Smart vending machine status online monitoring and management system

By using multi-layer shelves, high-definition cameras and weight sensors in smart vending machines, building product size models and adjusting the recognition range, the problem of low recognition efficiency of smart vending machines is solved, and the recognition accuracy and speed are improved.

CN118736724BActive Publication Date: 2024-11-08Hangzhou Qianjing Technology Co., Ltd. +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411208334.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-08
Estimated Expiration
2044-08-30

AI Technical Summary

Technical Problem

When existing smart vending machines identify goods, the recognition accuracy is affected by the size of the goods, distance from the camera and recognition range, resulting in poor recognition efficiency and slow speed.

Method used

An intelligent vending machine using multi-layer shelves, combining high-definition cameras, deep learning models and weight sensors, builds an independent image recognition model by building a product size model, adjusting the image recognition range and scaling ratio, and optimizing the product placement and recognition process.

Benefits of technology

It improves the accuracy and speed of product recognition, reduces the learning cost of recognition models, and ensures the clarity and integrity of product display in images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118736724B_ABST
    Figure CN118736724B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of intelligent accounting systems, and specifically to an intelligent vending machine status online monitoring and management system, including a vending machine with multiple layers of shelves and an image recognition module for intelligently identifying commodity types. A high-definition camera facing downward is provided on the top of the vending machine, and the vending machine also includes a placement management module and an auxiliary recognition module. The placement management module includes a commodity scanning unit and a recommended placement unit, and the auxiliary recognition module includes a weight sensing unit and an identification adjustment unit. The present invention establishes an independent image recognition model for each shelf by setting up an independent image library, thereby reducing the interference of other layers of goods on the image recognition model in identifying the commodities on the shelf on that layer. Compared with the image recognition model for unified identification, the learning cost of the image recognition model can be reduced, the model establishment speed can be improved, and the independent image recognition model can only identify the commodities on that layer, narrowing the comparison range and output range, which helps to improve the recognition accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent accounting systems, and in particular to an online monitoring and management system for the status of intelligent vending machines. Background Art

[0002] Smart vending machines are automated retail devices that combine advanced technologies such as the Internet, the Internet of Things, and smart hardware. They achieve fully automated sales of goods through built-in sensors and sensor systems, as well as intelligent recognition technologies (such as barcode scanning, RFID recognition, and visual recognition). When consumers select goods, the vending machine will automatically identify the product information, including type, price, and quantity, and support multiple payment methods (such as code scanning, card payment, and cash payment) to complete the transaction.

[0003] Smart vending machines are equipped with cameras or image sensors that can capture the process of consumers picking up goods in real time and identify the types and quantities of goods purchased by consumers through image recognition model algorithms. However, smart vending machines in the prior art directly identify goods by looking at the image of consumers picking up goods. This has the defect that the recognition accuracy is affected by the size of the goods, the distance from the camera, and the range of goods to be recognized, resulting in poor recognition efficiency and slow recognition speed. Summary of the invention

[0004] In view of the above-mentioned shortcomings of the prior art, the present invention provides an intelligent vending machine status online monitoring and management system, which can effectively solve the problems of poor efficiency and slow recognition speed of image recognition of goods for sale in the prior art.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0006] The present invention provides an intelligent vending machine status online monitoring and management system, which at least includes: a vending machine with multi-layer shelves and an image recognition module for intelligently identifying the types of goods, a high-definition camera facing downward is provided on the top of the vending machine, and also includes: a placement management module, an auxiliary recognition module, and a weight verification module;

[0007] An image database consisting of multiple product images is preset. The image data in the image database is algorithmically learned through a deep learning model to establish an image recognition model. The image recognition model identifies the products appearing in the picture during the purchase process and forms a purchase product set.

[0008] The placement management module includes a product scanning unit and a recommended placement unit for improving recognition accuracy. The product scanning unit scans and enters product data, and the recommended placement unit plans the placement of products based on the product data and recommends shelves with different layer heights for different products.

[0009] The auxiliary recognition module includes a weight sensing unit and a recognition adjustment unit. The weight sensing unit is provided with a weight sensor under each shelf for sensing the total weight of the goods on the shelf. The weight change difference of the goods on each shelf before and after purchase is calculated. The goods adjustment signal of the shelf is generated according to the data change of the weight sensor. At the same time, the frame of the captured pickup photo is adjusted according to the number of shelves on the shelf. Independent image recognition models are established for multiple goods on shelves on different levels. The recognition range of the image recognition model is adjusted according to the shelf layer corresponding to the goods adjustment signal.

[0010] The weight verification module verifies the total weight of the goods in the purchase commodity set, generates a normal purchase signal or an abnormal purchase signal, and saves the transaction record and the consumer purchase video when an abnormal signal is generated.

[0011] Furthermore, the scanning and input process of the commodity scanning unit is specifically as follows:

[0012] The goods to be placed in the vending machine are recorded as target goods. Before placing the target goods into the vending machine, the name, selling price and appearance pictures of the target goods at multiple angles are obtained, and the appearance pictures of the target goods at multiple angles are added to the image database, the outer packaging size of the target goods is measured, and a size model of the target goods is constructed. The size model is a rectangular parallelepiped, and its length, width and height correspond to the length, width and height measurement values ​​of the outer packaging of the target goods respectively. When the target goods is a non-standard rectangular parallelepiped, the maximum width and maximum length of its top view are used as the length and width values ​​of its size model, and the maximum height of its front view is used as the height value of its size model.

[0013] Furthermore, the placement recommendation process of the recommendation placement unit is specifically as follows:

[0014] Get the size model of each target product, calculate the area of ​​the six faces of the size model, and select the minimum area of ​​the six faces as the minimum recognition area of ​​the target product. , get the vertical distance between each shelf and the top of the vending machine , substitute the preset formula Calculate and obtain the reference value of the identification area , where a is a preset constant coefficient and k is the aspect ratio of the camera imaging. The reference value of the recognition area of ​​each shelf is multiplied by a preset imaging ratio value to obtain the minimum imaging area of ​​each shelf. , where i represents the number of shelves from bottom to top, when i=1 it represents the first layer, the minimum recognition area of ​​each target product , when the minimum recognition area Belong to the interval When , a recommended placement instruction is generated, and the i-th layer is output as the recommended placement layer number. According to the recommended placement layer number of each target product, a same-layer placement set of different placement layers is constructed. The same-layer placement set contains all target products with the same recommended placement layer number.

[0015] Furthermore, the weight sensing unit signal generation process is as follows:

[0016] The total weight of the goods on the shelf is recorded in real time by the weight sensor and recorded as the total weight of the goods. A maintenance time threshold is preset. When the vending machine cabinet door is in a closed state and the total weight of the goods remains unchanged and continues to exceed the maintenance time threshold, a stable signal of the goods is generated and the stable total weight is calculated. The difference between two consecutive stable total weights is calculated to obtain the weight change difference of the goods corresponding to the shelf on this layer.

[0017] When the vending machine door is open and the total weight of the goods changes, a goods adjustment signal is generated. Furthermore, when the goods adjustment signal is generated, the corresponding shelf layer number is output. , adjust the size of the picture captured by the high-definition camera, scale the real-time captured picture, and crop it based on the center of the original picture as the reference point, the size and aspect ratio of the original picture to obtain the recognition picture. The calculation formula of the scaling ratio N is: , where h is the preset proportional coefficient.

[0018] Furthermore, the same-layer placement set of the shelf corresponding to the goods adjustment signal is obtained, and the maximum value of the minimum recognition area of ​​all target goods in the same-layer placement set is selected as the adjustment reference value P, and substituted into the formula The scaling ratio N is further corrected and calculated to obtain the corrected scaling ratio N, where It is the reference value of the identification area corresponding to this shelf.

[0019] Furthermore, the identification and adjustment unit periodically takes photos of the goods during the time when the goods adjustment signal is generated and lasts, and the photo-taking and identification process is specifically as follows:

[0020] The duration interval of each cargo adjustment signal is recorded as a verification interval, and a shooting time interval T is preset. A pickup photo is captured every T time interval in the verification interval, and the pickup photo is processed to obtain an identification photo. All identification photos captured in the verification interval are identified by an image recognition model, and the goods identified in each identification photo are marked as sold goods. The same sold goods in adjacent identification photos are recorded as the same sold goods;

[0021] The same-layer placement set of each shelf is obtained, and the image database is screened according to the commodity types in each same-layer placement set, so as to obtain multiple image databases containing only commodities in the same-layer placement set, which are recorded as independent image libraries. Each independent image library corresponds to a same-layer placement set. Independent image recognition models are established for shelves on different layers with independent image libraries, and the shelf corresponding to the cargo adjustment signal is obtained. The identification photo is identified with the image recognition model corresponding to the shelf, and the output range of the image recognition model is limited to the commodity types contained in the same-layer placement set.

[0022] Furthermore, the weight checking module checks the total weight of the goods in the purchased goods set as follows:

[0023] Obtain information about the consumer who opens the door of the vending machine, record video through a high-definition camera when the door of the vending machine is opened, stop recording video when the door of the vending machine is closed, bind the recorded video to the consumer information, obtain the consumer's purchase commodity set and order information, calculate the total weight of all commodities in the purchase commodity set as the total purchase weight, obtain the difference in weight change of goods on each shelf and calculate the total to obtain the verified total weight, generate a normal purchase signal when the purchase total weight is equal to the verified total weight, retain the order information and clear the recorded video, and generate an abnormal purchase signal when the purchase total weight is not equal to the verified total weight, and save the video and order information recorded during the consumer's purchase process.

[0024] Compared with the known prior art, the technical solution provided by the present invention has the following beneficial effects:

[0025] 1. The present invention can better analyze commodities of different shapes by constructing size models for different target commodities, and establish a unified quantitative standard, so as to compare the sizes of different target commodities from the perspective of recognition and analysis, so as to allocate the target commodities to different shelf layers of the vending machine according to their sizes, so that in the subsequent process, when the image recognition model captures the image for recognition, the target commodity can occupy a sufficient proportion of the frame in the image, thereby improving the clarity of the target commodity displayed in the image, which helps to improve the recognition accuracy of the image recognition model.

[0026] 2. The present invention can adjust the zoom ratio of the captured image according to the shelf layer where the product is located, further increasing the proportion of the detected product in the detection image. Compared with ordinary zoom adjustment, the zoom ratio adjustment is limited according to the maximum minimum recognition area, which can magnify the image of the detected product in the detection image as much as possible while avoiding the problem of incomplete display of the product due to excessive zoom, further ensuring that the product can be stably identified.

[0027] 3. The recognition adjustment unit of the present invention establishes an independent image recognition model for each shelf, thereby reducing the interference of other layers of goods on the image recognition model in identifying the goods on this layer of the shelf. Compared with the unified recognition image recognition model, it can reduce the learning cost of the image recognition model and improve the model establishment speed. In addition, the independent image recognition model can only identify the goods on this layer, narrowing the comparison range and output range, which helps to improve the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 It is the overall module block diagram of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] The present invention will be further described below in conjunction with the embodiments.

[0032] See also Figure 1 , the intelligent vending machine status online monitoring and management system includes at least: a vending machine with multi-layer shelves and an image recognition module for intelligently identifying the types of goods, and also includes a placement management module

[0033] The image recognition module includes a high-definition camera installed on the top of the vending machine and facing downward. The high-definition camera takes photos when consumers pick up goods. The taken photos are compared and recognized through image recognition to obtain the types of goods purchased by consumers. The image recognition process is as follows:

[0034] An image database consisting of multiple product images is preset, each product image corresponds to a product on sale (the same product on sale corresponds to product images from multiple angles), and each product on sale corresponds to a product name and a product price. The image data in the image database is algorithmically learned through a deep learning model and an image recognition model is established. When a picture contains a product on sale contained in the image database, the image recognition model can identify the product on sale in the picture and output the corresponding product name and product price data. The photos taken during the consumer's pickup process are recorded as pickup photos. The pickup photos are identified through the image recognition model, and all the products on sale that appear in the pickup photos are obtained to form a purchase product set.

[0035] It should be noted that algorithm learning based on deep learning models is an existing technology, such as YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector) and CNN (Convolutional Neural Network). The algorithm will extract features of product images in the image database. These features may include the shape, color, texture, etc. of the product. By extracting these features as an important basis for distinguishing different products, the algorithm can determine the type of product corresponding to the image when identifying the same or similar feature combinations.

[0036] The placement management module includes a product scanning unit and a recommended placement unit. The product scanning unit scans and enters product data. The recommended placement unit plans the placement of products based on the product data and recommends shelves of different heights for different products. When products on different shelves are taken away, the image recognition module can accurately capture the complete picture and adjust the resolution to enlarge the image, so that the image recognition module can recognize the name and price of the product.

[0037] The scanning and input process of the commodity scanning unit is as follows:

[0038] The goods to be placed in the vending machine are recorded as target goods. Before placing the target goods into the vending machine, the name, selling price and appearance pictures of the target goods at multiple angles are obtained (the appearance pictures at multiple angles include but are not limited to the front photo, the back photo and the side photo of the target goods), and the appearance pictures of the target goods at multiple angles are added to the image database for learning the image recognition model, the outer packaging size of the target goods is measured, and a size model of the target goods is constructed. The size model is a rectangular parallelepiped, and its length, width and height correspond to the length, width and height measurement values ​​of the outer packaging of the target goods respectively. When the target goods is a non-standard rectangular parallelepiped, the maximum width and maximum length of its top view are used as the length and width values ​​of its size model, and the maximum height of its front view is used as the height value of its size model.

[0039] It should be noted that the size model aims to uniformly quantify the length, width and height of the target product, and abstractly simulate all products as rectangular blocks to facilitate area and volume analysis and comparison in the subsequent process. By comparing the size models of different target products, the smallest side of the target product can be determined, that is, the side with the smallest projected area.

[0040] The recommended placement process of the recommended placement unit is as follows:

[0041] Get the size model of each target product, calculate the area of ​​the six faces of the size model, and select the minimum area of ​​the six faces as the minimum recognition area of ​​the target product. , get the vertical distance between each shelf and the top of the vending machine , substitute the preset formula Calculate and obtain the reference value of the identification area , where a is a preset constant coefficient and k is the aspect ratio of the camera imaging. The reference value of the recognition area of ​​each shelf is multiplied by a preset imaging ratio value (a preset object screen ratio value, under which the object can occupy a larger frame in the picture captured by the high-definition camera for easy recognition by the image recognition module), and the minimum imaging area of ​​each shelf is obtained. , where i represents the number of shelves from bottom to top, when i=1 it represents the first layer, the minimum recognition area of ​​each target product , when the minimum recognition area Belong to the interval , generates a recommended placement instruction, outputs the i-th layer as the recommended placement layer number, and constructs the same-layer placement sets of different placement layers according to the recommended placement layer number of each target product. The same-layer placement set contains all target products with the same recommended placement layer number, that is, all target products recommended to be placed on the same layer are divided into the same same-layer placement set.

[0042] It should be noted that by constructing size models for different target commodities, it is possible to better analyze commodities of different shapes and establish a unified quantitative standard to facilitate size comparison of different target commodities from the perspective of recognition and analysis, so that the target commodities can be allocated to different shelf layers of the vending machine according to their sizes. In the subsequent process, when the image recognition model captures the image for recognition, the target commodity can occupy a sufficient proportion of the frame in the image, thereby improving the clarity of the target commodity displayed in the image, which helps to improve the recognition accuracy of the image recognition model.

[0043] The auxiliary recognition module includes a weight sensing unit and an identification adjustment unit. The weight sensing unit is provided with a weight sensor under each shelf to sense the total weight of the goods on the shelf. A goods adjustment signal for the shelf is generated according to the data changes of the weight sensor. At the same time, the captured pickup photo is adjusted according to the number of layers of the shelf, and the goods taken by the consumer are recorded as sold goods. The identification adjustment unit adjusts the recognition range of the image recognition model according to the same-layer placement set to which the sold goods belong, so that the image recognition model can narrow the comparison range, improve the comparison speed and the probability of misjudgment.

[0044] The weight sensing unit records the total weight of the goods on the shelf in real time through the weight sensor and records it as the total weight of the goods. A maintenance time threshold is preset (the value is 30 seconds in a specific embodiment). When the vending machine cabinet door is in a closed state and the total weight of the goods remains unchanged and continues to exceed the maintenance time threshold, a stable signal of the goods is generated and the stable total weight is calculated. The difference between two adjacent stable total weights is calculated to obtain the weight change difference of the goods corresponding to the shelf at this level;

[0045] When the door of the vending machine is in the open state and the total weight of the goods changes, a goods adjustment signal is generated. Furthermore, when the goods adjustment signal is generated, the corresponding shelf layer number i is output, the size of the image captured by the high-definition camera is adjusted, the real-time captured image is scaled, and the original image center is used as the reference point, the size of the original image and the aspect ratio are cropped to obtain the recognition image. The calculation formula of the scaling ratio N is: , where h is the preset proportional coefficient.

[0046] For example, in a specific embodiment, the final value of N is calculated to be 1.5, so the image captured by the high-definition camera is enlarged 1.5 times, and then the original image is cropped based on the center of the original image and the size and aspect ratio of the original image to obtain an enlarged image for easy recognition by the image recognition model.

[0047] Furthermore, the same-layer placement set of the shelf corresponding to the goods adjustment signal is obtained, and the maximum value of the minimum recognition area of ​​all target goods in the same-layer placement set is selected as the adjustment reference value P, and substituted into the formula The scaling ratio N is further corrected and calculated to obtain the corrected scaling ratio N, where It is the reference value of the identification area corresponding to this shelf.

[0048] By adjusting the zoom ratio of the captured image, the proportion of the detected product in the detection image can be further increased. Compared with ordinary zoom adjustment, limiting the adjustment of the zoom ratio according to the maximum and minimum recognition areas can magnify the image of the detected product in the detection image as much as possible while avoiding the problem of incomplete display of the product due to excessive zoom, further ensuring that the product can be stably identified.

[0049] The identification and adjustment unit periodically takes photos of the goods during the time when the goods adjustment signal is generated and lasts. The specific process of photo-taking and identification is as follows:

[0050] The duration interval of each cargo adjustment signal is recorded as a verification interval, and a shooting time interval T is preset (in a specific embodiment, the shooting time interval is 1 second). In the verification interval, a pickup photo is captured every time interval T, and the pickup photo is processed to obtain an identification photo (based on each pickup photo corresponding to the generation of the cargo adjustment signal). All identification photos captured in the verification interval are identified by an image recognition model, and the goods identified in each identification photo are marked as sold goods. The same sold goods in adjacent identification photos are recorded as the same sold goods (to avoid the same product appearing in multiple identification photos and being identified multiple times);

[0051] It should be noted that the purchase commodity set consists of one or more sold commodities from different shelves. The commodities on sale, target commodities and sold commodities all refer to commodities. The main difference between them is that they exist in different analysis steps to facilitate the distinction in different steps.

[0052] The same-layer placement set of each shelf is obtained, and the image database is screened according to the commodity types in each same-layer placement set, so as to obtain multiple image databases containing only commodities in the same-layer placement set, which are recorded as independent image libraries. Each independent image library corresponds to a same-layer placement set. Independent image recognition models are established for shelves on different layers with independent image libraries, and the shelf corresponding to the cargo adjustment signal is obtained. The identification photo is identified with the image recognition model corresponding to the shelf, and the output range of the image recognition model is limited to the commodity types contained in the same-layer placement set.

[0053] It should be noted that by setting up an independent image library to establish an independent image recognition model for each shelf, the interference of other layers of goods on the image recognition model in identifying the goods on that layer of the shelf can be reduced. Compared with the unified recognition image recognition model, the learning cost of the image recognition model can be reduced and the model establishment speed can be improved. In addition, the independent image recognition model can only identify the goods on that layer, narrowing the comparison range and output range, which helps to improve the recognition accuracy.

[0054] The weight verification module verifies the total weight of the goods in the purchase set and generates a normal purchase signal or an abnormal purchase signal. When an abnormal signal is generated, the transaction record and the consumer purchase video are saved for subsequent order verification. The operation process of the weight verification module is as follows:

[0055] The information of the consumer who opens the door of the vending machine is obtained, and a video is recorded through a high-definition camera when the door of the vending machine is opened. The video recording is stopped when the door of the vending machine is closed. The recorded video is bound to the consumer information, and the purchased goods set and order information of the consumer are obtained (the order information is automatically generated based on the purchased goods set when the consumer closes the door, and a complete purchase order is formed after the consumer completes the payment). The total weight of all goods in the purchased goods set is calculated as the total purchase weight, and the difference in the weight change of goods on each shelf is obtained and the total is calculated to obtain the verified total weight. When the purchase total weight is equal to the verified total weight, a normal purchase signal is generated, the order information is retained, and the recorded video is cleared, thereby retaining sufficient storage space. When the purchase total weight is not equal to the verified total weight, an abnormal purchase signal is generated, and the video and order information recorded during the purchase process of the consumer are saved to facilitate order verification in subsequent processes.

[0056] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent vending machine status online monitoring and management system, including a vending machine with multi-layer shelves and an image recognition module for intelligently identifying the types of goods. A high-definition camera facing downward is provided on the top of the vending machine, characterized in that: Also includes: Placement management module, auxiliary identification module, weight verification module; An image database consisting of multiple product images is preset. The image data in the image database is algorithmically learned through a deep learning model to establish an image recognition model. The image recognition model identifies the products appearing in the picture during the purchase process and forms a purchase product set. The placement management module includes a product scanning unit and a recommended placement unit for improving recognition accuracy. The product scanning unit scans and enters product data, and the recommended placement unit plans the placement of products based on the product data and recommends shelves with different layer heights for different products. The auxiliary recognition module includes a weight sensing unit and a recognition adjustment unit. The weight sensing unit is provided with a weight sensor under each shelf for sensing the total weight of the goods on the shelf. The weight change difference of the goods on each shelf before and after purchase is calculated. The goods adjustment signal of the shelf is generated according to the data change of the weight sensor. At the same time, the frame of the captured pickup photo is adjusted according to the number of shelves on the shelf. Independent image recognition models are established for multiple goods on shelves on different levels. The recognition range of the image recognition model is adjusted according to the shelf layer corresponding to the goods adjustment signal. The weight verification module verifies the total weight of the goods in the purchase commodity set, generates a normal purchase signal or an abnormal purchase signal, and saves the transaction record and the consumer purchase video when an abnormal signal is generated.

2. The intelligent vending machine status online monitoring and management system according to claim 1 is characterized in that: The scanning and input process of the commodity scanning unit is as follows: The goods to be placed in the vending machine are recorded as target goods. Before placing the target goods into the vending machine, the name, selling price and appearance pictures of the target goods at multiple angles are obtained, and the appearance pictures of the target goods at multiple angles are added to the image database, the outer packaging size of the target goods is measured, and a size model of the target goods is constructed. The size model is a rectangular parallelepiped, and its length, width and height correspond to the length, width and height measurement values ​​of the outer packaging of the target goods respectively. When the target goods is a non-standard rectangular parallelepiped, the maximum width and maximum length of its top view are used as the length and width values ​​of its size model, and the maximum height of its front view is used as the height value of its size model.

3. The intelligent vending machine status online monitoring and management system according to claim 2 is characterized in that: The specific process of the recommended placement unit recommending placement is as follows: Get the size model of each target product, calculate the area of ​​the six faces of the size model, and select the minimum area of ​​the six faces as the minimum recognition area of ​​the target product. , get the vertical distance between each shelf and the top of the vending machine , substitute the preset formula Calculate in and get the reference value of the identification area , where a is a preset constant coefficient and k is the aspect ratio of the camera imaging. The reference value of the recognition area of ​​each shelf is multiplied by a preset imaging ratio value to obtain the minimum imaging area of ​​each shelf. , where i represents the number of shelves from bottom to top, when i=1 it represents the first layer, the minimum recognition area of ​​each target product , when the minimum recognition area Belong to the interval When , a recommended placement instruction is generated, and the i-th layer is output as the recommended placement layer number. According to the recommended placement layer number of each target product, a same-layer placement set of different placement layers is constructed. The same-layer placement set contains all target products with the same recommended placement layer number.

4. The intelligent vending machine status online monitoring and management system according to claim 3 is characterized in that: The weight sensing unit signal generation process is as follows: The total weight of the goods on the shelf is recorded in real time by the weight sensor and recorded as the total weight of the goods. A maintenance time threshold is preset. When the vending machine cabinet door is in a closed state and the total weight of the goods remains unchanged and continues to exceed the maintenance time threshold, a stable signal of the goods is generated and the stable total weight is calculated. The difference between two consecutive stable total weights is calculated to obtain the weight change difference of the goods corresponding to the shelf on this layer. When the vending machine door is open and the total weight of the goods changes, a goods adjustment signal is generated. When the goods adjustment signal is generated, the corresponding shelf layer number is output. , adjust the size of the picture captured by the high-definition camera, scale the real-time captured picture, and crop it based on the center of the original picture as the reference point, the size and aspect ratio of the original picture to obtain the recognition picture. The calculation formula of the scaling ratio N is: , where h is the preset proportional coefficient.

5. The intelligent vending machine status online monitoring and management system according to claim 4 is characterized in that: Get the same-layer placement set of the shelf corresponding to the goods adjustment signal, select the maximum value of the minimum recognition area of ​​all target goods in the same-layer placement set as the adjustment reference value P, and substitute it into the formula The scaling ratio N is corrected and calculated to obtain the corrected scaling ratio N, where It is the reference value of the identification area corresponding to this shelf.

6. The intelligent vending machine status online monitoring and management system according to claim 1, characterized in that: The identification and adjustment unit periodically takes photos of the goods during the time when the goods adjustment signal is generated and lasts. The specific process of the photo identification is as follows: The duration interval of each cargo adjustment signal is recorded as a verification interval, and a shooting time interval T is preset. A pickup photo is captured every T time interval in the verification interval, and the pickup photo is processed to obtain an identification photo. All identification photos captured in the verification interval are identified by an image recognition model, and the goods identified in each identification photo are marked as sold goods. The same sold goods in adjacent identification photos are recorded as the same sold goods; The same-layer placement set of each shelf is obtained, and the image database is screened according to the commodity types in each same-layer placement set, so as to obtain multiple image databases containing only commodities in the same-layer placement set, which are recorded as independent image libraries. Each independent image library corresponds to a same-layer placement set. Independent image recognition models are established for shelves on different layers with independent image libraries, and the shelf corresponding to the cargo adjustment signal is obtained. The identification photo is identified with the image recognition model corresponding to the shelf, and the output range of the image recognition model is limited to the commodity types contained in the same-layer placement set.

7. The intelligent vending machine status online monitoring and management system according to claim 1, characterized in that: The weight checking module checks the total weight of the goods in the purchased goods set in detail as follows: Obtain information about the consumer who opens the door of the vending machine, record video through a high-definition camera when the door of the vending machine is opened, stop recording video when the door of the vending machine is closed, bind the recorded video to the consumer information, obtain the consumer's purchase commodity set and order information, calculate the total weight of all commodities in the purchase commodity set as the total purchase weight, obtain the difference in weight change of goods on each shelf and calculate the total to obtain the verified total weight, generate a normal purchase signal when the purchase total weight is equal to the verified total weight, retain the order information and clear the recorded video, and generate an abnormal purchase signal when the purchase total weight is not equal to the verified total weight, and save the video and order information recorded during the consumer's purchase process.