information and obtain the composition of order goods; use the goods numbers in the composition of order goods as indexes to search a goods management feature library to obtain goods recognition features, perform interference feature recognition within the order on the goods recognition features to determine interference recognition features; obtain interference adjacent goods and their adjacent interference features within the intelligent factory; perform feature selection processing on the goods recognition features to obtain robust features; use the robust features as recognition features and perform loading task recognition processing according to the quantity of goods. This solves the technical problem that existing loading quantity recognition cannot efficiently and accurately recognize the combination of goods in the face of diversified and customized production requirements, resulting in low loading accuracy and efficiency, and achieves the technical effect of improving the efficiency and accuracy of factory loading. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in sequence. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0016] Figure 1 Schematic flowchart of an automatic loading quantity recognition method for an intelligent factory provided by an embodiment of the present application;

[0017] Figure 2 Schematic structural diagram of an automatic loading quantity recognition system for an intelligent factory provided by an embodiment of the present application.

[0018] Description of reference numerals: Order goods composition acquisition module 10, interference recognition feature determination module 20, goods management feature library search module 30, goods recognition feature processing module 40, loading task recognition processing module 50. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The above description is only an overview of the technical solutions of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the following specifically illustrates the specific embodiments of this application.

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following will further describe this application in detail with reference to the accompanying drawings. The described embodiments should not be construed as limitations on this application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

[0021] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first / second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application.

[0022] An embodiment of this application provides an automatic loading quantity identification method for an intelligent factory, as Figure 1 shown. The method includes:

[0023] Step S100: Connect to the task platform to obtain loading task information, decompose the loading task information, and obtain the order goods composition. The order goods composition is the information of all goods that need to be assembled in the loading task, including the goods number and the quantity of goods.

[0024] Preferably, it is connected to the task platform through a network or other communication means to obtain detailed information about the loading tasks to be executed, which may include customer orders, shipping instructions, loading times, etc. Among them, the task platform is used to manage and distribute various production and logistics tasks within the factory and is usually integrated with a production management system (such as an ERP system) or an order management system (such as an OMS system) to ensure that all tasks can be updated and distributed in real time. Task decomposition refers to decomposing the loading task information obtained from the task platform to obtain the relevant order goods composition. Among them, the loading task contains the classification and quantity of the goods specified for assembly in the order. The order goods composition refers to all the goods information that needs to be assembled in the loading task. Each order usually contains different types and quantities of goods, including the goods number and quantity of each good. The goods number is the unique identifier of the good and is used to identify and manage specific goods in the system, such as barcodes, QR codes, etc. The goods quantity refers to the quantity of each type of good that needs to be loaded in the loading order. By reading the quantity information in the order, it is ensured that the correct quantity can be accurately loaded during loading.

[0025] Step S200: Use the goods number of the order goods composition as an index to search the goods management feature library to obtain the goods identification features, and perform in-order interference feature identification on the goods identification features to determine the interference identification features.

[0026] Preferably, use the numbers of the goods in each order as query keywords or indexes to search in the goods management feature library to find the feature information related to the goods. These features may include the physical characteristics of the goods (such as size, shape, color), packaging methods, identifiers, visual features (such as images or models), etc. Among them, the goods management feature library is a database or information library that stores the feature information related to each type of good. These identification features are the key information used by the system to distinguish different goods in actual operations. By searching in the feature library through the goods number, the identification features related to the good are extracted. For example, for a specific part, the identification features may be its unique shape and color information. Analyze the identification features of all the goods in the order to identify the features that may cause interference. Because in the actual loading process, there may be multiple goods in the order, and there may be similarities or overlaps in their features. These similar features may cause interference or misidentification in the identification process. For example, if two goods have the same color but different shapes, it is necessary to identify and distinguish the interference identification features to avoid confusing them during loading. Finally, determine the interference identification features to take corresponding measures during the loading process to ensure the accuracy and robustness of the final identification result, that is, to still maintain accurate goods identification in the presence of interference.

[0027] In a possible implementation, step S200 further includes step S210 of collecting the physical features, image features, and kit composition features of all incoming goods in the smart factory. The physical features include weight, density, and material. The image features include size, shape, color, texture, and light reflection features. The kit composition features include the number of components in the kit, the installation and assembly position. Step S220 encodes the physical features, image features, and kit composition features, converts them into a storable format, establishes a mapping relationship between the goods number of the incoming goods and the feature code, and constructs the goods management feature library. Among them, for the kit composition features, a multi-level mapping relationship from components to kits and from kits to goods numbers is established.

[0028] Preferably, the goods to be warehoused refer to all goods that are about to be stored in the warehouse in the intelligent factory, which can be raw materials, semi-finished products or finished products. For each good, its key feature information needs to be collected, including physical features, image features, and kit composition features. Specifically, physical features refer to information related to the physical properties of the goods, including the weight, density, and material of the goods. Image features are the appearance features of the goods collected through visual information, including the size, shape, color, texture, and light reflection features of the goods. Kit composition features refer to goods composed of multiple components or parts (such as mechanical equipment or electronic equipment, etc.), including the number of components in a kit (the number of components contained in a kit), installation, and assembly positions (the installation position and assembly sequence of components in the kit); then, the collected physical features, image features, and kit composition features are converted into a digital coding form. The coding can be a combination of numbers, letters, or other specific identifiers, which is convenient for the system to store, retrieve, and process. For example, color can be encoded through RGB or HSV values, size can be represented by specific numerical values, and material can be represented by a standardized code (such as metal can be encoded as "MTL", and plastic can be encoded as "PLS"). Convert the coded data into a format that the system can store and process, usually a structured data format, such as a table format in a database, XML, or JSON file, etc. Establish a mapping relationship between the goods number of the goods to be warehoused and the feature coding, that is, each goods number is associated with the feature coding of the goods, forming a mapping relationship. The feature mapping relationships of multiple goods form a goods management feature library, which records the mapping relationships of the physical features, image features, kit composition features, and their numbers of each good. Through this feature library, relevant goods features can be quickly found and processed when identifying goods, managing inventory, or performing loading tasks; among them, for the kit composition features, a multi-level mapping relationship from components to kits and from kits to goods numbers is established, that is, a mapping relationship is established for each component contained in the goods, clarifying the specific information of each component and its role in the kit. A good may be composed of multiple kits, and each kit is mapped to the final goods number. For example, a goods number may correspond to multiple kits, and each kit corresponds to multiple components. By establishing a multi-level mapping relationship, the system can quickly trace from the goods number to the detailed composition information of each kit and the component information in each kit.

[0029] In a possible implementation, step S220 further includes step S221 of partitioning according to the product number, physical characteristics, image characteristics, and kit composition characteristics; step S222 of integrating with the product number partition of the product management feature library using the product number as the unique index; step S223 of respectively establishing combined indexes with the physical characteristics, image characteristics, and kit composition characteristics as the main indexes and the specific characteristics under the features as the auxiliary indexes; and step S224 of respectively integrating the combined indexes corresponding to each feature category into the physical characteristics, image characteristics, and kit composition characteristics partitions of the product management feature library.

[0030] Preferably, partitioning is performed according to the product number, physical characteristics, image characteristics, and kit composition characteristics. A separate index is established for each category in the product management feature library, and these information are effectively stored and retrieved through multi-level indexes in the feature library to facilitate quick search and matching. Specifically, different types of information are logically divided into independent areas for more efficient management and retrieval, that is, divided into a product number partition, a physical characteristics partition, an image characteristics partition, and a kit composition characteristics partition; using the product number as the unique index, the product number and its related information are stored in the product number partition to ensure that all information of a certain product can be accessed quickly and uniquely; using the physical characteristics, image characteristics, and kit composition characteristics as the main indexes for their respective categories and the specific characteristics under the features as the auxiliary indexes, combined indexes are respectively established. Among them, the main index is the main index for quickly finding specific category data, and the auxiliary index is used to further refine the query based on the main index. For example, the physical characteristics are the main index of this category, and weight, density, material, etc. are the auxiliary indexes. The combination of the main index and the auxiliary index forms a combined index, that is, preliminary screening can be first performed according to the main index (such as physical characteristics), and then refined retrieval can be performed through the auxiliary index; the combined indexes established above are respectively integrated into different feature partitions, that is, the combined index of physical characteristics is integrated into the physical characteristics partition, the combined index of image characteristics is integrated into the image characteristics partition, and the combined index of kit composition characteristics is integrated into the kit composition characteristics partition.

[0031] In a possible implementation, step S200 further includes step S230 of calculating the feature similarity respectively according to the physical features, image features, and kit composition features to obtain the similarity of each feature category; step S240 of establishing an N×N matrix, where N is the number of product types, and each element represents the similarity of two products under a certain feature category; step S250 of filling the matrix according to the feature category similarity, and performing weighted average of the similarity for each pair of product combinations to obtain the comprehensive similarity value; step S260 of using the comprehensive similarity value to assign values to the matrix, and when the matrix element reaches the preset interference threshold, determining the corresponding feature as the interference recognition feature.

[0032] Preferably, calculate the feature similarity respectively according to physical features such as weight, density, and material, image features such as the size, shape, color, and texture of the product, and kit composition features such as the number of components and assembly positions of the product, to obtain the similarity value of each pair of products under different feature categories; then establish an N×N matrix based on these calculated feature similarities, where N represents the number of product types. For example, if there are 5 products, then the size of the matrix is 5×5, and each element of the matrix represents the similarity of two products under a certain feature category; fill the calculation results of the similarities of physical features, image features, and kit composition features into different parts of the matrix respectively. Each pair of product combinations will have a corresponding similarity value in the matrix. Perform weighted average of the similarity for each pair of product combinations to obtain the comprehensive similarity value, that is, assign different weights to physical features, image features, and kit composition features and calculate the comprehensive similarity value; then fill these values into the matrix. For example, if the comprehensive similarity of products A and B is 0.85, then the element at the position of A and B in the matrix will be assigned a value of 0.85. The preset interference threshold is a critical value used to judge whether two products are similar enough to cause recognition confusion or interference. If the similarity of two products exceeds the preset interference threshold, it indicates that some features of these two products may be too similar and are likely to cause interference in the recognition process. Then determine the corresponding feature as the interference recognition feature. For example, if the similarity of two products in image features such as color and size is too high, then the color and size features are marked as interference features.

[0033] In a possible implementation, step S250 further includes step S251, obtaining a set of loading exception cases, clustering the set of loading exception cases according to factory operation processes and item numbers, and constructing exception case clusters; step S252, through the exception case clusters, respectively fitting the exception features and exception probabilities for the factory operation processes and item numbers to obtain the exception features and exception probabilities of the factory operation processes and the exception features and exception probabilities of the item numbers; step S253, extracting the factory operation processes of the loading task according to the loading task information, and matching the exception features and exception probabilities of the processes; step S254, according to the item numbers composed of the order goods, matching the exception features and exception probabilities of the item numbers; step S255, according to the exception features and exception probabilities of the matched processes and the exception features and exception probabilities of the item numbers, performing exception probability matching on the features of each pair of goods combinations and their feature category similarities, and calculating the average exception probability, where the average exception probability is the average of the exception probabilities of each good in the goods combination in this feature category and the exception probability of the matched process; step S256, using the average exception probability to weight the comprehensive similarity, and using the weighting to perform similarity weighting calculation on each pair of goods combinations to obtain the comprehensive similarity value.

[0034] Preferably, the set of loading exception cases is a set of all exception situations recorded by the factory during the loading process. For example, misloading, missing loading, and incorrect item identification during the loading process can all be regarded as exception cases. According to the different operation processes of the factory (such as loading, inspection, packaging) and item numbers, these exception cases are clustered and analyzed, and similar exception cases are grouped together to identify common features. For example, if the same type of goods frequently has exceptions under the same process, they are clustered into one cluster, forming several exception case clusters, and each cluster represents common exceptions under similar operation processes or item numbers; exception features refer to features that often appear in a certain factory operation process or a certain item number. By analyzing the exception case clusters, the occurrence probabilities of exceptions for each process and item number are fitted to determine the probability of exception events occurring; when performing a loading task, extract the specific processes involved in the task, such as loading, sorting, packaging, etc., and then match the extracted processes with the previously analyzed exception features and probabilities to determine what the exception features of each process are and what the probability of exception occurrence is in the current loading task; according to the item numbers in the order, match the exception features and exception probabilities in the exception case clusters to find the common exception situations and their probabilities of this item in historical operations.

[0035] Preferably, then match the similarity of each pair of goods combinations with their corresponding process anomaly features and goods number anomaly features, and calculate the anomaly probabilities of these feature categories. That is, calculate the possibility of anomalies occurring under specific process and goods feature combinations. For example, if the similarity of two goods in terms of color is very high (0.9), and color recognition errors are common anomaly features, then the anomaly probability may be relatively high; weight-average the anomaly probabilities of each pair of goods' feature categories with the anomaly probabilities of their corresponding processes to calculate the mean anomaly probability, which represents the possibility of anomalies occurring in this specific operation. For example, for the comprehensive similarity calculation of goods a and goods b in terms of color features, assume that the historical confusion probability of goods a in this process flow is 0.4 and the color probability is 0.1, then the mean anomaly probability is 0.25. For goods b, the historical process confusion anomaly is 0.5 and the color confusion probability is 0.05, and the probability mean is 0.275. When weighting, it should be noted that the sum of the weights of the same pair of goods features is 1. Therefore, convert according to 0.25 and 0.275 to determine the weights. In this calculation, the weight of goods a = 0.25 / (0.25 + 0.275), and the weight of goods b = 0.275 / (0.25 + 0.275); adjust the comprehensive similarity value according to the mean anomaly probability of each pair of goods combinations. The higher the anomaly probability, the more likely it is that the features of this pair of goods are likely to cause problems during the operation. Therefore, the weight of the similarity may be reduced, and finally a weighted comprehensive similarity value is obtained, which not only considers the similarity of goods features but also combines the possibility of anomalies occurring.

[0036] Step S300: Search the goods management feature library using the goods identification features as the index to obtain the interfering neighboring goods and their neighboring interference features within the intelligent factory.

[0037] Preferably, the goods identification features are used as index keywords to further search in the goods management feature library, and other neighboring goods and their features related to the current goods identification features are found to ensure that goods with similar features to the target goods that may cause identification confusion can be identified, that is, the interfering neighboring goods and their neighboring interference features in the intelligent factory are obtained. The interfering neighboring goods refer to the goods in the intelligent factory that are adjacent to the target goods and have similar or interfering features. For example, the neighboring goods may be physically close to the target goods (such as their positions in the warehouse or on the production line), or they may have similar features such as color, shape, etc., and are similar to the target goods during the goods identification process, resulting in identification confusion or misjudgment. After finding the interfering neighboring goods, the interference features of these goods will be further extracted and considered in the subsequent identification process. The neighboring interference features refer to the features of the interfering neighboring goods, which may be similar to or overlap with the features of the target goods, thus causing interference in the goods identification. For example, if two goods are similar in color or close in shape, these features will be marked as interference features.

[0038] Step S400: Using the interference identification features and the neighboring interference features as constraint conditions, perform feature selection processing on the goods identification features to obtain robust features. The robust features are the core identification features or combinations of core identification features that are not affected by the interference features.

[0039] Preferably, an optimization search is used to find the feature or feature combination with the least interference. Specifically, the interference recognition feature and the adjacent interference feature are used as constraints, which means that when performing feature selection, the interference recognition feature and the adjacent interference feature are used as a kind of restrictive condition or filtering rule to restrict the features that are easily interfered, and preferentially select the features that are not affected by the interference recognition feature and the adjacent interference feature, avoiding misjudgment caused by interference and ensuring the robustness of the system. Feature selection processing refers to selecting the most effective features from multiple goods recognition features as the basis for identifying goods. The goal of feature selection is to reduce the data dimension while retaining the most discriminative and least interfered features. For example, the principal component analysis method is used to perform feature selection processing on the goods recognition features to obtain robust features. The selection of robust features ensures that the target goods can be correctly identified even in the presence of interference factors. For example, if the color is easily interfered, but the shape is unique and less interfered, then the shape can be used as a robust feature. Among them, the robust feature is the core recognition feature or core recognition feature combination that is not affected by the interference feature, which means that the robust feature can be a single feature (such as shape) or a combination of multiple features (such as shape + label). By combining different robust features, a high goods recognition accuracy can be maintained in various interference environments. The core recognition feature refers to the most discriminative and important feature in the process of goods recognition. These features can accurately and quickly identify goods and are not easily affected by interference factors in the environment. For example, in the case of similar colors, the size may become the core recognition feature because the size is relatively stable and not easily misidentified; sometimes a single feature cannot provide sufficient discrimination ability, and a combination of multiple features is required to form the core recognition feature combination. For example, the combination of color and shape may be more reliable than a single shape or color. By combining multiple non-interfered features, the target goods can be more effectively identified in a complex loading environment (such as goods stacking, adjacent goods being similar, etc.), reducing the goods recognition error rate and maintaining high accuracy and reliability of goods recognition.

[0040] In a possible implementation, step S400 further includes step S410 of removing the interference recognition feature and the adjacent interference feature as removal features from the goods recognition feature to construct a pure recognition feature set; step S420 of randomly sampling from the pure recognition feature set to construct a number of feature subsets, where the subsets include different numbers of features or feature combinations, and evaluating the recognition probability of each random feature or feature combination through a recognition evaluation model to obtain a recognition evaluation probability, and screening the random sampling feature with the highest recognition evaluation probability as an alternative robust feature; step S430 of determining whether the recognition evaluation probability reaches a preset recognition target, and if so, using the alternative robust feature as the robust feature; step S440 of, when it does not reach, using the interference recognition feature and the adjacent interference feature as additional features, randomly selecting from the additional features, adding the randomly added features to the feature subsets to obtain a newly constructed feature combination, and evaluating the recognition probability of the newly constructed feature combination using the recognition evaluation model until the recognition evaluation probability reaches the preset recognition target, and determining the robust feature.

[0041] Preferably, the determined interference recognition feature and adjacent interference feature are removed from the goods recognition feature set as removal features, and the remaining goods recognition features are constructed into a pure recognition feature set that no longer contains features that may cause confusion or interference; randomly select features from the pure recognition feature set to generate a number of different feature subsets, which may include different numbers of features or feature combinations. For example, one feature subset may only include size and color, and another subset may include shape, color, and weight, with the aim of finding the best robust feature combination among multiple feature combinations; evaluating the recognition probability of each random feature or feature combination through a recognition evaluation model refers to calculating the recognition probability of each feature subset through the evaluation model, that is, the success probability of the subset in goods recognition. The higher the recognition probability, the more accurately and effectively the feature subset can identify the goods. Among them, the recognition evaluation model is constructed based on machine learning (such as a classifier or regression model), and evaluates the performance of each feature subset by simulating the goods recognition process, and screening the subset with the highest recognition evaluation probability from all feature subsets as the alternative robust feature.

[0042] Preferably, it is determined whether the recognition evaluation probability reaches a preset recognition target, which is the recognition accuracy target preset by the system in advance. For example, the system requires the recognition accuracy rate to reach more than 95%. If the recognition probability of the alternative robust feature reaches or exceeds this preset target, the alternative robust feature will be officially selected as the robust feature for accurate identification of goods; if the recognition probability of the alternative feature fails to reach the preset target, new features will be introduced, that is, the interference recognition features and adjacent interference features that were previously eliminated will be used as potentially useful additional features. Some features will be randomly selected from the additional features and added to the feature subset to construct a new feature combination, enhance the recognition ability, and make up for the deficiencies of the current feature set. Then, through the recognition evaluation model, the newly constructed feature combinations will be evaluated, their recognition probabilities will be calculated, and it will be determined whether the new feature combinations are more effective than the previous ones. If the new feature combinations still fail to reach the preset recognition target, continue to randomly sample from the additional features, construct feature combinations and evaluate them until a feature combination that meets the target requirements is found. When the recognition evaluation probability of a certain feature combination reaches or exceeds the preset recognition target, the feature combination will be determined as the final robust feature.

[0043] In a possible implementation manner, step S440 further includes step S441 of adjusting the recognition weight of the newly constructed feature combination, where the interference feature weight of the randomly added feature is less than the recognition feature weight of the original feature combination; step S442 of evaluating the recognition probability of the newly constructed feature combination through the recognition evaluation model. If the recognition evaluation probability does not reach the preset recognition target, continue to add interference features to update the recognition feature combination; step S443, where the recognition weight is adjusted before each iteration to ensure that the interference feature weight of the randomly added feature is less than the recognition feature weight of the original feature combination, and repeat the iteration until the recognition evaluation probability reaches the preset recognition target.

[0044] Preferably, in order to avoid interference or weaken the influence of interference features, the recognition weight of the newly constructed feature combination is adjusted, that is, the weight of each feature in the newly constructed feature combination is adjusted to optimize the recognition effect, and it is ensured that the interference feature weight of the randomly added features is less than the recognition feature weight of the original feature combination, reducing the influence of interference features in the recognition process and preventing them from interfering too much with the accurate recognition of goods; the recognition probability of the adjusted new feature combination is calculated through the recognition evaluation model, that is, the accuracy and effect of these feature combinations in recognizing goods are judged. The higher the recognition probability, the better the effect of the feature combination in goods recognition; a low recognition probability indicates that the performance of the feature combination is poor; if the recognition probability of the newly constructed feature combination does not reach the system preset target (such as the recognition accuracy rate needs to reach a certain threshold), more features are randomly selected from the interference features and added to the current feature combination, and an attempt is made to improve the recognition effect by expanding the feature set. The weight is adjusted before adding new interference features in each iteration to ensure that the weight of the newly added interference features must always be less than the weight of the original recognition features, avoiding too many interference factors from affecting the final recognition effect. The feature combination is adjusted and the recognition probability is calculated each time until the preset recognition target is reached, the iteration is stopped, and the current feature combination is regarded as the final robust feature combination.

[0045] Step S500: Using the robust features as recognition features, perform loading task recognition processing according to the quantity of the goods.

[0046] Preferably, the obtained robust features are used as recognition features, that is, each item is matched according to these robust features to ensure that the item can be accurately recognized. For example, the shape and barcode of the item are used as the core recognition features. When the system scans the item, different items will be distinguished by these features, and the loading task recognition of the item quantity will be performed. The item quantity refers to that each order usually contains multiple different items and their corresponding quantity information. The loading task requires completing the loading task according to the specific quantity of each item in the order. For example, the order may require loading 10 parts A and 5 parts B. The loading process not only needs to identify each item, but also determine whether the loading task is completed according to the quantity information in the order. Specifically, the identified item features are compared with the quantity required in the order. For example, if the order requires loading 10 parts A, the system will automatically record the quantity of each item recognized during loading until 10 parts A are recognized cumulatively, then the loading task of this item is considered completed. If the recognized item quantity is less than or more than the requirement in the order, the system will issue a prompt or warning to ensure that there is no overloading or underloading of items. For example, if the order requires 5 parts B and the system recognizes that 6 parts B are loaded, the loader will be prompted to make adjustments. After each item is recognized and its quantity is confirmed, the system will update the execution status of the task to ensure real-time tracking of the progress of the entire loading task; when the recognition and quantity matching of all items are completed, it is confirmed that the loading task has been completed, ensuring the accuracy and integrity of each loading task recognition, thereby improving the automation level and efficiency of the intelligent factory.

[0047] In the above text, reference is made to Figure 1 A method for automatically identifying the loading quantity for an intelligent factory according to an embodiment of the present invention is described in detail. Next, a system for automatically identifying the loading quantity for an intelligent factory according to an embodiment of the present invention will be described with reference to Figure 2 A system for automatically identifying the loading quantity for an intelligent factory according to an embodiment of the present invention is used to solve the technical problem that the existing loading quantity identification cannot efficiently and accurately identify the item combination in the face of diverse and customized production requirements, resulting in low loading accuracy and efficiency, and achieves the technical effect of improving the efficiency and accuracy of factory loading. A system for automatically identifying the loading quantity for an intelligent factory includes: an order item composition acquisition module 10, an interference recognition feature determination module 20, an item management feature library search module 30, an item recognition feature processing module 40, and a loading task recognition processing module 50.

[0048] The order item composition acquisition module 10 is used to connect to the task platform to obtain loading task information, decompose the loading task information, and obtain the order item composition. The order item composition is all the item information that needs to be assembled in the loading task, including item numbers and item quantities;

[0049] The order item composition acquisition module 10 is used to connect to the task platform to obtain loading task information, decompose the loading task information, and obtain the order item composition. The order item composition is all the item information that needs to be assembled in the loading task, including item numbers and item quantities;

[0050] An interference recognition feature determination module 20, configured to use the product number composed of the order products as an index to search the product management feature library, obtain product recognition features, perform interference feature recognition within the order on the product recognition features, and determine interference recognition features;

[0051] A product management feature library search module 30, configured to use the product recognition features as an index to search the product management feature library, and obtain interfering adjacent products and their adjacent interference features within the intelligent factory;

[0052] A product recognition feature processing module 40, configured to use the interference recognition features and the adjacent interference features as constraint conditions to perform feature selection processing on the product recognition features, and obtain robust features, where the robust features are core recognition features or combinations of core recognition features that are not affected by interference features;

[0053] A loading task recognition processing module 50, configured to use the robust features as recognition features and perform loading task recognition processing according to the quantity of the products.

[0054] Next, the specific configuration of the interference recognition feature determination module 20 will be described in detail. The interference recognition feature determination module 20 may further include: collecting the physical features, image features, and kit composition features of all incoming products in the intelligent factory, where the physical features include weight, density, and material, the image features include size, shape, color, texture, and light reflection features, and the kit composition features include the number of kit components and the installation and assembly positions; encoding the physical features, image features, and kit composition features, converting them into a storable format, establishing a mapping relationship between the product numbers of the incoming products and the feature codes, and constructing the product management feature library, where a multi-level mapping relationship from components to kits and from kits to product numbers is established for the kit composition features.

[0055] Next, the specific configuration of the interference recognition feature determination module 20 will be further described in detail. The interference recognition feature determination module 20 may further include: partitioning according to the product numbers, physical features, image features, and kit composition features; integrating into the product number partition of the product management feature library with the product numbers as the only index; respectively establishing combined indexes with the physical features, image features, and kit composition features as the main indexes and the specific features under the features as the auxiliary indexes; and respectively integrating the combined indexes corresponding to each feature category into the physical feature, image feature, and kit composition feature partitions of the product management feature library.

[0056] Next, the specific configuration of the interference recognition feature determination module 20 will be further described in detail. The interference recognition feature determination module 20 may further include: calculating the feature similarity respectively according to the physical features, image features, and kit composition features to obtain the similarity of each feature category; establishing an N×N matrix, where N is the number of product types, and each element represents the similarity of two products under a certain feature category; filling the matrix according to the feature category similarity, performing similarity weighted averaging on each pair of product combinations to obtain a comprehensive similarity value; using the comprehensive similarity value to assign values to the matrix, and when the matrix element reaches a preset interference threshold, determining the corresponding feature as the interference recognition feature.

[0057] Next, the specific configuration of the interference recognition feature determination module 20 will be further described in detail. The interference recognition feature determination module 20 further includes: obtaining a set of loading abnormal cases, clustering the set of loading abnormal cases according to the factory operation process and product number to construct an abnormal case cluster; through the abnormal case cluster, respectively fitting the abnormal features and abnormal probabilities of the factory operation process and product number to obtain the abnormal features and abnormal probabilities of the factory operation process, and the abnormal features and abnormal probabilities of the product number; extracting the factory operation process of the loading task according to the loading task information, and matching the abnormal features and abnormal probabilities of the process; according to the product numbers of the order product composition, matching the abnormal features and abnormal probabilities of the product numbers; according to the abnormal features and abnormal probabilities of the matching process, and the abnormal features and abnormal probabilities of the product numbers, performing abnormal probability matching on the features of each pair of product combinations and their feature category similarities, and calculating the average abnormal probability, where the average abnormal probability is the average of the abnormal probabilities of each product in the product combination in this feature category and the abnormal probability of the matching process; using the average abnormal probability to weight the comprehensive similarity, and using the weighted value to perform similarity weighted calculation on each pair of product combinations to obtain the comprehensive similarity value.

[0058] Next, the specific configuration of the goods identification feature processing module 40 will be described in detail. The goods identification feature processing module 40 further includes: removing the interference identification feature and the adjacent interference feature as rejection features from the goods identification features to construct a pure identification feature set; randomly sampling from the pure identification feature set to construct a number of feature subsets, where the subsets include different numbers of features or feature combinations, and evaluating the recognition probability of each random feature or feature combination through a recognition evaluation model to obtain a recognition evaluation probability, and screening the random sampling feature with the highest recognition evaluation probability as an alternative robust feature; determining whether the recognition evaluation probability reaches a preset recognition target, and if so, using the alternative robust feature as the robust feature; when it does not reach, using the interference identification feature and the adjacent interference feature as additional features, randomly selecting from the additional features, and adding the randomly added features to the feature subset to obtain a newly constructed feature combination, and using the recognition evaluation model to evaluate the recognition probability of the newly constructed feature combination until the recognition evaluation probability reaches the preset recognition target, and determining the robust feature.

[0059] Next, the specific configuration of the goods identification feature processing module 40 will be further described in detail. The goods identification feature processing module 40 may further include: adjusting the recognition weight of the newly constructed feature combination, where the interference feature weight of the randomly added feature is less than the recognition feature weight of the original feature combination; evaluating the recognition probability of the newly constructed feature combination through the recognition evaluation model, and if the recognition evaluation probability does not reach the preset recognition target, continue to add interference features to update the recognition feature combination; where, before each iteration, the recognition weight is adjusted to ensure that the interference feature weight of the randomly added feature is less than the recognition feature weight of the original feature combination, and the iteration is repeated until the recognition evaluation probability reaches the preset recognition target.

[0060] The automatic loading quantity identification system for an intelligent factory provided by an embodiment of the present invention can execute the automatic loading quantity identification method for an intelligent factory provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method.

[0061] Although this application makes various references to certain modules in the system according to the embodiments of this application, however, any number of different modules can be used and run on the user terminal and / or the server. The included various units and modules are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0062] The above specific embodiments do not constitute a limitation on the protection scope of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. A method for automatically identifying loading quantity for a smart factory, characterized in that: The method comprises: Connecting to the task platform to obtain loading task information, decomposing the loading task information, and obtaining the order goods composition, wherein the order goods composition is the information of all goods that need to be assembled in the loading task, including the goods number and the goods quantity; Using the product number of the order product composition as an index, searching the product management feature library to obtain product identification features, performing order interference feature identification on the product identification features, and determining the interference identification features; Using the commodity identification feature as an index, searching the commodity management feature library to obtain interfering neighboring commodities and their neighboring interference features in the smart factory; Taking the interference identification feature and the adjacent interference feature as constraint conditions, performing feature selection processing on the product identification feature to obtain a robust feature, wherein the robust feature is a core identification feature or a combination of core identification features that is not affected by the interference feature; Using the robust features as identification features, performing loading task identification processing according to the quantity of goods; The searching of the product management feature library also includes: Collect the physical characteristics, image characteristics, and kit composition characteristics of all incoming goods in the smart factory. The physical characteristics include weight, density, and material. The image characteristics include size, shape, color, texture, and light reflection characteristics. The kit composition characteristics include the number of kit components and installation and assembly positions. Encode the physical features, image features, and kit composition features, convert them into a storable format, establish a mapping relationship between the product number and the feature code of the incoming goods, and construct the product management feature library, wherein a multi-level mapping relationship from component to kit and from kit to product number is established for the kit composition features; Performing order interference feature identification on the product identification feature to determine the interference identification feature includes: According to the physical features, image features, and kit composition features, feature similarity calculations are performed respectively to obtain similarities of each feature category; Create an N×N matrix, where N is the number of product categories and each element represents the similarity between two products under a certain feature category; Filling the matrix according to the similarity of the feature categories, performing weighted average of the similarity of each pair of product combinations, and obtaining a comprehensive similarity value; The matrix is ​​assigned values ​​using the comprehensive similarity value, and when a matrix element reaches a preset interference threshold, a corresponding feature is determined as the interference identification feature.

2. The method for automatically identifying the loading quantity for a smart factory according to claim 1, characterized in that: Constructing the product management feature library also includes: Partitioning is performed according to the product number, physical features, image features, and kit composition features; The product number is used as a unique index and integrated into the product number partition of the product management feature library; The physical feature, the image feature, and the kit component feature are respectively used as the primary index, and the specific features under the feature are used as the auxiliary index, and a combined index is respectively established; The combined index corresponding to each feature category is respectively integrated into the physical feature, image feature, and kit composition feature partitions of the product management feature library.

3. The method for automatically identifying the loading quantity for a smart factory according to claim 1, characterized in that: The obtaining of the comprehensive similarity value further includes: Obtaining a set of loading abnormality cases, clustering the set of loading abnormality cases according to factory operation procedures and product numbers, and constructing abnormal case clusters; Through the abnormal case cluster, the abnormal characteristics and abnormal probability of the factory operation process and the product number are fitted respectively to obtain the abnormal characteristics and abnormal probability of the factory operation process and the abnormal characteristics and abnormal probability of the product number; Extracting the factory operation process of the loading task according to the loading task information, and matching the abnormal characteristics and abnormal probability of the process; According to the product number of the order product composition, match the abnormal characteristics and abnormal probability of the product number; According to the abnormal features and abnormal probabilities of the matching process and the abnormal features and abnormal probabilities of the product numbers, abnormal probability matching is performed on each pair of product combinations and the similarity of their feature categories, and the average abnormal probability is calculated. The average abnormal probability is the average of the abnormal probabilities of each product in the product combination in the feature category and the abnormal probability of the matching process; The comprehensive similarity is weighted using the abnormal probability mean, and the weighted similarity calculation is performed on each pair of product combinations to obtain the comprehensive similarity value.

4. The method for automatically identifying the loading quantity for a smart factory according to claim 1, characterized in that: Taking the interference identification feature and the adjacent interference feature as constraint conditions, performing feature selection processing on the product identification feature to obtain a robust feature includes: Eliminate the interference identification feature and the adjacent interference feature as elimination features from the product identification feature to construct a pure identification feature set; Randomly sampling from the pure recognition feature set to construct several feature subsets, the subsets include different numbers of features or feature combinations, performing recognition probability evaluation on each random feature or feature combination through the recognition evaluation model to obtain the recognition evaluation probability, and selecting the randomly sampled feature with the largest recognition evaluation probability as a candidate robust feature; Determine whether the recognition evaluation probability reaches a preset recognition target, and if so, use the candidate robust feature as the robust feature; When it is not reached, the interference identification feature and the adjacent interference feature are used as additional features, randomly selected from the additional features, and added to the feature subset using the randomly added features to obtain a new feature combination, and the recognition probability of the new feature combination is evaluated using the recognition evaluation model until the recognition evaluation probability reaches the preset recognition target, thereby determining the robust feature.

5. The method for automatically identifying the loading quantity for a smart factory as claimed in claim 4, characterized in that: Using the recognition evaluation model to evaluate the recognition probability of the newly constructed feature combination until the recognition evaluation probability reaches the preset recognition target, further comprising: Adjusting the recognition weight of the newly constructed feature combination, wherein the interference feature weight of the randomly added feature is less than the recognition feature weight of the original feature combination; The recognition evaluation model is used to evaluate the recognition probability of the newly constructed feature combination. If the recognition evaluation probability does not reach the preset recognition target, interference features are continuously added to update the recognition feature combination. Among them, the recognition weight adjustment is performed before each iteration to ensure that the interference feature weight of the randomly added feature is smaller than the recognition feature weight of the original feature combination, and the iteration is repeated until the recognition evaluation probability reaches the preset recognition target.

6. An automatic loading quantity identification system for smart factories, characterized in that: The system is used to implement the method for automatically identifying the loading quantity for a smart factory according to any one of claims 1 to 5, and the system comprises: An order goods composition acquisition module is used to connect to the task platform to obtain loading task information, perform task decomposition on the loading task information, and obtain order goods composition, wherein the order goods composition is information on all goods that need to be assembled in the loading task, including the goods number and the goods quantity; An interference identification feature determination module is used to use the product number of the order product composition as an index to search the product management feature library, obtain the product identification feature, perform order interference feature identification on the product identification feature, and determine the interference identification feature; A product management feature library search module, used to search the product management feature library with the product identification feature as an index to obtain interfering neighboring products and their neighboring interference features in the smart factory; A product identification feature processing module, used to use the interference identification feature and the adjacent interference feature as constraints, perform feature selection processing on the product identification feature, and obtain a robust feature, wherein the robust feature is a core identification feature or a combination of core identification features that is not affected by the interference feature; The loading task identification processing module is used to use the robust features as identification features and perform loading task identification processing according to the quantity of goods.

Images