Intelligent Inbound Classification Method and System for Warehouse Goods Based on Volume and Weight Characteristics

Abstract

This invention relates to the technical field of intelligent sorting and storage of goods, and discloses a method and system for intelligent warehousing and classification of goods based on volume and weight characteristics. The intelligent warehousing and classification method is applied to electronic sorting control equipment and specifically includes the following steps: S11: Receiving information about goods entering the warehouse warehousing port; a 3D scanning device, weighing device, and information entry terminal deployed at the warehouse warehousing port work collaboratively to perform raw data acquisition for each piece of goods entering the warehouse. The raw volume data, raw weight data, and basic attribute information collected are packaged into raw data records. This invention utilizes a multi-dimensional compensation dynamic calibration process, including inverse volume estimation based on material density, equivalent volume calculation for fluid particulate goods, temperature and humidity weight compensation, and volume-weight correlation calibration, effectively eliminating measurement deviations caused by irregular shapes, material differences, and environmental factors.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent sorting and storage of goods, and more specifically, to a method and system for intelligent warehousing and classification of goods based on volume and weight characteristics. Background Technology

[0002] With the rapid development of e-commerce, intelligent manufacturing and cold chain logistics, warehouse management systems are continuously evolving towards automation and intelligence. As the starting point of the warehousing operation chain, the accuracy and efficiency of goods classification directly affect the overall efficiency of subsequent location allocation, picking route planning and inventory counting.

[0003] Currently, the warehousing cargo classification technologies commonly used in the industry are mainly divided into two categories: one is a simple classification method based on a single dimension, that is, grouping goods only based on volume data or only based on weight data. This type of method is simple to implement but does not make full use of information and is very easy to ignore the inherent relationship between volume and weight, resulting in a serious deviation between the classification results and actual storage needs; the other is a manual-assisted classification method, in which operators judge the category based on the appearance or labels of the goods based on experience. This is not only inefficient but also greatly affected by subjective factors, making it difficult to guarantee the consistency and accuracy of classification.

[0004] In recent years, a few studies have attempted to achieve multidimensional classification by simply superimposing or linearly combining volume and weight, such as forming two-dimensional grid partitions by setting fixed volume and weight thresholds. However, such schemes still fall under the category of static multi-threshold classification, failing to consider the impact of irregular cargo shapes, material density differences, and environmental temperature and humidity on weighing accuracy. They also fail to address measurement deviations in special scenarios such as hollow cargo, fluid granular cargo, and irregularly shaped cargo. Furthermore, most existing warehouse classification systems adopt a classification-termination architecture, where classification results are only used for labeling goods or entering data into a database. There is a lack of real-time linkage mechanisms between these systems and subsequent steps such as intelligent location allocation, dynamic planning of picking tasks, and category-based inventory checks, significantly reducing the application value of classification information. For example, existing technologies such as Chinese patent CN116821763A, which discloses a warehouse goods classification method, mention multidimensional features but do not involve accurate calibration of volume and weight or adaptive classification algorithms. Chinese patent CN104021426A focuses on location optimization based on multidimensional elements but does not focus on the dynamic adaptation capabilities of the inbound classification process itself.

[0005] In summary, current warehousing and sorting technologies still have significant shortcomings in terms of data collection accuracy, adaptive classification rules, end-to-end collaborative efficiency, and the ability to handle abnormal goods. These shortcomings hinder the intelligent upgrading of warehouse management systems and the improvement of overall operational efficiency. There is an urgent need for a new warehousing and sorting technology solution that can achieve dynamic calibration of volume and weight and adaptive classification. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for intelligent warehousing and classification of goods based on volume and weight characteristics. Through a multi-dimensional compensation and dynamic calibration process, including inverse volume estimation based on material density, equivalent volume calculation for fluid particulate goods, temperature and humidity weight compensation, and volume-weight correlation calibration, the method effectively eliminates measurement deviations caused by irregular shapes, material differences, and environmental factors.

[0007] This invention is implemented as follows: a smart warehousing and classification method for stored goods based on volume and weight characteristics, applied to electronic sorting control equipment, specifically including the following steps: S11: Receives information about goods entering the warehouse. The 3D scanning equipment, weighing equipment and information entry terminal deployed at the warehouse entry port work together to perform raw data collection operations on each piece of goods entering the warehouse. After the raw volume data, raw weight data and basic attribute information are collected, they are packaged into raw data records, and after adding timestamps and goods identification codes, they are transmitted to the data processing module. S12: After receiving the original data record, the data processing module starts a dynamic calibration process based on multi-dimensional compensation to eliminate the influence of cargo shape irregularity, material density difference, and environmental temperature and humidity on weighing accuracy, and generates calibrated accurate volume data, accurate weight data and weighted volume weight characteristic data. S13: Input the obtained precise volume data, precise weight data, and weighted volume-weight feature vector into the machine learning model deployed in the adaptive classification module. The machine learning model first automatically matches the corresponding category sub-rule library based on the goods category in the basic attribute information. Then, it uses dynamic threshold adjustment to compare the calibrated precise volume data and precise weight data with the threshold in the category sub-rule library. At the same time, it introduces the weighted volume-weight feature vector for joint judgment and outputs the initial classification result of the goods. S14: The initial classification results are synchronized to the warehouse management system in real time via the API interface, triggering linkage operations with location allocation, picking planning, and inventory counting, forming a closed-loop management from inbound classification to subsequent operations. S15: During the closed-loop management process from warehousing and classification to subsequent operations, anomaly identification is performed. The volume data, weight data, and three-dimensional contour data of the goods are monitored in real time. The volume data is compared with the preset upper limit threshold for warehousing volume, and the weight data is compared with the preset upper limit threshold for weight. Oversize warning and overweight warning are triggered respectively. After anomaly identification is completed, the process enters the hierarchical fallback process to achieve reliable classification of all goods.

[0008] Furthermore, in S11, the 3D scanning equipment, weighing equipment, and information entry terminals deployed at the warehouse inbound port work together to perform raw data collection operations on each inbound item, including: The 3D scanning device uses multi-view scanning to scan the outline of the goods from at least three different spatial angles and obtain complete 3D point cloud data of the goods. Based on the three-dimensional point cloud data, the original volume data of the cargo is calculated to generate a three-dimensional contour shape feature vector of the cargo. The weighing device performs at least three independent repeated weighing operations when the goods are stationary, records the weighing reading for each operation, and uses the arithmetic mean of the three readings as the original weight data of the goods. The information entry terminal obtains basic attribute information of goods by scanning a code or manually entering the information. The basic attribute information includes at least the category of goods, material type, fragility level, and expected storage duration.

[0009] Furthermore, in S12, after receiving the raw data record, the data processing module initiates a dynamic calibration process based on multi-dimensional compensation to eliminate the effects of cargo shape irregularities, material density differences, and environmental temperature and humidity on weighing accuracy, including: First, volume calibration is performed. Based on the material type in the basic attribute information, the theoretical density range of the goods is obtained. The theoretical weight range is calculated by combining the original volume data of the goods. The theoretical weight range is compared with the original weight data. If the deviation exceeds the preset threshold, the goods are determined to be hollow or loosely structured. At this point, the effective volume is calculated based on the material density to replace the original volume data. Meanwhile, the three-dimensional contour shape feature vector is matched with the standard shape template library. For deformed or irregular goods with a matching degree lower than the preset value, compensation calculation is performed to generate corrected and accurate volume data. For goods whose material type is identified as fluid or particulate, the system automatically switches to the equivalent volume calculation model. The theoretical volume is calculated in reverse based on the accurate weight data and standard density as the effective cargo volume, while the outer packaging volume is reserved for subsequent cargo space matching. Next, weight calibration is performed. Temperature and humidity sensors deployed around the weighing equipment collect the current ambient temperature and humidity values. A pre-trained temperature and humidity-weight compensation model is called to establish the mapping relationship between environmental factors and weighing deviation. The weight compensation coefficient is calculated based on the current temperature and humidity values. The arithmetic mean of the original weight data is compensated and corrected to generate accurate weight data after correction. Finally, perform volume-weight correlation calibration. According to the category and fragility level of the goods, dynamically allocate volume weight factors and weight weight factors, establish a weighted volume weight feature vector, calculate the ratio of volume to weight, compare this ratio with the standard ratio range of the same category, and give a secondary calibration prompt for the goods outside the range.

[0010] Further, in S13, input the obtained accurate volume data, accurate weight data, and weighted volume weight feature vector into the machine learning model deployed in the adaptive classification module, including: The machine learning model uses the gradient boosting decision tree algorithm and has been pre-trained based on the historical inbound dataset. The historical inbound dataset at least includes the volume, weight, category, material, subsequent storage location turnover efficiency, and picking frequency labels of past goods; The machine learning model automatically mines the implicit mapping relationship between different volume weight features and the optimal classification results through learning, and generates an initial classification rule set. The classification rule set includes multiple classification levels and the volume thresholds, weight thresholds, and volume-weight joint thresholds corresponding to each level.

[0011] Further, introduce the weighted volume weight feature vector for joint determination at the same time, and output the initial classification result of the goods, including: When the cumulative historical inbound dataset of the same category reaches the preset sample size, the machine learning model starts the online learning sub-step, automatically extracts the feature data of newly inbound goods and their subsequent warehousing feedback data, uses the feedback data as the reward signal for reinforcement learning, and updates the classification thresholds and classification levels in the current classification rule set to make the classification rules adapt to the dynamic changes of the warehousing scenario.

[0012] Further, in S14, synchronize the initial classification result information to the warehousing management system in real time through the API interface, triggering linkage operations with location allocation, picking planning, and inventory counting, including: In the location allocation linkage, the location management unit of the warehousing management system traverses the available location information in the warehouse location database according to the received goods classification level, accurate volume data, and accurate weight data. The available location information at least includes the bearing capacity limit, spatial length, width, and height dimensions, the floor where the location is located, and the distance from the location to the shipping outlet, automatically calculates the optimal location for each classified goods, and pushes the allocation result to the inbound handling equipment. Among them, heavy parts are forcibly allocated to the bottom bearing locations, high-frequency picking light small parts are allocated to the middle and upper floors close to the shipping outlet, and irregular volume parts are allocated to locations with special partitions; In the picking planning linkage, the initial classification results are synchronized to the picking system. The picking system dynamically plans the optimal picking route for a single batch of picking tasks based on the classification labels of various goods and the allocated storage location information. At the same time, goods with high-frequency picking classification labels are given higher access priority in the route planning, and the route planning results are pushed to the mobile terminals of picking personnel or automated guided vehicles. In the inventory counting linkage, the initial classification results are synchronized to the inventory counting module. The inventory counting area and inventory counting cycle are automatically divided according to the classification category of goods. A high-precision fixed-point inventory counting method is adopted for heavy parts, and a batch rapid inventory counting method based on RFID is adopted for light and small parts. The inventory counting tasks are assigned according to the classification results to realize classification inventory driven by volume and weight characteristics.

[0013] Furthermore, in S15, anomaly detection is performed during the closed-loop management process from inbound sorting to subsequent operations, including: The obtained three-dimensional contour shape feature vector is compared point by point with the preset standard shape template in the warehouse management system to calculate the contour deviation rate. When the deviation rate exceeds the damage threshold, an appearance damage warning is triggered. Calculate the volume and weight (V / W) value. When the V / W value deviates significantly from the standard value of the same category by more than ±30%, a ratio imbalance warning is triggered and the product is identified as abnormal.

[0014] Furthermore, after anomaly identification is completed, a tiered catch-all process is initiated to ensure reliable classification of all goods, including: For items that simultaneously meet the following conditions: triggering only the proportional imbalance warning but not the oversize or overweight warning; having an appearance profile deviation rate between 5-15%; and having a volume and weight deviation within ±5%, the warehouse management system automatically performs parameter adjustment and reclassification operations. It dynamically corrects the volume calibration coefficient or weight compensation coefficient according to the anomaly type and re-executes adaptive classification. At the same time, it marks the anomaly type in the cargo record for subsequent tracking and analysis. For severely abnormal goods that trigger any of the following warnings: oversize warning, overweight warning, appearance profile deviation rate exceeding 15%, or volume weight deviation exceeding ±10%, the warehouse management system immediately interrupts the automatic classification process, generates a manual review work order containing a picture of the goods, the type of abnormality, the original data, and the calibrated data, pushes the work order to the warehouse management terminal, and highlights it in a pop-up window on the management terminal interface, guiding the reviewer to perform remeasurement, manual classification, or goods return. At the same time, the inbound circulation status of the goods is locked until the manual review is completed.

[0015] Compared with existing technologies, the intelligent warehousing and classification method and system for stored goods based on volume and weight characteristics provided by this invention have the following advantages: 1. Through a multi-dimensional compensation dynamic calibration process, including inverse volume estimation based on material density, equivalent volume calculation for fluid particulate goods, temperature and humidity weight compensation, and volume-weight correlation calibration, measurement deviations caused by irregular cargo shape, material differences, and environmental factors are effectively eliminated. The gradient boosting decision tree model and online learning in the adaptive classification module enable dynamic updates to classification thresholds and levels, allowing classification rules to adapt to different warehousing scenarios such as e-commerce, industry, and cold chain. The end-to-end linkage module synchronizes classification results in real time to location allocation, picking planning, and inventory counting, forming a closed-loop management system. Simultaneously, anomaly identification and tiered fallback processing automatically correct parameters and reclassify minor anomalies, while generating manual review work orders for severe anomalies. This solves the problem of weak anomaly handling capabilities and improves the accuracy, efficiency, and reliability of inbound classification. 2. By calibrating models for hollow goods, irregularly shaped goods, and fluid granular goods, the characterization error of volumetric and weight features is reduced by 30%. Combined with the joint judgment of adaptive classification algorithms, the classification accuracy is significantly improved. At the same time, the operational efficiency of the entire warehousing process is substantially optimized. Since the classification results directly drive intelligent location matching, dynamic planning of picking routes, and inventory counting by category and zone, the frequency of location adjustment after receiving goods is reduced, the picking distance is shortened, and the inventory counting time is reduced, thus improving the overall warehousing operational efficiency. In addition, the warehousing management system has long-term self-optimization capabilities. With the help of online learning units, it continuously updates the classification threshold and rule set using subsequent warehousing feedback data. At the same time, it achieves closed-loop data management for anomaly handling through anomaly traceability records and hierarchical fallback mechanisms, which greatly reduces the cost of manual intervention and the risk of omissions and misclassifications, providing solid technical support for the intelligent upgrade of the warehousing management system.

[0016] A smart warehousing and classification system for stored goods based on volumetric and weight characteristics, used to execute the aforementioned smart warehousing and classification method for stored goods, the system comprising: The information acquisition module is used to collect the three-dimensional point cloud, weight, and attribute information of the goods. The data processing module is used to perform multi-dimensional compensation dynamic calibration and generate accurate volume, weight and weighted feature data; The adaptive classification module is used to output initial classification results through a machine learning model; The end-to-end linkage module is used to synchronize the classification results to the warehouse management system to drive location allocation, picking planning and inventory counting; The exception handling module is used to monitor data in real time and perform tiered fallback processing. The control module is used to coordinate the timing of each module and manual intervention.

[0017] Specifically, the adaptive classification module includes: The rule matching unit is used to automatically match the corresponding category sub-rule library based on the product category; The dynamic threshold adjustment unit is used to make joint judgments based on the calibrated volume, weight, and weighted feature vector and the sub-rule base threshold. Online learning units are used to update classification thresholds and classification levels using subsequent warehouse feedback data. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the intelligent warehousing and classification method for warehouse goods based on volume and weight characteristics proposed in this invention. Figure 2 This is a flowchart illustrating the process of initiating a dynamic calibration procedure based on multidimensional compensation in the intelligent warehousing and classification method for warehouse goods based on volume and weight characteristics proposed in this invention. Figure 3 This is a schematic diagram of the intelligent warehousing and classification system for stored goods based on volume and weight characteristics proposed in this invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] The implementation of the present invention will be described in detail below with reference to specific embodiments.

[0021] In the accompanying drawings of this embodiment, the same or similar reference numerals correspond to the same or similar components. In the description of this invention, it should be understood that if terms such as "upper," "lower," "left," and "right" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting this invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0022] Reference Figure 1-2 As shown, the intelligent warehousing and classification method for stored goods based on volume and weight characteristics, applied to electronic sorting control equipment, specifically includes the following steps: S11: Receives information about goods entering the warehouse. The 3D scanning equipment, weighing equipment, and information entry terminal deployed at the warehouse entry port work together to perform raw data collection for each piece of goods entering the warehouse. The raw volume data, raw weight data, and basic attribute information collected are packaged into raw data records, and after adding a timestamp and goods identification code, they are transmitted to the data processing module. The original physical attributes of the goods are obtained through multi-source sensing devices to provide basic data for subsequent accurate calibration. In S11, multi-view scanning, multiple weighings, and information entry ensure the comprehensiveness and redundancy of raw data, laying the foundation for dynamic calibration. This involves the collaborative work of 3D scanning equipment, weighing equipment, and information entry terminals deployed at the warehouse receiving port to perform raw data collection operations on each incoming item, including: The 3D scanning equipment uses multi-view scanning to scan the outline of the goods from at least three different spatial angles and obtain complete 3D point cloud data of the goods. The original volume data of the cargo is calculated based on 3D point cloud data to generate a 3D contour shape feature vector of the cargo. The weighing equipment performs at least three independent repeated weighing operations while the goods are stationary, records the weighing reading for each operation, and uses the arithmetic mean of the three readings as the original weight data of the goods. The information entry terminal obtains the basic attribute information of the goods by scanning a code or manually entering it. The basic attribute information includes at least the category of goods, material type, fragility level and expected storage time. S12: Through multi-dimensional compensation dynamic calibration, the effects of cargo irregularity, material density difference and environmental temperature and humidity on measurement accuracy are specifically eliminated, significantly improving the accuracy of volumetric weight data. After receiving the original data record, the data processing module starts the multi-dimensional compensation-based dynamic calibration process to eliminate the effects of cargo irregularity, material density difference and environmental temperature and humidity on weighing accuracy, and generates calibrated accurate volume data, accurate weight data and weighted volumetric weight feature data. In S12, the volume calibration stage focuses on addressing volume distortion issues related to hollow, loose, deformed, and fluid / particulate goods; the weight calibration stage utilizes a temperature and humidity compensation model to eliminate environmental drift; and the correlation calibration dynamically allocates volume and weight weights according to the product category, making the feature vector more closely match actual warehousing needs. After receiving the raw data records, the data processing module initiates a dynamic calibration process based on multi-dimensional compensation to eliminate the impact of irregular cargo shape, material density differences, and environmental temperature and humidity on weighing accuracy, including: First, volume calibration is performed. Based on the material type in the basic attribute information, the theoretical density range of the goods is obtained. The theoretical weight range is calculated by combining the original volume data of the goods. The theoretical weight range is then compared with the original weight data. If the deviation exceeds the preset threshold, the goods are determined to be hollow or loosely structured. At this point, the effective volume is calculated based on the material density to replace the original volume data. Meanwhile, the three-dimensional contour shape feature vector is matched with the standard shape template library. For deformed or irregular goods with a matching degree lower than the preset value, compensation calculation is performed to generate corrected and accurate volume data. For goods whose material type is identified as fluid or particulate, the system automatically switches to the equivalent volume calculation model. The theoretical volume is calculated in reverse based on the accurate weight data and standard density as the effective cargo volume, while the outer packaging volume is reserved for subsequent cargo space matching. Next, weight calibration is performed. Temperature and humidity sensors deployed around the weighing equipment collect the current ambient temperature and humidity values. A pre-trained temperature and humidity-weight compensation model is called to establish the mapping relationship between environmental factors and weighing deviation. The weight compensation coefficient is calculated based on the current temperature and humidity values. The arithmetic mean of the original weight data is compensated and corrected to generate accurate weight data after correction. Finally, volume-weight correlation calibration is performed. Based on the cargo category and fragility level, volume weight factor and weight weight factor are dynamically allocated to establish a weighted volume-weight feature vector, and the ratio of volume to weight is calculated. This ratio is compared with the standard ratio range of the same category, and a secondary calibration prompt is given for cargo that exceeds the range. The formulas for calculating the effective volume and compensating for deformation are as follows:

[0023] If determined to be hollow / loose or fluid / particulate, execute Equation 1 to calculate accurate volume data; if M <M th， For deformed / irregular shapes, execute Equation 2 to calculate accurate volume data; In the formula: V corrected Corrected accurate volume data (m³); W raw : Arithmetic mean (kg) of the original weight data; ρ typ Typical or standard density of material (kg / m³). V raw : The original volume (m³) obtained from the 3D scan; k: Compensation strength coefficient (empirical value, such as 0.1-0.3); M: The degree of matching between the 3D contour and the standard template (0-1, 1 is a perfect match); M thMatching degree preset threshold (e.g., 0.7; values ​​below this are considered deformed or irregular). Temperature, humidity, and weight compensation formula: ;

[0024] W comp : Compensated accurate weight data (kg); : The arithmetic mean (kg) of at least three independent weighings; T: Current ambient temperature (°C); H: Current ambient relative humidity (%RH); T0: Standard ambient temperature (typical value 20℃); H0: Standard ambient humidity (typical value 50%); α: Temperature compensation coefficient (1 / ℃), fitted by a pre-trained model; β: Humidity compensation coefficient (1 / %RH), fitted by a pre-trained model; Weighted volumetric weight eigenvector and ratio formula: ; ;

[0025] Where the volume weight w v Nonlinear dynamic allocation is adopted: ;

[0026] In the formula, F is the weighted volume weight feature vector, used for joint determination in the classification model; V final : Final accurate volume (m³) after volume calibration; W final Final accurate weight (kg) after weight calibration; w v ,w w Volume weighting factor and weight weighting factor; R: Volumetric weight ratio (m³ / kg), used for comparison with the standard ratio range; I shape Shape irregularity index (0-1), obtained from principal component analysis of 3D point cloud; U density Material density uniformity coefficient (0-1), with lower values ​​for liquid / powder and higher values ​​for solid blocks; F fragile Fragility rating (0-1), the higher the rating, the more fragile the item, and the lower the volume weight of fragile items; a, b, c: Scenario adaptation coefficients, which are preset with different values ​​based on historical data fitting; S13: Utilizing a pre-trained gradient boosting decision tree model, the calibrated multidimensional features are dynamically compared with the category sub-rule library to achieve adaptive and high-precision preliminary classification of goods. Specifically, the obtained precise volume data, precise weight data, and weighted volume-weight feature vector are input into the machine learning model deployed in the adaptive classification module. The machine learning model first automatically matches the corresponding category sub-rule library based on the goods category in the basic attribute information. Then, dynamic threshold adjustment is used to compare the calibrated precise volume data and precise weight data with the threshold in the category sub-rule library. At the same time, the weighted volume-weight feature vector is introduced for joint judgment, and the initial classification result of the goods is output. In S13, the model automatically generates the optimal classification rule set by learning the correlation between volumetric weight and subsequent efficiency in historical data, avoiding the bias of manually setting thresholds. The obtained accurate volumetric data, accurate weight data, and weighted volumetric weight feature vector are input into the machine learning model deployed in the adaptive classification module, including: The machine learning model uses the gradient boosting decision tree algorithm and has been pre-trained based on the historical inbound dataset. The historical inbound dataset includes at least the volume, weight, category, material, subsequent storage location turnover efficiency, and picking frequency tags of the goods in the past. The machine learning model learns to automatically discover the implicit mapping relationship between different volume and weight features and the optimal classification result, and generates an initial classification rule set. The classification rule set includes multiple classification levels and the corresponding volume threshold, weight threshold and joint volume and weight threshold for each level. Through an online learning mechanism, the classification rules can continuously adapt to dynamic changes in the warehousing scenario (such as seasons and promotional activities), achieving adaptive optimization without manual intervention. A weighted volumetric and weight feature vector is introduced for joint determination, outputting the initial classification results of the goods, including: Once the historical inbound dataset of the same product category reaches the preset sample size, the machine learning model starts the online learning sub-step, automatically extracting the feature data of newly inbound goods and their subsequent warehousing feedback data. The feedback data is used as a reward signal for reinforcement learning to update the classification threshold and classification level in the current classification rule set, so that the classification rules can adapt to the dynamic changes in the warehousing scenario. By calibrating models for hollow goods, irregularly shaped goods, and fluid granular goods, the characterization error of volumetric and weight features is reduced by 30%. Combined with the joint judgment of adaptive classification algorithms, the classification accuracy is greatly improved. At the same time, the operational efficiency of the entire warehousing process is substantially optimized. Since the classification results directly drive intelligent location matching, dynamic planning of picking routes, and inventory counting by category and zone, the frequency of location adjustment after receiving the goods is reduced, the picking distance is shortened, and the inventory counting time is reduced, thus improving the overall warehousing operational efficiency. In addition, the warehousing management system has long-term self-optimization capabilities, and continuously updates the classification threshold and rule set with subsequent warehousing feedback data through online learning units. S14: The classification results are synchronized to the warehouse management system in real time through the API interface, driving the three subsequent operations of location allocation, picking route planning and inventory counting, forming a closed loop from inbound to outbound, which significantly improves the overall operational efficiency. Specifically, the initial classification result information is synchronized to the warehouse management system in real time through the API interface, triggering the linkage operation with location allocation, picking planning and inventory counting, forming a closed loop management from inbound classification to subsequent operations through linkage operation; In S14, location allocation automatically matches the optimal storage location based on classification level, volume, and weight, avoiding safety hazards such as placing heavy items on high shelves; picking planning utilizes classification tags to optimize paths and access priorities; inventory is conducted by category and zone, improving inventory efficiency, including: In the warehouse location allocation linkage, the warehouse management system's location management unit, based on the received goods classification level, accurate volume data, and accurate weight data, traverses the available location information in the warehouse location database. The available location information includes at least the location's load-bearing capacity, length, width, and height dimensions, the floor where the location is located, and its distance from the outbound exit. It automatically calculates the optimal location for each category of goods and pushes the allocation results to the inbound handling equipment. Heavy items are forcibly allocated to the bottom load-bearing locations, high-frequency picking light and small items are allocated to mid-to-high-level locations near the outbound exit, and irregularly sized items are allocated to locations with dedicated partitions. In the picking planning linkage, the initial classification results are synchronized to the picking system. The picking system dynamically plans the optimal picking route for a single batch of picking tasks based on the classification labels of various goods and the allocated storage location information. At the same time, goods with high-frequency picking classification labels are given higher access priority in the route planning, and the route planning results are pushed to the mobile terminals of picking personnel or automated guided vehicles. In the integrated inventory counting process, the initial classification results are synchronized to the inventory counting module. The inventory counting area and cycle are automatically divided according to the classification of goods. A high-precision fixed-point inventory counting method is used for heavy items, while a batch rapid inventory counting method based on RFID is used for light and small items. The inventory counting tasks are assigned according to the classification results, realizing classification and inventory counting driven by volume and weight characteristics. Through anomaly traceability records and a hierarchical fallback mechanism, a closed-loop data management system for anomaly handling is realized, which greatly reduces the cost of manual intervention and the risk of omissions and misclassifications, providing solid technical support for the intelligent upgrade of the warehouse management system. S15: Throughout the closed-loop management process, an anomaly identification and tiered fallback mechanism is continuously implemented to ensure that abnormal goods such as oversized, overweight, damaged, disproportionate, and deformed goods during transit are reliably handled, avoiding omissions or misclassifications. Specifically, anomaly identification is implemented during the closed-loop management process from warehousing classification to subsequent operations. The volume data, weight data, and three-dimensional contour data of the goods are monitored in real time. The volume data is compared with the preset upper limit threshold for warehousing volume, and the weight data is compared with the preset upper limit threshold for weight. Oversized and overweight warnings are triggered respectively. After anomaly identification is completed, tiered fallback processing is initiated to achieve reliable classification of all goods.

[0027] The aforementioned early warning formula includes: Oversize warning judgment formula: ;

[0028] V final : Final accurate volume data (m³) after volume calibration; V max The preset upper limit threshold for warehouse storage volume (m³) is set according to the warehouse equipment and storage location specifications; Alert size : Oversized warning indicator, True indicates that a warning has been triggered; ;

[0029] W final : Final accurate weight data (kg) after weight calibration; W max The preset upper limit of weight (kg) is set based on the rated load of the handling equipment and the load-bearing capacity of the storage location; Alert weight : Overload warning sign, True indicates that the warning has been triggered; In S15 of this embodiment, through contour comparison and volume-weight ratio analysis, abnormalities such as appearance damage and imbalances in the proportions of bulky / confined goods are automatically identified, expanding the dimensions of anomaly detection, including: The obtained three-dimensional contour shape feature vector is compared point by point with the preset standard shape template in the warehouse management system to calculate the contour deviation rate. When the deviation rate exceeds the damage threshold, an appearance damage warning is triggered. Calculate the volume and weight (V / W) value. When the V / W value deviates significantly from the standard value of the same category by more than ±30%, a ratio imbalance warning is triggered and the product is identified as abnormal.

[0030] In this embodiment, the tiered fallback process works as follows: minor anomalies are automatically corrected and reclassified, then marked for tracking; severe anomalies immediately interrupt the process, generate a manual review work order, and lock the goods to ensure that abnormal goods do not flow into subsequent stages. After anomaly identification is completed, the tiered fallback process is initiated to achieve reliable classification of all goods, including: For items that simultaneously meet the following conditions: triggering only the proportional imbalance warning but not the oversize or overweight warning; having an appearance profile deviation rate between 5-15%; and having a volume and weight deviation within ±5%, the warehouse management system automatically performs parameter adjustment and reclassification operations. It dynamically corrects the volume calibration coefficient or weight compensation coefficient according to the anomaly type and re-executes adaptive classification. At the same time, it marks the anomaly type in the cargo record for subsequent tracking and analysis. For severely abnormal goods that trigger any of the following warnings: oversize warning, overweight warning, appearance profile deviation rate exceeding 15%, or volume weight deviation exceeding ±10%, the warehouse management system immediately interrupts the automatic classification process, generates a manual review work order containing a picture of the goods, the type of abnormality, the original data, and the calibrated data, pushes the work order to the warehouse management terminal, and highlights it in a pop-up window on the management terminal interface, guiding the reviewer to perform remeasurement, manual classification, or goods return. At the same time, the inbound circulation status of the goods is locked until the manual review is completed.

[0031] This technical solution employs a multi-dimensional compensation dynamic calibration process, including inverse volume estimation based on material density, equivalent volume calculation for fluid and particulate goods, temperature and humidity weight compensation, and volume-weight correlation calibration. This effectively eliminates measurement deviations caused by irregular cargo shapes, material differences, and environmental factors. Through a gradient boosting decision tree model and online learning in the adaptive classification module, the classification threshold and hierarchy are dynamically updated, enabling classification rules to adapt to different warehousing scenarios such as e-commerce, industry, and cold chain. A full-process linkage module synchronizes classification results in real-time to location allocation, picking planning, and inventory counting, forming a closed-loop management system. Simultaneously, anomaly identification and tiered fallback processing automatically correct parameters and reclassify minor anomalies, while generating manual review work orders for severe anomalies. This addresses the weakness in handling abnormal goods, improving the accuracy, efficiency, and reliability of inbound classification.

[0032] Reference Figure 3As shown, a smart warehousing and classification system for stored goods based on volume and weight characteristics is used to execute the aforementioned smart warehousing and classification method for stored goods. The system includes: The information acquisition module is used to collect the three-dimensional point cloud, weight, and attribute information of the goods. The data processing module is used to perform multi-dimensional compensation dynamic calibration, generate accurate volume, weight and weighted feature data, infer effective volume based on material density, calculate equivalent volume for fluid particulate cargo, perform temperature and humidity weight compensation and volume-weight correlation calibration, effectively eliminating measurement deviations caused by irregular cargo shape, material differences and environmental factors. The adaptive classification module is used to output the initial classification results through the machine learning model. The classification threshold and level are dynamically updated through the gradient boosting decision tree model and online learning in the adaptive classification module, so that the classification rules can adapt to different warehousing scenarios such as e-commerce, industry, and cold chain. The classification results are synchronized to the location allocation, picking planning and inventory counting in real time through the full-process linkage module, forming a closed-loop management. The end-to-end linkage module is used to synchronize the classification results to the warehouse management system to drive location allocation, picking planning and inventory counting; The anomaly handling module is used to monitor data in real time and perform tiered fallback processing. Anomaly identification and tiered fallback processing enable automatic parameter correction and reclassification for minor anomalies, and generate manual review work orders for severe anomalies. This solves the problem of weak handling capacity for abnormal goods and improves the accuracy, efficiency and reliability of inbound classification. The control module is used to coordinate the timing of each module and manual intervention.

[0033] Specifically, the adaptive classification module includes: The rule matching unit is used to automatically match the corresponding category sub-rule library based on the product category; The dynamic threshold adjustment unit is used to make joint judgments based on the calibrated volume, weight, and weighted feature vectors and the sub-rule base thresholds. By calibrating models for hollow goods, irregularly shaped goods, and fluid particulate goods, the characterization error of volume and weight features is reduced by 30%. Combined with the joint judgment of the adaptive classification algorithm, the classification accuracy is greatly improved. At the same time, the operational efficiency of the entire warehousing process is substantially optimized. Since the classification results directly drive intelligent location matching, dynamic planning of picking paths, and inventory counting by category and zone, the frequency of location adjustment after receiving the goods is reduced, the picking distance is shortened, and the inventory counting time is reduced, thus improving the overall warehousing operational efficiency. The online learning unit is used to update the classification thresholds and classification levels using subsequent warehouse feedback data. The warehouse management system has long-term self-optimization capabilities. With the help of the online learning unit, the classification thresholds and rule sets are continuously updated using subsequent warehouse feedback data. At the same time, the system achieves closed-loop data management for anomaly handling through anomaly traceability records and a hierarchical fallback mechanism, which greatly reduces the cost of manual intervention and the risk of omissions and misclassifications, providing solid technical support for the intelligent upgrade of the warehouse management system.

[0034] In this embodiment, the entire operation process can be automated by computer control. In each operation stage, sensors can be set up to provide signal feedback and ensure that the steps are performed sequentially. These are all conventional knowledge of current automation control, and will not be elaborated on in this embodiment.

[0035] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart warehousing and classification method for stored goods based on volumetric and weight characteristics, characterized in that, Applied to electronic sorting control equipment, it specifically includes the following steps: S11: Receives information about goods entering the warehouse. The 3D scanning equipment, weighing equipment and information entry terminal deployed at the warehouse entry port work together to perform raw data collection operations on each piece of goods entering the warehouse. After the raw volume data, raw weight data and basic attribute information are collected, they are packaged into raw data records, and after adding timestamps and goods identification codes, they are transmitted to the data processing module. S12: After receiving the original data record, the data processing module starts a dynamic calibration process based on multi-dimensional compensation to eliminate the influence of cargo shape irregularity, material density difference, and environmental temperature and humidity on weighing accuracy, and generates calibrated accurate volume data, accurate weight data and weighted volume weight characteristic data. S13: Input the obtained precise volume data, precise weight data, and weighted volume-weight feature vector into the machine learning model deployed in the adaptive classification module. The machine learning model first automatically matches the corresponding category sub-rule library based on the goods category in the basic attribute information. Then, it uses dynamic threshold adjustment to compare the calibrated precise volume data and precise weight data with the threshold in the category sub-rule library. At the same time, it introduces the weighted volume-weight feature vector for joint judgment and outputs the initial classification result of the goods. S14: The initial classification results are synchronized to the warehouse management system in real time via the API interface, triggering linkage operations with location allocation, picking planning, and inventory counting, forming a closed-loop management from inbound classification to subsequent operations. S15: During the closed-loop management process from warehousing and classification to subsequent operations, anomaly identification is performed. The volume data, weight data, and three-dimensional contour data of the goods are monitored in real time. The volume data is compared with the preset upper limit threshold for warehousing volume, and the weight data is compared with the preset upper limit threshold for weight. Oversize warning and overweight warning are triggered respectively. After anomaly identification is completed, the process enters the hierarchical fallback process to achieve reliable classification of all goods.

2. The intelligent warehousing and classification method for stored goods based on volume and weight characteristics as described in claim 1, characterized in that, In S11, 3D scanning equipment, weighing equipment, and information entry terminals deployed at the warehouse inbound port work together to perform raw data collection operations on each inbound item, including: The 3D scanning device uses multi-view scanning to scan the outline of the goods from at least three different spatial angles and obtain complete 3D point cloud data of the goods. Based on the three-dimensional point cloud data, the original volume data of the cargo is calculated to generate a three-dimensional contour shape feature vector of the cargo. The weighing device performs at least three independent repeated weighing operations when the goods are stationary, records the weighing reading for each operation, and uses the arithmetic mean of the three readings as the original weight data of the goods. The information entry terminal obtains basic attribute information of goods by scanning a code or manually entering the information. The basic attribute information includes at least the category of goods, material type, fragility level, and expected storage duration.

3. The intelligent warehousing and classification method for stored goods based on volume and weight characteristics as described in claim 2, characterized in that, In S12, after receiving the raw data record, the data processing module initiates a dynamic calibration process based on multi-dimensional compensation to eliminate the effects of cargo shape irregularities, material density differences, and environmental temperature and humidity on weighing accuracy, including: First, volume calibration is performed. Based on the material type in the basic attribute information, the theoretical density range of the goods is obtained. The theoretical weight range is calculated by combining the original volume data of the goods. The theoretical weight range is compared with the original weight data. If the deviation exceeds the preset threshold, the goods are determined to be hollow or loosely structured. At this point, the effective volume is calculated based on the material density to replace the original volume data. Meanwhile, the three-dimensional contour shape feature vector is matched with the standard shape template library. For deformed or irregular goods with a matching degree lower than the preset value, compensation calculation is performed to generate corrected and accurate volume data. For goods whose material type is identified as fluid or particulate, the system automatically switches to the equivalent volume calculation model. The theoretical volume is calculated in reverse based on the accurate weight data and standard density as the effective cargo volume, while the outer packaging volume is reserved for subsequent cargo space matching. Next, weight calibration is performed. Temperature and humidity sensors deployed around the weighing equipment collect the current ambient temperature and humidity values. A pre-trained temperature and humidity-weight compensation model is called to establish the mapping relationship between environmental factors and weighing deviation. The weight compensation coefficient is calculated based on the current temperature and humidity values. The arithmetic mean of the original weight data is compensated and corrected to generate accurate weight data after correction. Finally, a volume-weight correlation calibration is performed. Based on the product category and fragility level, volume weight factors and weight weight factors are dynamically allocated to establish a weighted volume-weight feature vector. The volume-to-weight ratio is calculated and compared with the standard ratio range for the same product category. For products that exceed the range, a secondary calibration prompt is given.

4. The intelligent warehousing and classification method for stored goods based on volume and weight characteristics as described in claim 3, characterized in that, In S13, the obtained precise volume data, precise weight data, and weighted volume-weight feature vector are input into the machine learning model deployed in the adaptive classification module, including: The machine learning model uses the gradient boosting decision tree algorithm and has been pre-trained based on the historical inbound dataset. The historical inbound dataset includes at least the volume, weight, category, material, subsequent storage location turnover efficiency, and picking frequency tags of the goods in the past. The machine learning model learns to automatically mine the implicit mapping relationship between different volume and weight features and the optimal classification result, and generates an initial classification rule set. The classification rule set includes multiple classification levels and the corresponding volume threshold, weight threshold and joint volume and weight threshold for each level.

5. The intelligent warehousing and classification method for stored goods based on volumetric and weight characteristics as described in claim 4, characterized in that, Simultaneously, the weighted volumetric weight feature vector is introduced for joint determination, outputting the initial classification result of the goods, including: Once the historical inbound dataset of the same product category reaches a preset sample size, the machine learning model initiates an online learning sub-step, automatically extracting the feature data of newly inbound goods and their subsequent warehousing feedback data. The feedback data is used as a reward signal for reinforcement learning to update the classification threshold and classification level in the current classification rule set, so that the classification rules can adapt to the dynamic changes in the warehousing scenario.

6. The intelligent warehousing and classification method for stored goods based on volume and weight characteristics as described in claim 5, characterized in that, In S14, the initial classification results are synchronized to the warehouse management system in real time via the API interface, triggering linked operations with location allocation, picking planning, and inventory counting, including: In the warehouse location allocation linkage, the warehouse management system's location management unit, based on the received goods classification level, accurate volume data, and accurate weight data, traverses the available location information in the warehouse location database. The available location information includes at least the location's load-bearing capacity, length, width, and height dimensions, the floor where the location is located, and its distance from the outbound exit. It automatically calculates the optimal location for each category of goods and pushes the allocation results to the inbound handling equipment. Heavy items are forcibly allocated to the bottom load-bearing locations, high-frequency picking light and small items are allocated to mid-to-high-level locations near the outbound exit, and irregularly sized items are allocated to locations with dedicated partitions. In the picking planning linkage, the initial classification results are synchronized to the picking system. The picking system dynamically plans the optimal picking route for a single batch of picking tasks based on the classification labels of various goods and the allocated storage location information. At the same time, goods with high-frequency picking classification labels are given higher access priority in the route planning, and the route planning results are pushed to the mobile terminals of picking personnel or automated guided vehicles. In the inventory counting linkage, the initial classification results are synchronized to the inventory counting module. The inventory counting area and inventory counting cycle are automatically divided according to the classification category of goods. A high-precision fixed-point inventory counting method is adopted for heavy parts, and a batch rapid inventory counting method based on RFID is adopted for light and small parts. The inventory counting tasks are assigned according to the classification results to realize classification inventory driven by volume and weight characteristics.

7. The intelligent warehousing and classification method for stored goods based on volumetric and weight characteristics as described in claim 6, characterized in that, In S15, anomaly detection is performed during the closed-loop management process from inbound sorting to subsequent operations, including: The obtained three-dimensional contour shape feature vector is compared point by point with the preset standard shape template in the warehouse management system to calculate the contour deviation rate. When the deviation rate exceeds the damage threshold, an appearance damage warning is triggered. Calculate the volume and weight (V / W) value. When the V / W value deviates significantly from the standard value of the same category by more than ±30%, a ratio imbalance warning is triggered and the product is identified as abnormal.

8. The intelligent warehousing and classification method for stored goods based on volume and weight characteristics as described in claim 7, characterized in that, After anomaly identification is completed, a tiered fallback process is initiated to ensure reliable classification of all goods, including: For items that simultaneously meet the following conditions: triggering only the proportional imbalance warning but not the oversize or overweight warning; having an appearance profile deviation rate between 5-15%; and having a volume and weight deviation within ±5%, the warehouse management system automatically performs parameter adjustment and reclassification operations. It dynamically corrects the volume calibration coefficient or weight compensation coefficient according to the anomaly type and re-executes adaptive classification. At the same time, it marks the anomaly type in the cargo record for subsequent tracking and analysis. For severely abnormal goods that trigger any of the following warnings: oversize warning, overweight warning, appearance profile deviation rate exceeding 15%, or volume weight deviation exceeding ±10%, the warehouse management system immediately interrupts the automatic classification process, generates a manual review work order containing a picture of the goods, the type of abnormality, the original data, and the calibrated data, pushes the work order to the warehouse management terminal, and highlights it in a pop-up window on the management terminal interface, guiding the reviewer to perform remeasurement, manual classification, or goods return. At the same time, the inbound circulation status of the goods is locked until the manual review is completed.

9. A smart warehousing and classification system for stored goods based on volume and weight characteristics, characterized in that, The system is used to execute the intelligent warehousing and classification method for stored goods according to any one of claims 1-8, the system comprising: The information acquisition module is used to collect the three-dimensional point cloud, weight, and attribute information of the goods. The data processing module is used to perform multi-dimensional compensation dynamic calibration and generate accurate volume, weight and weighted feature data; The adaptive classification module is used to output initial classification results through a machine learning model; The end-to-end linkage module is used to synchronize the classification results to the warehouse management system to drive location allocation, picking planning and inventory counting; The exception handling module is used to monitor data in real time and perform tiered fallback processing. The control module is used to coordinate the timing of each module and manual intervention.

10. The intelligent warehousing and classification system for stored goods based on volume and weight characteristics as described in claim 9, characterized in that, The adaptive classification module includes: The rule matching unit is used to automatically match the corresponding category sub-rule library based on the product category; The dynamic threshold adjustment unit is used to make joint judgments based on the calibrated volume, weight, and weighted feature vector and the sub-rule base threshold. Online learning units are used to update classification thresholds and classification levels using subsequent warehouse feedback data.

Images