Verification warehousing method, system and equipment based on commodity information and medium

Abstract

The invention relates to a verification warehousing method, system and device based on commodity information and a medium, and the method comprises the following steps: collecting multi-dimensional data of warehoused commodities in a warehouse, and extracting a multi-dimensional feature vector of the multi-dimensional data of the commodities; constructing a feature spectrogram of each dimension based on the multi-dimensional feature vectors of all the commodities, and fusing the multi-dimensional feature vector of each commodity and the feature spectrogram of each dimension to obtain a fused feature vector of the commodity; constructing a commodity standard feature library based on the fused feature vector of each commodity; collecting multi-dimensional data of a to-be-stored commodity, extracting a fusion feature vector of the to-be-stored commodity through the above steps, inputting the fusion feature vector into the constructed feature enhancement model, and performing feature enhancement to obtain an enhanced fusion feature of the to-be-stored commodity; and calculating a wrong warehousing probability of the current to-be-warehoused commodity based on the enhanced fusion feature of the to-be-warehoused commodity and a commodity standard feature library, and refusing warehousing when the wrong warehousing probability of the to-be-warehoused commodity exceeds a preset threshold.

Description

Technical Field

[0001] The present invention relates to a method, system, device and medium for verifying and warehousing based on commodity information, and belongs to the technical field of warehousing management. Background Art

[0002] With the continuous expansion of the scale of B2C and C2C e-commerce, the warehousing center needs to complete the verification and warehousing of a large number of commodities within an extremely short cycle. Once "similar but not the same commodity" is wrongly warehoused, subsequent picking, shipping and after-sales links will all result in chain losses, manifested as: The costs of consumer returns and exchanges, merchant reverse logistics and customer service increase significantly; the commodity SKUs, batches or purchase order numbers in the system are wrongly bound, resulting in distorted subsequent demand forecasting and replenishment decisions; in case of a quality accident, the true supply batch cannot be accurately traced, bringing regulatory and compliance risks.

[0003] Most warehouses use order barcodes or product QR codes for matching verification. This approach relies on single-character information and cannot effectively distinguish commodities with similar packaging or coding errors; when the supply chain simultaneously circulates parallel coding systems generated by multiple suppliers, the problems of misreading and duplicate codes are particularly prominent. Summary of the Invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a method, system, device and medium for verifying and warehousing based on commodity information.

[0005] The technical solution of the present invention is as follows: On the one hand, the present invention provides a method for verifying and warehousing based on commodity information, including the following steps: Collect multi-dimensional data of the commodities already warehoused in the warehouse, and extract the multi-dimensional feature vectors of the commodity multi-dimensional data; Based on the multi-dimensional feature vectors of all commodities, construct a feature spectrum diagram for each dimension, and after fusing the multi-dimensional feature vectors of each commodity with the feature spectrum diagram of each dimension, obtain the fused feature vector of the commodity; Construct a commodity standard feature library based on the fused feature vectors of each commodity; After collecting the multi-dimensional data of the commodity to be warehoused and extracting the fused feature vector of the commodity to be warehoused through the above steps, input it into the constructed feature enhancement model for feature enhancement to obtain the enhanced fused feature of the commodity to be warehoused; Based on the enhanced fused feature of the commodity to be warehoused and the commodity standard feature library, calculate the wrong warehousing probability of the current commodity to be warehoused. When the wrong warehousing probability of the commodity to be warehoused exceeds the preset threshold, it is determined that the current commodity to be warehoused does not belong to this warehouse and the warehousing is rejected.

[0006] Preferably, the multi-dimensional data of the commodity includes image data, information data, and historical warehousing time data; The information data includes the commodity purchase order number, SKU, and commodity batch; Extract the features of the image data through the convolutional neural network of the EfficientNet network structure to obtain the image feature vector of the commodity; Extract the features of the information data through the Transformer network to obtain the information feature vector of the commodity; Extract the warehousing cycle features corresponding to the historical warehousing time data based on the long short-term memory network to obtain the warehousing cycle feature vector of the commodity; Construct the multi-dimensional feature vector of the commodity based on the image feature vector, information feature vector, and warehousing cycle feature vector of the commodity.

[0007] Preferably, the specific steps for constructing the feature spectrogram of each dimension based on the multi-dimensional feature vectors of all commodities are as follows: For the feature vectors of each dimension, use the feature vectors of each commodity in this dimension as nodes, connect the two nodes with an edge, and set the correlation between the two nodes in the current dimension feature space as the edge weight, and finally obtain the feature spectrogram of the current dimension.

[0008] Preferably, construct an adjacency matrix for the feature spectrogram of each dimension, as shown in the following formula: ; Where: Represents the adjacency matrix of the th dimension feature spectrogram; Represents the th node and the th node in the adjacency matrix of the th dimension feature spectrogram; Represents the feature vector of the th commodity in the th dimension; Represents the feature vector of the th commodity in the th dimension; Represents the variance of the feature vectors of the th dimension; Construct the Laplacian matrix of each dimension feature spectrogram based on the adjacency matrix of each dimension feature spectrogram.

[0009] Preferably, construct a dynamic spectral graph convolutional fusion network to fuse the Laplacian matrix of each dimension feature spectrogram with the multi-dimensional feature vector of each commodity, as shown in the following formula: ; Where: Represents the fused feature vector of the current commodity; Represents the activation function; Represents the dimension set; Represents the trainable weight matrix for the Represents the Laplacian matrix of the feature spectrogram for the Represents the spectral dynamic filter for the Represents the transpose operation; Represents the feature vector of the current commodity in the th dimension.

[0010] Preferably, the steps for constructing the enhanced fused feature of the commodity to be warehoused are as follows: Collect the weight, volume, and density of the commodity to be warehoused, and construct them into the physical feature vector of the commodity to be warehoused; Embed the physical feature vector of the commodity to be warehoused into its fused feature vector to obtain the preliminary enhanced feature vector; Construct a feature enhancement model based on the attention mechanism, and perform secondary enhancement on the preliminary enhanced feature vector through the feature enhancement model to obtain the enhanced fused feature of the commodity to be warehoused, as shown in the following formula: ; ; Where: Represents the enhanced fused feature of the commodity to be warehoused; Represents the query vector of the preliminary enhanced feature vector; Represents the key vector of the preliminary enhanced feature vector; Represents the value vector of the preliminary enhanced feature vector; Represents the dimension of the preliminary enhanced feature vector; Represents the physical interaction matrix; Represents the weight coefficient; Represents the interaction weight matrix of the preliminary enhanced feature vector; Represents the bias term of the physical interaction matrix; Represents the interaction weight matrix of the physical feature vector of the commodity to be warehoused; Represents the query vector of the physical feature vector of the commodity to be warehoused; Represents the key vector of the physical feature vector of the commodity to be warehoused; Represents the Hadamard product.

[0011] Preferably, the calculation formula for the mis-warehousing probability of the commodity to be warehoused is: ; Where: represents the incorrect warehousing probability of the goods to be warehoused; represents the number of samples in the commodity standard feature library; represents the th sample in the commodity standard feature library; represents the similarity normalization coefficient of the th sample in the commodity standard feature library; represents the th sample attenuation width in the commodity standard feature library; represents the global multiplier; represents the th sample balance coefficient in the commodity standard feature library.

[0012] On the other hand, the present invention also provides a verification warehousing system based on commodity information, including a data acquisition module, a feature fusion module, a standard feature library construction module, a feature enhancement module, and an incorrect warehousing probability calculation module; The data acquisition module is used to collect multi-dimensional data of the goods already warehoused in the warehouse and extract the multi-dimensional feature vectors of the multi-dimensional data of the goods; The feature fusion module is used to construct a feature spectrogram for each dimension based on the multi-dimensional feature vectors of all goods, and fuse the multi-dimensional feature vectors of each good with the feature spectrogram of each dimension to obtain the fused feature vector of the good; The standard feature library construction module is used to construct a commodity standard feature library based on the fused feature vectors of each good; After the feature enhancement module collects the multi-dimensional data of the goods to be warehoused and extracts the fused feature vector of the goods to be warehoused through the feature fusion module, it inputs the constructed feature enhancement model for feature enhancement to obtain the enhanced fused feature of the goods to be warehoused; The incorrect warehousing probability calculation module is used to calculate the incorrect warehousing probability of the current goods to be warehoused based on the enhanced fused feature of the goods to be warehoused and the commodity standard feature library. When the incorrect warehousing probability of the goods to be warehoused exceeds the preset threshold, it is determined that the current goods to be warehoused do not belong to the warehouse and the warehousing is rejected.

[0013] On yet another aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method described in the present invention is implemented.

[0014] On yet another aspect, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method described in the present invention is implemented.

[0015] The present invention has the following beneficial effects: 1. The present invention significantly improves the discrimination accuracy through multi-modal fusion and physical quantity enhancement, which is particularly effective for the scenario of "products with extremely similar appearances but different specifications".

[0016] 2. The present invention extracts multi-dimensional features from data such as product images and texts, generates a global spectral map for each feature dimension, and fuses the features of a single product with the spectral maps of each dimension to obtain a unified high-dimensional fusion vector, eliminating the information fragmentation between modalities and significantly enhancing the recognition accuracy of fine-grained differences (such as the texture of packaging materials or slight differences in batch codes). Since the fusion expression captures the global feature distribution, when a single modality is occluded, reflected, or missing, the system can still maintain stable determination based on other dimensions.

[0017] 3. After injecting measured physical features such as weight, volume, and density into the vector, products with extremely similar appearances but different specifications can be quickly distinguished. In difficult visual recognition scenarios such as low light or reflection, the physical signals provide a second "discrimination path" to ensure that the overall detection rate is not affected by environmental fluctuations.

[0018] 4. The present invention constructs a standard feature library for standard products. Every time a new SKU is added to or removed from the inventory, only the corresponding vector needs to be written into or removed from the standard product standard library to complete the update, without the need for overall model retraining, ensuring long-term maintainability. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] It should be understood that the step numbers used in the text are only for convenient description and do not limit the execution order of the steps.

[0022] It should be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms of "a", "an", and "the" are intended to include the plural forms.

[0023] The terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0024] The term "and / or" refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0025] Example 1: A method for verifying and storing goods based on commodity information, comprising the following steps: Collect multi-dimensional data of the goods already stored in the warehouse, and extract the multi-dimensional feature vectors of the commodity multi-dimensional data; Based on the multi-dimensional feature vectors of all goods, construct a feature spectrogram for each dimension, and after fusing the multi-dimensional feature vectors of each commodity with the feature spectrograms of each dimension, obtain the fused feature vector of the commodity; Construct a commodity standard feature library based on the fused feature vectors of each commodity; After collecting the multi-dimensional data of the goods to be stored in the warehouse and extracting the fused feature vector of the goods to be stored in the warehouse through the above steps, input it into the constructed feature enhancement model for feature enhancement to obtain the enhanced fused feature of the goods to be stored in the warehouse; Based on the enhanced fused feature of the goods to be stored in the warehouse and the commodity standard feature library, calculate the incorrect storage probability of the current goods to be stored in the warehouse. When the incorrect storage probability of the goods to be stored in the warehouse exceeds the preset threshold, it is determined that the current goods to be stored in the warehouse do not belong to this warehouse, and the storage is rejected.

[0026] As a preferred implementation manner of this embodiment, the multi-dimensional data of the commodity includes image data, information data, and historical storage time data; The information data includes the commodity purchase order number, SKU, and commodity batch; Extract the features of the image data through a convolutional neural network with an EfficientNet network structure to obtain the image feature vector of the commodity; Extract the features of the information data through a Transformer network to obtain the information feature vector of the commodity; Extract the storage cycle features corresponding to the historical storage time data based on a long short-term memory network to obtain the storage cycle feature vector of the commodity; Construct the multi-dimensional feature vector of the commodity based on the image feature vector, information feature vector, and storage cycle feature vector of the commodity.

[0027] As a preferred implementation manner of this embodiment, the specific steps for constructing a feature spectrogram for each dimension based on the multi-dimensional feature vectors of all goods are: For the feature vectors of each dimension, take the feature vectors of each commodity in this dimension as nodes, connect the two nodes with an edge, and set the correlation between the two nodes in the feature space of the current dimension as the edge weight, and finally obtain the feature spectrogram of the current dimension.

[0028] As a preferred implementation manner of this embodiment, construct an adjacency matrix for the feature spectrogram of each dimension, as shown in the following formula: ; Where: represents the adjacency matrix of the feature spectrogram of the th dimension; represents that the th node in the adjacency matrix of the feature spectrogram of the th dimension is adjacent to the th node; represents the feature vector of the th commodity in the th dimension; represents the feature vector of the th commodity in the th dimension; represents the variance of the feature vectors of the th dimension; Based on the adjacency matrix of the feature spectrogram of each dimension, construct the Laplacian matrix of the feature spectrogram of each dimension, as shown in the following formula: ; Where: represents the degree matrix of the adjacency matrix, and its diagonal elements are the sum of each row of the adjacency matrix.

[0029] As a preferred implementation manner of this embodiment, construct a dynamic spectrogram convolution fusion network to fuse the Laplacian matrix of the feature spectrogram of each dimension with the multi-dimensional feature vectors of each commodity, as shown in the following formula: ; Where: represents the fused feature vector of the current commodity; represents the activation function; represents the dimension set; represents the trainable weight matrix of the th dimension; represents the Laplacian matrix of the feature spectrogram of the th dimension; represents the spectral dynamic filter of the th dimension; represents the transpose operation; represents the feature vector of the current commodity in the th dimension; The construction steps of the spectral graph dynamic filter are as follows: Extract the statistical vector corresponding to the Laplacian matrix of each dimension feature spectral graph, as shown in the following formula: ; Where: represents the statistical vector of the Laplacian matrix of the current dimension feature spectral graph; represents the selected sample set of the Laplacian matrix. In this embodiment, the first 60 samples of the Laplacian matrix are selected as statistical data; represents the average value of the selected samples; represents the variance of the selected samples; represents the total number of nodes; represents the degree of the th node; represents the eigenvector corresponding to the th node; is fed into an extremely lightweight MLP to generate the gating parameters of the current dimension spectral graph dynamic filter, as shown in the following formula: ; Where: represents the low-pass bandwidth parameter of the current dimension spectral graph dynamic filter; represents the high-pass gain parameter of the current dimension spectral graph dynamic filter; represents the weight matrix of the extremely lightweight MLP, which is used to map the statistical vector in the four-dimensional space to the two-dimensional gating space; represents the bias term of the extremely lightweight MLP; represents the activation function; Construct the filter kernel based on the gating parameters of the current dimension spectral graph dynamic filter, as shown in the following formula: ; Where: represents the sample set of the Laplacian matrix; represents for the filter kernel; In actual inference, no explicit multiplication is performed, but the order Chebyshev polynomial is approximated as ; ; Where: represents for the filter kernel; represents The approximate order of the Chebyshev polynomial of the first kind, which also represents the convolution order. The larger it is, the farther neighbors can be captured, but the calculation and the risk of overfitting also increase; Denote the kernel coefficient of the Chebyshev polynomial of the th order, which changes dynamically with Denote the recursive Chebyshev polynomial of the Denote the after normalization, , where, Denote the maximum sample value in Denote the identity matrix, which scales the frequency domain of to ; Preset optimal sampling points, and sample in , then Denote the sample value corresponding to the th optimal sampling point.

[0030] As a preferred implementation manner of this embodiment, the steps for constructing the enhanced fusion feature of the goods to be warehoused are as follows: Collect the weight, volume and density of the goods to be warehoused, and construct them into a physical feature vector of the goods to be warehoused; Embed the physical feature vector of the goods to be warehoused into its fusion feature vector to obtain a preliminary enhanced feature vector; Construct a feature enhancement model based on the attention mechanism, and perform secondary enhancement on the preliminary enhanced feature vector through the feature enhancement model to obtain the enhanced fusion feature of the goods to be warehoused, as shown in the following formula: ; ; Where: Denote the enhanced fusion feature of the goods to be warehoused; Denote the query vector of the preliminary enhanced feature vector; Denote the key vector of the preliminary enhanced feature vector; Denote the value vector of the preliminary enhanced feature vector; Denote the dimension of the preliminary enhanced feature vector; Denote the physical interaction matrix; Denote the weight coefficient; Denote the interaction weight matrix of the preliminary enhanced feature vector; Denote the bias term of the physical interaction matrix; Denote the interaction weight matrix of the physical feature vector of the goods to be warehoused; Denote the query vector of the physical feature vector of the goods to be warehoused; The key vector representing the physical feature vector of the goods to be warehoused; Denote the Hadamard product.

[0031] As a preferred implementation manner of this embodiment, the calculation formula for the wrong warehousing probability of the goods to be warehoused is: ; Where: Represents the wrong warehousing probability of the goods to be warehoused; Represents the number of samples in the commodity standard feature library; Represents the th sample in the commodity standard feature library; Represents the similarity normalization coefficient of the th sample in the commodity standard feature library; Represents the attenuation width of the th sample in the commodity standard feature library. The smaller the attenuation width, the sharper the attenuation for a slightly larger sample distance. The larger the attenuation width, the greater the influence on a farther sample is still retained; Represents the global multiplier, which is used to smooth and overall consider all sample distances to prevent extreme values caused by a single nearest neighbor and is set based on experimental data; Represents the balance coefficient of the th sample in the commodity standard feature library, which balances the contributions of each sample distance and is set based on the training iteration of the machine learning model; The larger the value of the similarity normalization coefficient, the greater the influence of the sample on the wrong warehousing probability, as shown in the following formula: ; Where: Represents the Euclidean distance between and Represents the feature distance variance, , Represents the feature average distance, ; Represents the Euclidean distance between and Optionally, when the wrong warehousing probability of the goods to be warehoused exceeds the preset threshold, by constructing a misjudgment recognition algorithm, it is judged whether the wrong warehousing probability of the current goods to be warehoused is misjudged. The specific steps are as follows: Calculate the comprehensive confidence score of the wrong warehousing probability of the current goods to be warehoused based on the misjudgment recognition algorithm, as shown in the following formula: ; Where: Represents the comprehensive confidence score of the wrong warehousing probability of the current goods to be warehoused; Indicates the logistics track consistency index; Indicates the learnable weight corresponding to the logistics track consistency sub-index; Indicates the credibility loss, and the optimal value is obtained through machine learning model training. , Indicates the logit function. Indicates the learning rate; Set the comprehensive confidence score threshold. When the comprehensive confidence score is less than the comprehensive confidence score threshold, it is determined that the mis-inventory probability of the current product to be warehoused is a misjudgment, and warehousing is allowed, and the corresponding enhanced fusion features are input into the product standard feature library. In this embodiment, the calculation steps of the logistics track consistency index are as follows: Collect the logistics track of the current product to be warehoused and query the logistics track of the product closest to the product to be warehoused in the product feature standard library. The logistics track is a string of longitude and latitude coordinates with timestamps sorted in chronological order. Then the logistics track of the current product to be warehoused is expressed as: , where Indicates the longitude of the th coordinate, latitude of the th coordinate, timestamp of the th coordinate; Indicates the total number of longitude and latitude coordinates of the logistics track of the product closest to the product to be warehoused in the product feature standard library. , where Indicates the longitude of the th coordinate, latitude of the th coordinate, timestamp of the th coordinate; Calculate the spatial deviation distance and time synchronization deviation between the product to be warehoused and the product corresponding to the product feature standard library based on the logistics track, as shown in the following formula: ; ; Among them: Indicates the spatial deviation distance; Indicates finding the coordinate point closest to the coordinate point in , and obtaining distance The nearest error; It means that for each in select the one with the largest nearest error from its distance to ; Symmetric in reverse with to avoid being ignored due to extra loops; It represents the time synchronization deviation; It represents the differentiable dynamic time warping distance; It represents the smoothing coefficient. If it is large, it is smoother and more robust but the resolution decreases. When it approaches 0, it reverts to the classical dynamic time warping distance; It represents the scale factor, ; Based on the spatial deviation distance and the time synchronization deviation, calculate the logistics trajectory consistency index, as shown in the following formula: ; Among them: It represents the spatial deviation distance weight; It represents the time synchronization deviation weight; It represents the sensitivity index, which is fine-tuned according to the season.

[0032] Example 2: A verification warehousing system based on commodity information, including a data acquisition module, a feature fusion module, a standard feature library construction module, a feature enhancement module, and an incorrect warehousing probability calculation module; The data acquisition module is used to collect multi-dimensional data of the commodities already warehoused in the warehouse and extract the multi-dimensional feature vectors of the commodity multi-dimensional data; The feature fusion module is used to construct a feature spectrogram for each dimension based on the multi-dimensional feature vectors of all commodities, and fuse the multi-dimensional feature vectors of each commodity with the feature spectrogram of each dimension to obtain the fused feature vector of the commodity; The standard feature library construction module is used to construct a commodity standard feature library based on the fused feature vectors of each commodity; The feature enhancement module is used to collect the multi-dimensional data of the commodity to be warehoused, extract the fused feature vector of the commodity to be warehoused through the feature fusion module, and then input it into the constructed feature enhancement model for feature enhancement to obtain the enhanced fused feature of the commodity to be warehoused; The incorrect warehousing probability calculation module is used to calculate the incorrect warehousing probability of the current commodity to be warehoused based on the enhanced fused feature of the commodity to be warehoused and the commodity standard feature library. When the incorrect warehousing probability of the commodity to be warehoused exceeds the preset threshold, it is determined that the current commodity to be warehoused does not belong to this warehouse and the warehousing is rejected.

[0033] This system is used to implement the method in the first embodiment, which will not be elaborated here.

[0034] Embodiment Three: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method described in any embodiment of the present invention is implemented.

[0035] Embodiment Four: This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method described in any embodiment of the present invention is implemented.

[0036] In the embodiments of the present application, "at least one" means one or more, and "a plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent the case where A exists alone, A and B exist simultaneously, or B exists alone. Where A and B may be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one of the following" and its similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c may represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c may be single or multiple.

[0037] Those of ordinary skill in the art can realize that the units and algorithm steps described in the embodiments disclosed herein can be implemented by a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0038] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be elaborated here.

[0039] In several embodiments provided by the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs.

[0040] The above are only the embodiments of the present invention, and thus do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A verification and warehousing method based on commodity information, characterized in that, It includes the following steps: Collect multi-dimensional data of the goods that have been warehoused in the warehouse, and extract the multi-dimensional feature vectors of the multi-dimensional data of the goods; Based on the multi-dimensional feature vectors of all goods, construct a feature spectrogram for each dimension, and after fusing the multi-dimensional feature vectors of each good with the feature spectrograms of each dimension, obtain the fused feature vector of the good; Construct a standard feature library of goods based on the fused feature vectors of each good; After collecting the multi-dimensional data of the goods to be warehoused and extracting the fused feature vector of the goods to be warehoused through the above steps, input it into the constructed feature enhancement model for feature enhancement to obtain the enhanced fused feature of the goods to be warehoused; Based on the enhanced fused feature of the goods to be warehoused and the standard feature library of goods, calculate the wrong warehousing probability of the current goods to be warehoused. When the wrong warehousing probability of the goods to be warehoused exceeds the preset threshold, it is determined that the current goods to be warehoused do not belong to this warehouse and the warehousing is rejected.

2. The verification and warehousing method based on commodity information according to claim 1, wherein The multi-dimensional data of the goods includes image data, information data, and historical warehousing time data; The information data includes the purchase order number of the goods, SKU, and the batch number of the goods; Extract the features of the image data through a convolutional neural network with the EfficientNet network structure to obtain the image feature vector of the goods; Extract the features of the information data through the Transformer network to obtain the information feature vector of the goods; Based on the long short-term memory network, extract the warehousing cycle features corresponding to the historical warehousing time data to obtain the warehousing cycle feature vector of the goods; Construct the multi-dimensional feature vector of the goods based on the image feature vector, information feature vector, and warehousing cycle feature vector of the goods.

3. The method for verifying and warehousing based on commodity information according to claim 1, characterized in that The specific steps for constructing the feature spectrogram of each dimension based on the multi-dimensional feature vectors of all goods are as follows: For the feature vectors of each dimension, use the feature vectors of each good in this dimension as nodes, connect the two nodes with an edge, and set the correlation between the two nodes in the current dimension feature space as the edge weight, and finally obtain the feature spectrogram of the current dimension.

4. The verification and warehousing method based on commodity information according to claim 3, wherein, Construct an adjacency matrix for the feature spectrogram of each dimension, as shown in the following formula: ； Wherein: represents the adjacency matrix of the th dimensional feature spectrogram; represents the th node and the th node in the adjacency matrix of the th dimensional feature spectrogram; represents the eigenvector of the th product in the th dimension; represents the eigenvector of the th product in the th dimension; represents the variance of the eigenvector of the th dimension; Construct the Laplacian matrix of the feature spectrogram of each dimension based on the adjacency matrix of the feature spectrogram of each dimension.

5. The verification and warehousing method based on commodity information according to claim 4, wherein Construct a dynamic spectrogram convolution fusion network to fuse the Laplacian matrix of the feature spectrogram of each dimension with the multi-dimensional feature vectors of each good, as shown in the following formula: ； Wherein: represents the fusion feature vector of the current commodity; represents the activation function; represents the dimension set; represents the trainable weight matrix of the th dimension; represents the Laplacian matrix of the feature spectrogram of the th dimension; represents the spectral dynamic filter of the th dimension; represents the transpose operation; represents the feature vector of the current commodity in the th dimension.

6. The verification and warehousing method based on commodity information according to claim 5, characterized in that The construction steps of the enhanced fused feature of the goods to be warehoused are as follows: Collect the weight, volume, and density of the goods to be warehoused, and construct them into a physical feature vector of the goods to be warehoused; Embed the physical feature vector of the goods to be warehoused into its fused feature vector to obtain a preliminary enhanced feature vector; Construct a feature enhancement model based on the attention mechanism, and perform secondary enhancement on the preliminary enhanced feature vector through the feature enhancement model to obtain the enhanced fused feature of the goods to be warehoused, as shown in the following formula: ； ； Wherein: represents the enhanced fusion feature of the goods to be warehoused; represents the query vector of the preliminary enhanced feature vector; represents the key vector of the preliminary enhanced feature vector; represents the value vector of the preliminary enhanced feature vector; represents the dimension of the preliminary enhanced feature vector; represents the physical interaction matrix; represents the weight coefficient; represents the interaction weight matrix of the preliminary enhanced feature vector; represents the bias term of the physical interaction matrix; represents the interaction weight matrix of the physical feature vector of the goods to be warehoused; represents the query vector of the physical feature vector of the goods to be warehoused; represents the key vector of the physical feature vector of the goods to be warehoused; represents the Hadamard product.

7. A method for verifying and warehousing based on commodity information according to claim 6, characterized in that The calculation formula for the wrong warehousing probability of the goods to be warehoused is: ； Wherein: represents the wrong storage probability of the goods to be warehoused; represents the sample quantity of the goods standard feature library; represents the th sample of the goods standard feature library; represents the similarity normalization coefficient of the th sample of the goods standard feature library; represents the attenuation width of the th sample of the goods standard feature library; represents the global multiplier; represents the balance coefficient of the th sample of the goods standard feature library.

8. A verification and warehousing system based on commodity information, characterized in that, It includes a data collection module, a feature fusion module, a standard feature library construction module, a feature enhancement module, and a wrong warehousing probability calculation module; The data collection module is used to collect multi-dimensional data of the goods that have been warehoused in the warehouse, and extract the multi-dimensional feature vectors of the multi-dimensional data of the goods; The feature fusion module is used to construct a feature spectrogram for each dimension based on the multi-dimensional feature vectors of all commodities, and fuse the multi-dimensional feature vectors of each commodity with the feature spectrogram of each dimension to obtain the fused feature vector of the commodity; The standard feature library construction module is used to construct a commodity standard feature library based on the fused feature vectors of each commodity; The feature enhancement module is used to collect the multi-dimensional data of the commodity to be warehoused, extract the fused feature vector of the commodity to be warehoused through the feature fusion module, and then input it into the constructed feature enhancement model for feature enhancement to obtain the enhanced fused feature of the commodity to be warehoused; The mis-warehousing probability calculation module is used to calculate the mis-warehousing probability of the current commodity to be warehoused based on the enhanced fused feature of the commodity to be warehoused and the commodity standard feature library. When the mis-warehousing probability of the commodity to be warehoused exceeds a preset threshold, it is determined that the current commodity to be warehoused does not belong to this warehouse and warehousing is rejected.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method according to any one of claims 1 to 7.

Images