A method for item identification in warehouse management

Abstract

This invention relates to an item identification method for warehouse management, belonging to the field of image recognition technology. The method includes: constructing a background model image set H; segmenting items and background in the images using the RPCA method; standardizing the item region images; identifying the item type; and outputting the identification result, achieving the purpose of identifying items with damaged tags. The beneficial effects of this invention are that it provides an item identification method for damaged warehouse item tags, with lightweight computation to meet the edge computing needs of IoT nodes, and high identification accuracy.

Description

Technical Field

[0001] This invention belongs to the field of image recognition technology, specifically relating to a method for identifying items in warehouse management. Background Technology

[0002] In warehouse management, item identification during storage and retrieval primarily relies on RFID tags, barcodes, or QR codes. However, due to prolonged use, these tags often wear out or become damaged, making it difficult to determine their type (number) and whether they belong to the warehouse. In recent years, image recognition methods have been introduced as a supplement to identify items with damaged tags, effectively avoiding empty or fake tags. However, these methods rely on deep learning and other classification techniques to directly identify the images of the items being identified. This computational complexity and high computational load make it difficult to meet the lightweight requirements of edge computing for IoT nodes, and the accuracy is only around 80%, which is insufficient. Therefore, a lightweight item identification method is urgently needed. Summary of the Invention

[0003] (a) Technical problems to be solved

[0004] The technical problem to be solved by this invention is how to provide an item identification method for warehouse management to solve the problem of identifying items with damaged tags.

[0005] (II) Technical Solution

[0006] To address the aforementioned technical problems, this invention proposes an item identification method for warehouse management, which includes the following steps:

[0007] S1. Image Acquisition: Acquire the image I to be recognized;

[0008] S2. Constructing the background model image set H: The construction or updating of the background model image set H is determined by the power-on flag, the quality of the items on the acquisition platform, and the time t since the last background acquisition. The background model image set H is updated in both the power-on and during the process.

[0009] S3. Based on the background model image set H, the Robust Principal Component Analysis (RPCA) method is used to segment the object region image I in the image to be identified from the background. e ;

[0010] S4, Item Area Image I e Standardization yields image I g ;

[0011] S5. Item Category Recognition: This involves using a pre-trained classifier for identification, specifically image I... g The data is fed into a classifier, which outputs the identified item categories.

[0012] S6. Output recognition results

[0013] Based on the comparison between the identified item type and the item list in the warehouse, the item status and name are output.

[0014] Further, in step S2, constructing the background model image set H upon power-on includes: when a power-on flag is received, outputting the prompt "Please keep the platform free of items", and detecting the quality of items on the platform. When the quality of items on the platform is equal to or close to 0, it indicates that there are no items on the platform. Then, the background update time t is recorded, the platform camera continuously acquires n background images, and finally, the background model image set H is output.

[0015] Furthermore, in step S2, constructing the background model image set H during the process includes: detecting the quality of platform items; when the quality of platform items is equal to or close to 0, and the time t since the last background update is greater than T0, then recording the background update time t; the platform camera continuously acquires n background images; and finally outputs the background model image set H.

[0016] Furthermore, the time threshold T0 is set to 2 hours.

[0017] Furthermore, n = 20.

[0018] Furthermore, step S3 specifically includes the following steps:

[0019] S31: Vectorize each image in the background model image set H and the image to be identified I. Combine the above vectors column by column to form the observation matrix D, where the last column of D is the vectorized data of the image to be identified I. Then, use the RPCA method to decompose the observation matrix D into A+E, where A is the output sparse matrix and E is a low-rank matrix. The last column of the sparse matrix A contains the information of the item to be identified.

[0020] S32: Rearrange the last column of the sparse matrix A according to the size of the image I to be identified to obtain image I1, where I1 is the region image of the item;

[0021] S33: Perform binarization on the object region image I1 to obtain a binary image I. B ;

[0022] S34: For binarized image I B Perform the closing operation to obtain I c ;

[0023] S35: For image I c Perform an opening operation to obtain image I d ;

[0024] S36: MABR calculates image I based on the minimum area principle.d Find the minimum bounding rectangle M of the largest connected region in the equation, and obtain the coordinates of the center point of M, as well as the length and width of M.

[0025] S37: Based on the center point coordinates of M and the length and width of M, extract the image corresponding to the position of the smallest bounding rectangle M in the image I to be identified, and obtain the object region image I. e .

[0026] Furthermore, in step S33, the binarization threshold T is calculated using the Otsu method (maximum inter-class variance method).

[0027] Further, in step S4, the item region image I e Standardization includes angle standardization and Z-score standardization; angle standardization is the process of adjusting the image of the extracted object region (I) e Based on its texture features, rotate it to a uniform angle to obtain image I. f Then for I f Image I is obtained by performing Z-score normalization. g .

[0028] Furthermore, in step S5, the classifier is a type of support vector machine (SVM) or neural network. Before using this method, there is also a step of training the classifier. When training the classifier, the sample images used need to be processed sequentially through the above steps S1, S2, S3, and S4 before they can be sent to the classifier for training.

[0029] Furthermore, step S5 specifically includes: if it is a legal item, output the item name and "belongs to items entering / leaving the warehouse"; if it is an illegal item, output "belongs to illegal items".

[0030] (III) Beneficial Effects

[0031] This invention proposes an item identification method for warehouse management. The advantages of this invention are that it provides an item identification method for damaged item tags in warehouses, with lightweight computational requirements that can meet the edge computing needs of IoT nodes, and high identification accuracy. Attached Figure Description

[0032] Figure 1 A flowchart of an item identification method for warehouse management provided by the present invention;

[0033] Figure 2 A diagram illustrating the construction process of the background model image set provided by this invention;

[0034] Figure 3 This invention provides a flowchart for segmenting objects and background in an image using the RPCA method. Detailed Implementation

[0035] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0036] This invention discloses an item identification method for warehouse management. The method first constructs a background model image set H, then uses the RPCA method to segment items and background in the images, standardizes item region images, identifies item types, and outputs the identification results, achieving the goal of identifying items with damaged tags. The construction of the background model image set H is determined by the power-on flag, the quality of items on the acquisition platform, and the time t since the last background acquisition. The background model image set H is updated during power-on and throughout the process. Next, the RPCA method is used to segment items and background in the images. The observation matrix D is obtained by vectorizing each image in the background model image set H and the image to be identified I, and combining them column-wise. RPCA decomposes the observation matrix D into a sparse matrix A and a low-rank matrix E. The last column of the sparse matrix A contains information about the item to be identified. The last column of matrix A is rearranged according to the size of image I to obtain the item region image. Secondly, the item region image is standardized, including angle standardization and Z-score standardization. Finally, a classifier is used to identify the item type and output the identification result. The training sample images used by the classifier for item type identification need to undergo the background segmentation, region image extraction, and region image standardization processes described above. The output identification result is the item name and item status.

[0037] To overcome the shortcomings of existing technologies, the present invention aims to provide an item identification method for warehouse management. This method uses the RPCA method to segment items and background in an image, standardize the image of the item region, identify the item type, and output the identification results, thereby achieving the purpose of identifying items with damaged labels.

[0038] The present invention provides a method for item identification in warehouse management, comprising the following steps: image acquisition (acquiring image I to be identified), constructing a background model image set H, segmenting the items and background in the image using the RPCA method, standardizing the item region image, identifying the item type, and outputting the identification result.

[0039] The purpose of this invention is to provide a method for constructing a background model image set H, the construction process of which is as follows: Figure 2As shown, the background model image set H is determined by the power-on flag, the quality of items on the acquisition platform, and the time t since the last background acquisition. The background model image set H is updated both during power-on and during the process. During power-on, when constructing the background model image set H, upon receiving the power-on flag, a prompt is output: "Please keep the platform empty." The quality of items on the platform is checked; if the quality is equal to or close to 0, indicating no items on the platform, the background update time t is recorded. The platform camera continuously acquires 20 background images, and finally, the background model image set H is output. During the process, when constructing the background model image set H, the quality of items on the platform is checked. If the quality is equal to or close to 0, and the time t since the last background acquisition is greater than T0 (generally, the time threshold T0 can be 2 hours, or its value can be determined based on the degree of change in light intensity over time), the background update time t is recorded. The platform camera continuously acquires 20 background images, and finally, the background model image set H is output. Further, the statement that the platform camera continuously acquires 20 background images is not limited to 20; other numbers are all within the protection range.

[0040] The purpose of this invention is to provide an object-background segmentation method based on the RPCA method. The process is as follows: Each image in the background model image set H and the image to be identified I are vectorized. These vectors are combined column-wise to form an observation matrix D. RPCA decomposes the observation matrix D into a sparse matrix A and a low-rank matrix E. The sparse matrix A contains information about the object to be identified. The last column of matrix A is rearranged according to the size of image I to obtain image I1, which is the region image of the object. The object region image is then binarized, followed by a closing operation and an opening operation. The minimum bounding rectangle of the object region is calculated, and the object region image in the image to be identified is extracted. The extraction of the object region image in the image to be identified is based on the coordinates of the minimum bounding rectangle. The RPCA method decomposes a matrix into a sparse matrix A and a low-rank matrix E. Since matrix E is low-rank, the similarity between its columns is high, which matches the high similarity between background images; therefore, the low-rank matrix E represents the background. The rank of the sparse matrix A is relatively high, which is consistent with the characteristics of the target image. Therefore, the sparse matrix A represents the target region image.

[0041] The purpose of this invention is to provide components for the standardization of object region images, including angle standardization and Z-score standardization. Angle standardization involves rotating the extracted object region image to a uniform angle based on its texture features, and then applying the Z-score method for standardization.

[0042] The purpose of this invention is to provide the classifier with sample images that have undergone the above-mentioned background segmentation, region image extraction, and region image standardization.

[0043] The beneficial effect of this invention is that it provides a method for identifying items with damaged labels in warehouses, and its lightweight computational requirements can meet the edge computing needs of IoT nodes.

[0044] Example 1:

[0045] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0046] The present invention provides a method for item identification in warehouse management, the process of which is as follows: Figure 1 As shown, the feature includes the following steps: image acquisition, construction of a background model image set H, segmentation of objects and background in the image using the RPCA method, standardization of object region images, object type identification, and output of identification results.

[0047] S1. Image Acquisition: Acquire the image I to be recognized;

[0048] S2, Construct the background model image set H

[0049] The construction process of the background model image set H is as follows: Figure 2 As shown, the construction or updating of the background model image set H is determined by the power-on flag, the quality of items on the acquisition platform, and the time t since the last background acquisition. The background model image set H is updated in both the power-on and during-process scenarios.

[0050] Furthermore, a background model image set H is constructed upon power-on. When a power-on flag is received, a prompt is output: "Please keep the platform empty of items." The quality of items on the platform is also detected. When the quality of items on the platform is equal to or close to 0, it indicates that there are no items on the platform. The background update time t is recorded. The platform camera continuously acquires n background images, and finally, the background model image set H is output.

[0051] Furthermore, during the process, a background model image set H is constructed, and the quality of platform items is detected. When the quality of platform items is equal to or close to 0, and the time t since the last background update is greater than T0 (generally, the time threshold T0 can be 2 hours, or the value of T0 can be determined based on the degree of change in light intensity over time), the background update time t is recorded. The platform camera continuously acquires n background images, and finally, the background model image set H is output. Further, of the n background images continuously acquired by the platform camera, n = 20 images, and the remaining numbers are all within the protection range.

[0052] S3. Based on the background model image set H, robust principal component analysis (RPCA, a type of low-rank recovery algorithm) is used to segment the objects and background in the image to be identified I, obtaining the object region image I. e .

[0053] The RPCA method decomposes a matrix into a sparse matrix A and a low-rank matrix E. Since matrix E is low-rank, the columns have high similarity, which aligns with the high similarity between background images; therefore, low-rank matrix E represents the background. The sparse matrix A, on the other hand, has a relatively high rank, consistent with the characteristics of the target image; therefore, the sparse matrix represents the target region image.

[0054] The flowchart of the RPCA method for segmenting objects and background in image I to be identified is as follows. Figure 3 As shown. The process is as follows:

[0055] S31: Vectorize each image in the background model image set H and the image to be identified I. Combine the above vectors column by column to form the observation matrix D, where the last column of D is the vectorized data of the image to be identified I. Then, use the RPCA method to decompose the observation matrix D into A+E, where A is the output sparse matrix and E is a low-rank matrix. The last column of the sparse matrix A contains the information of the item to be identified.

[0056] S32: Rearrange the last column of the sparse matrix A according to the size of the image I to be identified to obtain image I1, where I1 is the region image of the item;

[0057] S33: Perform binarization on the object region image I1 to obtain a binary image I. B The binarization threshold T can be calculated using the Otsu method;

[0058] S34: For binarized image I B Perform the closing operation to obtain I c ;

[0059] S35: For image I c Perform an opening operation to obtain image I d ;

[0060] S36: MABR calculates image I based on the minimum area principle. d Find the minimum bounding rectangle M of the largest connected region in the equation, and obtain the coordinates of the center point of M, as well as the length and width of M.

[0061] S37: Based on the center point coordinates of M and the length and width of M, extract the image corresponding to the position of the smallest bounding rectangle M in the image I to be identified, and obtain the object region image I. e .

[0062] S4, Item Area Image I e standardization.

[0063] The item area image I e Standardization includes angle standardization and Z-score standardization. Angle standardization is the process of adjusting the image of the extracted object region (I)...e Based on its texture features, rotate it to a uniform angle to obtain image I. f Then for I f Image I is obtained by performing Z-score normalization. g .

[0064] S5, Item Type Recognition

[0065] Recognition is performed using a pre-trained classifier, i.e., image I g The data is fed into a classifier, which outputs the identified item type. The classifier can be a Support Vector Machine (SVM), a neural network, or similar mechanism. Furthermore, before using this method, a classifier training step is included. During classifier training, the sample images used need to be processed sequentially through steps S1, S2, S3, and S4 before being fed into the classifier for training.

[0066] S6. Output recognition results

[0067] Based on the comparison between the identified item type and the item list in the warehouse, output the item status and name. If it is a legal item, output the item name and "belongs to inbound / outbound items"; if it is an illegal item, output "belongs to illegal items".

[0068] Example 2:

[0069] A method for item identification in warehouse management includes the following steps: image acquisition (acquiring image I to be identified), constructing a background model image set H, segmenting items and background in the image using the RPCA method, standardizing the item region image, identifying item type, and outputting the identification result.

[0070] Furthermore, the background model image set H is determined by the power-on indicator, the quality of items on the data acquisition platform, and the time t since the last background data acquisition. The background model image set H is updated both during power-on and during the process.

[0071] Furthermore, upon receiving the power-on flag, the system outputs the message "Please keep the platform empty of items" and checks the quality of items on the platform. When the quality of items on the platform is equal to or close to 0, it indicates that there are no items on the platform. The system then records the background update time t, and the platform camera continuously acquires 20 background images. Finally, the system outputs the background model image set H.

[0072] Furthermore, the quality of items on the platform is detected. When the quality of the items is equal to or close to 0, and the time t since the last background acquisition is greater than T0 (generally, the time threshold T0 can be 2 hours, or the value of T0 can be determined based on the degree of change in light intensity over time), the background update time t is recorded. The platform camera continuously acquires 20 background images, and finally outputs a background model image set H. Further, the requirement that the platform camera continuously acquires 20 background images is not limited to 20; other numbers are all within the protection range.

[0073] Further, each image in the background model image set H and the image to be identified I are vectorized. These vectors are then combined column-wise to form an observation matrix D. RPCA decomposes the observation matrix D into a sparse matrix and a low-rank matrix E. The sparse matrix A contains information about the object to be identified. The last column of matrix A is rearranged according to the size of image I to obtain image I1, which is the region image of the object. The object region image is then binarized, followed by a closing operation and an opening operation. The minimum bounding rectangle of the object region is calculated, and the object region image is extracted from the image to be identified I. The extraction of the object region image from the image to be identified I is based on the coordinates of the minimum bounding rectangle.

[0074] Furthermore, the object region image is standardized, including angle standardization and Z-score standardization. Angle standardization is the process of adjusting the extracted object region image I... e Based on its texture features, rotate it to a uniform angle to obtain image I. f Then for I f Image I is obtained by performing Z-score normalization. g

[0075] The beneficial effect of this invention is that it provides a method for identifying items with damaged labels in warehouses, which has a lightweight computational load that can meet the edge computing needs of IoT nodes, and has a high recognition accuracy.

[0076] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for item identification in warehouse management, characterized in that, The method includes the following steps: S1. Image Acquisition: Acquire the image I to be recognized; S2. Constructing the background model image set H: The construction or updating of the background model image set H is determined by the power-on flag, the quality of the items on the acquisition platform, and the time t since the last background acquisition. The background model image set H is updated in both the power-on and during the process. S3. Based on the background model image set H, the Robust Principal Component Analysis (RPCA) method is used to segment the object region image I in the image to be identified from the background. e ; S4, Item Area Image I e Standardization yields image I g ; S5. Item Category Recognition: This involves using a pre-trained classifier for identification, specifically image I... g The data is fed into a classifier, which outputs the identified item categories. S6. Output recognition results Based on the comparison between the identified item type and the item list in the warehouse, output the item status and name; in, Step S3 specifically includes the following steps: S31: Vectorize each image in the background model image set H and the image to be identified I. Combine the above vectors column by column to form the observation matrix D, where the last column of D is the vectorized data of the image to be identified I. Then, use the RPCA method to decompose the observation matrix D into A+E, where A is the output sparse matrix and E is a low-rank matrix. The last column of the sparse matrix A contains the information of the item to be identified. S32: Rearrange the last column of the sparse matrix A according to the size of the image I to be identified to obtain image I1, where I1 is the region image of the item; S33: Perform binarization on the object region image I1 to obtain a binary image I. B ; S34: For binarized image I B Perform the closing operation to obtain I c ; S35: For image I c Perform an opening operation to obtain image I d ; S36: MABR calculates image I based on the minimum area principle. d Find the minimum bounding rectangle M of the largest connected region in the equation, and obtain the coordinates of the center point of M, as well as the length and width of M. S37: Based on the center point coordinates of M and the length and width of M, extract the image corresponding to the position of the smallest bounding rectangle M in the image I to be identified, and obtain the object region image I. e .

2. The item identification method for warehouse management as described in claim 1, characterized in that, In step S2, constructing the background model image set H upon power-on includes: when a power-on flag is received, outputting the prompt "Please keep the platform empty", and detecting the quality of items on the platform. When the quality of items on the platform is equal to or close to 0, it indicates that there are no items on the platform. Then, the background update time t is recorded, the platform camera continuously acquires n background images, and finally, the background model image set H is output.

3. The item identification method for warehouse management as described in claim 1, characterized in that, In step S2, constructing the background model image set H during the process includes: detecting the quality of platform items; when the quality of platform items is equal to or close to 0, and the time t since the last background update is greater than T0, then recording the background update time t; the platform camera continuously acquires n background images; and finally outputting the background model image set H.

4. The item identification method for warehouse management as described in claim 3, characterized in that, The time threshold T0 is set to 2 hours.

5. The item identification method for warehouse management as described in any one of claims 2-4, characterized in that, n=20。 6. The item identification method for warehouse management as described in claim 1, characterized in that, In step S33, the binarization threshold T is calculated using the Otsu method (maximum inter-class variance method).

7. The item identification method for warehouse management as described in claim 1, characterized in that, The item region image I in step S4 e Standardization includes angle standardization and Z-score standardization; angle standardization is the process of adjusting the image of the extracted object region (I) e Based on its texture features, rotate it to a uniform angle to obtain image I. f Then for I f Image I is obtained by performing Z-score normalization. g .

8. The item identification method for warehouse management as described in claim 7, characterized in that, In step S5, the classifier is a type of support vector machine (SVM) or neural network. Before using this method, there is also a step of training the classifier. When training the classifier, the sample images used need to be processed sequentially through the above steps S1, S2, S3, and S4 before they can be sent to the classifier for training.

9. The item identification method for warehouse management as described in claim 7, characterized in that, Step S5 specifically includes: if it is a legal item, output the item name and "belongs to inbound / outbound items"; if it is an illegal item, output "belongs to illegal items".

Images