Artificial intelligence recognition scale system, recognition method, computer device, computer-readable storage medium, and computer program product
The AI recognition scale system with autonomous incremental learning addresses unreliable product recognition by autonomously adapting to new products, enhancing reliability and reducing human intervention, thus improving efficiency and accuracy in product recognition and payment processes.
Patent Information
- Application Number
- JP2023554371
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2022-11-16
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing AI vision recognition scales struggle with newly added or differently presented products, leading to unreliable or incorrect recognition results, necessitating human intervention or specialized training, which is inefficient and costly.
An AI recognition scale system employing autonomous incremental learning, comprising a weighing system, vision sensor, and recognition system, that autonomously performs online learning for new products and erroneous recognitions, using a recognition algorithm module, incremental learning algorithm module, and data module to enhance reliability and robustness.
The system outputs highly reliable and robust results after one or a few learning sessions, reducing human intervention and offline optimization needs, improving efficiency and accuracy in product recognition and payment processes.
Smart Images

Figure 2024543756000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to the technical field of artificial intelligence, and in particular to an artificial intelligence recognition scale system and recognition method based on autonomous incremental learning. [Background technology]
[0002] This section provides a background or context for the embodiments of the invention described in the claims. Nothing herein is admitted to be prior art by reason of its inclusion in this section.
[0003] In the field of offline product sales, the main payment method is to scan the product barcode to recognize the product and then perform the product payment. However, for non-standard products without product barcodes, such as fruits and vegetables, customers usually need to manually send the selected products to a specific weighing platform to be weighed, select the product type, and print or attach a barcode tag containing the corresponding information such as type, unit price, and price. In the process, a lot of operation effort is required, which is costly and inefficient, and the waiting time of consumers is inevitably increased.
[0004] However, with the development of artificial intelligence, and especially with the application of deep learning, it is possible to use computer vision technology to intelligently recognize various products that require weight measurement, thereby improving the efficiency of weighing and settlement. In recent years, a technology called AI (artificial intelligence) vision recognition scale has emerged that can intelligently recognize and weigh products without barcodes, and can simplify product settlement based on the results of product recognition and weighing. Summary of the Invention [Problem to be solved by the invention]
[0005] However, when a general AI vision recognition scale is currently used, it may encounter newly added products or products that have not been targeted so far, or may encounter various different aspects of products during weighing and payment, such as products in bags, products not in bags, small packages, large packages, etc. In such cases, problems may occur in which the AI vision recognition scale outputs candidate results with very low reliability or outputs erroneous recognition results, and ultimately the scale may not be able to recognize the products. In such cases, manual intervention by an administrator is required, or the AI recognition model needs to be newly trained and optimized by a specialized technician or a background server. [Means for solving the problem]
[0006] An embodiment of the present invention provides an artificial intelligence recognition scale system based on autonomous incremental learning, which can autonomously perform incremental learning online for different aspects of different products, newly added products, and incorrectly recognized products, and the output results have high reliability and are robust and adaptable. The system includes a weighing system, a weight sensor, a vision sensor, and a recognition system, where:
[0007] the weighing system is for receiving a weight of the target product, receiving and transmitting an image of the target product to the recognition system, receiving at least one candidate product identifier for the target product, identifying a product identifier for the target product from the at least one candidate product identifier and transmitting the product identifier to the recognition system, and generating a product tag based on the weight of the target product and the product identifier; The weight sensor is for acquiring the weight of the target product and transmitting it to the weighing system; The vision sensor is used to capture the image of the target product and transmit it to the weighing system. The recognition system includes a recognition algorithm module, an incremental learning algorithm module, and a data module; the recognition algorithm module is for recognizing an image of a target product, obtaining a feature vector of the image, calculating a similarity between the feature vector and a feature vector in a feature vector database, obtaining at least one feature vector exceeding a similarity threshold, identifying a product identifier candidate corresponding to the at least one feature vector, and outputting the candidate to a weighing system; The incremental learning algorithm module is for, after receiving the identified product identifier of the target product, judging whether incremental learning is necessary, and if so, incrementally processing the image, extracting a feature vector, merging the extracted feature vector with the product identifier of the target product, and then storing it in the feature vector database of the data module.
[0008] An embodiment of the present invention provides a recognition method for an artificial intelligence recognition scale based on autonomous incremental learning, which is used in an artificial intelligence recognition scale system based on autonomous incremental learning, which can autonomously perform incremental learning online for different aspects of different products, newly added products, and incorrectly recognized products, and has high reliability of output results and strong robustness and adaptability, and the method includes: Recognizing an image of a target product and obtaining a feature vector of the image; calculating a similarity between the feature vector and a feature vector in a feature vector database to obtain at least one feature vector that exceeds a similarity threshold; identifying and outputting candidate item identifiers corresponding to the at least one feature vector to a metering system, the metering system being used to receive item identifiers for target items identified by a user from among the at least one candidate item identifier; and determining whether incremental learning is required after receiving the identified product identifier of the target product; if necessary, incrementally processing the image before extracting a feature vector; and The method includes merging the extracted feature vector with a product identifier for the target product before storing it in a feature vector database of the data module.
[0009] In an embodiment of the present invention, the weighing system is for receiving a weight of the target product, receiving an image of the target product and sending it to the recognition system, receiving at least one candidate product identifier for the target product, identifying a product identifier of the target product from the at least one candidate product identifier and sending it to the recognition system, and generating a product tag based on the weight and product identifier of the target product; the weight sensor is for obtaining a weight of the target product and sending it to the weighing system; the vision sensor is for obtaining an image of the target product and transmitting it to the weighing system; the recognition system includes: a recognition algorithm module for recognizing the image of the target product, obtaining a feature vector of the image, calculating a similarity between the feature vector and a feature vector in a feature vector database, obtaining at least one feature vector that exceeds a similarity threshold, identifying a product identifier candidate corresponding to the at least one feature vector, and outputting it to the weighing system; and an incremental learning algorithm module for determining whether incremental learning is necessary after receiving the identified product identifier of the target product, and if so, incrementally processing the image, extracting a feature vector, merging the extracted feature vector with the product identifier of the target product, and then storing it in the feature vector database of the data module. Compared to the prior art, it is possible to simultaneously and autonomously perform incremental learning online for different aspects of different products, newly added products, and misrecognized products, and after one or a few learning rounds, it can output highly reliable and accurate results, and has strong robustness and adaptability. [Brief description of the drawings]
[0010] In order to more clearly describe the embodiments of the present invention or the technical solutions of the prior art, the following briefly introduces the accompanying drawings necessary for describing the embodiments or the prior art. Obviously, the accompanying drawings described below are only some embodiments of the present invention, and those skilled in the art can obtain other accompanying drawings based on these accompanying drawings without creative efforts. In the accompanying drawings, [Figure 1] FIG. 1 is a schematic diagram of an artificial intelligence cognitive scale system based on autonomous incremental learning in an embodiment of the present invention. [Diagram 2] FIG. 1 is a schematic diagram of a weighing system in an embodiment of the present invention. [Diagram 3] FIG. 2 is a schematic diagram of a recognition algorithm module in an embodiment of the present invention. [Figure 4] 2 is a flowchart of a recognition method for an artificial intelligence recognition scale based on autonomous incremental learning in an embodiment of the present invention; [Diagram 5] FIG. 2 is a schematic diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0011] In order to make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will further describe the embodiments of the present invention with reference to the accompanying drawings, in which the exemplary embodiments of the present invention and the description thereof are for interpreting the present invention but are not intended to limit the present invention.
[0012] FIG. 1 is a schematic diagram of an artificial intelligence recognition scale system based on autonomous incremental learning in an embodiment of the present invention, which includes a weighing system 10, a weight sensor 20, a vision sensor 30 and a recognition system 40, where: the weighing system 10 is for receiving a weight of the target product, receiving an image of the target product and then transmitting it to the recognition system, receiving at least one candidate product identifier of the target product, identifying a product identifier of the target product from the at least one candidate product identifier and transmitting it to the recognition system, and generating a product tag based on the weight of the target product and the product identifier; The weight sensor 20 is for acquiring the weight of the target product and transmitting it to the weighing system; The vision sensor 30 is for acquiring an image of the target product and transmitting it to the weighing system; The recognition system 40 includes a recognition algorithm module 401, an incremental learning algorithm module 402, and a data module 403. the recognition algorithm module 401 is for recognizing an image of a target product, obtaining a feature vector of the image, calculating a similarity between the feature vector and a feature vector in a feature vector database, obtaining at least one feature vector exceeding a similarity threshold, identifying a product identifier candidate corresponding to the at least one feature vector, and outputting the candidate to a weighing system; The incremental learning algorithm module 402 is for, after receiving the identified product identifier of the target product, judging whether incremental learning is necessary, and if so, incrementally processing the image, extracting a feature vector, merging the extracted feature vector with the product identifier of the target product, and then storing it in the feature vector database of the data module 403.
[0013] Compared with the conventional technology, the system proposed in the embodiment of the present invention can simultaneously and autonomously perform incremental learning online for different aspects of different products, newly added products, and misrecognized products, and can output highly reliable and accurate results after one or a few learning rounds, and has strong robustness and adaptability.
[0014] The weighing system in the system proposed in the embodiment of the present invention can achieve several automated functions. First, obtain the image of the target product, and use an artificial intelligence algorithm to recognize the image. Then, from the product identification result, determine the product identifier of the target product. Then, obtain the corresponding price in the database and multiply it by the weight of the target product to obtain the payment price and generate a product tag. Through the above-mentioned automated recognition and obtaining the corresponding price data, the efficiency of weighing and payment for products can be improved, and the labor costs for product payment can be reduced.
[0015] According to the recognition system in the system proposed in the embodiment of the present invention, incremental learning can be performed autonomously online, and incremental learning can be performed online for different aspects of products, aspects that have not been targeted so far, newly added products, and erroneously recognized products in cooperation with the feature vector database. In a specific manner, incremental learning is performed for low-confidence and accurate candidate results, learning for high-confidence and erroneous results is suppressed, and incremental learning can be performed online for newly added products. According to the system proposed in the embodiment of the present invention, the robustness of the recognition system can be improved autonomously online, frequent misrecognition of the same type of product in the subsequent use process can be avoided, the number of human interference operations in the use process of the artificial intelligence recognition scale can be reduced, and frequent offline optimization of the recognition model by the background training service platform can be avoided.
[0016] In one embodiment, the product identifier includes a product name and a product code.
[0017] Specifically, the weighing system, for example, an electronic scale platform, runs on a Windows system, a Linux system, or an Android system. FIG. 2 is a schematic diagram of a weighing system according to an embodiment of the present invention. In one embodiment, the weighing system includes: a positioning assembly 101 for positioning the connected weight sensor, vision sensor, recognition system, and register system; a touch screen 102 for displaying the product identifier and weight of the target product; an interactive module 103 for receiving a weight of the target product, receiving an image of the target product and then transmitting it to the recognition system, receiving at least one candidate product identifier of the target product, identifying a product identifier of the target product from the at least one candidate product identifier and transmitting it to the recognition system, performing a search operation after receiving a search command input by a user, invoking the recognition system to perform recognition of the target product after receiving a re-recognition command from the user, and receiving the product identifier confirmed by the user, the interactive module 103 also being capable of printing a product tag; and a payment module 104 for generating a product tag based on the weight and product identifier of the target product and for making a payment for the target product through a connected cash register system.
[0018] In an embodiment of the present invention, the weight sensor is a hardware device with a balancer as its core, which can sense the weight and weight changes of an object, and can be directly connected to a weighing system and transmit weight information to the weighing system.
[0019] The vision sensor is commonly camera hardware. In one embodiment, the vision sensor is further used to capture and transmit a video stream of the target item by the weighing system to a recognition system, and a recognition algorithm module of the recognition system is further used to extract an image of the target item from the video stream.
[0020] In one embodiment, the interactive module of the weighing system further comprises: If there is no candidate product identifier for the received target product, a product identifier input by a user is received; This is used to make the product identifier entered by the user the identified product identifier.
[0021] In an embodiment of the present invention, the recognition system may be installed in a weighing system based on a Windows system, a Linux system or an Android system. The recognition system is based on vision image and artificial intelligence technology, including traditional digital image processing algorithms and deep neural network models in artificial intelligence technology, and can recognize target objects.
[0022] 3 is a schematic diagram of a recognition algorithm module in an embodiment of the present invention. In one embodiment, the recognition algorithm module includes: A target detection module 4011 for recognizing a product region in an image of a target product; an image segmentation module 4012 for segmenting a product area from an image of a target product; In practical use, one may choose to capture the product region by one of target detection and image segmentation methods.
[0023] The image recognition algorithm module 4013 further includes: recognizing the product region, obtaining a feature vector of the image, calculating a similarity between the feature vector and a feature vector in a feature vector database, obtaining at least one feature vector that exceeds a similarity threshold (generally taking several top ranked feature vectors), identifying product identifier candidates corresponding to the at least one feature vector, and outputting them to the weighing system.
[0024] Here, the target detection module 4011 is realized by a single-stage or two-stage detection algorithm. The image segmentation module 4012 is realized by an instance segmentation and semantic segmentation algorithm. The image recognition algorithm module 4013 includes AI model weights, algorithm reasoning, pre-processing programs, post-processing programs, etc.
[0025] In an embodiment of the present invention, the data module is a commonly seen structured database, and stores the product code, product name, and feature vector for search of each product.
[0026] In one embodiment, the incremental learning algorithm module specifically: If the identified product identifier of the target product is ranked first among the product identifier candidates corresponding to the at least one recognized feature vector, and the confidence of the target product is lower than a confidence threshold; or If the identified product identifier of the target product is not ranked first among the candidate product identifiers corresponding to the at least one recognized feature vector, or This is used to determine the need for incremental learning when there are no recognized product identifier candidates.
[0027] In one embodiment, the incremental learning algorithm module specifically: performing incremental processing on the image of the target product, the incremental processing including at least one of rotation, flipping, and color transformation; calling a recognition algorithm module to recognize the incrementally processed image and obtain a feature vector; The image is incrementally processed and then used to extract the feature vector, in a step of excluding items from the feature vector whose degree of differentiation is less than a differentiation threshold to obtain the feature vector to be extracted.
[0028] In one embodiment, the incremental learning algorithm module specifically: Add 1 to the first record of the number of times the specific product identifier of the target product was selected; If the identified product identifier of the target product is not ranked first, add 1 to the second record; The reliability of the target product is calculated based on the first record and the second record.
[0029] Specifically, the incremental learning algorithm module 402 merges the extracted feature vector with the product identifier of the target product, and then stores it in the feature vector database of the data module 403, so that the merged feature vector database is newly loaded when the next new recognition process is executed. The candidate results output in the next new recognition process are comprehensively judged based on the number of times they have been selected and the number of times they have been misrecognized, and the reliability of the output is autonomously processed. For candidates that have been selected many times and have few times of misrecognition, the reliability is increased, and for candidates that have been selected few times and have many times of misrecognition, the reliability is reduced.
[0030] In the embodiment of the present invention, a recognition method for artificial intelligence recognition scale based on autonomous incremental learning is further proposed, which is used in the above-mentioned artificial intelligence recognition scale system based on autonomous incremental learning, and its principle is similar to that of the artificial intelligence recognition scale system based on autonomous incremental learning, so it will not be described here.
[0031] FIG. 4 is a flowchart of an AI cognitive scale recognition method based on autonomous incremental learning in an embodiment of the present invention. In the AI cognitive scale recognition method based on autonomous incremental learning, Step 401: receiving and recognizing an image of a target item sent from a weighing system, and obtaining a feature vector of the image; Step 402: calculating the similarity between the feature vector and a feature vector in a feature vector database to obtain at least one feature vector that exceeds a similarity threshold; identifying and outputting candidate item identifiers corresponding to the at least one feature vector to a weighing system, the weighing system receiving a product identifier of a target item identified by a user from the at least one candidate item identifier and being used to generate a product tag based on a weight of the target item and the identified product identifier; Step 404: after receiving the identified product identifier of the target product transmitted from the weighing system, determine whether incremental learning is required; If necessary, a step 405 of incrementally processing the image and then extracting a feature vector; and Step 406 includes merging the extracted feature vector with the product identifier of the target product before storing it in the feature vector database of the data module.
[0032] In one embodiment, the step of receiving and recognizing an image of a target item transmitted from a weighing system and obtaining a feature vector of the image comprises: Recognizing a product area in an image of a target product or extracting a product area from the image of the target product; and This involves recognizing the product area and obtaining a feature vector of the image.
[0033] In one embodiment, the step of determining whether incremental learning is necessary comprises: If the identified product identifier of the target product is ranked first among the product identifier candidates corresponding to the at least one recognized feature vector, and the confidence of the target product is lower than a confidence threshold; or If the identified product identifier of the target product is not ranked first among the candidate product identifiers corresponding to the at least one recognized feature vector, or Determining that there is a need for incremental learning when there are no recognized product identifier candidates.
[0034] In one embodiment, the step of incrementally processing the image and then extracting a feature vector comprises: performing incremental processing on the image of the target product, the incremental processing including at least one of rotating, flipping, and color transforming; calling a recognition algorithm module to recognize the incrementally processed image to obtain a feature vector; and This includes excluding items from the feature vectors whose degree of differentiation is less than a differentiation threshold to obtain the feature vector to be extracted.
[0035] In one embodiment, the method further comprises: adding one to a first record of the number of times the identified product identifier of the target product was selected; if the identified product identifier of the target product is not ranked first, adding 1 to the second record; and The method further includes calculating the reliability of the target product based on the first record and the second record.
[0036] As described above, the system and recognition method proposed in the embodiment of the present invention provide the following beneficial effects.
[0037] First, the system configuration is clearer and simpler, without any complicated assembly. The addition of the recognition system enables autonomous incremental learning, which avoids the need for a separately developed and docked background model training service platform, improving the use efficiency and accuracy of the intelligent weighing platform in daily life, and also improving the production and development efficiency of intelligent weighing devices.
[0038] Second, by effectively utilizing vision image information and artificial intelligence algorithms, goods can be automatically recognized and corresponding price data can be obtained, so that the efficiency of weighing and paying for goods can be improved and the human costs involved in paying for goods can be reduced.
[0039] Third, after the incremental learning algorithm module performs incremental learning, the system can newly load the merged feature vector database when performing a new recognition process next time, so that when recognizing the target product, the output candidate results have more accurate sorting and confidence. Incremental learning can be performed online for different aspects of the product, aspects that have not been targeted so far, newly added products, and products that have been incorrectly recognized, improving the robustness of the recognition system and avoiding incorrect recognition of the same type of products, reducing the number of manual interference operations during the use of the AI recognition scale, avoiding frequent offline optimization support for the recognition model by the background training server, and reducing the maintenance costs of the product.
[0040] In an embodiment of the present invention, a computer device is further proposed. Figure 5 is a schematic diagram of the computer device in the embodiment of the present invention. The computer device 500 includes a memory 510, a processor 520, and a computer program 530 that is stored in the memory 510 and can be run by the processor 520, and when the computer program 530 is executed by the processor 520, the above-mentioned recognition method of the artificial intelligence recognition scale based on autonomous incremental learning is realized.
[0041] In an embodiment of the present invention, a computer-readable storage medium is further proposed, which stores a computer program, and when the computer program is executed by a processor, realizes the above-mentioned recognition method of the artificial intelligence recognition scale based on autonomous incremental learning.
[0042] In an embodiment of the present invention, a computer program product is further proposed, the computer program product including a computer program, which, when executed by a processor, realizes the above-mentioned recognition method of the artificial intelligence recognition scale based on autonomous incremental learning.
[0043] As will be appreciated by those skilled in the art, embodiments of the present invention may be provided as a method, a system, or a computer program product. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied in one or more computer usable storage media (including, but not limited to, disk memory, CD-ROM, optical memory, etc.) containing computer usable program code.
[0044] The present invention has been described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, or a combination of processes and / or blocks in the flowcharts and / or block diagrams, may be realized by computer program commands. These computer program commands may be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor, or other programmable data processing apparatus to generate an apparatus for implementing a function specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams according to the commands executed by the processor of the computer or other programmable data processing apparatus.
[0045] These computer program commands may be stored in a computer readable memory that can guide a computer or other programmable data processing apparatus to operate in a particular manner, such that the commands stored in the computer readable memory cause a product including a command device to implement the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.
[0046] These computer program commands may be loaded into a computer or other programmable data processing apparatus to cause the computer or other programmable apparatus to execute a series of operational steps to produce a computer-implemented process, and the commands executed by the computer or other programmable apparatus provide steps for implementing the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.
[0047] In the above-described specific embodiments, the objectives, technical solutions and beneficial effects of the present invention are further described in detail. It should be understood that the above-described contents are only specific embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any amendments, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention should be within the protection scope of the present invention.
Claims
1. An artificial intelligence cognitive scale system based on autonomous incremental learning, comprising: A weighing system, a weight sensor, a vision sensor, and a recognition system, the weighing system is for receiving a weight of the target product, receiving and transmitting an image of the target product to the recognition system, receiving at least one candidate product identifier for the target product, identifying a product identifier for the target product from the at least one candidate product identifier and transmitting the product identifier to the recognition system, and generating a product tag based on the weight of the target product and the product identifier; The weight sensor is for acquiring the weight of the target product and transmitting it to the weighing system; The vision sensor is used to capture the image of the target product and transmit it to the weighing system. The recognition system includes a recognition algorithm module, an incremental learning algorithm module, and a data module; the recognition algorithm module is for recognizing an image of a target product, obtaining a feature vector of the image, calculating a similarity between the feature vector and a feature vector in a feature vector database, obtaining at least one feature vector exceeding a similarity threshold, identifying a product identifier candidate corresponding to the at least one feature vector, and outputting the candidate to a weighing system; The incremental learning algorithm module is for, after receiving the identified product identifier of the target product, judging whether incremental learning is necessary, and if so, incrementally processing the image, extracting a feature vector, and merging the extracted feature vector with the product identifier of the target product, and then storing the feature vector database of the data module; A system characterized by:
2. The weighing system is a positioning assembly for positioning the connected weight sensor, vision sensor, recognition system, and register system; a touch screen for displaying a product identifier and a weight of the target product; an interactive module for receiving a weight of the target product, receiving an image of the target product and then transmitting it to the recognition system, receiving at least one candidate product identifier of the target product, identifying a product identifier of the target product from the at least one candidate product identifier and transmitting it to the recognition system, performing a search operation after receiving a search command input by a user, invoking the recognition system to perform recognition of the target product after receiving a re-recognition command from the user, and receiving the product identifier confirmed by the user; and a payment module for generating a product tag based on a weight and a product identifier of the target product and for making a payment for the target product through a connected cash register system.
2. The system of claim 1 .
3. The interactive module of the weighing system furthermore: If there is no candidate product identifier for the received target product, a product identifier input by a user is received; 2. The system of claim 1, wherein the system is adapted to determine whether a product identifier entered by a user is the identified product identifier.
4. The recognition algorithm module is a target detection module for recognizing a product region in an image of a target product; an image segmentation module for extracting a product region from an image of a target product; an image recognition algorithm module for recognizing an item region, obtaining a feature vector of the image, calculating a similarity between the feature vector and a feature vector in a feature vector database, obtaining at least one feature vector that exceeds a similarity threshold, and identifying item identifier candidates corresponding to the at least one feature vector, and outputting the candidate identifiers to a weighing system.
2. The system of claim 1 .
5. Specifically, the incremental learning algorithm module: If the identified product identifier of the target product is ranked first among the candidate product identifiers corresponding to the at least one recognized feature vector, and the confidence of the target product is lower than a confidence threshold; or If the identified product identifier of the target product is not ranked first among the candidate product identifiers corresponding to the at least one recognized feature vector, or If there are no recognized product identifier candidates, Used to determine the need for incremental learning, 2. The system of claim 1 .
6. Specifically, the incremental learning algorithm module: performing incremental processing on the image of the target product, the incremental processing including at least one of rotation, flipping, and color transformation; calling a recognition algorithm module to recognize the incrementally processed image and obtain a feature vector; A step of excluding items whose differentiation degree is smaller than a differentiation threshold from among the feature vectors to obtain a feature vector to be extracted, 2. The system of claim 1, wherein the image is incrementally processed before being used to extract feature vectors.
7. Specifically, the incremental learning algorithm module: add one to the first record of the number of times the specific product identifier of the target product was selected; If the identified product identifier of the target product is not ranked first, add 1 to the second record; The system of claim 1, further comprising: a first record and a second record for calculating a reliability of the target product based on the first record and the second record.
8. The vision sensor is further used to capture a video stream of the target item and transmit it to the recognition system by the weighing system; The recognition algorithm module of the recognition system is further adapted to extract an image of the target item from the video stream.
2. The system of claim 1 .
9. The system of claim 1 , wherein the product identifier includes a product name and a product code.
10. A method for detecting an artificial intelligence scale based on autonomous incremental learning, comprising: For use in the system according to any one of claims 1 to 9, receiving and recognizing an image of a target item transmitted from a weighing system and obtaining a feature vector of the image; calculating a similarity between the feature vector and a feature vector in a feature vector database to obtain at least one feature vector that exceeds a similarity threshold; identifying and outputting candidate item identifiers corresponding to the at least one feature vector to a weighing system, the weighing system being adapted to receive a product identifier for a target item identified by a user from among the at least one candidate item identifier and to generate a product tag based on a weight of the target item and the identified product identifier; determining whether incremental learning is required after receiving a specific product identifier of the target product transmitted from the weighing system; if necessary, incrementally processing the image before extracting a feature vector; and merging the extracted feature vector with a product identifier of the target product before storing it in a feature vector database of a data module; The present invention relates to a method for recognizing an artificial intelligence scale based on autonomous incremental learning.
11. receiving and recognizing an image of a target item transmitted from a weighing system and obtaining a feature vector of the image; Recognizing a product area in an image of a target product or extracting a product area from the image of the target product; and Recognizing the product region and obtaining a feature vector of the image; 11. The method of claim 10.
12. The step of determining whether incremental learning is necessary includes: If the identified product identifier of the target product is ranked first among the product identifier candidates corresponding to the at least one recognized feature vector, and the confidence of the target product is lower than a confidence threshold; or If the identified product identifier of the target product is not ranked first among the candidate product identifiers corresponding to the at least one recognized feature vector, or If there are no recognized product identifier candidates, including determining that there is a need for incremental learning; 11. The method of claim 10.
13. The step of incrementally processing the image and then extracting a feature vector comprises: performing incremental processing on the image of the target product, the incremental processing including at least one of rotating, flipping, and color transforming; calling a recognition algorithm module to recognize the incrementally processed image to obtain a feature vector; and excluding items from the feature vectors whose differentiation degree is less than a differentiation threshold to obtain the feature vector to be extracted; 11. The method of claim 10.
14. adding one to a first record of the number of times the identified product identifier of the target product was selected; if the identified product identifier of the target product is not ranked first, adding 1 to the second record; and The method of claim 10, further comprising: calculating a reliability of the target product based on the first record and the second record.
15. A computer device including a memory, a processor, and a computer program stored in the memory and operable on the processor, comprising: A computing apparatus characterized in that it implements the method according to any one of claims 10 to 14 when the computer program is executed by the processor.
16. A computer program is stored in the The computer program, when executed by a processor, performs the steps of the method according to any one of claims 10 to 14. A computer-readable storage medium comprising:
17. This includes computer programs, The computer program, when executed by a processor, causes the method according to any one of claims 10 to 14 to be implemented.
23. A computer program product comprising:
Citation Information
Patent Citations
Intelligent weighing device and intelligent weighing method based on computer vision technology
CN112466068A
Commodity settlement method and device, electronic equipment and medium
CN114078299A
Commodity identification system optimization method and device, equipment and storage medium
CN114140696A
Commodity recognition system and commodity recognition program
JP2015022538A
