Technical Exploration and Research on Smart Canteens

Amid the wave of digital transformation in university logistics services, the intelligent renovation of campus dining scenarios has become a crucial breakthrough to enhance the satisfaction of teachers and students. The long-standing problems in traditional canteens, such as low service efficiency, unbalanced resource allocation, and insufficient information transparency, have found innovative solutions with the empowerment of mobile internet technology and cloud computing architecture. The intelligent ordering system built on the SpringBoot and Vue frameworks has restructured the dining service process, integrated Internet of Things devices, and applied big data analysis, forming a complete closed loop from online reservation to offline fulfillment, providing a new digital model for university dining services.

I. In-depth Analysis of System Architecture
The technical architecture design of this system fully considers the particularity of the campus scenario and builds an elastic and scalable service system based on the modular design concept. The front-end application is developed based on Vue3's composable API, taking advantage of its reactive programming features to achieve dynamic data binding. The unified design language of the ElementPlus component library ensures the consistency of the cross-platform interface. For mobile usage scenarios, the front-end has specially optimized touch interaction logic, developed convenient functions such as gesture sliding to select meals and voice search for dishes, and introduced Service Worker technology to achieve offline caching, ensuring that core functions can still be used normally in the event of network fluctuations.

The backend service is built as a microservice cluster using the SpringBoot 2.7 framework, splitting the core business into six independent modules: user service, order service, payment service, recommendation service, inventory service, and data analysis service. Each microservice is equipped with an independent Hystrix circuit breaker, which automatically switches to a backup node when a service instance fails, ensuring the overall system availability reaches 99.95%. Service-to-service communication adopts the Feign declarative client, combined with Ribbon to achieve client-side load balancing, effectively distributing server pressure. In terms of distributed transaction processing, the system innovatively combines the AT mode of Seata with the eventual consistency solution of message queues, designing compensation transaction mechanisms for critical operations such as payment and inventory modification, ensuring that data states are traceable and repairable in case of exceptions.

The data storage layer adopts a multi-engine hybrid architecture: The MySQL 8.0 cluster achieves master-slave synchronization through GTID, uses the InnoDB engine to handle transactional data, and horizontally shards the order table, with each shard configured with a read-write separation strategy; the Redis 6.2 cluster is deployed in sentinel mode, caching hot dish information and user session data, and implementing a distributed lock mechanism through Redisson; the time series database InfluxDB is specifically used to store device operation logs and operation flows, providing a structured data source for later big data analysis. File storage selects MinIO to build a private object storage cluster, using erasure coding technology to achieve data redundancy, and combines with CDN to accelerate the loading speed of images, optimizing the loading time of dish detail page images from 2.3 seconds to 0.6 seconds.

II. Technical Implementation Path of Core Functions
The intelligent recommendation engine, as the core innovation point of the system, adopts a multi-layer recommendation strategy fusion technology. The basic layer builds a user-based collaborative filtering (UserCF) model, calculates the user preference matrix through the cosine similarity algorithm, and generates a recommendation list by real-time analysis of the order data of the last 30 days. The enhancement layer introduces a time context-aware mechanism, focusing on different category recommendations during breakfast, lunch, and dinner periods, and dynamically adjusts the weight of the soup category based on temperature and humidity data obtained from the weather API. At the health management level, the system connects to the school hospital's physical examination system to obtain user health records, applies a rule engine to execute 128 dietary rules formulated by nutritionists, automatically filters out allergenic ingredients, and marks high-sugar and high-salt meals. Measured data shows that this recommendation system has reduced the average ordering time by 58% and increased the cross-category trial rate by 23%.

The high-concurrency order processing system has designed a three-dimensional solution. At the front-end level, fine-grained traffic control is implemented: a 500ms debounce mechanism is added to button clicks, and the inventory display adopts a hybrid strategy of local cache and timed refresh to reduce the impact of invalid requests. On the server side, Redisson distributed locks are used to manage inventory deduction, and a pre-reserved inventory pool mechanism is developed for flash sale type meals, ensuring atomic operations through Lua scripts. After an order is created, it is immediately pushed to the RabbitMQ message queue, where multiple consumers handle persistence operations in parallel, and the Sentinel flow control framework is used to achieve peak shaving and valley filling. At the database level, ShardingSphere 5.1 is used for database and table sharding. The order table is sharded by the user ID hash, and a composite index of (user ID, creation time) is established, which increases the query efficiency by 4 times.

The payment system has established a three-tier guarantee system: The instant payment channel integrates the official SDKs of Alipay and WeChat, and adopts a dual verification mechanism of asynchronous notification and active query; the campus card channel connects to the campus card system through an encrypted tunnel and develops a network disconnection resumption function to ensure the reliability of offline payments; the credit payment system introduces the Sesame Credit Score assessment model and provides credit services to users with a score of 650 or above. The payment compensation service monitors the transaction status in real time, initiates an automatic reconciliation process for orders that have not been returned within the time limit, and ensures zero error in funds through the T+1 reconciliation file at the end of the day. Data from the system's six-month operation shows that the payment success rate is 99.97%, and the average time for error handling is only 8 minutes.

III. Technological Innovation and Scenario Integration
The intelligent tray recognition system has pioneered a new model of unmanned settlement. The hardware end deploys customized visual recognition terminals, equipped with Jetson Xavier NX edge computing devices, and runs an improved YOLOv5 model to achieve food detection. The software algorithm integrates HSV color space analysis and contour feature extraction technology, which can distinguish similar-looking food categories. The physical layer integrates high-precision weighing sensors. When the visual recognition confidence is lower than 95%, weight verification is automatically initiated. The dual-modal recognition increases the overall accuracy rate to 99.2%. This system compresses the food collection and settlement process from the traditional manual mode of 45 seconds to 7 seconds, reducing labor costs by 80%.

The health management system has established a multi-dimensional digital nutritionist system. By connecting with wearable devices such as smart bracelets, it collects real-time physiological data such as the user's physical activity and sleep quality, and generates personalized health reports based on daily food intake. A deep learning model analyzes the user's diet records over 30 consecutive days to automatically detect the risk of nutritional imbalance. When insufficient vitamin C intake is detected, the system will prioritize recommending meals rich in this component. Customized functions are developed for special groups: for diabetic patients, the system automatically calculates the GI value of meals before they eat; for pregnant and postpartum women, it provides prompts on folic acid intake; and for fitness enthusiasts, it offers protein supplementation suggestions.

The material management system uses Internet of Things (IoT) technology to achieve full-chain traceability. In the warehousing stage, RFID electronic tags are deployed, and the inbound and outbound operations are automatically recorded on the blockchain evidence platform. The demand forecasting module adopts LSTM neural networks to analyze historical sales data and combines the campus calendar (such as exam weeks and holidays) for sales volume prediction, with an accuracy rate 34% higher than traditional methods. The logistics dispatching system integrates the Gaode Map API and plans the optimal delivery route based on real-time traffic conditions, increasing transportation efficiency by 18%. The supplier management platform establishes a dynamic evaluation system that automatically generates supplier ratings based on six indicators, including the pass rate of food material inspection and on-time delivery rate.

IV. Performance Optimization and Security Assurance
The server-side performance optimization was implemented through a full-chain transformation. At the JVM level, the G1 garbage collector was adopted, and the -XX:MaxGCPauseMillis parameter was adjusted to 200 to control the pause time. The ratio of the Young and Old generations was optimized to 1:3, reducing the frequency of Full GC from five times a day to once a week. In terms of SQL optimization, 12 complex queries were reconstructed based on the analysis of slow query logs, and a covering index was added to the user behavior table, reducing the response time of paginated queries from 850ms to 120ms. The cache strategy was implemented with a hierarchical storage approach: hot data was stored in the Redis cluster with a 15-minute expiration time, and local cache was implemented using Caffeine with an LFU eviction policy. The hit rate of the secondary cache remained stable at over 89%.

Front-end performance has been enhanced through multiple innovative measures. Tree Shaking was enabled during the build process to remove unused code, and the vendor file size was reduced by 38% through a Webpack splitting strategy. All image resources were converted to WebP format and combined with lazy loading technology to reduce the first-screen resource load from 3.2MB to 1.1MB. Virtual scrolling technology was enabled during runtime, maintaining a frame rate of over 55FPS for lists with tens of thousands of data items. A compatibility mode was specially developed for low-end devices, automatically downgrading animation effects and disabling non-core functions, which improved the operation smoothness of devices like Redmi 9A by 70%.

The security protection system establishes a five-dimensional defense matrix. The TLS 1.3 protocol is enabled for the entire site at the transport layer, and sensitive fields are encrypted and stored using the national cipher SM4 algorithm. The authentication system implements dynamic two-factor verification, shortening the validity period of access tokens to 10 minutes and binding refresh tokens to device fingerprints. The permission management is based on the RBAC model and extended in the space dimension to distinguish the operation permissions of different cafeteria stalls. The risk control system trains a GBDT algorithm model to detect abnormal order placement behaviors in real time (such as multiple payments by the same account within one minute), and automatically triggers a manual review process. In terms of data security, incremental backups are performed at 3 a.m. every day, and important data is backed up in a remote location. The drill shows that the RTO is controlled within 15 minutes.

V. Implementation Outcomes and Ecological Value
The operational data of the system in the pilot universities indicate that digital transformation has brought about significant benefits. The service throughput during lunch peak hours has increased from 120 people per hour to 320 people per hour, and the efficiency per square meter of the stalls has grown by 2.8 times. The rate of meal preparation errors has dropped from 13% to 0.7%, and disputes caused by information asymmetry have decreased by 91%. Through data analysis, the structure of dishes has been optimized, and the elimination of slow-selling items has reduced the rate of food material waste by 42%. The operation dashboard provided by the merchant-end App helps managers accurately identify peak sales periods and popular categories. After a certain spicy hot pot stall adjusted its business hours based on the system's suggestions, its average daily sales increased by 65%.

User research data shows that digital transformation has been highly recognized. Among the 2,300 valid feedbacks collected, 94% of students believe that the intelligent recommendation has saved their ordering time, 87% of faculty and staff appreciate the health management function, and 92% of merchants say that the system has improved their operational efficiency. The specially developed accessibility mode has received high praise from visually impaired users, and the voice navigation function has made it possible for them to order independently. The contactless meal pickup function launched during the epidemic has reduced the density of people flow in the cafeteria by 40%, becoming an important technical support for campus epidemic prevention.

The ecological value created by this system is continuously expanding. After being connected with the unmanned supermarket system on campus, users can exchange their dining points for goods; once the data is integrated with the academic affairs system, meals for students who are late for laboratory classes will be reserved automatically; the data on kitchen waste collected by the environmental protection module provides a basis for the logistics department to formulate emission reduction plans. Currently, three universities have carried out secondary development based on this framework, adding scenario-based services such as classroom reservation meal delivery and customized tea breaks for academic conferences, which verifies the scalability of the technical framework.

VI. Future Evolution Directions and Technological Outlook
The system will continue to integrate cutting-edge technologies to enhance service capabilities. It is planned to introduce a large language model to build a knowledge graph of catering, allowing users to query food information and obtain nutritional advice through natural conversations. The environmental perception system will deploy UWB positioning base stations to monitor the density of people in the cafeteria in real time and provide the best path planning for new students. The carbon neutrality module will develop a carbon footprint calculator to accurately calculate the carbon emissions of each meal and guide green consumption behavior.

In terms of the hardware ecosystem, the facial recognition function of the smart meal pick-up cabinet is currently under testing and will be integrated with the campus security system in the future. The unmanned delivery vehicle project has completed the verification of the path planning algorithm and is about to provide the "last 100 meters" meal delivery service in the dormitory area. In the AR technology upgrade plan, users can scan the meal plate to view the traceability information of the ingredients, and the 3D holographic projection will display the meal preparation process.

The construction experience of this system is shaping industry standards. Its core technical modules, such as microservice governance solutions, real-time data analysis models, and hybrid cloud deployment strategies, have applied for six software copyrights. The project team, in collaboration with the catering association, has compiled the "White Paper on the Construction of Smart Canteens in Colleges and Universities", providing a systematic implementation guide for the digital transformation of the industry. With the popularization of 5G, edge computing and other technologies, campus catering services will gradually evolve into intelligent entities featuring human-machine collaboration and virtual-real integration, continuously driving digital innovation in higher education scenarios.