Goods selling cabinet commodity recognition algorithm based on artificial intelligence model

Abstract

The invention relates to the technical field of vending cabinet commodity identification, in particular to a vending cabinet commodity identification algorithm based on an artificial intelligence model. Based on single-camera video data, a consumer interaction graph is constructed after continuous inter-frame feature tracking, local feature aggregation is carried out by using a graph neural network, then potential purchase intention indexes of each commodity are predicted by capturing global space-time dependence, consumer behaviors are structurally described by using the interaction graph, detailed features of each interaction are reserved, and the commodity purchase intention index is predicted. Multi-level information fusion is realized through graph convolution and Transform, and high-precision behavior modeling and purchase intention prediction are realized on the premise of only depending on single visual data, so that the commodity layout is iteratively optimized, the commodities with high purchase intention are preferentially distributed to more reasonable display positions, display configuration of the commodities in the vending cabinet is realized, and the display efficiency of the vending cabinet is improved. And the purchase experience of the user is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vending machine commodity recognition, and in particular to a vending machine commodity recognition algorithm based on an artificial intelligence model. Background Art

[0002] In modern retail environments, the display of goods in vending machines has a significant impact on consumers' purchasing decisions. How to optimize the product placement strategy to maximize sales conversion rate has become an important research direction for smart retail systems. Currently, smart retail mainly relies on technologies such as product identification, consumer behavior analysis, and inventory management, but there are still many challenges in practical applications, especially in how to accurately obtain consumers' purchase intention index and adjust the product display plan based on this.

[0003] At present, product display optimization mainly relies on empirical rules and sales data analysis. For example, sales counters usually adjust shelf display based on the historical sales data of best-selling and slow-moving products combined with marketing strategies (such as promotional activities and bundling sales). This approach has the following problems: Lag: Optimization methods based on sales data can only analyze product sales after the fact, but cannot predict consumers’ purchasing intentions before they make decisions, thus lacking the ability to make real-time adjustments. Consumer behavior factors are not taken into account: The current optimization plan is mainly based on sales data and does not fully utilize the behavioral characteristics of consumers during the shopping process (such as touch interaction) to dynamically optimize display. Summary of the invention

[0004] In view of the above-mentioned shortcomings of the prior art, the present invention provides a vending machine commodity recognition algorithm based on an artificial intelligence model, which can effectively solve the problem in the prior art that it is impossible to dynamically optimize the display of vending machine commodities in combination with consumers' behavioral characteristics.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: The present invention provides a vending machine commodity recognition algorithm based on an artificial intelligence model, comprising the following steps: Detect the product area and the consumer's hand area in the product image; The interaction between consumers and products is constructed into a graph structure to form an interaction graph, including: The intersection-and-union ratio, curvature of the hand motion trajectory, and acceleration are calculated in the candidate interaction events, where: The curvature of the hand motion trajectory is calculated using the fitting curve and its derivative; The Bayesian estimation method is introduced to combine the acceleration of the previous frame and the observed data of the speed change of the current frame to establish the prior distribution and observation probability, and realize the maximum a posteriori estimation of acceleration. Based on the Mann filtering method, the process noise covariance and the observation noise covariance are dynamically adjusted according to the prediction error to obtain the updated acceleration. Determine the comprehensive score based on the intersection-and-union ratio, the curvature of the hand motion trajectory, and the updated acceleration to determine the interaction events of the products in the image frame; Treat interaction events as nodes in the graph, use time intervals and spatial distances to construct edges, and construct a graph structure through nodes and edges to achieve the construction of an interaction graph. The interaction graph is input into the graph neural network for node feature aggregation. The potential purchase intention index of each product is output by capturing the global spatiotemporal dependency, and the purchase intention index is used to optimize the product display.

[0006] Furthermore, when the consumer's hand area is obtained, the feature points are tracked to determine the optical flow matching points, as follows: Select a set of feature points in the image frame , Indicates the first The coordinates of the feature points are , Indicates the total number of feature points selected in the image frame; Solve the motion displacement of each feature point in the next frame ; Assume that the motion of the feature point is in its local neighborhood window Keep consistent within the local neighborhood window and minimize the error: in, represents the neighborhood window centered on the feature point, , Respectively represent the feature points in and Directional displacement; By finding the optimal motion displacement, the optical flow matching points are obtained to form matching point pairs; Input feature point matching set; Calculate candidate transformation matrices; Calculate the transformation errors of all optical flow matching points, and re-determine the optical flow matching points based on the transformation errors.

[0007] Furthermore, the method for determining the curvature of the hand motion trajectory is: Assume that the sequence of hand center positions in multiple consecutive frames is ; Indicates at time The coordinates of the hand center in the image frame, Indicates the starting time, Indicates the number of consecutive frames collected; Perform robust curve fitting: in, represents the fitting curve, represents the coordinate function of the fitting curve, represents the second-order derivative of the fitted curve, represents the smoothing parameter; Calculate the curvature of the hand motion trajectory: in, , Respectively , about The derivative of , Respectively , The second derivative of .

[0008] Furthermore, the maximum a posteriori estimate of acceleration is determined as follows: Determine the prior estimate of the current frame , represents the estimated acceleration of the previous frame, represents the process noise covariance; Observation probability : in, represents the observation noise covariance, represents the time interval between frames, represents normal distribution; Compute the posterior acceleration distribution , represents the normalization constant; Determine the maximum a posteriori estimate: in, Indicates at time The acceleration value obtained by maximum a posteriori estimation at , Respectively and The inverse value of .

[0009] Furthermore, the method for obtaining the updated acceleration is: Determine the predicted acceleration for the current frame , Indicates at time The acceleration value calculated by maximum a posteriori estimation (MAP); Update the covariance: , Indicates the predicted time The acceleration estimation error covariance is Calculate Kalman gain ; Calculate updated hand acceleration , Represents the acceleration observation value obtained through observation in the current frame.

[0010] After the interaction event is generated, the method to form the interaction graph is: Each interaction event generates an interaction event node ; If the interaction event node Interaction event nodes Corresponding to the interaction events in adjacent frames, the edge is constructed , construct an interaction graph based on interaction event nodes and edges.

[0011] Interaction event nodes , which contains: Product category, product location, hand contact duration, and hand movement speed.

[0012] Furthermore, the characteristics of the edge include: Node time interval , , Represents the interaction event nodes , The moment of appearance; Spatial distance , , Represents interaction event nodes The coordinates of , Represents interaction event nodes The coordinates of .

[0013] Furthermore, the method of inputting the interaction graph into the graph neural network for node feature aggregation is: Suppose the nodes participating in feature aggregation in the interaction graph are The initial features are ; The update formula of graph convolution is: in, Indicates Layer Node Features, Representation Node The set of neighbor nodes of and Respectively represent The weight matrices and biases of the layers, Represents the activation function.

[0014] Furthermore, the method of outputting the potential purchase intention index of a product by capturing the global spatiotemporal dependency is as follows: For each node Calculate the purchase intention index: in, Indicates The purchase intention index of nodes, express The The feature vector of each node, , They represent the weight matrix and bias of the fully connected layer respectively.

[0015] Compared with the known prior art, the technical solution provided by the present invention has the following beneficial effects: Based on single-camera video data, the consumer interaction graph is constructed by tracking features between continuous frames, and then local features are aggregated using a graph neural network. Subsequently, the potential purchase intention index of each product is predicted by capturing global spatiotemporal dependencies. The interaction graph is used to structuredly describe consumer behavior, retaining detailed features of each interaction (such as product category, location, hand touch duration, and movement speed), and multi-level information fusion is achieved through graph convolution and Transformer. High-precision behavior modeling and purchase intention prediction are achieved under the premise of relying only on single visual data, so as to iteratively optimize the product layout, prioritize high purchase intention products to more reasonable display locations, realize the display configuration of sales counter products, and improve the user purchasing experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] 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.

[0017] Figure 1 It is a schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION

[0018] 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.

[0019] With the rapid development of smart retail and unmanned vending technology, automated vending machines have become an important part of the retail sector. Traditional vending machine management mainly relies on the following methods: Single visual identity: Using a single camera or multiple cameras to capture static images, a convolutional neural network (CNN) is used for target detection and recognition. This method can achieve good results under ideal lighting and unobstructed conditions, but in actual applications it often faces problems such as uneven lighting, occlusion, and limited viewing angle, resulting in a decrease in recognition rate.

[0020] Multimodal sensor fusion Some systems compensate for the shortcomings of a single sensor by fusing data from multiple sensors, such as vision, weight, and RFID. However, the core flaw of this solution is its high technical complexity and rising costs. In addition, information redundancy or noise interference may occur during the fusion process of multi-sensor data, which reduces the overall robustness and maintainability of the system.

[0021] In unmanned vending machines, product identification is not only for inventory counting, but more importantly, it is to achieve the following goals by obtaining real-time interaction data between consumers and products: Consumer Behavior Analysis: By recording every time a consumer touches and operates a product in front of a vending machine, we can conduct a detailed analysis of the consumer's purchasing path, hesitation behavior, preferences, etc., providing data support for subsequent personalized recommendations, promotional strategies, and product display optimization.

[0022] Display optimization and marketing strategy adjustment: According to consumer behavior data (such as potential purchase intention, interaction frequency, length of stay, etc.), dynamically adjust the display location of goods in the vending machine to achieve reasonable display of goods in the vending machine, ensure the normal sale of goods, and also improve the user's purchase experience (by reasonably displaying goods with high purchase intention, it is convenient for users to quickly find and purchase goods) However, in practical applications, existing technologies mainly face the following challenges: Environmental and operational complexity: Factors such as lighting, occlusion, and multi-angle viewing angles inside the vending machine will affect the accuracy of visual recognition; the diversity and continuity of consumers' hand movements during operation also pose challenges to behavior modeling.

[0023] In view of the above problems, the solution provided by the present invention is: Robust tracking of consumer hand movements improves robustness under conditions of occlusion and lighting changes; by defining interaction events and constructing a spatiotemporal interaction graph, each contact between the consumer's hand and the product is converted into a graph node, and the spatiotemporal correlation information is recorded using edges. This graph structure is intuitive and has stronger interpretability, and can effectively reflect the complex decision-making path of consumers; combined with real-time PWI data, the simulated annealing algorithm is used to guide the adjustment of product display, dynamically optimize product exposure, and simultaneously improve the user's convenient experience of purchasing products in the sales counter.

[0024] The present invention will be further described below in conjunction with the embodiments.

[0025] Example 1 (see Figure 1 ): The sales counter commodity recognition algorithm based on the artificial intelligence model includes the following steps: Step 1: Detect all product areas and consumer hand areas from the product image captured by a single camera, and perform frame-to-frame tracking to ensure the continuity and stability of the detection results, as follows: Product area detection: Get the product image at time An image frame , the bounding box and category information are determined by a pre-trained object detection model such as YOLOv5 or EfficientDet: , Indicates The bounding box (coordinates) of the products, Indicates Category labels of products (such as cola, potato chips, etc.), Represents an image frame The number of items detected in .

[0026] Consumer hand area detection: Detect the consumer's hand area on the image frame based on the target detection algorithm and obtain the hand bounding box , I will not go into details here; The Lucas-Kanade optical flow algorithm is used to track the feature points in the consumer's hand area, and abnormal matches are eliminated through RANSAC to robustly track the hand movement, ensuring that stable two-dimensional trajectory information can still be obtained when there is partial occlusion and lighting changes, providing accurate spatiotemporal position information for the subsequent construction of the interaction map, as follows: Select a set of feature points in the image frame , Indicates the first The feature points (feature points are extracted using Shi-Tomasi corner detection or Harris corner detection. Corner detection selects points with significant gradient changes in the image, such as the edges and corners of products) have coordinates of , Indicates the total number of feature points selected in the image frame; Use the Lucas-Kanade algorithm to solve the motion displacement of each feature point in the next frame , , Represents the feature points exist and Directional displacement, thereby determining whether goods have been taken away or put in; Assume that the motion of the feature point is within its local neighborhood window If the error is consistent within the neighborhood window, the Lucas-Kanade method minimizes the error within the neighborhood window: in, Represents a neighborhood window centered on the feature point, usually 3×3 or 5×5, by finding an optimal motion displacement , so that the pixel difference in the local neighborhood window is minimized, that is, the best matching position of the feature point in the next frame is found, that is, the feature point The position in the next frame is , that is, the optical flow matching point is obtained , thus forming a matching point pair .

[0027] It should be noted that the motion displacement calculated by Lucas-Kanade may contain incorrect optical flow matching points. This is because: Occlusion and noise: some feature points are occluded between two frames, resulting in invalid calculated optical flow matching points; Non-rigid deformations: If the image contains non-rigid objects (such as human bodies, fluids, etc.), then the assumed local consistency may not hold, resulting in false matches.

[0028] Lighting changes: Since the Lucas-Kanade method assumes constant luminosity (i.e., the pixel value at the same point remains unchanged), there may be brightness changes in reality, which affects the calculation of optical flow matching points; Therefore, based on RANSAC, we can eliminate erroneous optical flow matching points and ensure the accuracy of subsequent transformation matrix estimation: Input feature point matching set ; Calculate the candidate transformation matrix, including: Compute the homography matrix : ; Calculate the basis matrix : ; Calculate the transformation error of all optical flow matching points: , thereby eliminating the optical flow matching points whose transformation error exceeds the threshold, ensuring the detection accuracy.

[0029] Step 2: Construct each interaction between a consumer and a product (such as a hand approaching or touching a product) into a graph structure and record the time and space information to form an interaction graph that reflects the consumer behavior path, including: In the image frame For each candidate interaction event (a candidate where the consumer's hand area overlaps with the product area in the image frame), the following indicators are calculated: Calculate the intersection and union ratio of the two rectangular areas based on the bounding box of the product and the bounding box of the hand ; in, represents the intersection area of ​​two rectangles, represents the area of ​​the union of two rectangles; Curvature of hand motion trajectory (Describes the degree of curvature of the hand's movement trajectory, which can reflect the slight changes in consumers' movements when picking up or browsing): Assume that the sequence of hand center positions in multiple consecutive frames is ; Indicates at time The coordinates of the hand center in the image frame, Indicates the starting time, Indicates the number of consecutive frames collected; Perform robust curve fitting: in, represents the fitting curve, , Represents the coordinate function of the fitting curve, describing the hand motion trajectory obtained by fitting. It represents the second-order derivative of the fitting curve, which represents the acceleration information of the curve, that is, the change in the curvature of the curve, and is used to measure the smoothness of the curve. represents the smoothing parameter; The local curvature is approximated using the second-order derivative: in, , Respectively , about The derivative of the curve reflects the time The instantaneous velocity component of , Respectively , The second derivative of describes the acceleration or bending change of the curve. It represents the curvature of the hand motion trajectory. It should be noted that the curvature calculation uses the analytical expression of the fitting curve and its derivative to avoid the influence of noise in discrete difference, so as to obtain a more stable and continuous curvature of the hand motion trajectory to accurately describe the geometric characteristics of the hand motion trajectory.

[0030] Hand acceleration (Reflects changes in hand movement speed, helping to distinguish quick glances from actual picking behavior):

[0031] in, Indicates the hand movement speed of the image frame (current frame).

[0032] It should be noted that, considering that the calculation of hand acceleration only depends on the difference between adjacent frames, it is easily affected by noise, motion blur and changes in dynamic interaction patterns, resulting in unstable acceleration. Therefore, the Bayesian estimation method is introduced to establish the prior distribution and observation probability by combining the acceleration of the previous frame and the observation data of the speed change of the current frame, so as to achieve the maximum a posteriori estimation of the acceleration; at the same time, in order to further smooth and optimize this estimation, the adaptive Kalman filtering method is used to dynamically adjust the process noise covariance and the observation noise covariance according to the prediction error to adaptively update the acceleration. The estimates are as follows: In time The estimated acceleration is used as a priori estimate of the current frame , represents the estimated acceleration of the previous frame, represents the process noise covariance, describing the uncertainty of acceleration estimation; Observation probability , which means based on the current speed The inferred acceleration distribution: in, represents the observation noise covariance, describing the error in velocity measurement, represents the time interval between frames, represents normal distribution; Combining prior distribution and observation probability to calculate posterior acceleration distribution , represents the normalization constant, that is, the probability of observing the current velocity under all possible acceleration values, ensuring that the integral of the posterior acceleration distribution is 1; The solution of the maximum a posteriori estimate is: in, Indicates at time The acceleration value obtained by maximum a posteriori estimation when , that is, the optimal acceleration estimate obtained after combining the prior distribution and the observation probability, , Respectively and The inverse value of Predicted acceleration for the current frame , Indicates at time The acceleration value calculated by maximum a posteriori estimation (MAP); Update the covariance: , Indicates the predicted time The acceleration estimation error covariance is Calculate Kalman gain , Determines the weighted ratio of predicted values ​​to observed values ​​in the update phase, reflecting the relative relationship between prediction uncertainty and observation noise; Calculate updated hand acceleration , Represents the acceleration observation value obtained through observation in the current frame.

[0033] Through the above, the smoothness and accuracy of acceleration calculation can be maintained when facing environmental noise, motion blur and different interaction modes (such as grabbing, pushing or sliding), providing more stable and reliable basic data for subsequent consumer behavior analysis and product display optimization.

[0034] Furthermore, , , Perform weighted sum calculation to get a comprehensive score ,like If the threshold is exceeded, it is determined that an interaction event of the product has occurred in the image frame; Based on the intersection-over-union ratio (which measures the degree of overlap between the consumer's hand area and the product area), the curvature of the hand motion trajectory, and the acceleration characteristics, it captures more subtle behavioral changes of consumers during the interaction process, such as: When consumers scan products quickly, although there may be some overlap, the curvature and acceleration of their motion trajectories vary greatly, and the overall score will be lower; When consumers pick up or browse carefully, the overlap is higher and the movement is smoother, resulting in a higher score, which can more accurately reflect whether an interactive event occurs with the product in the image frame.

[0035] Take each interaction event as a node in the graph, use the time interval and spatial distance to construct the edge, and form a graph structure through nodes and edges to form an interaction graph. Then we have: Each interaction event is generated as an interaction event node , represents an interaction event, which includes: Product category, product location, hand touch duration (interaction duration within multiple consecutive frames), and hand movement speed; If the interaction event node Interaction event nodes Corresponding to the interaction events in adjacent frames, the edge is constructed , the characteristics of the edge include: Node Time Interval , , Represents interaction event nodes , The moment of appearance; Spatial distance , , Represents interaction event nodes The coordinates of , Represents interaction event nodes It should be noted that the interaction diagram structures the multiple interactive behaviors of consumers in front of the vending machine, which not only retains the detailed characteristics of each interactive event, but also records the spatiotemporal relationship between the behaviors.

[0036] Furthermore, after obtaining the interaction graph as described above, the interaction graph can be input into the graph neural network for node feature aggregation, and then combined with the Transformer module to capture the global spatiotemporal dependency, and finally output the potential purchase intention index PWI of each product. The index is used to guide the optimization of product display and realize the management of sales cabinet products, including: For each node in the interaction graph, graph convolution is used to update the features and aggregate local information, then: Suppose the nodes participating in feature aggregation in the interaction graph are The initial features are ; The update formula of graph convolution is: in, Indicates Layer Node Features, Representation Node The set of neighbor nodes of and Respectively represent The weight matrices and biases of the layers, Represents an activation function, such as Relu; It is worth mentioning that the graph convolution operation can aggregate the information of neighboring nodes, so that the features of each node not only contain its own information, but also integrate the context of its adjacent interaction events for subsequent product management.

[0037] The node feature sequence updated by graph convolution is input into the Transformer module, and the long-distance dependencies and temporal dynamics between nodes are captured through the self-attention mechanism, then: Arrange node features into sequences , Represents the number of all nodes in the interaction graph, that is, all interaction events; Self-attention calculation: in, Form a The node feature matrix of is the dimension of each node feature, Represents a healthy vector The dimension of The length of each column vector of , , , Represent the learnable weight matrices, with dimensions of ; Calculate the similarity matrix : Calculate all query vectors With all key vectors The similarity score between them, that is, the attention score matrix, is used to obtain the fusion node features of the entire graph , Represents the Transformer module, which processes node features through the self-attention mechanism to capture long-distance dependencies and global timing information. After Transformer processing, the features of each node are integrated with the global context information, allowing the model to analyze interactive behaviors more accurately.

[0038] Therefore, the fully connected layer is used to calculate the purchase intention index of each product: For each node (Represents a product interaction) Calculate the purchase intention index:

[0039] in, Indicates The purchase intention index of nodes, express The The feature vector of each node, , Represent the weight matrix and bias of the fully connected layer respectively, represents the activation function, and the sigmoid function can be used to normalize the output value to the range of [0, 1]. It should be noted that the purchase intention index reflects the consumer's potential purchase intention for a certain product. A higher A value of indicates that the consumer behavior data shows a high level of attention and purchase intention for the product.

[0040] Finally, based on the calculation The geometric simulated annealing algorithm determines the optimal display scheme for goods in the vending cabinet, that is, the optimal solution guides the display and recommendation strategy of goods in the vending cabinet, including: Define the merchandise collection for the vending machine , which represents all the goods to be displayed in the vending cabinet, Indicates the total number of products; Define a collection of display locations , which represents the fixed display positions in the sales counter (e.g. different heights, left and right positions on the counter), and the number of positions is the same as the number of goods; Defining layout solutions : Layout solution is a permutation vector or mapping representing each product Assigned to which position , you can use Indicates that Display products Placed in position ; Constructing the objective function , where the goal is to maximize the overall effect; in, , represents a collection of products with higher PWI values, represents the threshold value, , Respectively represent the corresponding weight coefficients to balance the impact of the two parts of the score, Display products In Location The exposure score is usually related to the prominence of the location. For example, the upper middle area of ​​the vending machine has a higher score. Display products The strategy score reflects factors such as inventory and profit; By simulating the physical annealing process to gradually reduce the system temperature, accept some inferior solutions to jump out of the local optimum, and thus approach the global optimal solution, we have: Initial layout solution ; Set a higher temperature ; Defining the Temperature Attenuation Coefficient :After each iteration, the temperature is updated to , the current temperature Multiplying by the attenuation coefficient to get a new temperature value indicates that the temperature decreases exponentially, thereby gradually reducing the probability of accepting an inferior solution, and eventually the algorithm converges to a solution that approximates the global optimal solution; Iterations : The number of iterations at each temperature; Shutdown criteria: Current temperature drops to a certain low threshold or reaches the maximum number of iterations; Solve the current layout Make a slight perturbation to generate a new solution. For example, randomly select two items and exchange their positions. and Get a new neighborhood solution ; Solve the current layout and neighborhood solution Calculate the objective function value and ; , and similarly we get ; Calculate the energy difference ; like Less than or equal to 0, it means accepting the neighborhood solution ; like If it is greater than 0, it means that the neighborhood solution is accepted with probability , probability ; Generate random numbers , range 0-1, if , then accept , otherwise keep the original solution, so through PW calculation, simulated annealing algorithm finally finds an optimal placement solution, making high The product gets the best exposure, according to Reasonable display and distribution of goods will help subsequent users quickly find and purchase goods in the sales cabinet, thereby improving the user's purchasing experience.

[0041] 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. The vending machine commodity recognition algorithm based on artificial intelligence model is characterized by: The steps include: Detect the product area and the consumer's hand area in the product image; The graph structure is constructed based on the interaction between consumers and products to form an interaction graph, including: The intersection-and-union ratio, curvature of the hand motion trajectory, and acceleration are calculated in the candidate interaction events, where: The curvature of the hand motion trajectory is calculated using the fitting curve and its derivative; The Bayesian estimation method is introduced to combine the acceleration of the previous frame and the observed data of the speed change of the current frame to establish the prior distribution and observation probability, and realize the maximum a posteriori estimation of acceleration. Based on the Mann filtering method, the process noise covariance and the observation noise covariance are dynamically adjusted according to the prediction error to obtain the updated acceleration. Determine the comprehensive score based on the intersection-and-union ratio, the curvature of the hand motion trajectory, and the updated acceleration to determine the interaction events of the products in the image frame; Treat interaction events as nodes in the graph, use time intervals and spatial distances to construct edges, and construct a graph structure through nodes and edges to achieve the construction of an interaction graph. The interaction graph is input into the graph neural network for node feature aggregation. The potential purchase intention index of each product is output by capturing the global spatiotemporal dependency, and the purchase intention index is used to optimize the product display.

2. The vending machine commodity recognition algorithm based on artificial intelligence model according to claim 1 is characterized in that: When the consumer's hand area is obtained, the feature points are tracked to determine the optical flow matching points, as follows: Select a set of feature points in the image frame , Indicates the first The coordinates of the feature points are , Indicates the total number of feature points selected in the image frame; Solve the motion displacement of each feature point in the next frame ; Assume that the motion of the feature point is in its local neighborhood window Keep consistent within the local neighborhood window and minimize the error: in, represents the neighborhood window centered on the feature point, , Respectively represent the feature points in and Directional displacement; By finding the optimal motion displacement, the optical flow matching points are obtained to form matching point pairs; Input feature point matching set; Calculate candidate transformation matrices; Calculate the transformation errors of all optical flow matching points, and re-determine the optical flow matching points based on the transformation errors.

3. The vending machine commodity identification algorithm based on artificial intelligence model according to claim 2 is characterized in that: The method for determining the curvature of the hand motion trajectory is: Assume that the sequence of hand center positions in multiple consecutive frames is ; Indicates at time The coordinates of the hand center in the image frame, Indicates the starting time, Indicates the number of consecutive frames collected; Perform robust curve fitting: in, represents the fitting curve, represents the coordinate function of the fitting curve, represents the second-order derivative of the fitted curve, represents the smoothing parameter; Calculate the curvature of the hand motion trajectory: in, , Respectively , about The derivative of , Respectively , The second derivative of .

4. The vending machine commodity identification algorithm based on artificial intelligence model according to claim 3 is characterized in that: The maximum a posteriori estimate of the acceleration is determined by: Determine the prior estimate of the current frame , represents the estimated acceleration of the previous frame, represents the process noise covariance; Observation probability : in, represents the observation noise covariance, represents the time interval between frames, represents normal distribution; Compute the posterior acceleration distribution , represents the normalization constant; Determine the maximum a posteriori estimate: in, Indicates at time The acceleration value obtained by maximum a posteriori estimation at , Respectively and The inverse value of .

5. The vending machine commodity identification algorithm based on artificial intelligence model according to claim 4 is characterized in that: The method to get the updated acceleration is: Determine the predicted acceleration for the current frame , Indicates at time The acceleration value calculated by maximum a posteriori estimation (MAP); Update the covariance: , Indicates the predicted time The acceleration estimation error covariance is Calculate Kalman gain ; Calculate updated hand acceleration , Represents the acceleration observation value obtained through observation in the current frame.

6. The vending machine commodity identification algorithm based on artificial intelligence model according to claim 5 is characterized in that: After the interaction event is generated, the method of forming the interaction graph is: Each interaction event generates an interaction event node ; If the interaction event node Interaction event nodes Corresponding to the interaction events in adjacent frames, the edge is constructed , construct an interaction graph based on interaction event nodes and edges.

7. The vending machine commodity identification algorithm based on artificial intelligence model according to claim 6 is characterized in that: The interaction event node , which contains: Product category, product location, hand contact duration, and hand movement speed.

8. The vending machine commodity identification algorithm based on artificial intelligence model according to claim 6 is characterized in that: The characteristics of the edge include: Node Time Interval , , Represents interaction event nodes , The moment of appearance; Spatial distance , , Represents interaction event nodes The coordinates of , Represents interaction event nodes The coordinates of .

9. The vending machine commodity identification algorithm based on artificial intelligence model according to claim 1 is characterized in that: The method of inputting the interaction graph into the graph neural network for node feature aggregation is: Suppose the nodes participating in feature aggregation in the interaction graph are The initial features are ; The update formula of graph convolution is: in, Indicates Layer Node Features, Representation Node The set of neighbor nodes of and Respectively represent The weight matrices and biases of the layers, Represents the activation function.

10. The vending machine commodity identification algorithm based on artificial intelligence model according to claim 9 is characterized in that: The method of capturing the global spatiotemporal dependency and outputting the potential purchase intention index of a product is as follows: For each node Calculate the purchase intention index: in, Indicates The purchase intention index of nodes, express The The feature vector of a node, , They represent the weight matrix and bias of the fully connected layer respectively.

Images