Intelligent retail shelf monitoring system based on artificial intelligence image recognition

Abstract

The invention relates to the technical field of artificial intelligence and machine learning, in particular to an intelligent retail shelf monitoring system based on artificial intelligence image recognition. Comprising a data acquisition unit, an image recognition unit, a data processing and analysis unit, a feedback and decision support unit and a communication unit, wherein the data acquisition unit is used for acquiring state information of commodities on a shelf through a camera and a sensor which are mounted on the shelf. Multi-modal information is fused through the image recognition unit, a lightweight model is developed, a meta-learning mechanism is introduced, the recognition efficiency and precision are improved, the computing resource requirement is reduced, the model adaptability is enhanced, prediction, self-adaptive adjustment and customer subdivision of the future sales trend are generated by using a time sequence analysis model and other external factors, and the market competitiveness is improved. The sales data value is mined, the trend is accurately predicted, the inventory is dynamically optimized, retailers are helped to better plan purchasing plans and sales promotion activities, and the market competitiveness is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and machine learning, and in particular to an intelligent retail shelf monitoring system using artificial intelligence image recognition. Background Art

[0002] The intelligent retail shelf monitoring system with artificial intelligence image recognition is a system that uses computer vision technology, machine learning algorithms, and the Internet of Things to conduct real-time monitoring and data analysis of retail shelves. Shelf images are collected by cameras and other devices installed on or around the shelves, and then image recognition algorithms are used to identify and analyze the products in the images, thereby realizing functions such as inventory management, display compliance testing, and customer behavior analysis. As competition in the retail industry becomes increasingly fierce, companies' demand for refined management is constantly increasing. Accurately understanding the display, inventory, and out-of-stock conditions of shelf products is crucial to improving sales efficiency, optimizing inventory management, and enhancing the customer shopping experience. In existing technologies, traditional manual inspections of retail shelves are inefficient and have large errors. They cannot meet the refined management needs of modern retail, and it is difficult for retail companies to efficiently optimize inventory, conduct precision marketing, and improve their competitiveness.

[0003] Based on this, the present invention provides an intelligent retail shelf monitoring system based on artificial intelligence image recognition to solve the technical problems raised above. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent retail shelf monitoring system based on artificial intelligence image recognition. The present invention integrates multimodal information through image recognition units, develops lightweight models and introduces meta-learning mechanisms to improve recognition efficiency and accuracy, reduce computing resource requirements and enhance model adaptability. It also generates forecasts of future sales trends by utilizing time series analysis models and other external factors, adaptively adjusts and segments customers, mines the value of sales data, accurately predicts trends, and dynamically optimizes inventory, helping retailers better plan procurement plans and promotional activities and enhance market competitiveness.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The present invention provides an intelligent retail shelf monitoring system with artificial intelligence image recognition, comprising a data acquisition unit, an image recognition unit, a data processing and analysis unit, a feedback and decision support unit, and a communication unit, wherein:

[0007] The data acquisition unit is used to collect status information of goods on the shelf through cameras and sensors installed on the shelf;

[0008] The image recognition unit is used to achieve efficient and accurate recognition of products and their status by fusing multimodal information, developing lightweight models, and introducing meta-learning mechanisms;

[0009] The data processing and analysis unit is responsible for processing the data obtained from the image recognition unit and generating multi-dimensional analysis of inventory levels, sales trends, and customer behavior by integrating predictive analysis models, adaptive mechanisms, and customer segmentation technologies;

[0010] The feedback and decision support unit is used to provide merchants and staff with a visual operation interface for displaying the analysis results of the monitoring system;

[0011] The communication unit is used to realize data transmission and communication between various units of the system and between the system and external devices or systems.

[0012] The status information includes the product placement location, inventory quantity, appearance status, and display status.

[0013] The data acquisition unit includes a camera module, a sensor module, and a data preprocessing module, wherein:

[0014] The camera module is used to continuously capture images of goods on the shelf with cameras at different angles to provide visual information input;

[0015] The sensor module is used to collect status information on the quantity, location, and status of goods through various sensors;

[0016] The data preprocessing module is used to preprocess the collected images and sensor data.

[0017] The image recognition unit includes a multimodal information fusion module, a lightweight model development module, and a meta-learning mechanism module, wherein:

[0018] The multimodal information fusion module is used to fuse the image information obtained by the camera module with the commodity weight and image information obtained by the sensor module;

[0019] The lightweight model development module is used to optimize the traditional convolutional neural network model by using model compression and pruning technology to develop a lightweight convolutional neural network model;

[0020] The meta-learning mechanism module is used to introduce a meta-learning mechanism to enable the model to quickly adapt to the recognition of new products.

[0021] The lightweight model development module uses model compression and pruning technology to optimize the traditional convolutional neural network model and develop a lightweight convolutional neural network model. The specific operations are as follows:

[0022] A1: Evaluate the key performance indicators of existing models to determine the optimization baseline;

[0023] A2: Identify and remove unimportant weights or neurons, and apply appropriate pruning strategies to reduce model redundancy;

[0024] A3: Convert model parameters into low-precision numerical representations, using quantization to reduce storage requirements and accelerate inference.

[0025] A4: Use a larger and more complex teacher model to guide the student model's learning and design an appropriate loss function to improve the student model's performance;

[0026] A5: Fine-tune the lightweight model to recover the performance degradation caused by pruning and quantization and improve task adaptability.

[0027] A6: Fully verify and test the optimized lightweight model to ensure it meets the expected performance standards and resource consumption requirements on the target platform.

[0028] The data processing and analysis unit includes a data cleaning and integration module, a prediction analysis model module, an adaptive mechanism module, and a customer segmentation module, wherein:

[0029] The data cleaning and integration module is used to clean the data obtained from the image recognition unit, remove noise, duplication and erroneous data, and integrate data of different formats and sources to unify the data format and standard;

[0030] The forecast analysis model module generates forecasts of future sales trends based on the time series analysis model and historical sales data and other external factors;

[0031] The adaptive mechanism module is used to automatically adjust the analysis strategy according to market dynamics and update inventory management and replenishment suggestions in real time;

[0032] The customer segmentation module is used to divide different customer groups and identify their preferences through in-depth analysis of customer purchasing behavior, thereby supporting personalized marketing and services.

[0033] The forecast analysis model module generates a forecast of future sales trends based on the time series analysis model and historical sales data and other external factors. The specific operations are as follows:

[0034] B1: Smoothing and outlier correction of historical sales data;

[0035] B2: Use the logistic growth model to decompose the basic trend items:

[0036]

[0037] In the formula, C is the market carrying capacity; k is the growth rate parameter; m is the trend turning point; ∈ t is the random error term;

[0038] B3: Fitting the periodicity through Fourier series, the seasonal term is modeled as:

[0039]

[0040] Where p is the sales cycle day / week / year, a n 、b n are the Fourier series coefficients;

[0041] B4: Weighted integration of promotional activities, weather and other factors:

[0042] βX t =β1x1,t+β2x2,t

[0043] Where x1,t is the intensity of promotion activities; x2,t is the weather impact index;

[0044] B5: Output the final predicted value:

[0045]

[0046] B6: Retrain model parameters every 28 days sliding window.

[0047] The customer segmentation module divides different customer groups and identifies their preferences through in-depth analysis of customer purchasing behavior, supporting personalized marketing and services. The specific operations are as follows:

[0048] C1: Extract shopping trajectory features, including product picking frequency, shelf dwell time, and purchase combination correlation;

[0049] C2: Clustering is performed using the DBSCAN algorithm. The distance measurement formula is:

[0050] d(x,y) = α·purchase interval + β·category familiarity

[0051] Where x, y are the behavioral feature vectors of two customers; α, β are trainable weight coefficients, α = 0.6, β = 0.4;

[0052] C3: Generate preference labels for each group: When the purchase rate of discounted products is greater than 60%, it is price-sensitive; when the purchase rate within 7 days of listing is greater than 40%, it is new product trial;

[0053] C4: Associate customer segments with product display recommendations, including recommendations for exclusive shelf locations for high-value customers and dynamic pricing threshold adjustments.

[0054] The feedback and decision support unit includes a visualization display module, a decision suggestion generation module, and an interactive operation module, wherein:

[0055] The visualization display module is used to display the analysis results obtained by the data processing and analysis unit on the visualization operation interface in the form of intuitive charts, reports, maps, etc.;

[0056] The decision suggestion generation module is used to provide specific decision suggestions to merchants and staff based on the data analysis results and using preset rules and algorithms;

[0057] The interactive operation module is used to support merchants and staff to interact with the visual operation interface.

[0058] The communication unit includes an internal communication module and an external communication module, wherein:

[0059] The internal communication module is used to ensure data transmission and communication between various units in the system;

[0060] The external communication module is responsible for data transmission and communication between the system and external devices or systems.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] The present invention integrates multimodal information through image recognition units, develops lightweight models and introduces meta-learning mechanisms to improve recognition efficiency and accuracy, reduce computing resource requirements and enhance model adaptability. It also generates forecasts of future sales trends by utilizing time series analysis models and other external factors, adaptively adjusts and segments customers, mines the value of sales data, accurately predicts trends, and dynamically optimizes inventory, helping retailers better plan procurement plans and promotional activities and enhance market competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a system diagram of an intelligent retail shelf monitoring system with artificial intelligence image recognition according to the present invention.

[0064] Figure 2 This is a diagram of the internal structure of the image recognition unit in an intelligent retail shelf monitoring system with artificial intelligence image recognition according to the present invention.

[0065] Description of Figure Numbers:

[0066] 100. Data acquisition unit; 101. Camera module; 102. Sensor module; 103. Data preprocessing module; 200. Image recognition unit; 201. Multimodal information fusion module; 202. Lightweight model development module; 203. Meta-learning mechanism module; 300. Data processing and analysis unit; 301. Data cleaning and integration module; 302. Predictive analysis model module; 303. Adaptive mechanism module; 304. Customer segmentation module; 400. Feedback and decision support unit; 401. Visualization display module; 402. Decision suggestion generation module; 403. Interactive operation module; 500. Communication unit; 501. Internal communication module; 502. External communication module. DETAILED DESCRIPTION

[0067] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only 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 making any creative efforts shall fall within the scope of protection of the present invention.

[0068] Example:

[0069] like Figure 1-Figure 2 As shown, this embodiment provides an intelligent retail shelf monitoring system using artificial intelligence image recognition, comprising a data acquisition unit 100, an image recognition unit 200, a data processing and analysis unit 300, a feedback and decision support unit 400, and a communication unit 500. The data acquisition unit 100 is configured to collect status information of goods on the shelf using cameras and sensors installed on the shelf. This status information includes the product placement, inventory quantity, appearance, and display status. The image recognition unit 200 is configured to achieve efficient and accurate identification of goods and their status by fusing multimodal information, developing lightweight models, and introducing meta-learning mechanisms. The data processing and analysis unit 300 is responsible for processing the data obtained from the image recognition unit 200 and generating multi-dimensional analysis of inventory levels, sales trends, and customer behavior by integrating predictive analysis models, adaptive mechanisms, and customer segmentation technologies. The feedback and decision support unit 400 is configured to provide merchants and staff with a visual operating interface for displaying the analysis results of the monitoring system. The communication unit 500 is configured to facilitate data transmission and communication between various units of the system and between the system and external devices or systems.

[0070] In this embodiment, it should be noted that the data acquisition unit 100 collects status information including the location of the goods, inventory quantity, appearance status, and display status through cameras and sensors installed on the shelves, and transmits it to the image recognition unit 200. The unit integrates multimodal information, develops lightweight models, and introduces meta-learning mechanisms to efficiently and accurately identify the goods and their status, and transmits the identified data to the data processing and analysis unit 300. It generates multi-dimensional analysis results on inventory levels, sales trends, and customer behaviors by integrating predictive analysis models, adaptive mechanisms, and customer segmentation technologies, and then displays them to merchants and staff through a visual operation interface by the feedback and decision support unit 400. During the whole process, the communication unit 500 realizes data transmission and communication between the various units of the system and between the system and external devices or systems.

[0071] In the present invention, the data acquisition unit 100 includes a camera module 101, a sensor module 102, and a data preprocessing module 103, wherein: the camera module 101 is used to continuously capture images of goods on the shelf with cameras at different angles to provide visual information input; the sensor module 102 is used to collect status information such as the quantity, location, and status of the goods through various sensors; and the data preprocessing module 103 is used to preprocess the collected images and sensor data.

[0072] In this embodiment, it should be noted that the camera module 101 continuously captures product images from different angles to provide visual information, and the sensor module 102 collects information on the quantity, location, and status of the products through various sensors. The data collected by the two are uniformly transmitted to the data preprocessing module 103, which preprocesses the image and sensor data.

[0073] Furthermore, it should be noted that sensors include weight sensors and RFID readers. Preprocessing includes noise reduction, grayscale conversion, and normalization to remove noise interference and adjust image brightness and contrast. Sensor data is filtered and calibrated to ensure accuracy and reliability, preparing for subsequent analysis. Visual information includes the appearance, location, and arrangement of products.

[0074] In the present invention, the image recognition unit 200 includes a multimodal information fusion module 201, a lightweight model development module 202, and a meta-learning mechanism module 203, wherein: the multimodal information fusion module 201 is used to fuse the image information obtained by the camera module 101 with the information of the product weight and image obtained by the sensor module 102; the lightweight model development module 202 is used to optimize the traditional convolutional neural network model by using model compression and pruning technology to develop a lightweight convolutional neural network model; the specific operations are as follows: A1: Evaluate the key performance indicators of the existing model to determine the optimization basis A2: Identify and remove unimportant weights or neurons, and apply appropriate pruning strategies to reduce model redundancy. A3: Convert model parameters to low-precision numerical representations, using quantization methods to reduce storage requirements and accelerate inference. A4: Utilize larger and more complex teacher models to guide student model learning, and design appropriate loss functions to improve student model performance. A5: Fine-tune the lightweight model to recover performance degradation caused by pruning and quantization, and improve task adaptability. A6: Thoroughly verify and test the optimized lightweight model to ensure it meets expected performance standards and resource consumption requirements on the target platform. Meta-learning mechanism module 203 is used to introduce a meta-learning mechanism, enabling the model to quickly adapt to the recognition of new products.

[0075] In this embodiment, it should be noted that the multimodal information fusion module 201 integrates the information obtained by the camera module 101 and the sensor module 102 to provide richer data for image recognition; the lightweight model development module 202 optimizes the traditional convolutional neural network model through a series of operations such as performance evaluation, pruning, quantization, knowledge distillation, fine-tuning, and verification testing to reduce computing resource requirements; the meta-learning mechanism module 203 introduces a meta-learning mechanism to enable the model to quickly adapt to new product recognition.

[0076] Furthermore, it should be noted that the meta-learning mechanism module 203 adopts a model-agnostic meta-learning algorithm, and the specific formula is: MAML finds a universal initial parameter θ in multiple task training. Update parameters, where α is the learning rate, It is task T i The gradient of the loss function with respect to the parameter θ is obtained by Optimize the initial parameters so that the model can adapt quickly when identifying new products with a small amount of data gradient updates.

[0077] The prototype network calculates the category prototype by Where d is the distance metric function, and the distance between the query sample and the prototype is calculated to complete the classification.

[0078] In the present invention, the data processing and analysis unit 300 includes a data cleaning and integration module 301, a prediction analysis model module 302, an adaptive mechanism module 303, and a customer segmentation module 304. The data cleaning and integration module 301 is used to clean the data obtained from the image recognition unit 200, remove noise, duplication, and erroneous data, and integrate data of different formats and sources to unify the data format and standard. The prediction analysis model module 302 generates a forecast of future sales trends based on a time series analysis model and historical sales data and other external factors. The specific operations are as follows:

[0079] B1: Smoothing and outlier correction of historical sales data;

[0080] B2: Use the logistic growth model to decompose the basic trend items:

[0081]

[0082] In the formula, C is the market carrying capacity; k is the growth rate parameter; m is the trend turning point; ∈ t is the random error term;

[0083] B3: Fitting the periodicity through Fourier series, the seasonal term is modeled as:

[0084]

[0085] Where p is the sales cycle day / week / year, a n 、b n are the Fourier series coefficients;

[0086] B4: Weighted integration of promotional activities, weather and other factors:

[0087] βX t =β1x1,t+β2x2,t

[0088] Where x1,t is the intensity of promotion activities; x2,t is the weather impact index;

[0089] B5: Output the final predicted value:

[0090]

[0091] B6: Retrain model parameters using a sliding window every 28 days. The adaptive mechanism module 303 is used to automatically adjust analysis strategies based on market dynamics, updating inventory management and replenishment recommendations in real time. The customer segmentation module 304 is used to analyze customer purchasing behavior in depth, identify different customer groups, and support personalized marketing and services. The specific operations are as follows: C1: Extract shopping trajectory features, including product pickup frequency, shelf dwell time, and purchase combination correlation; C2: Use the DBSCAN algorithm for clustering, and the distance metric formula is:

[0092] d(x,y) = α·purchase interval + β·category familiarity

[0093] Where x and y are the behavioral feature vectors of two customers; α and β are trainable weight coefficients, α = 0.6 and β = 0.4; C3: Generates preference labels for each group: when the purchase rate of discounted products is greater than 60%, it is price-sensitive; when the purchase rate within 7 days of listing is greater than 40%, it is new product trial; C4: Associates customer grouping with product display suggestions, including recommending exclusive shelf locations for high-value customers and adjusting dynamic pricing thresholds.

[0094] In this embodiment, it should be noted that the data cleaning and integration module 301 first purifies and unifies the format of the data obtained by the image recognition unit 200 to provide a high-quality data foundation for the prediction analysis model module 302, the adaptive mechanism module 303 and the customer segmentation module 304; the prediction analysis model module 302 predicts sales trends based on models such as time series analysis, and the results provide a basis for the adaptive mechanism module 303 to adjust inventory management and replenishment recommendations. At the same time, the customer segmentation module 304 divides customers into groups and identifies preferences by analyzing customer behavior.

[0095] Furthermore, it should be noted that web crawlers, API interfaces, and other technologies are used to obtain market dynamics data in real time. The specific operations of the adaptive mechanism module 303 are: ① Real-time monitoring of six market dynamic indicators, including promotion intensity, competitive product prices, social media sentiment, weather warning levels, real-time customer flow, and historical sales deviations over the same period, and triggering analysis through preset thresholds; ② Using a reinforcement learning framework, the Q-learning algorithm dynamically optimizes each signal weight wi and model parameter θ every hour. The specific formula for the model parameters is: Where w i is the real-time weight of the i-th type signal; R i is the expected equity of the corresponding strategy; the specific formula of the weight is: ③ Calculate the replenishment threshold based on a dynamic formula. Replenishment threshold calculation: In the formula, the inventory pressure coefficient α = 0.3, the demand trend coefficient β = 0.2, and the emergency replenishment trigger rule with dual conditions is set. And; I current<0.4I max ④ Conduct A / B testing every 4 hours and use statistical significance tests (p<0.05) to determine the effectiveness of the strategy; ⑤ Automatically roll back to the safe strategy when a sudden drop in sales, a sudden drop in turnover rate, or resource overload is detected; Safe strategy: sales drop >20% within 2 hours of implementation; sudden drop in inventory turnover rate >30%; system resource utilization rate continuously >90%.

[0096] In the present invention, the feedback and decision support unit 400 includes a visualization display module 401, a decision suggestion generation module 402, and an interactive operation module 403, wherein: the visualization display module 401 is used to display the analysis results obtained by the data processing and analysis unit 300 in the form of intuitive charts, reports, maps, etc. on the visualization operation interface; the decision suggestion generation module 402 is used to provide specific decision suggestions to merchants and staff based on the data analysis results, using preset rules and algorithms; the interactive operation module 403 is used to support merchants and staff to interact with the visualization operation interface.

[0097] In this embodiment, it should be noted that the visualization display module 401 presents the analysis results of the data processing and analysis unit 300 in an intuitive form such as charts and reports on the operation interface, the decision recommendation generation module 402 uses rule algorithms based on these results to output specific decision recommendations, and the interactive operation module 403 supports user interaction with the visualization interface.

[0098] In the present invention, the communication unit 500 includes an internal communication module 501 and an external communication module 502, wherein: the internal communication module 501 is used to ensure data transmission and communication between various units in the system; the external communication module 502 is responsible for data transmission and communication between the system and external devices or systems.

[0099] In this embodiment, it should be noted that the internal communication module 501 ensures smooth data transmission and communication between the various units in the intelligent retail shelf monitoring system, such as the data acquisition unit 100, the image recognition unit 200, the data processing and analysis unit 300, and the feedback and decision support unit 400, while the external communication module 502 is responsible for data interaction between the system and external devices or systems.

[0100] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0101] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent retail shelf monitoring system based on artificial intelligence image recognition, characterized in that: The system comprises a data acquisition unit (100), an image recognition unit (200), a data processing and analysis unit (300), a feedback and decision support unit (400), and a communication unit (500), wherein: The data collection unit (100) is used to collect status information of commodities on the shelf through cameras and sensors installed on the shelf; The image recognition unit (200) is used to achieve efficient and accurate recognition of goods and their status by fusing multimodal information, developing lightweight models and introducing meta-learning mechanisms; The data processing and analysis unit (300) is responsible for processing the data obtained from the image recognition unit (200) and generating a multi-dimensional analysis of inventory levels, sales trends, and customer behavior by integrating a predictive analysis model, an adaptive mechanism, and customer segmentation technology; The feedback and decision support unit (400) is used to provide merchants and staff with a visual operation interface for displaying the analysis results of the monitoring system; The communication unit (500) is used to realize data transmission and communication between various units of the system and between the system and external equipment or systems.

2. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 1, characterized in that: The status information includes the product placement location, inventory quantity, appearance status, and display status.

3. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 2, characterized in that: The data acquisition unit (100) comprises a camera module (101), a sensor module (102), and a data pre-processing module (103), wherein: The camera module (101) is used to continuously capture images of commodities on the shelf with cameras at different angles to provide visual information input; The sensor module (102) is used to collect status information on the quantity, location, and status of commodities through various sensors; The data preprocessing module (103) is used to preprocess the collected images and sensor data.

4. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 1, characterized in that: The image recognition unit (200) includes a multimodal information fusion module (201), a lightweight model development module (202), and a meta-learning mechanism module (203), wherein: The multimodal information fusion module (201) is used to fuse the image information acquired by the camera module (101) with the commodity weight and image information acquired by the sensor module (102); The lightweight model development module (202) is used to optimize the traditional convolutional neural network model by using model compression and pruning technology to develop a lightweight convolutional neural network model; The meta-learning mechanism module (203) is used to introduce a meta-learning mechanism so that the model can quickly adapt to the recognition of new products.

5. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 4 is characterized in that: The lightweight model development module (202) uses model compression and pruning technology to optimize the traditional convolutional neural network model and develop a lightweight convolutional neural network model. The specific operations are as follows: A1: Evaluate the key performance indicators of existing models to determine the optimization baseline; A2: Identify and remove unimportant weights or neurons, and apply appropriate pruning strategies to reduce model redundancy; A3: Convert model parameters into low-precision numerical representations, using quantization to reduce storage requirements and accelerate inference. A4: Use a larger and more complex teacher model to guide the student model's learning and design an appropriate loss function to improve the student model's performance; A5: Fine-tune the lightweight model to recover the performance degradation caused by pruning and quantization and improve task adaptability. A6: Fully verify and test the optimized lightweight model to ensure it meets the expected performance standards and resource consumption requirements on the target platform.

6. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 1, characterized in that: The data processing and analysis unit (300) includes a data cleaning and integration module (301), a prediction analysis model module (302), an adaptive mechanism module (303), and a customer segmentation module (304), wherein: The data cleaning and integration module (301) is used to clean the data obtained from the image recognition unit (200), remove noise, duplication and erroneous data, and integrate data of different formats and sources to unify the data format and standard; The forecast analysis model module (302) generates a forecast of future sales trends based on a time series analysis model and according to historical sales data and other external factors; The adaptive mechanism module (303) is used to automatically adjust the analysis strategy according to market dynamics and update inventory management and replenishment suggestions in real time; The customer segmentation module (304) is used to divide different customer groups and identify their preferences through in-depth analysis of customer purchasing behavior, thereby supporting personalized marketing and services.

7. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 6, characterized in that: The forecast analysis model module (302) generates a forecast of future sales trends based on the time series analysis model and according to historical sales data and other external factors. The specific operations are as follows: B1: Smoothing and outlier correction of historical sales data; B2: Use the logistic growth model to decompose the basic trend items: In the formula, C is the market carrying capacity; k is the growth rate parameter; m is the trend turning point; ∈ t is the random error term; B3: Fitting the periodicity through Fourier series, the seasonal term is modeled as: Where p is the sales cycle day / week / year, a n 、b n are the Fourier series coefficients; B4: Weighted integration of promotional activities, weather and other factors: βX t =β1x1,t+β2x2,t Where x1,t is the intensity of promotion activities; x2,t is the weather impact index; B5: Output the final predicted value: B6: Retrain model parameters every 28 days sliding window.

8. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 6, characterized in that: The customer segmentation module (304) divides different customer groups and identifies their preferences through in-depth analysis of customer purchasing behavior, supporting personalized marketing and services. The specific operations are as follows: C1: Extract shopping trajectory features, including product picking frequency, shelf dwell time, and purchase combination correlation; C2: Clustering is performed using the DBSCAN algorithm. The distance measurement formula is: d(x,y) = α·purchase interval + β·category familiarity Where x, y are the behavioral feature vectors of two customers; α, β are trainable weight coefficients, α = 0.6, β = 0.4; C3: Generate preference labels for each group: When the purchase rate of discounted products is greater than 60%, it is price-sensitive; when the purchase rate within 7 days of listing is greater than 40%, it is new product trial; C4: Associate customer segments with product display recommendations, including recommendations for exclusive shelf locations for high-value customers and dynamic pricing threshold adjustments.

9. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 1, characterized in that: The feedback and decision support unit (400) includes a visualization display module (401), a decision suggestion generation module (402), and an interactive operation module (403), wherein: The visualization display module (401) is used to display the analysis results obtained by the data processing and analysis unit (300) on a visualization operation interface in the form of intuitive charts, reports, maps, etc. The decision suggestion generation module (402) is used to provide specific decision suggestions to merchants and staff based on the data analysis results and using preset rules and algorithms; The interactive operation module (403) is used to support merchants and staff to interact with the visual operation interface.

10. The intelligent retail shelf monitoring system based on artificial intelligence image recognition according to claim 1, characterized in that: The communication unit (500) comprises an internal communication module (501) and an external communication module (502), wherein: The internal communication module (501) is used to ensure data transmission and communication between various units in the system; The external communication module (502) is responsible for data transmission and communication between the system and external devices or systems.

Images