Intelligent anti-theft take-out storage device and using method thereof

Abstract

According to the intelligent anti-theft take-out storage device and the using method thereof, through real-time monitoring, intelligent linkage and algorithm optimization, accurate tracking and abnormal alarm of a take-out position are achieved, a suspect picture can be transmitted in time, and all-around safety guarantee is provided for the take-out industry; the device not only has a strong anti-theft function, but also enhances the user interaction experience and improves the take-out delivery safety and efficiency through seamless connection with the Internet platform. Meanwhile, the reliability and the stability of the equipment are ensured through the comprehensive design improvement, and the equipment is suitable for various working environments.

Description

Technical Field

[0001] The present invention relates to the field of intelligent anti-theft devices and Internet of Things application technologies, and more specifically, to an intelligent anti-theft takeaway storage device and a method for using the same. Background Art

[0002] With the booming food delivery market and consumers’ increasing reliance on food delivery services, safety issues during food delivery, especially food theft, misdelivery, and insufficient safety monitoring during delivery, have become key factors restricting the healthy development of the industry. Traditional solutions, such as adding fixed food delivery cabinets, have alleviated the problem of food theft to a certain extent, but they face challenges such as high deployment costs, large space resource usage, and difficulty in flexible deployment in various complex environments. In addition, these solutions often lack intelligent monitoring and early warning mechanisms, and are unable to identify and respond to abnormal situations in food delivery in real time and accurately. Summary of the invention

[0003] In order to overcome the above-mentioned defects of being unable to detect takeaway theft in real time and accurately identify and respond to takeaway theft, the present invention provides an intelligent anti-theft takeaway storage device and a method of using the same. The device aims to solve the problems of theft, misdelivery and security monitoring in the takeaway delivery process through intelligent monitoring, real-time alarms and data analysis, thereby improving the safety of takeaway delivery and user experience.

[0004] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0005] The present invention provides an intelligent anti-theft takeaway storage device, comprising a main body, a steering gear, a first protection component, a second protection component, a camera module, a first sensor, a buzzer, and a second sensor;

[0006] The steering gear is arranged at the center of one side of the upper surface of the main body, and is used to open and close the first protection component and the second protection component;

[0007] The first protection component and the second protection component are symmetrically arranged about the steering gear axis and are connected to one side of the upper surface of the main body through a hinge;

[0008] The camera module is symmetrically arranged on the first protection component and the second protection component about the steering gear axis, and is used to obtain identity information;

[0009] The first sensor is arranged on the main body and is used to obtain position information in real time;

[0010] The buzzer is arranged on one side of the main body and is used to emit an alarm sound;

[0011] The second sensor is arranged on the main body and is used for detecting temperature;

[0012] The main body is used to control the camera module, the steering gear, the first sensor, the buzzer and the second sensor.

[0013] Preferably, the main body includes a PCB board and a third protective component; the PCB board is connected to the third protective component by bolts; the upper surface of the third protective component is provided with a first through hole and a second through hole; the second sensor is arranged on the upper surface of the third protective component and is electrically connected to the PCB board; the buzzer is arranged on one side of the third protective component and is electrically connected to the PCB board.

[0014] Preferably, a solar panel and an LED light strip are provided on the upper surfaces of the first protective component and the second protective component; the solar panel is used to provide power supply; and the LED light strip is used to display the number when the user picks up the meal.

[0015] Preferably, the PCB board includes a power management module, a communication interface module, a first sensor, a heating module, a relay module, a servo drive module and a main control unit; the power management module is electrically connected to the main control unit; the communication interface module is electrically connected to the main control unit; the first sensor is electrically connected to the main control unit; the output end of the second sensor is connected to the input end of the heating module; the output end of the heating module is connected to the main control unit; the relay module is electrically connected to the main control unit; the main control unit is electrically connected to the servo drive module, and the output end of the servo drive module is connected to the input end of the servo.

[0016] Preferably, refrigeration components are provided on the sides of the first protection component and the second protection component, and the refrigeration components are used to reduce the temperature.

[0017] Preferably, a battery pack is further included, and the battery pack is arranged on a PCB board, and the output end of the solar panel module is connected to the input end of the battery pack to provide power supply.

[0018] Preferably, the display screen is an OLED display screen.

[0019] Preferably, an anti-slip pad is provided at the bottom of the main body.

[0020] The present invention also provides a method for using the above-mentioned intelligent anti-theft takeaway storage device, comprising:

[0021] The rider or merchant carries the food to the destination;

[0022] Using a mobile device to input the user's identity information through an Internet platform, and placing food on the upper surface of the third protective component;

[0023] After the food is placed, the device enters the monitoring mode, monitors the location and temperature of the food in real time, and turns on or off the heating module and refrigeration components according to the temperature of the food;

[0024] The probability of food theft at the destination is obtained through the trained multi-scale residual model, and the probability of food theft at the destination is uploaded to the Internet platform;

[0025] If the user authorizes through the Internet platform, the corresponding number of the food will be displayed through the LED light strip, and the first protective component or the second protective component will be opened accordingly, then the food can be taken away. If the food is moved without authorization from the Internet platform, the buzzer will sound an alarm, and the camera module will capture the user's identity information and synchronize the alarm information and the user's identity information to the Internet platform.

[0026] Preferably, obtaining a trained multi-scale residual model includes:

[0027] Obtain user order data, destination environmental characteristics, and user behavior data;

[0028] The user's order data, the environmental characteristics of the destination, and the user behavior data are input into the constructed multi-scale residual model for training. When the preset number of training times is reached, a trained multi-scale residual model is obtained.

[0029] Preferably, the order data includes meal delivery time period, meal information and delivery date; the destination environment characteristics include destination location information, location safety rating and monitoring facility conditions; and the user behavior data includes the user's meal pick-up time and meal evaluation records.

[0030] Preferably, the constructed multi-scale residual model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a first normalized activation layer, a second normalized activation layer, a third normalized activation layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a first maximum pooling layer, a second maximum pooling layer, a third maximum pooling layer, a splicing layer and an output layer;

[0031] The output end of the input layer is connected to the input ends of the first convolutional layer and the splicing layer; the output end of the first convolutional layer is connected to the input end of the first normalized activation layer; the output end of the first normalized activation layer is connected to the input end of the fourth convolutional layer; the output end of the fourth convolutional layer is connected to the input end of the first maximum pooling layer; the output end of the first maximum pooling layer is connected to the input end of the splicing layer;

[0032] The output end of the input layer is connected to the input ends of the second convolutional layer and the splicing layer; the output end of the second convolutional layer is connected to the input end of the second normalized activation layer; the output end of the second normalized activation layer is connected to the input end of the fifth convolutional layer; the output end of the fifth convolutional layer is connected to the input end of the second maximum pooling layer; the output end of the second maximum pooling layer is connected to the input end of the splicing layer;

[0033] The output end of the input layer is connected to the input ends of the third convolutional layer and the splicing layer; the output end of the third convolutional layer is connected to the input end of the third normalized activation layer; the output end of the third normalized activation layer is connected to the input end of the sixth convolutional layer; the output end of the sixth convolutional layer is connected to the input end of the third maximum pooling layer; the output end of the third maximum pooling layer is connected to the input end of the splicing layer;

[0034] The output end of the splicing layer is connected to the input end of the output layer.

[0035] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0036] The present invention proposes an intelligent anti-theft takeaway storage device and its use method. Through real-time monitoring, intelligent linkage and algorithm optimization, it can achieve accurate tracking of takeaway locations and abnormal alarms, and can instantly transmit suspect images, providing all-round security protection for the takeaway industry. The device not only has a powerful anti-theft function, but also enhances the user's interactive experience through seamless connection with the Internet platform, and improves the safety and efficiency of takeaway delivery. At the same time, its comprehensive design improvements ensure the reliability and stability of the equipment and adapt to various working environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a schematic diagram of the structure of the intelligent anti-theft takeaway storage device described in Example 1;

[0038] Figure 2 This is a schematic diagram of the structure of the PCB board described in Example 2;

[0039] Figure 3 This is a schematic diagram of the structure of the intelligent anti-theft takeaway storage device described in Example 2;

[0040] Figure 4 is a schematic structural diagram of the third protection component described in Example 2;

[0041] Figure 5 is a schematic structural diagram of the third protection component described in Example 2;

[0042] Figure 6 This is a schematic diagram of the structure of the first protection component and the second protection component on the side in Example 2;

[0043] Figure 7 Schematic diagram of the structure of the multi-scale residual model described in Example 3;

[0044] Figure 8 This is a flowchart of the multi-scale residual model training described in Example 3. DETAILED DESCRIPTION

[0045] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0046] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0047] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0048] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0049] Example 1

[0050] This embodiment provides an intelligent anti-theft takeaway storage device, such as Figure 1 As shown, it includes a main body 1, a steering gear 2, a first protective component 3, a second protective component 4, a camera module 5, a first sensor 6, a buzzer 7 and a second sensor 8;

[0051] The steering gear 2 is arranged at the center of one side of the upper surface of the main body 1, and is used to open and close the first protection component 3 and the second protection component 4;

[0052] The first protection component 3 and the second protection component 4 are symmetrically arranged about the steering gear 2 axis, and are connected to one side of the upper surface of the main body 1 through a hinge;

[0053] The camera module 5 is symmetrically arranged on the first protection component 3 and the second protection component 4 about the steering gear 2 axis, and is used to obtain identity information;

[0054] The first sensor 6 is arranged on the main body 1 and is used to obtain position information in real time;

[0055] The buzzer 7 is arranged on one side of the main body 1 and is used to emit an alarm sound;

[0056] The second sensor 8 is arranged on the main body 1 and is used to detect the temperature;

[0057] The main body 1 is used to control the camera module 5, the steering engine 2, the first sensor 6, the buzzer 7 and the second sensor 8.

[0058] This embodiment achieves accurate tracking of takeout locations and abnormal alarms through real-time monitoring, intelligent linkage and algorithm optimization, and can instantly transmit suspect images, providing all-round security for the takeout industry; the device not only has a powerful anti-theft function, but also enhances the user interaction experience through seamless connection to the Internet platform, and improves the safety and efficiency of takeout delivery. At the same time, its comprehensive design improvements ensure the reliability and stability of the equipment and adapt to various working environments.

[0059] Example 2

[0060] This embodiment provides an intelligent anti-theft takeaway storage device, including a main body 1, a steering gear 2, a first protection component 3, a second protection component 4, a camera module 5, a first sensor 6, a buzzer 7 and a second sensor 8;

[0061] The steering gear 2 is arranged at the center of one side of the upper surface of the main body 1, and is used to open and close the first protection component 3 and the second protection component 4;

[0062] The first protection component 3 and the second protection component 4 are symmetrically arranged about the steering gear 2 axis, and are connected to one side of the upper surface of the main body 1 through a hinge;

[0063] The camera module 5 is symmetrically arranged on the first protection component 3 and the second protection component 4 about the steering gear 2 axis, and is used to obtain identity information;

[0064] The first sensor 6 is arranged on the main body 1 and is used to obtain position information in real time;

[0065] The buzzer 7 is arranged on one side of the main body 1 and is used to emit an alarm sound;

[0066] The second sensor 8 is arranged on the main body 1 and is used to detect the temperature;

[0067] The main body 1 is used to control the camera module 5, the steering engine 2, the first sensor 6, the buzzer 7 and the second sensor 8.

[0068] The main body 1 includes a PCB board 9 and a third protective component 10; the PCB board 9 and the third protective component 10 are connected by bolts; Figure 4 and Figure 5 As shown, the upper surface of the third protective component 10 is provided with a first through hole 11 and a second through hole 12; the second sensor 8 is arranged on the upper surface of the third protective component 10 and is electrically connected to the PCB board 9; the buzzer 7 is arranged on one side of the third protective component 10 and is electrically connected to the PCB board 9.

[0069] like Figure 3As shown, the upper surfaces of the first protective component 3 and the second protective component 4 are provided with a solar panel 13 and an LED light strip 14; the solar panel 13 is used to provide power supply; the LED light strip 14 is used to display the number when the user picks up the meal.

[0070] like Figure 2 As shown, the PCB board (9) includes a power management module 15, a communication interface module 16, a first sensor 6, a heating module 17, a relay module 18, a steering gear drive module 19 and a main control unit 20; the power management module 15 is electrically connected to the main control unit 20; the communication interface module 16 is electrically connected to the main control unit 20; the first sensor 6 is electrically connected to the main control unit 20; the output end of the second sensor is connected to the input end of the heating module 17; the output end of the heating module 17 is connected to the main control unit 20; the relay module 18 is electrically connected to the main control unit 20; the main control unit 20 is electrically connected to the steering gear drive module 19, and the output end of the steering gear drive module 19 is connected to the input end of the steering gear 2.

[0071] like Figure 6 As shown, refrigeration components 21 are disposed on the sides of the first protection component 3 and the second protection component 4, and the refrigeration components 21 are used to reduce the temperature.

[0072] It also includes a battery pack 22, which is arranged on the PCB board 9 and is used to provide power supply.

[0073] The display screen is an OLED display screen.

[0074] The bottom of the main body 1 is provided with an anti-slip pad.

[0075] In a specific embodiment, high-quality FR-4 material is selected as the basis of the PCB board, which ensures the stability and durability of the entire system with its excellent electrical performance and mechanical strength. The production process of the PCB board strictly follows the advanced printed circuit board manufacturing process, including precise design, high-quality photolithography, fine etching, precise drilling and meticulous welding steps to ensure that each PCB board can achieve optimal electrical connection and mechanical strength.

[0076] In order to achieve efficient power management, the TP4056 lithium battery charging management chip is used. The chip supports direct charging through the micro USB interface and has overcharge, over-discharge and short-circuit protection functions, ensuring the safety of battery use and the stability of power supply. At the same time, the AMS1117 voltage regulator chip is introduced to stably convert the input voltage to the 5V voltage required by the system, providing stable power support for various electronic components. During the installation process, the chip is soldered to the PCB board in strict accordance with the specifications, and carefully connected to the power input and other circuit components to ensure accurate connections.

[0077] In order to ensure the accuracy and stability of signal transmission between the sensor and the communication module, optimized wiring specifications and precise installation techniques are adopted. When connecting the ultrasonic sensor, camera module and ESP8266 wireless communication module, special attention is paid to the correspondence of the pins and the correctness of the wiring to avoid signal interference or loss due to incorrect connection. At the same time, the connection port is also reinforced to ensure that it will not loosen or fall off during long-term use.

[0078] After completing the PCB board manufacturing and the installation of the power management module, the welding, connection and fixing of the various modules of the system began. This includes the installation of the 0.96-inch OLED display, STM32F103C8T6 microcontroller, buzzer, ultrasonic sensor, OV7670 camera module and external battery module. Soldering and connection were carried out one by one according to the established steps, and preliminary testing and debugging were carried out after each step was completed to ensure that each module can work properly and work together to complete the expected functions. Finally, the entire system was fully tested and debugged to ensure that it can operate stably and reliably in actual applications.

[0079] TP4056 lithium battery charging management chip: supports 4.2V charging cut-off voltage, maximum charging current can reach 1A, and has temperature monitoring function to ensure the safety and efficiency of the battery charging process.

[0080] AMS1117 voltage regulator chip: stabilizes the input voltage at 5V, the maximum output current is 1A, and the low voltage dropout characteristic ensures voltage stability.

[0081] ESP8266 wireless communication module: It has 802.11b / g / n Wi-Fi function, operates in the frequency range of 2.4GHz to 2.5GHz, and provides a maximum data transmission rate of 72.2Mbps, meeting the needs of real-time data uploading and receiving cloud commands.

[0082] STM32F103C8T6 microcontroller: Running at 72MHz, it has 128KB of flash memory and 20KB of SRAM, which is enough to process data from sensors and control the display.

[0083] Ultrasonic sensor: With a measurement range of 2cm to 400cm and an accuracy of 3mm, it ensures accurate monitoring of the location of objects.

[0084] OV7670 camera module: provides 0.3-megapixel image resolution and supports frame rates up to 30fps, suitable for real-time video surveillance applications.

[0085] The embedded hardware of the entire project consists of a sensor part, a temperature control part, a communication part and a power management part, which can achieve efficient detection and early warning of takeout theft and temperature maintenance of takeout.

[0086] The main control uses the STM32F103C8T6 minimum system board which is widely used in various circuit hardware scenarios. It has the characteristics of high performance, low power consumption and access to rich peripherals. The main control is the "brain" of the hardware in the takeaway anti-theft device. It uses I / O ports, USART, SPI, I2C and other serial communication interfaces to control ultrasonic sensors, temperature sensors, motor drive modules, cooling / heating modules, buzzers, WiFi and other modules.

[0087] The sensor part uses the HC-SR04 ultrasonic sensor to detect whether the takeaway has been taken away, the DS18B20 temperature sensor senses the current temperature of the takeaway, the camera monitors the image inside the lid in real time, and the positioning module transmits the current position of the takeaway anti-theft device; the temperature control part uses semiconductor cooling chips to lower the temperature of the takeaway, and the PTC heating chip to increase the temperature of the takeaway; the communication part uses the WiFi module to connect to the OneNet platform to push information to users; the power management part uses a 12V lithium battery pack to store electrical energy and power the hardware as a whole, the solar panel charges the lithium battery autonomously, the DC power conversion module converts the 12V power supply to various voltage values ​​for output, and the relay module controls the on and off of the LED light strip; the remaining TB6612 motor driver module drives the MG90S servo to control the opening and closing of the lid, the LED light strip is used to prompt the user the location of the takeaway, and the buzzer sounds an alarm when the takeaway is stolen.

[0088] The HC-SR04 ultrasonic sensor is used to detect whether the takeaway has been taken away. It is a non-contact distance measurement sensor that uses the CS-100A ultrasonic distance measurement SOC chip, which has the characteristics of high performance, industrial grade, wide voltage, and low price. It works on the principle of ultrasonic echolocation, sending out ultrasonic pulses and receiving the reflected signals to calculate the distance.

[0089] The DS18B20 temperature sensor is a commonly used high-precision single-bus digital temperature measurement chip, which is used to measure the temperature of takeout food, so as to control the temperature of takeout food. It has the characteristics of small size, low hardware cost, strong anti-interference ability and high precision.

[0090] The camera and positioning module are both independent modules, powered by built-in lithium batteries. By using these two modules, users can view real-time images and takeaway locations on their mobile phones and computers.

[0091] Temperature control part: semiconductor cooling sheets and PTC heating sheets are used to keep takeouts with different hot and cold requirements warm. Both use non-contact temperature control, that is, heat energy is transferred to the takeouts in the form of heating / cooling air, which is safer.

[0092] The outer side of the semiconductor cooling plate is a heat dissipation surface with a cooling fan, and the inner side is a cooling surface with a small fan to blow cold air to quickly reduce the temperature of the takeaway. The heating plate has a high temperature resistance, a fast heating speed, no redness and no open flame, and is safe and guaranteed.

[0093] The ESP8266 module is used to establish a communication bridge with the OneNet IoT platform. Users and riders can exchange information through the platform, thereby completing the information transmission between users and takeaway cabinets. The ESP8266 module has a built-in high-efficiency antenna and integrates powerful WiFi communication functions. It can upload data to the cloud in real time and receive cloud commands. It has the advantages of stable performance, low cost and rich development resources. Through the wiring port, each sensor and ESP8266 communication module are connected, so that each module can communicate with STM32 and exchange data, complete the detection of takeaway storage, alarm to the outside world and connection with the OneNet IoT platform.

[0094] Once an abnormal situation occurs, the takeaway anti-theft device will activate its camera module to capture and record the suspect's image. These data and images are uploaded to the OneNet platform through the ESP8266 wireless communication module, which records and processes all security information from the device. After the data is uploaded, the WeChat applet will receive the alarm information from OneNet in real time, and push the warning message and the suspect's image to the user's mobile phone, so that the user can immediately understand the safety status of the takeaway.

[0095] The DC power conversion module is directly connected to the 12V lithium battery pack, converting the 12V power supply into the working voltage 5V and 3.3V output required by some modules. In addition, the power conversion module can also play a shunting role to ensure uniform power supply. The solar module is connected to the 5V solar panel to convert light energy into electrical energy, with a constant output current of 1.5A, charging the lithium battery pack, realizing the autonomous power supply of the takeaway anti-theft device and extending the battery life. When the LED light needs to be turned on, the main control gives the relay a high level, and the current path supplies power to the LED light strip to make it light up.

[0096] Example 3

[0097] This embodiment provides a method for using the intelligent anti-theft takeaway storage device described in Embodiment 1 or Embodiment 2, including:

[0098] The rider or merchant carries the food to the destination;

[0099] Use a mobile device to input the user's identity information through the Internet platform, and place the food on the upper surface of the third protective component 10;

[0100] After the food is placed, the device enters the monitoring mode, monitors the location and temperature of the food in real time, and turns on or off the heating module 17 and the refrigeration component 21 according to the temperature of the food;

[0101] The probability of food theft at the destination is obtained through the trained multi-scale residual model, and the probability of food theft at the destination is uploaded to the Internet platform;

[0102] If the user authorizes through the Internet platform, the corresponding number of the food is displayed through the LED light strip 14, and the first protective component 3 or the second protective component 4 is opened accordingly, then the food can be taken away. If the food is not moved without authorization from the Internet platform, the buzzer 7 will sound an alarm, and the camera module 5 will capture the user's identity information, and synchronize the alarm information and the user's identity information to the Internet platform.

[0103] Get the trained multi-scale residual model, including:

[0104] Obtain user order data, destination environmental characteristics and user behavior data;

[0105] The user's order data, the environmental characteristics of the destination, and the user behavior data are input into the constructed multi-scale residual model for training. When the preset number of training times is reached, a trained multi-scale residual model is obtained.

[0106] The order data includes the meal delivery time period, meal information and delivery date; the destination environment characteristics include destination location information, location safety rating and monitoring facility conditions; the user behavior data includes the user's meal pick-up time and meal evaluation records.

[0107] like Figure 7 As shown, the constructed multi-scale residual model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a first normalized activation layer, a second normalized activation layer, a third normalized activation layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a first maximum pooling layer, a second maximum pooling layer, a third maximum pooling layer, a splicing layer and an output layer;

[0108] The output end of the input layer is connected to the input ends of the first convolutional layer and the splicing layer; the output end of the first convolutional layer is connected to the input end of the first normalized activation layer; the output end of the first normalized activation layer is connected to the input end of the fourth convolutional layer; the output end of the fourth convolutional layer is connected to the input end of the first maximum pooling layer; the output end of the first maximum pooling layer is connected to the input end of the splicing layer;

[0109] The output end of the input layer is connected to the input ends of the second convolutional layer and the splicing layer; the output end of the second convolutional layer is connected to the input end of the second normalized activation layer; the output end of the second normalized activation layer is connected to the input end of the fifth convolutional layer; the output end of the fifth convolutional layer is connected to the input end of the second maximum pooling layer; the output end of the second maximum pooling layer is connected to the input end of the splicing layer;

[0110] The output end of the input layer is connected to the input ends of the third convolutional layer and the splicing layer; the output end of the third convolutional layer is connected to the input end of the third normalized activation layer; the output end of the third normalized activation layer is connected to the input end of the sixth convolutional layer; the output end of the sixth convolutional layer is connected to the input end of the third maximum pooling layer; the output end of the third maximum pooling layer is connected to the input end of the splicing layer;

[0111] The output end of the splicing layer is connected to the input end of the output layer.

[0112] In a specific embodiment, the software algorithm is based on the MAML meta-learning framework, which improves the adaptability and accuracy of the system by optimizing model parameters and intelligent analysis. MAML (Model-Agnostic Meta-Learning) is a general meta-learning framework that is particularly suitable for solving small sample learning problems. It optimizes the initial parameters of the model so that it can quickly achieve good performance when faced with new tasks with only a small amount of gradient updates. At the same time, it can extract general and transferable knowledge or learning strategies by learning a series of "tasks" with similar properties (in this project, it refers to the prediction of takeaway theft in different regions), so that the model can quickly achieve higher performance when faced with new tasks with limited data, which is convenient for product promotion and customization.

[0113] The data includes order details (delivery time period, day of the week, pick-up location), environmental characteristics (location safety rating, monitoring facilities, complaints), and user behavior data (pick-up habits, evaluation records, and wait time for takeout).

[0114] The data is divided into multiple independent tasks, called "subjects", according to regions (such as dormitory buildings, office areas, school gates, etc.). Each subject represents the takeaway theft prediction problem in a specific region. Each task contains its own training set, validation set, and test set, simulating the learning and evaluation scenarios under limited data in a new region.

[0115] Each piece of data in the data set corresponds to nine features, including: order details (delivery time period, day of the week, pick-up location), environmental features (location safety rating, monitoring facilities, complaints), and user behavior data (pick-up habits, evaluation records, and wait time for takeout). First, the data is expanded from 1*9 sequence data to 9*9 (that is, 9 rows of the same data), and then converted into a 9*9*32 data block through linear mapping. In other words, each piece of data corresponds to a 9*9*32 data block, which is convenient for subsequent convolution processing.

[0116] In order to extract multi-level features and enhance feature representation at the same time, a multi-scale residual model suitable for this task is proposed instead of using a single machine learning model. The model consists of three independent multi-scale residual blocks. 3*3, 5*5 and 9*9 convolutions are used to extract local and global feature information respectively. Finally, the output of each block is concatenated as the output of the model, and the dimension is still 9*9*32. The multi-scale residual block can extract and integrate feature information of different scales in a single network layer, which is convenient for identifying and analyzing fine-grained and global features in the data. The residual learning mechanism allows direct information transfer, alleviating the gradient vanishing problem in deep networks, and the multi-scale structure further optimizes this process, so that information of different scales can flow through the network effectively and efficiently. By introducing multi-scale fusion in the residual structure, the model can more effectively learn cross-layer feature associations and promote information transfer between low-level features and high-level semantic features. This helps to generate richer and more discriminative feature representations and improve the recognition and classification performance of the model.

[0117] Note that the amount of data is small. In order to allow the takeaway anti-theft device to be deployed in different areas in a personalized manner, it may be difficult to achieve the traditional training strategy. Therefore, the MAML meta-learning strategy is adopted. The training step is mainly divided into three stages: 1. Warm-up stage: At this time, all training data are passed into the model to train the feature extractor (convolutional layer of the model) and the classifier (the last layer FC). Learn the data patterns common to all source domain subjects 2. Meta-training stage: In this step, the training data will be divided into a support set and a query set. The support set is used to perform the first gradient update on the model, with the aim of learning the common data features between different subjects. Then, the query set and the parameters obtained from the support set are used for the second gradient update. The Adam optimizer is used to directly adjust the parameters of the multi-scale residual model to enhance the adaptability of the model to unseen subjects. 3. Meta-testing stage: The target domain requires a small amount of labeled data for migration, which is also divided into a support set and a query set. The support set is labeled data. First, the parameters trained by meta-training are initialized. The support set is used to update the feature extractor. Since the query set is unlabeled, the classifier is not updated. Instead, the classifier fine-tuned in the meta-training stage is used to classify the query set. Finally, the accuracy on the test set is evaluated to verify the generalization of the model on new subjects.

[0118] Note that the data collected through the questionnaire on takeout is relatively small. If we want to deploy takeout anti-theft devices in different areas in a personalized way, it may be difficult to achieve using traditional training strategies. Therefore, the MAML meta-learning strategy is adopted.

[0119] like Figure 8 As shown, the training step is mainly divided into three stages:

[0120] Warm-up phase: All training data is passed into the model to train the feature extractor (convolutional layer of the model) and classifier (the last FC layer). Learn the data patterns common to all source domain subjects

[0121] Meta-training phase: In this step, the training data is first divided into a support set and a query set. The support set is used to perform the first gradient update on the model, with the goal of learning the common data features between different subjects. Then, the query set and the parameters obtained from the support set are used for the second gradient update, and the Adam optimizer is used to directly adjust the multi-scale residual model parameters to enhance the model's adaptability to unseen subjects.

[0122] Meta-testing phase: The target domain requires a small amount of labeled data for migration, which is also divided into a support set and a query set. The support set is labeled data. Initialize the parameters trained by meta-training first, and use the support set to update the feature extractor. Since the query set is unlabeled, the classifier is not updated. Instead, the classifier fine-tuned in the meta-training phase is used to classify the query set. Finally, evaluate the accuracy of the test set to verify the generalization of the model on the new subject.

[0123] The multi-scale residual model is used to replace the traditional convolutional network for feature extraction. For example, there is an image input. First, the original image is processed by 3*3 convolution to obtain a feature map. The feature map is normalized through the BN layer, and then nonlinearly activated through the ReLU activation function. Then, further feature extraction is performed through 3*3 convolution to obtain a new feature map, and finally the maximum pooling operation is performed on each row of pixels in the image. Maximum pooling is to select the largest pixel in each row and delete all others to retain important features, and finally obtain a one-dimensional vector. Then, the original image is similarly processed through two paths of 5*5 convolution and 9*9 convolution to obtain two one-dimensional vectors, and then the three one-dimensional vectors are concatenated (concat), and then the output is obtained through the fully connected layer dimensionality reduction. The residual means that the initial input is added to the feature map output by the second convolution layer. This ensures that even if the model does not learn anything in the later stage, it can maintain the initial state, at least it will not regress and prevent the model from crashing.

[0124] Training data is a lot of pictures, divided into support set and query set. There is no need to delve into these two terms. Then use all the training data to train the multi-scale residual model. At this time, we learn the data patterns common to all source domain subjects. This is called the warm-up phase. Then use the support set and pass it into the model for a gradient update, which means to perform a training. The purpose is to learn the common data features between different subjects. Then use the query set and pass it into the model for a second gradient update to enhance the adaptability of the model to unseen subjects. The above is the training process. Next, test data is tested. There are only a few labeled data in the test data. Labeling means that, for example, there is a picture of a cat and there is a label file telling the model that this picture is a cat. No label means there is only one picture and no labels. Divide all the test data, the labeled data into the support set, and the unlabeled data into the query set. Initialize the model parameters trained above. The support set is used to update the feature extractor of the model. Since the query set has no labels, the classifier is not updated. Instead, the classifier fine-tuned in the meta-training phase is used to classify the query set. The feature extractor of the model is an operation such as convolution, and the classifier is to reduce the high-dimensional data, such as a 30-dimensional vector, into 2 dimensions (if it is a binary classification). Then the 2 dimensions represent the probabilities of the two categories respectively, and the prediction result of the model is the category with a higher probability.

[0125] The initialization parameters of the model are randomly initialized. In this paper, it is a multi-scale residual model, and the tasks here can be regarded as the data related to theft of takeout in different regions. For the purpose of customization, it aims to collect a large amount of data related to theft of takeout in one area, train a core model, and then achieve good effect migration through a small amount of data on theft of takeout in other regions. Therefore, Task 1 has many batch samples and is trained many times, and finally the parameters of the model are trained from θ1 to θ k , then the subsequent Task 2 and Task n only need to use a small amount of data (such as Batch Samplek) to directly update the parameters once to get better results. The following formula is the traditional model parameter update formula, which is implemented by the back propagation algorithm. It mainly applies the idea of ​​MAML few-sample migration to customized occasions. The task learning rate \alpha and meta-learning rate \beta of this project are both 1e-3. The batch size is 64. k is obtained by dividing the total number of samples by the batch size.

[0126] The same or similar reference numerals correspond to the same or similar components;

[0127] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;

[0128] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. An intelligent anti-theft takeaway storage device, characterized in that: It comprises a main body (1), a steering gear (2), a first protective component (3), a second protective component (4), a camera module (5), a first sensor (6), a buzzer (7) and a second sensor (8); The steering gear (2) is arranged at a central position on one side of the upper surface of the main body (1) and is used to open and close the first protection component (3) and the second protection component (4); The first protective component (3) and the second protective component (4) are symmetrically arranged about the steering gear (2) axis, and are connected to one side of the upper surface of the main body (1) via a hinge; The camera module (5) is symmetrically arranged on the first protection component (3) and the second protection component (4) about the steering gear (2) axis, and is used to obtain identity information; The first sensor (6) is arranged on the main body (1) and is used to obtain position information in real time; The buzzer (7) is arranged on one side of the main body (1) and is used to emit an alarm sound; The second sensor (8) is arranged on the main body (1) and is used to detect temperature; The main body (1) is used to control the camera module (5), the steering engine (2), the first sensor (6), the buzzer (7) and the second sensor (8).

2. The intelligent anti-theft takeaway storage device according to claim 1, characterized in that: The main body (1) comprises a PCB board (9) and a third protective component (10); the PCB board (9) and the third protective component (10) are connected by bolts; the upper surface of the third protective component (10) is provided with a first through hole (11) and a second through hole (12); the second sensor (8) is arranged on the upper surface of the third protective component (10) and is electrically connected to the PCB board (9); the buzzer (7) is arranged on one side of the third protective component (10) and is electrically connected to the PCB board (9).

3. The intelligent anti-theft takeaway storage device according to claim 1, characterized in that: A solar panel (13) and an LED light strip (14) are provided on the upper surfaces of the first protective component (3) and the second protective component (4); the solar panel (13) is used to provide power supply; and the LED light strip (14) is used to display the number when the user picks up the meal.

4. The intelligent anti-theft takeaway storage device according to claim 3 is characterized in that: The PCB board (9) comprises a power management module (15), a communication interface module (16), a first sensor (6), a heating module (17), a relay module (18), a steering gear drive module (19) and a main control unit (20); the power management module (15) is electrically connected to the main control unit (20); the communication interface module (16) is electrically connected to the main control unit (20); the first sensor (6) is electrically connected to the main control unit (20); the output end of the second sensor is connected to the input end of the heating module (17); the output end of the heating module (17) is connected to the main control unit (20); the relay module (18) is electrically connected to the main control unit (20); the main control unit (20) is electrically connected to the steering gear drive module (19), and the output end of the steering gear drive module (19) is connected to the input end of the steering gear (2).

5. The intelligent anti-theft takeaway storage device according to claim 1, characterized in that: A refrigeration component (21) is provided on the side surfaces of the first protection component (3) and the second protection component (4), and the refrigeration component (21) is used to reduce the temperature.

6. The intelligent anti-theft takeaway storage device according to claim 4, characterized in that: It also includes a battery pack (22), which is arranged on the PCB board (9) and is used to provide power supply.

7. A method for using the intelligent anti-theft takeaway storage device according to any one of claims 1 to 6, characterized in that: include: The rider or merchant carries the food to the destination; Using a mobile device to input user identity information through an Internet platform, and placing food on the upper surface of the third protective component (10); After the food is placed, the device enters a monitoring mode, monitors the location and temperature of the food in real time, and turns on or off the heating module (17) and the refrigeration component (21) according to the temperature of the food; The probability of food theft at the destination is obtained through the trained multi-scale residual model, and the probability of food theft at the destination is uploaded to the Internet platform; If the user authorizes through the Internet platform, the number corresponding to the food is displayed through the LED light strip (14), and the first protective component (3) or the second protective component (4) is opened accordingly, then the food can be taken away. If the food is not moved without authorization from the Internet platform, the buzzer (7) will sound an alarm, and the camera module (5) will capture the user's identity information, and synchronize the alarm information and the user's identity information to the Internet platform.

8. The method for using the intelligent anti-theft takeaway storage device according to claim 7, characterized in that: Get the trained multi-scale residual model, including: Obtain user order data, destination environmental characteristics and user behavior data; The user's order data, the environmental characteristics of the destination, and the user behavior data are input into the constructed multi-scale residual model for training. When the preset number of training times is reached, a trained multi-scale residual model is obtained.

9. The method for using the intelligent anti-theft takeaway storage device according to claim 8, characterized in that: The order data includes the meal delivery time period, meal information and delivery date; the destination environment characteristics include destination location information, location safety rating and monitoring facility conditions; the user behavior data includes the user's meal pick-up time and meal evaluation records.

10. The method for using the intelligent anti-theft takeaway storage device according to claim 8, characterized in that: The constructed multi-scale residual model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a first normalized activation layer, a second normalized activation layer, a third normalized activation layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a first maximum pooling layer, a second maximum pooling layer, a third maximum pooling layer, a splicing layer and an output layer; The output end of the input layer is connected to the input ends of the first convolutional layer and the splicing layer; the output end of the first convolutional layer is connected to the input end of the first normalized activation layer; the output end of the first normalized activation layer is connected to the input end of the fourth convolutional layer; the output end of the fourth convolutional layer is connected to the input end of the first maximum pooling layer; the output end of the first maximum pooling layer is connected to the input end of the splicing layer; The output end of the input layer is connected to the input ends of the second convolutional layer and the splicing layer; the output end of the second convolutional layer is connected to the input end of the second normalized activation layer; the output end of the second normalized activation layer is connected to the input end of the fifth convolutional layer; the output end of the fifth convolutional layer is connected to the input end of the second maximum pooling layer; the output end of the second maximum pooling layer is connected to the input end of the splicing layer; The output end of the input layer is connected to the input ends of the third convolutional layer and the splicing layer; the output end of the third convolutional layer is connected to the input end of the third normalized activation layer; the output end of the third normalized activation layer is connected to the input end of the sixth convolutional layer; the output end of the sixth convolutional layer is connected to the input end of the third maximum pooling layer; the output end of the third maximum pooling layer is connected to the input end of the splicing layer; The output end of the splicing layer is connected to the input end of the output layer.

Images