Vending machine system

Abstract

The invention relates to the technical field of image processing, in particular to an unmanned vending machine system comprising an image acquisition module used for acquiring commodity images in a vending machine; the image transmission module is used for transmitting the image acquired by the image acquisition module to the management control module; the attitude sensor is used for monitoring and calculating door opening and closing angles of the vending machine and transmitting data to the management control module; the management control module is used for receiving and analyzing image data captured by the camera and door opening and closing angle information provided by the attitude sensor so as to monitor the working state and the commodity display condition of the vending machine; and the database is used for storing image data. According to the vending machine, the camera-control all-in-one machine is used, real-time monitoring and remote control over the state of the vending machine are achieved, the angle-adjustable function of a camera meets different shooting requirements, and the flexibility and applicability of the system are improved; the attitude sensor is combined with the accelerometer and the gyroscope, so that the state of the vending machine is accurately monitored and controlled.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an unmanned vending machine system. Background Art

[0002] Unmanned vending machines are automated retail devices that use advanced technology to sell goods without human intervention. These machines are usually operated by touch screens or buttons, allowing customers to browse a list of goods, select goods, and make payments. Payment methods may include cash, credit cards, debit cards, or mobile payment platforms such as Alipay and WeChat Pay. IoT technology enables vending machines to monitor inventory levels in real time, automatically replenish goods, and be maintained through remote management systems. Secondly, big data analysis allows companies to collect consumer purchase data, gain insights into consumer habits, and optimize product mix and pricing strategies. In addition, the popularity of mobile payments and electronic currencies has reduced transaction costs and increased the convenience of transactions. Traditional vending machine monitoring and control processes may be relatively simple, lacking remote monitoring functions, and users cannot immediately understand the inventory status or machine failures in the vending machine; at the same time, vending machines in different locations may face different shooting challenges, and fixed-angle cameras are difficult to adapt to all scenarios. Summary of the invention

[0003] The main purpose of the present invention is to provide an unmanned vending machine system to solve the problems of relatively simple monitoring and control processes and camera limitations proposed in the relevant technology.

[0004] In order to achieve the above object, according to one aspect of the present invention, an unmanned vending machine system is provided, comprising: Image acquisition module: used to collect product images in vending machines; Image transmission module: used to transmit the image collected by the image acquisition module to the management control module; Posture sensor: used to monitor and calculate the opening and closing angles of the vending machine and transmit the data to the management control module; Management and control module: used to receive and analyze image data captured by the camera and the door opening and closing angle information provided by the posture sensor to monitor the working status of the vending machine and the display of goods; Database: used to store image data.

[0005] Furthermore, the database includes a real-time database and a historical database. The real-time database is used to store the images of commodities to be identified transmitted by the image transmission module; the historical database is used to store the images of commodities saved in advance.

[0006] Furthermore, the image acquisition module is based on MIPI dual-camera. There are two requirements for the MIPI dual-camera. The first is to capture both the inside and outside of the cabinet; the second is to only capture the outside of the cabinet. The two lens angles of the MIPI dual-camera need to be adjustable from 15 to 30 degrees with reference to the vertical line on the left; and from 25 to 35 degrees with reference to the horizontal line on the right.

[0007] Furthermore, the attitude sensor includes an accelerometer and a gyroscope. The attitude sensor calculates the door opening and closing angle of the vending machine based on a complementary filtering algorithm. The calculation steps are as follows: S1: Use the accelerometer to get the tilt angle: ; in, is the tilt angle, is the y-axis component of the accelerometer; is the z-axis component of the accelerometer; is the noise term in the accelerometer data; S2: Use the gyroscope to get the angle change: ; in, is the estimated angle; is the angular velocity data; is the sampling time interval; is the noise term in the gyroscope data; S3: Angle estimation after complementary filtering of the results of the accelerometer and gyroscope: ; in, is the angle estimation value after complementary filtering; is the filter coefficient.

[0008] Furthermore, the steps of the management control module monitoring the working status of the vending machine are as follows: S4: Set the angle threshold for door opening and closing; S5: judging whether the current door state is open or closed according to the calculated angle information and the set threshold; S6: Implement corresponding control logic according to the door status judgment result; S7: According to the control logic, trigger the opening or closing action of the vending machine door.

[0009] Furthermore, the management and control module includes an image processing module, an image recognition module and a monitoring and analysis module; The image processing module is used to pre-process the transmitted images; The image recognition module is used to identify and judge the products in the image; The monitoring and analysis module is used to monitor and analyze the working status of the vending machine and the display of goods.

[0010] Furthermore, the image recognition module recognizes and judges the input image based on the SIFT algorithm, and the steps are as follows: S8: Extract SIFT features from the image, including detecting key points in the image, calculating the gradient information of the areas around these key points, and generating a 128-dimensional SIFT descriptor for each key point; S9: For all images to be matched, SIFT features are extracted and stored, and a feature library containing SIFT features of all images is established; S10: Compare each feature descriptor of the query image with all descriptors in the feature library; S11: Finally, the matching results are visualized to show the correspondence between the query image and the matching image, thereby achieving the purpose of image recognition matching.

[0011] Furthermore, in S8, the key point detection is based on the Gaussian difference pyramid, which finds the extreme points in the image by performing Gaussian blur and difference operations on the image at different scales. The expression is: ; ; in, The key points after Gaussian blur and difference operation; is the layer of the Gaussian pyramid; is the input product image to be detected; k is the scale factor; is the scale parameter; is a Gaussian function; * indicates a convolution operation.

[0012] Further, in S8, the gradient information of the areas around these key points is calculated, and the gradient information includes horizontal gradient, vertical gradient, gradient amplitude and gradient direction, and the expression is: ; ; ; ; in, is the horizontal gradient; is the vertical gradient; For the key point; is the gradient amplitude of the pixel; is the gradient direction of the pixel; is the key point of the input product image to be detected.

[0013] Further, in S10, each feature descriptor of the query image is compared with all descriptors in the feature library based on the Euclidean distance, which is expressed as: The descriptor is ,in, Represents the i-th element in the descriptor vector; The distance between two descriptors is expressed as: ; in, They respectively represent the descriptors of the input product images to be detected; Represents the corresponding feature descriptor in the feature library; Represents the i-th component in the input product image vector to be detected; The i-th component in the corresponding vector in the feature library; represents the weight of the i-th component; The threshold of the distance between descriptors and the threshold of the number of successful matches are set in advance. If the distance between a pair of feature descriptors is less than the set threshold, the two feature points are successfully matched. The number of pairs of successfully matched feature points is counted. If it exceeds a certain ratio or number threshold, the two images are judged to be the same.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The present invention uses an all-in-one camera and control machine to support remote monitoring and management, realize real-time monitoring and remote control of the vending machine status, and the camera angle adjustable function meets different shooting requirements, thereby improving the flexibility and applicability of the system; the attitude sensor is combined with an accelerometer and a gyroscope, and a complementary filtering algorithm is used to calculate the door opening and closing angles to ensure accurate monitoring and control of the vending machine status; and the SIFT algorithm is used to identify and match images, which can extract key features and perform effective matching to realize automatic identification and judgment of goods; the present invention has higher intelligence, automation and precise identification capabilities, can effectively monitor the status of vending machines, identify goods, improve service efficiency, and provide users with a more convenient shopping experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A system block diagram of the present invention as a whole; Figure 2 is a system block diagram of the management control module in the present invention; Illustration Description: 1. Image acquisition module; 2. Image transmission module; 3. Posture sensor; 4. Management and control module; 41. Image processing module; 42. Image recognition module; 43. Monitoring and analysis module; 5. Database; 51. Real-time database; 52. Historical database. DETAILED DESCRIPTION

[0016] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0017] See also Figure 1-Figure 2 , this embodiment provides an unmanned vending machine system, including: Image acquisition module 1: used to acquire commodity images in the vending machine; Image transmission module 2: used to transmit the image acquired by image acquisition module 1 to management control module 4; Attitude sensor 3: used to monitor and calculate the opening and closing angle of the vending machine and transmit the data to the management control module 4; Management control module 4: used to receive and analyze the image data captured by the camera and the door opening and closing angle information provided by the posture sensor 3, so as to monitor the working status of the vending machine and the display of goods; Database 5: used to store image data.

[0018] The database 5 includes a real-time database 51 and a historical database 52. The real-time database 51 is used to store the images of commodities to be identified transmitted by the image transmission module 2; the historical database 52 is used to store the images of commodities saved in advance.

[0019] Image acquisition module 1 is based on MIPI dual camera. There are two requirements for MIPI dual camera. The first is to capture both the inside and outside of the cabinet. The second is to capture only the outside of the cabinet. The two lens angles of MIPI dual camera need to be adjustable from 15 to 30 degrees with reference to the vertical line on the left side and from 25 to 35 degrees with reference to the horizontal line on the right side.

[0020] The attitude sensor 3 includes an accelerometer and a gyroscope. The attitude sensor 3 calculates the door opening and closing angle of the vending machine based on a complementary filtering algorithm. The calculation steps are as follows: S1: Use the accelerometer to get the tilt angle: ; in, is the tilt angle, is the y-axis component of the accelerometer; is the z-axis component of the accelerometer; is the noise term in the accelerometer data; S2: Use the gyroscope to get the angle change: ; in, is the estimated angle; is the angular velocity data; is the sampling time interval; is the noise term in the gyroscope data; S3: Angle estimation after complementary filtering of the results of the accelerometer and gyroscope: ; in, is the angle estimation value after complementary filtering; is the filter coefficient.

[0021] The steps of the management control module 4 monitoring the working status of the vending machine are as follows: S4: Set the angle threshold for door opening and closing; S5: judging whether the current door state is open or closed according to the calculated angle information and the set threshold; S6: Implement corresponding control logic according to the door status judgment result; S7: According to the control logic, trigger the opening or closing action of the vending machine door.

[0022] The management control module 4 includes an image processing module 41, an image recognition module 42 and a monitoring and analysis module 43; The image processing module 41 is used to pre-process the transmitted image; The image recognition module 42 is used to identify and judge the commodity in the image; The monitoring and analysis module 43 is used to monitor and analyze the working status of the vending machine and the display of goods.

[0023] The image recognition module 42 recognizes and judges the input image based on the SIFT algorithm, and the steps are as follows: S8: Extract SIFT features from the image, including detecting key points in the image, calculating the gradient information of the areas around these key points, and generating a 128-dimensional SIFT descriptor for each key point; S9: For all images to be matched, SIFT features are extracted and stored, and a feature library containing SIFT features of all images is established; S10: Compare each feature descriptor of the query image with all descriptors in the feature library; S11: Finally, the matching results are visualized to show the correspondence between the query image and the matching image, thereby achieving the purpose of image recognition matching.

[0024] In S8, key point detection is based on Gaussian difference pyramid, which finds extreme points in the image by performing Gaussian blur and difference operations on the image at different scales. The expression is: ; ; in, The key points after Gaussian blur and difference operation; is the layer of the Gaussian pyramid; is the input product image to be detected; k is the scale factor; is the scale parameter; is a Gaussian function; * indicates a convolution operation.

[0025] In S8, the gradient information of the areas around these key points is calculated. The gradient information includes horizontal gradient, vertical gradient, gradient amplitude and gradient direction. The expression is: ; ; ; ; in, is the horizontal gradient; is the vertical gradient; For the key point; is the gradient amplitude of the pixel; is the gradient direction of the pixel; is the key point of the input product image to be detected.

[0026] In S10, each feature descriptor of the query image is compared with all descriptors in the feature library based on the Euclidean distance, which is expressed as: The descriptor is ,in, Represents the i-th element in the descriptor vector; The distance between two descriptors is expressed as: ; in, They respectively represent the descriptors of the input product images to be detected; Represents the corresponding feature descriptor in the feature library; Represents the i-th component in the input product image vector to be detected; The i-th component in the corresponding vector in the feature library; represents the weight of the i-th component; The threshold of the distance between descriptors and the threshold of the number of successful matches are set in advance. If the distance between a pair of feature descriptors is less than the set threshold, the two feature points are successfully matched. The number of pairs of successfully matched feature points is counted. If it exceeds a certain ratio or number threshold, the two images are judged to be the same.

[0027] In this embodiment, the image acquisition module 1 will collect the commodity images in the vending machine, and the image transmission module 2 will transmit the collected images to the image processing module 41. At the same time, the posture sensor 3 will detect and calculate the door opening and closing angles of the vending machine and transmit the data to the monitoring and analysis module 43. The image processing module 41 will pre-process the transmitted images and store them in the real-time database 51 after processing. The image recognition module 42 will compare and judge the images in the real-time database 51 with the images in the historical database 52 to see whether the commodities displayed for sale are consistent with the items required by the manufacturer, and transmit the final judgment results to the monitoring and analysis module 43 to monitor whether the working status of the vending machine is consistent with the sale of commodities. The specific method is as follows: The attitude sensor 3 includes an accelerometer and a gyroscope. The selected attitude sensor is installed on the door of the vending machine to ensure that the sensor can accurately sense the movement and angle change of the door. The data provided by the sensor is used to calculate the opening and closing angle of the vending machine door through a complementary filtering algorithm. The complementary filtering algorithm is a method commonly used to fuse accelerometer and gyroscope data for attitude estimation. The data of the accelerometer and gyroscope are fused to obtain a more accurate and stable attitude estimation result. The calculation steps are as follows: First, use the accelerometer to get the tilt angle: ; in, is the tilt angle, is the y-axis component of the accelerometer; is the z-axis component of the accelerometer; is the noise term in the accelerometer data; Then use the gyroscope to get the angle change: ; in, is the estimated angle; is the angular velocity data; is the sampling time interval; is the noise term in the gyroscope data; Finally, the angle estimate is obtained by complementary filtering of the results of the accelerometer and gyroscope: ; in, is the angle estimation value after complementary filtering; is the filter coefficient. The filter coefficient is used to balance the weight of the accelerometer and gyroscope data in attitude estimation. By continuously updating the value of θ and combining the data of the accelerometer and gyroscope, a more stable and accurate attitude estimation result can be obtained.

[0028] In practical applications, sensor data may be affected by noise. The noise term in accelerometer data is used to simulate the uncertainty and noise of accelerometer measurements. Accelerometers are usually used to measure gravity acceleration and the acceleration of objects in space. Therefore, they may be affected by vibration, shock or other environmental factors, resulting in a certain degree of noise in the measurements. The noise term in gyroscope data is used to simulate the uncertainty and noise of gyroscope measurements. Gyroscopes are usually used to measure angular velocity, that is, the speed at which an object rotates around each axis. Gyroscope data may be affected by mechanical vibration, temperature changes and other factors, resulting in a certain degree of noise in the measurements. Introducing noise terms in the complementary filtering algorithm can more effectively process the noise in sensor data and improve the accuracy and stability of attitude estimation.

[0029] Furthermore, the steps of monitoring the working state of the vending machine are as follows: first set the angle thresholds for the door opening and closing in advance; determine whether the current door state is open or closed based on the angle information calculated above and the set threshold; implement the corresponding control logic based on the door state judgment result; for example: if the door angle exceeds the opening threshold and the current door state is closed, the door opening action is triggered; if the door angle is lower than the closing threshold and the current door state is open, the door closing action is triggered. According to the control logic, the opening or closing action of the vending machine door is triggered; the angle information of the vending machine door is continuously monitored, and the door state is updated and the corresponding control action is executed as needed.

[0030] The above steps can realize real-time monitoring and control of the vending machine door status, ensuring that the door opening and closing actions of the vending machine can be carried out as expected during operation, thereby improving the stability, safety and efficiency of the system. Through refined door control logic, the system can detect and handle abnormal situations in a timely manner, ensure the normal operation of the vending machine, and improve user experience and management efficiency.

[0031] Image acquisition module 1 is based on MIPI dual-camera. There are two requirements for MIPI dual-camera cameras. The first is to capture the inside and outside of the cabinet; the second is to capture only the outside of the cabinet. The two lens angles of MIPI dual-camera need to meet the requirements that the left side is adjustable by 15-30 degrees with reference to the vertical line; the right side is adjustable by 25-35 degrees with reference to the horizontal line. In the MIPI dual-camera system, the naked image data captured by the camera is directly transmitted to the main control CPU for processing, which can quickly process data without the participation of additional processing chips. By using two cameras, different areas inside and outside the cabinet can be covered. One camera focuses on shooting the items inside the cabinet, and the other camera captures the environment outside the cabinet. This setting can ensure that information is obtained from multiple perspectives, thereby improving the monitoring capability and accuracy of the overall system. At the same time, the installation angles of the two cameras can be adjusted according to needs to ensure that the cameras can effectively capture specific areas. The angle adjustment of the left camera relative to the vertical line can ensure clear imaging of the items inside the cabinet; and the angle adjustment of the right camera relative to the horizontal line is conducive to capturing activities outside the cabinet.

[0032] After receiving the collected data, the image recognition module 42 recognizes and judges the input image based on the SIFT algorithm, and the steps are as follows: extract SIFT features in the image, including detecting key points in the image, calculating the gradient information of the areas around these key points, and generating a 128-dimensional SIFT descriptor for each key point; for all images to be matched, SIFT features are extracted and stored, and a feature library containing SIFT features of all images is established; each feature descriptor of the query image is compared with all descriptors in the feature library; and finally, the matching results are visualized to show the correspondence between the query image and the matching image, thereby achieving the purpose of image recognition matching.

[0033] Among them, the key point detection is based on the Gaussian difference pyramid. The Gaussian difference pyramid finds the extreme points in the image by performing Gaussian blur and difference operations on the image at different scales. The expression is: ; ; in, is the key point after Gaussian blur and difference operation, which means the feature response image obtained after Gaussian blur and difference operation on the original image at different scales, and is used to find the extreme points in the image, that is, the potential key points; is the layer of the Gaussian pyramid, which means the smoothed image layer obtained by performing Gaussian blur operation on the original image at a specific scale. Using different layers of the Gaussian pyramid to calculate the difference instead of directly performing the difference operation on the original image can improve the stability and accuracy of the key points, so that the SIFT algorithm can better detect the key points in the image at different scales; is the input product image to be detected; k is the scale factor, which is used to adjust the distance between different layers in the scale space and affect the hierarchical structure of the pyramid; is the scale parameter, which is used to control the standard deviation of the Gaussian function and affects the degree and feature scale of Gaussian blur; is a Gaussian function used to perform Gaussian blur operation on the product image to be detected; * indicates convolution operation.

[0034] Among them, the gradient information includes horizontal gradient, vertical gradient, gradient amplitude and gradient direction, and the expression is: ; ; ; ; in, is the horizontal gradient; is the vertical gradient; For the key point; is the gradient amplitude of the pixel; is the gradient direction of the pixel; is the key point of the input product image to be detected.

[0035] The SIFT descriptor is usually a vector containing 128 elements, which is used to describe the features of the area around the key point. These elements are calculated based on the gradient information around the key point. Each element represents the contribution of the gradient information in a specific direction in the descriptor. Each feature descriptor of the query image is compared with all the descriptors in the feature library based on the Euclidean distance, which is expressed as: The descriptor is ,in, Represents the i-th element in the descriptor vector; The distance between two descriptors is expressed as: ; in, They respectively represent the descriptors of the input product images to be detected; Represents the corresponding feature descriptor in the feature library; Represents the i-th component in the input product image vector to be detected; The i-th component in the corresponding vector in the feature library; represents the weight of the i-th component.

[0036] The threshold of the distance between descriptors and the threshold of the number of successful matches are set in advance. If the distance between a pair of feature descriptors is less than the set threshold, the two feature points are matched successfully. The number of pairs of successfully matched feature points is counted. If it exceeds a certain ratio or number threshold, the two images are judged to be the same, indicating that the items placed for sale in the cabinet are consistent with the items required by the cabinet manufacturer.

[0037] At the same time, combined with the data from the posture sensor 3, the inventory level of the goods can be determined. For example, if the frequency of a certain product appearing in the image decreases, it may mean that the product is being consumed and needs to be replenished. Secondly, by analyzing the product images, it is possible to understand whether the placement of the products meets the best display effect, such as whether they are arranged in order by category or brand, whether there are any problems such as occlusion, and thus guide the adjustment of the product display. Finally, the image and posture data collected over a long period of time can be used to analyze sales trends and predict which products are more popular and which products may need to be promoted or removed from the shelves.

[0038] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any indirect modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. The unmanned vending machine system is characterized by: include: Image acquisition module (1): used to acquire images of commodities in the vending machine; An image transmission module (2): used for transmitting the image acquired by the image acquisition module (1) to the management control module (4); A posture sensor (3): used to monitor and calculate the opening and closing angle of the door of the vending machine and transmit the data to the management control module (4); Management control module (4): used to receive and analyze image data captured by the camera and door opening and closing angle information provided by the posture sensor (3) to monitor the working status of the vending machine and the display of goods; Database (5): used to store image data.

2. The unmanned vending machine system according to claim 1, characterized in that: The database (5) comprises a real-time database (51) and a historical database (52). The real-time database (51) is used to store images of commodities to be identified transmitted by the image transmission module (2); and the historical database (52) is used to store images of commodities saved in advance.

3. The unmanned vending machine system according to claim 1, characterized in that: The image acquisition module (1) is based on the MIPI dual camera. The requirements for the MIPI dual camera include two aspects: the angles of the two lenses of the MIPI dual camera must be adjustable from 15 to 30 degrees with reference to the vertical line on the left side, and from 25 to 35 degrees with reference to the horizontal line on the right side.

4. The unmanned vending machine system according to claim 1, characterized in that: The attitude sensor (3) includes an accelerometer and a gyroscope. The attitude sensor (3) calculates the door opening and closing angle of the vending machine based on a complementary filtering algorithm. The calculation steps are as follows: S1: Tilt angle obtained using the accelerometer: ； in, is the tilt angle, is the y-axis component of the accelerometer; is the z-axis component of the accelerometer; is the noise term in the accelerometer data; S2: Angle change obtained using the gyroscope: ； in, is the estimated angle; is the angular velocity data; is the sampling time interval; is the noise term in the gyroscope data; S3: Angle estimation after complementary filtering of the results of the accelerometer and gyroscope: ； in, is the angle estimation value after complementary filtering; is the filter coefficient.

5. The unmanned vending machine system according to claim 1, characterized in that: The steps of the management control module (4) monitoring the working status of the vending machine are as follows: S4: Set the angle threshold for door opening and closing; S5: judging whether the current door state is open or closed according to the calculated angle information and the set threshold; S6: Implement corresponding control logic according to the door status judgment result; S7: According to the control logic, trigger the opening or closing action of the vending machine door.

6. The unmanned vending machine system according to claim 1, characterized in that: The management control module (4) comprises an image processing module (41), an image recognition module (42) and a monitoring and analysis module (43); The image processing module (41) is used to pre-process the transmitted image; The image recognition module (42) is used to identify and judge the commodity in the image; The monitoring and analysis module (43) is used to monitor and analyze the working status of the vending machine and the display of goods.

7. The unmanned vending machine system according to claim 6, characterized in that: The image recognition module (42) recognizes and judges the input image based on the SIFT algorithm, and the steps are as follows: S8: Extract SIFT features from the image, including detecting key points in the image, calculating the gradient information of the areas around these key points, and generating a 128-dimensional SIFT descriptor for each key point; S9: For all images to be matched, SIFT features are extracted and stored, and a feature library containing SIFT features of all images is established; S10: Compare each feature descriptor of the query image with all descriptors in the feature library; S11: Finally, the matching results are visualized to show the correspondence between the query image and the matching image, thereby achieving the purpose of image recognition matching.

8. The unmanned vending machine system according to claim 7, characterized in that: In S8, key point detection is based on Gaussian difference pyramid, which finds extreme points in the image by performing Gaussian blur and difference operations on the image at different scales. The expression is: ； ； in, The key points after Gaussian blur and difference operation; is the layer of the Gaussian pyramid; is the input product image to be detected; k is the scale factor; is the scale parameter; is a Gaussian function; * indicates a convolution operation.

9. The unmanned vending machine system according to claim 7, characterized in that: In S8, the gradient information of the areas around these key points is calculated. The gradient information includes horizontal gradient, vertical gradient, gradient amplitude and gradient direction. The expression is: ； ； ； ； in, is the horizontal gradient; is the vertical gradient; For the key point; is the gradient amplitude of the pixel; is the gradient direction of the pixel; is the key point of the input product image to be detected.

10. The unmanned vending machine system according to claim 7, characterized in that: In S10, each feature descriptor of the query image is compared with all descriptors in the feature library based on the Euclidean distance, which is expressed as: The descriptor is ,in, Represents the i-th element in the descriptor vector; The distance between two descriptors is expressed as: ； in, They respectively represent the descriptors of the input product images to be detected; Represents the corresponding feature descriptor in the feature library; Represents the i-th component in the input product image vector to be detected; The i-th component in the corresponding vector in the feature library; represents the weight of the i-th component; The threshold of the distance between descriptors and the threshold of the number of successful matches are set in advance. If the distance between a pair of feature descriptors is less than the set threshold, the two feature points are successfully matched. The number of pairs of successfully matched feature points is counted. If it exceeds a certain ratio or number threshold, the two images are judged to be the same.

Images