Kitchen safety monitoring method and system based on multi-modal AI large model

Abstract

The invention provides a kitchen safety monitoring method and system based on a multi-mode AI large model. Belongs to the technical field of kitchen safety monitoring. The method comprises the following steps: collecting multi-modal information in a kitchen in real time through a sensor network installed in the kitchen; carrying out the preprocessing of the collected multi-mode information; carrying out fusion processing on the preprocessed data based on a multi-modal AI large model, and extracting key features in each piece of modal information; and through a deep learning algorithm, information of different modes is associated and matched to form comprehensive understanding of operation behaviors in the kitchen. Through real-time acquisition and analysis of multi-modal information, comprehensive monitoring of operation behaviors, equipment states and environmental factors in a kitchen is realized.

Description

Technical Field

[0001] The present invention proposes a kitchen safety monitoring method and system based on a multimodal AI large model, belonging to the technical field of kitchen safety monitoring. Background Art

[0002] Traditional kitchen safety monitoring methods mainly rely on manual monitoring and single sensor technology, which have problems such as limited monitoring range, delayed response, and high false alarm rate. With the rapid development of artificial intelligence technology, especially the widespread application of multimodal AI large models, new solutions have been provided for kitchen safety monitoring. However, the application of existing multimodal AI technology in the field of kitchen safety monitoring is still in its early stages, with problems such as insufficient modal information fusion, low monitoring accuracy, and poor adaptability. Summary of the invention

[0003] The present invention provides a kitchen safety monitoring method and system based on a multimodal AI large model to solve the problems mentioned in the above background technology: The present invention proposes a kitchen safety monitoring method based on a multimodal AI large model, the method comprising: S1. Collect multimodal information in the kitchen in real time through a sensor network installed in the kitchen; and pre-process the collected multimodal information; S2. Based on the multimodal AI big model, the pre-processed data is fused and processed to extract the key features of each modal information; and through the deep learning algorithm, the information of different modalities is associated and matched to form a comprehensive understanding of the operation behavior in the kitchen; S3. Based on the fused multimodal information, the operation behavior in the kitchen is monitored in real time, and the operator's actions, expressions, sounds and other behaviors are analyzed. Based on the analysis results, it is determined whether there are any violations or safety hazards; the equipment status and environmental parameters in the kitchen are monitored in real time to discover and warn of potential failures or dangers; S4. When any violation or safety hazard is detected, the system immediately triggers the early warning mechanism and sends early warning information to management personnel or operators in various ways. The system automatically takes corresponding intervention measures according to the severity of the early warning information. S5. By continuously collecting and analyzing kitchen safety monitoring data, the multimodal AI large model is continuously optimized and upgraded, and the system functions are expanded and improved based on user feedback and needs.

[0004] The kitchen safety monitoring system based on the multimodal AI large model proposed in the present invention comprises: Data acquisition module: collects multimodal information in the kitchen in real time through the sensor network installed in the kitchen; and pre-processes the collected multimodal information; Feature extraction module: Based on the multimodal AI big model, the pre-processed data is fused and processed to extract the key features of each modal information; and through the deep learning algorithm, the information of different modalities is associated and matched to form a comprehensive understanding of the operation behavior in the kitchen; Behavior monitoring module: Based on the fused multimodal information, it monitors the operation behaviors in the kitchen in real time, analyzes the operator's actions, expressions, sounds and other behaviors, and judges whether there are any violations or safety hazards based on the analysis results; and monitors the equipment status and environmental parameters in the kitchen in real time to discover and warn of potential faults or dangers; Mechanism trigger module: When a violation or safety hazard is detected, the system immediately triggers the early warning mechanism and sends early warning information to management personnel or operators in a variety of ways; and according to the severity of the early warning information, the system automatically takes corresponding intervention measures; Optimization and upgrade module: By continuously collecting and analyzing kitchen safety monitoring data, the multimodal AI large model is continuously optimized and upgraded, and the system functions are expanded and improved based on user feedback and needs.

[0005] Beneficial effects of the present invention: through the real-time collection and analysis of multimodal information (including images, sounds, environmental parameters, etc.), comprehensive monitoring of operation behaviors, equipment status and environmental factors in the kitchen is realized. Through the fusion and analysis of multimodal data, it is possible to more accurately identify safety hazards in the kitchen, and timely discover and warn potential risks; the edge computing node is used to preliminarily preprocess the data, and combined with intelligent scheduling and dynamic transmission strategies, the data transmission process is made more efficient and reliable. Especially in data transmission, the edge computing node can optimize the processing according to the urgency and importance of the data, improve the real-time performance and response speed of the system, and reduce the risk of transmission delay and packet loss; the kitchen operation behavior is analyzed by deep learning algorithm, and the behavior pattern of the operator can be identified in real time to determine whether there is illegal operation or potential safety hazards. Based on different behavior patterns and safety risks, the system implements a dynamic early warning mechanism, and automatically takes intervention measures of different levels according to the degree of danger; by analyzing the facial expressions and sound characteristics of the operator, the system can judge his emotional state, providing additional basis for judging whether there are potential safety hazards. At the same time, the system monitors the operating status and environmental parameters of kitchen equipment in real time, and can promptly detect equipment failures or environmental anomalies to ensure safe operation of the kitchen; the system's built-in early warning mechanism provides a variety of early warning information templates based on monitoring results, risk levels and preset strategies, and can customize the type, content and sending method of early warning information according to the needs of managers or operators. Through the response strategy library, the system can intelligently select and execute the best response strategy based on historical data and current monitoring conditions, take preventive measures in advance, and reduce the probability of potential accidents; the system can dynamically adjust and optimize by continuously collecting and analyzing monitoring data. The multimodal AI large model is continuously trained and optimized through reinforcement learning algorithms to improve the accuracy and adaptability of the model, ensuring that the system can perform best in different working environments; big data analysis technology helps the system to deeply mine monitoring data, discover potential safety hazards and improvement points, and present analysis results to users through data visualization, thereby providing managers with more accurate and comprehensive decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a step diagram of the method of the present invention; Figure 2 This is a system module diagram of the present invention. DETAILED DESCRIPTION

[0007] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0008] One embodiment of the present invention, as Figure 1As shown, a kitchen safety monitoring method based on a multimodal AI large model includes: S1. Collect multimodal information such as text, image, and sound in the kitchen in real time through a sensor network installed in the kitchen (such as cameras, microphones, temperature sensors, edge computing nodes, and central processing units, etc.); and pre-process the collected multimodal information; S2. Based on the multimodal AI big model, the pre-processed data is fused and processed to extract the key features of each modal information; and through the deep learning algorithm, the information of different modalities is associated and matched to form a comprehensive understanding of the operation behavior in the kitchen; S3. Based on the fused multimodal information, the operation behavior in the kitchen is monitored in real time, and the operator's actions, expressions, sounds and other behaviors are analyzed. Based on the analysis results, it is determined whether there are any violations or safety hazards; the equipment status and environmental parameters in the kitchen are monitored in real time to discover and warn of potential failures or dangers; S4. When any violation or safety hazard is detected, the system immediately triggers the early warning mechanism and sends early warning information to management personnel or operators through various means such as sound and light alarms, text messages, and emails. Depending on the severity of the early warning information, the system automatically takes corresponding intervention measures, such as suspending equipment operation, cutting off power supply, etc. S5. By continuously collecting and analyzing kitchen safety monitoring data, the multimodal AI large model is continuously optimized and upgraded, and the system functions are expanded and improved based on user feedback and needs.

[0009] The working principle of the above technical solution is as follows: cameras, microphones, temperature sensors and other sensors, as well as edge computing nodes and central processing units are arranged in the kitchen to form a complete sensor network; these sensors are responsible for capturing multimodal information in the kitchen in real time, such as text (such as cooking guides, food labels, etc.), images (such as operator's actions, kitchen environment, etc.), sounds (such as operator's voice, operating sound of kitchen equipment, etc.); the collected multimodal information is preliminarily processed, such as denoising, enhancement, compression, etc., to improve the efficiency and accuracy of subsequent processing; the image information is processed by clarity enhancement, color correction, etc.; the sound information is processed by denoising, volume adjustment, etc.; the temperature and other sensor data are calibrated and formatted; based on The multimodal AI large model fuses the pre-processed information of different modalities to form a unified information representation; uses the attention mechanism, multimodal fusion layer and other technologies in the deep learning algorithm to effectively integrate information of different modalities such as images, sounds, and texts; extracts key features from the fused multimodal information, which can reflect the essential attributes of the operation behavior in the kitchen; for example, extracts the action features and expression features of the operator from the image; extracts the voice features of the operator and the operating sound features of the kitchen equipment from the sound; through the association learning, matching learning and other technologies in the deep learning algorithm, the information of different modalities is associated and matched to form a comprehensive understanding of the operation behavior in the kitchen; for example, the operator's action The characteristics are associated with the operating sound characteristics of the kitchen equipment to determine whether the operator's behavior complies with safety regulations; based on the fused multimodal information, the operating behavior in the kitchen is monitored in real time; by analyzing the operator's movements, expressions, sounds and other behavioral characteristics, it is determined whether there are any violations or safety hazards; at the same time, the equipment status and environmental parameters in the kitchen are monitored in real time; by analyzing the equipment operating status data (such as temperature, pressure, current, etc.) and environmental parameter data (such as temperature, humidity, smoke concentration, etc.), potential faults or dangers are discovered and warned; when violations or safety hazards are detected, the system immediately triggers the warning mechanism; warning information is sent to management personnel or operators through various methods such as sound and light alarms, text messages, and emails. , reminding them to take corresponding measures; according to the severity of the warning information, the system automatically takes corresponding intervention measures; for example, when the unsafe behavior of the operator is detected, the system can suspend the operation of related equipment; when the fire hazard in the kitchen is detected, the system can automatically cut off the power supply, etc.; by continuously collecting and analyzing kitchen safety monitoring data, the multimodal AI large model is continuously optimized and upgraded; using online learning, incremental learning and other technologies in machine learning, the parameters and weights of the model are continuously updated to improve the accuracy and generalization ability of the model; according to user feedback and needs, the system functions are expanded and improved; for example, new sensor types are added to capture more modal information; the early warning mechanism is optimized to improve the accuracy and timeliness of the early warning;Add remote monitoring and control functions to facilitate managers to remotely manage the kitchen. ;

[0010] The effects of the above technical solution are as follows: by integrating information collected by multiple sensors such as cameras, microphones, temperature sensors, etc., comprehensive coverage of multimodal information such as text, images, and sounds in the kitchen is achieved; the fusion processing of multimodal information can extract key features from each modal information, and associate and match them through deep learning algorithms, so as to more accurately understand the operation behavior in the kitchen; the system can monitor the operator's actions, expressions, sounds and other behaviors in real time, as well as changes in the equipment status and environmental parameters in the kitchen; once a violation or safety hazard is detected, the system will immediately trigger the early warning mechanism and send a warning message to the management personnel or operators, thereby improving the timeliness and accuracy of monitoring; based on the deep learning algorithm, the system can intelligently identify the operator's violation or safety hazard, such as not wearing protective equipment such as chef hats and masks, or abnormal situations such as equipment failure; the system can also automatically take corresponding intervention measures according to the severity of the warning information, such as suspending equipment operation, cutting off power, etc., thereby avoiding potential safety risks; through the automated processing process, the system can automatically record and analyze kitchen safety monitoring data, and continuously optimize and upgrade the multimodal AI large model; it not only reduces the burden on management personnel, but also improves the efficiency and accuracy of safety management ; The system can comprehensively monitor the operation behavior and equipment status in the kitchen and record relevant data and events; this provides managers with rich data support to facilitate their subsequent analysis and decision-making; through the recorded data and events, managers can trace back to the time and place of any violations or safety hazards; it helps them find the root cause of the problem and take effective measures to rectify and prevent it; the method can monitor in real time whether the behavior of operators meets the standard, so as to prompt them to work in accordance with the prescribed operating procedures; it helps to reduce safety risks and quality problems caused by improper operation; through the assistance and reminders of the system, managers can have a more comprehensive understanding of the safety status of the kitchen and the behavior of operators; it helps them to improve their safety management awareness and ability and better perform their management responsibilities; the method integrates advanced technologies such as multimodal AI large models and deep learning algorithms, and provides new technical means and solutions for kitchen safety management; this helps to promote the innovation and development of kitchen safety management technology and improve the overall safety level; through real-time monitoring and early warning mechanisms, the method realizes dynamic management and control of kitchen safety; it helps to break the traditional static management model and promote kitchen safety management to develop in a more intelligent and dynamic direction.

[0011] In one embodiment of the present invention, the S1 includes: S11, collecting multimodal information in the kitchen in real time through sensor equipment installed in the kitchen, and sending the collected multimodal information to corresponding edge computing nodes for preliminary preprocessing; wherein the edge computing nodes are mapped one-to-one according to sensor types, for example, a camera corresponds to one node, and a temperature sensor corresponds to one node; S12, the edge computing node transmits the preliminarily preprocessed multimodal data to the central control unit through a multi-channel transmission protocol; S13. The central control unit preprocesses the preliminarily preprocessed data received from the edge computing node.

[0012] The working principle of the above technical solution is as follows: a variety of sensor devices are installed in the kitchen, such as cameras, microphones, temperature sensors, etc., to collect multimodal information such as text, images, and sounds in the kitchen in real time; these sensor devices can capture various details in the kitchen, such as the operator's movements, expressions, and sounds, as well as environmental parameters such as temperature and humidity in the kitchen; the collected multimodal information will be sent to the corresponding edge computing node for preliminary preprocessing; the edge computing node is mapped one-to-one according to the sensor type to ensure that each type of information can be properly processed; for example, the image information collected by the camera will be sent to the corresponding image processing node, the sound information collected by the microphone will be sent to the sound processing node, and the temperature information collected by the temperature sensor will be sent to the temperature processing node; The edge computing node performs preliminary preprocessing on the received multimodal information, such as data cleaning, format conversion, noise removal, etc. The preprocessed data is more standardized and accurate, providing a basis for subsequent processing and analysis. The edge computing node transmits the preprocessed multimodal data to the central control unit through a multi-channel transmission protocol. The multi-channel transmission protocol can ensure the stability and reliability of data during transmission and avoid data loss or damage. As the core part of the system, the central control unit is responsible for receiving multimodal data from the edge computing nodes. The central control unit has powerful data processing and analysis capabilities and can further process and analyze the received data. The central control unit integrates the received multimodal data and uniformly manages and stores the data from different edge computing nodes. The integrated data is further cleaned and verified to remove duplicate data, abnormal data and other useless information. The accuracy and consistency of the data are ensured to provide a reliable data basis for subsequent processing and analysis. The data is converted into a suitable format according to the needs of subsequent processing and analysis. This helps to improve the efficiency of data processing and analysis and ensure the normal operation of the system.

[0013] The effects of the above technical solution are as follows: the sensor equipment can collect multimodal information in the kitchen in real time, such as images, sounds, temperature, etc., ensuring the timeliness and accuracy of the data; the setting of the edge computing node enables the data to be pre-processed quickly after collection, reducing the delay and redundancy of data transmission; through the multi-channel transmission protocol, the edge computing node can efficiently and stably transmit the pre-processed data to the central control unit; this protocol supports the transmission of multiple data types and formats, ensuring the integrity and consistency of the data; after receiving the data, the central control unit will perform further pre-processing, such as data cleaning, format conversion, etc. This provides a more standardized and accurate data basis for subsequent data analysis and monitoring; the edge computing nodes correspond one-to-one according to the sensor type, realizing distributed data processing; this processing method reduces the processing burden of the central control unit and reduces the overall resource consumption of the system; through the multi-channel transmission protocol and the preprocessing of the central control unit, the data can be transmitted and stored in a more efficient and compact way; this reduces the occupation of data storage space and reduces the operating cost of the system; the technical solution adopts a modular design, and the edge computing nodes and the central control unit can be flexibly configured and expanded according to actual needs; this enables the system to adapt to kitchens of different sizes and different safety monitoring needs; the modular design also makes the maintenance and upgrade of the system simpler and more convenient; when it is necessary to add new sensor equipment or upgrade system functions, only the corresponding modules need to be replaced or upgraded; the technical solution can monitor the operation behavior and equipment status in the kitchen in real time, and once violations or safety hazards are found, the system will immediately trigger the early warning mechanism; this helps to take timely measures to avoid safety accidents and improve the intelligent level of kitchen safety monitoring; through further analysis and mining of the preprocessed data, the system can provide valuable decision support information for managers; this helps managers better understand the safety status of the kitchen and formulate more scientific and reasonable safety management strategies.

[0014] In one embodiment of the present invention, the S12 includes: S121. The edge computing node evaluates the size, urgency, and type (such as video, audio, temperature data, etc.) of the preliminarily preprocessed multimodal data based on preset standards, and dynamically allocates the optimal transmission channel for each data type based on the evaluation results; for example, video data may require higher bandwidth and lower latency, while temperature data may require lower bandwidth but requires continuous and stable transmission; S122. Before data transmission, the edge computing node compresses the multimodal data using a data compression algorithm and sets a priority based on the urgency and importance of the data. S123, the edge computing node sends the compressed and sorted data to multiple transmission channels in parallel through a multi-channel transmission protocol. At the same time, the central control unit monitors the usage of each channel and dynamically adjusts data allocation; S124. Based on the intelligent scheduling algorithm, the transmission strategy is dynamically adjusted according to the real-time network conditions (such as delay and packet loss rate), such as adjusting the data packet size, changing the transmission rate, etc., and based on the error recovery mechanism, when a data transmission error is detected, the retransmission or data repair process is immediately started; the real-time network conditions are obtained by the following formula:

[0015] Where NCS represents the network status score; α and β represent positive constant weight factors, which are used to adjust the impact of delay and packet loss rate on the network status score respectively; and Represents a positive decay parameter, which is used to control the decay speed of the exponential function; D represents the current network delay; represents the maximum acceptable value of network delay; P represents the current network packet loss rate; r and b represent positive constant parameters used to adjust the nonlinear effect of packet loss rate on the score.

[0016] S125. The central control unit regularly collects performance indicators (such as bandwidth utilization and delay variation) of each transmission channel, and performs in-depth analysis based on the performance indicators to identify potential bottlenecks and redundancies; and automatically adjusts channel configuration according to the analysis results.

[0017] The working principle of the above technical solution is as follows: after the edge computing node pre-processes the multimodal data, it will evaluate the size, urgency and type of the data according to preset standards; for example, video data usually requires higher bandwidth and lower latency to ensure transmission quality and real-time performance because it contains rich image information; while temperature data may have lower bandwidth requirements but require continuous and stable transmission; based on the evaluation results, the edge computing node will dynamically allocate the best transmission channel for each data type; this means that different data types will be allocated to the most suitable transmission channel according to their characteristics and requirements to ensure the efficiency and accuracy of data transmission; before data transmission, the edge computing node will use a data compression algorithm to compress the multimodal data; the compressed data will be smaller in size and faster in transmission speed, while reducing the network bandwidth occupation; according to the urgency and importance of the data, the edge computing node will set the priority of the data; For example, critical data such as smoke alarms and fire warnings will be set as high priority to ensure that they can be processed in a timely manner even when the network is congested; edge computing nodes will send compressed and sorted data in parallel to multiple transmission channels through multi-channel transmission protocols (such as MPTCP, QUIC, etc.); this parallel transmission method can make full use of network bandwidth and improve the efficiency and reliability of data transmission; The central control unit will monitor the usage of each transmission channel and dynamically adjust data allocation according to the actual situation. Through the load balancing strategy, it can avoid the situation where some channels are overloaded while other channels are idle, ensuring the balance and stability of data transmission. Based on the intelligent scheduling algorithm, the central control unit will dynamically adjust the transmission strategy according to the real-time network conditions (such as delay, packet loss rate). For example, when the network delay is high, the data packet size can be adjusted or the transmission rate can be reduced to ensure the stability of data transmission. When the network condition is good, the data packet size can be increased or the transmission rate can be increased to improve the efficiency of data transmission. When a data transmission error is detected, the central control unit will immediately start the retransmission or data repair process. Through this error recovery mechanism, the integrity and accuracy of the data can be ensured, and the adverse effects of data loss or damage on the system can be avoided. The central control unit will regularly collect the performance indicators of each transmission channel (such as bandwidth utilization, delay change) and conduct in-depth analysis. Through performance monitoring, potential bottlenecks and redundancy problems can be discovered in time, providing a basis for optimizing the configuration of transmission channels. According to the analysis results of performance monitoring, the central control unit will automatically adjust the channel configuration. For example, when the bandwidth utilization of a channel is too high, you can increase the number of channels to share the load; when the delay of a channel varies greatly, you can adjust its parameter settings to optimize performance.

[0018] The effects of the above technical solution are as follows: by dynamically allocating the optimal transmission channel for each data type based on the evaluation of data size, urgency and type, the targeted and efficient data transmission is ensured; the waste of transmission resources is avoided and the utilization rate of transmission channels is improved; multimodal data is compressed to reduce the data volume and improve the transmission speed; priority is set according to the urgency and importance of the data to ensure that key data can be transmitted first, further improving the efficiency of data transmission; compressed and sorted data is sent to multiple transmission channels in parallel through a multi-channel transmission protocol, realizing parallel transmission and load balancing of data; It avoids transmission interruption caused by overload of a single channel and improves the stability and reliability of data transmission. Based on the intelligent scheduling algorithm, the transmission strategy is dynamically adjusted according to the real-time network status, such as adjusting the packet size, changing the transmission rate, etc., to ensure the adaptability and stability of data transmission. When a data transmission error is detected, the retransmission or data repair process is immediately started, further enhancing the stability and reliability of data transmission. The central control unit regularly collects the performance indicators of each transmission channel and conducts in-depth analysis to identify potential bottlenecks and redundancies. According to the analysis results, the channel configuration is automatically adjusted, such as increasing or decreasing the number of channels, or adjusting the parameter settings of specific channels, which optimizes resource allocation and reduces the system operation cost. The technical solution adopts a modular design, and the edge computing nodes and the central control unit can be flexibly configured and expanded according to actual needs. When it is necessary to add new sensor devices or upgrade system functions, only the corresponding modules need to be replaced or upgraded, which improves the flexibility and scalability of the system. The above formula comprehensively considers the two key performance indicators of network delay and packet loss rate, which can more comprehensively and accurately reflect the real-time status of the current network. By adjusting the weight factor, the relative importance of delay and packet loss rate in the network status score can be flexibly controlled to meet the needs of different application scenarios. The decay parameter in the formula can control the decay speed of the exponential function, so that the score can be dynamically adjusted as the network conditions change. This dynamic adaptability helps the intelligent scheduling algorithm respond to network fluctuations in a timely manner and avoid transmission problems caused by deteriorating network conditions. By introducing constant parameters, the formula can handle the nonlinear effect of packet loss rate on network condition scores. This means that when the network packet loss rate is low, the score may be less sensitive to changes in the packet loss rate; when the packet loss rate is high, the score will drop rapidly, thus more accurately reflecting the poor condition of the network. Based on the network condition score calculated by this formula, the intelligent scheduling algorithm can dynamically adjust the transmission strategy, such as adjusting the packet size, changing the transmission rate, etc. This intelligent adjustment helps optimize transmission efficiency, reduce transmission errors, and improve the utilization of network resources. When a data transmission error is detected, the intelligent scheduling algorithm can immediately start the retransmission or data repair process.Combined with the network status score, the algorithm can select the retransmission strategy more intelligently, such as choosing to retransmit when the current network condition is better, or adjusting the priority and number of retransmissions based on the score.

[0019] In one embodiment of the present invention, the S121 includes: The edge computing node analyzes the size and complexity of each data type. For example, high-definition video data is usually large and complex to process, while simple temperature or humidity data is small and simple to process. The data is divided into three levels: high, medium, and low according to the urgency and importance of the data. For example, smoke alarms and fire warnings are high-urgency data and need to be transmitted immediately, while daily surveillance videos may be medium or low-urgency data. The edge computing node maintains a transmission channel characteristic database, including key indicators such as bandwidth, delay, and stability of each channel. Based on the data feature analysis results and transmission channel characteristics, the edge computing node runs an optimal channel selection algorithm. According to the selected transmission channel, the edge computing node formulates a specific transmission strategy for each data type, including transmission priority, data packet size, transmission rate, etc. During the data transmission process, the edge computing node monitors the usage and performance indicators of each transmission channel in real time. If a channel is found to be congested or the performance is degraded, the dynamic adjustment mechanism is immediately triggered; Based on the real-time monitoring data, the edge computing node fine-tunes the transmission strategy and updates the transmission channel characteristic database through the performance data.

[0020] The working principle of the above technical solution is as follows: the edge computing node first analyzes the size and complexity of each type of data received. For example, high-definition video data usually has a large amount of data and is complex to process due to its high resolution and frame rate; while simple temperature or humidity data has a small amount of data and is relatively simple to process; according to the urgency and importance of the data, the edge computing node divides the data into three levels: high, medium, and low. For example, key safety data such as smoke alarms and fire warnings are high-urgency data and need to be transmitted immediately to ensure timely response; while data such as daily monitoring videos may be medium or low-urgency data and can be transmitted when the network conditions are good; the edge computing node maintains a transmission channel characteristic database, which includes key indicators such as bandwidth, delay, and stability of each transmission channel. These indicators reflect the performance and applicable scenarios of the transmission channel; as the network environment and transmission channel performance change, the edge computing node will update the transmission channel characteristic database regularly or in real time to ensure that the information in the database is always accurate and reliable; based on the data feature analysis results and the transmission channel characteristic database, the edge computing node runs an optimal channel selection algorithm. The input of the algorithm includes factors such as the urgency, size, complexity of the data, and the bandwidth, latency, and stability of the transmission channel. The algorithm comprehensively considers the above factors and selects the most appropriate transmission channel for each data type. For example, for high-definition video data with high urgency and large data volume, the algorithm may choose a transmission channel with high bandwidth and low latency; while for temperature data with low urgency and small data volume, the algorithm may choose a transmission channel with good stability but slightly lower bandwidth; based on the selected transmission channel, the edge computing node formulates specific transmission strategies for each data type, including transmission priority, packet size, transmission rate, etc. These strategies are designed to ensure that data can be transmitted efficiently and stably on the selected channel; during data transmission, the edge computing node monitors the usage and performance indicators of each transmission channel in real time. If a channel is found to be congested or performance degraded, the edge computing node will immediately trigger a dynamic adjustment mechanism to fine-tune the transmission strategy. For example, the packet size can be increased to increase the transmission rate, or the transmission rate can be changed to adapt to changes in network conditions; based on real-time monitoring data, the edge computing node will also update the transmission channel characteristic database. This helps the algorithm to more accurately reflect changes in the current network environment and channel performance when selecting transmission channels in the future.

[0021] The effect of the above technical solution is: by comprehensively considering factors such as the urgency, size, complexity of the data, and the bandwidth, delay and stability of the transmission channel, the edge computing node can accurately select the most suitable transmission channel for each data type. This precise matching avoids the waste of transmission resources and improves the efficiency of data transmission; during the data transmission process, the edge computing node monitors the usage and performance indicators of each transmission channel in real time. Once a channel is found to be congested or the performance is degraded, the dynamic adjustment mechanism is immediately triggered to fine-tune the transmission strategy. This dynamic adjustment ensures the continuity and stability of data transmission and further improves the transmission efficiency; when selecting the transmission channel, the edge computing node not only considers the urgency and size of the data, but also considers key indicators such as the bandwidth, delay and stability of the channel. This multi-factor selection method ensures the reliability and stability of data transmission; by real-time monitoring of the usage and performance indicators of each transmission channel, and updating the transmission channel characteristic database based on these data, the edge computing node can timely discover and solve potential problems. This real-time monitoring and performance update mechanism enhances the reliability of data transmission and reduces the risk of data transmission failure; by accurately matching the transmission channel and dynamically adjusting the transmission strategy, the edge computing node can reasonably use the transmission resources and avoid waste of resources. This resource optimization configuration method reduces the operating cost of the system and improves resource utilization. The technical solution adopts a dynamic adjustment mechanism, which can fine-tune the transmission strategy according to the real-time network status and transmission requirements. This flexibility enables the system to better adapt to different application scenarios and network environments, improving the overall performance of the system. By dividing the data into three levels of high, medium and low, and selecting the transmission channel according to the urgency and importance, the edge computing node can ensure that critical data is transmitted first. This priority guarantee mechanism improves the user experience and service quality, especially in application scenarios that require real-time response. Through real-time monitoring and dynamic adjustment mechanisms, edge computing nodes can provide stable data transmission services. This stability is crucial for application scenarios that require continuous data transmission (such as video surveillance, telemedicine, etc.).

[0022] In one embodiment of the present invention, the S123 includes: The edge computing node initializes and allocates transmission channels based on currently available transmission channel resources (such as MPTCP connections, QUIC sessions, etc.), including setting initial parameters of the channel, such as bandwidth cap, latency threshold, etc. According to the characteristics of the transmission channel, the edge computing node divides the compressed and sorted multimodal data into packets of appropriate size, and adds metadata information to each packet, such as data type, priority, source address, and destination address; The edge computing node sends the segmented data packets to multiple transmission channels in parallel through a multi-channel transmission protocol, and monitors the usage of each transmission channel in real time through the central control unit, including key indicators such as bandwidth occupancy, delay change, and packet loss rate. Based on real-time monitoring data, the central control unit dynamically adjusts the data allocation on each channel; when a channel is congested or its performance degrades, the central control unit will reduce the amount of data sent to the channel and transfer part of the data to other idle or better-performing channels; the central control unit also regularly evaluates whether the channel configuration is optimal based on the channel's performance indicators; if it is found that a channel is improperly configured (such as the bandwidth upper limit is set too low or the delay threshold is set too high), the central control unit will automatically adjust the parameters.

[0023] The working principle of the above technical solution is as follows: The edge computing node first identifies the currently available transmission channel resources, such as MPTCP (Multi-Path TCP) connections, QUIC (Quick UDP Internet Connection) sessions, etc. These channel resources provide physical or logical paths for data transmission; For each available transmission channel, the edge computing node sets initial parameters, such as bandwidth upper limit, delay threshold, etc. These parameters define the performance characteristics and restrictions of the channel, which helps with subsequent data transmission decisions; The edge computing node divides the compressed and sorted multimodal data into packets of appropriate size. The size of the data packet is usually determined based on the bandwidth and delay characteristics of the transmission channel to ensure the efficiency and reliability of data transmission; Each data packet is added with metadata information, including data type (such as video, audio, text, etc.), priority (such as high, medium, low), source address and destination address, etc. These metadata are essential for the correct transmission and subsequent processing of data; The edge computing node sends the divided data packets in parallel to multiple transmission channels through a multi-channel transmission protocol. This parallel transmission method makes full use of the diversity of network resources and improves the throughput and reliability of data transmission; The central control unit monitors the usage of each transmission channel in real time, including key indicators such as bandwidth occupancy, delay change, and packet loss rate. These monitoring data provide a basis for subsequent dynamic adjustments; Based on real-time monitoring data, the central control unit dynamically adjusts the data allocation on each channel. When a channel is congested or performance degrades, the central control unit will reduce the amount of data sent to the channel and transfer part of the data to other idle or better-performing channels. This dynamic adjustment mechanism ensures the continuity and stability of data transmission; The central control unit also regularly evaluates whether the channel configuration is optimal based on the channel's performance indicators. If it is found that the configuration of a channel is unreasonable (such as the bandwidth upper limit is set too low or the delay threshold is set too high), the central control unit will automatically adjust the parameters. This optimized configuration mechanism improves the utilization efficiency and performance of the transmission channel.

[0024] The effect of the above technical solution is: by utilizing multiple transmission channel resources such as MPTCP connection and QUIC session, the edge computing node can send data packets in parallel, thereby significantly improving the throughput of data transmission. This parallel transmission mode enables data to be transmitted simultaneously on multiple paths, reducing the transmission delay caused by the congestion of a single channel; the central control unit can dynamically adjust the data allocation on each channel based on real-time monitoring data. When a channel is congested or the performance is degraded, it will reduce the amount of data sent to the channel and transfer part of the data to other idle or better-performing channels. This dynamic adjustment mechanism ensures the continuity and stability of data transmission and improves the overall transmission efficiency; the central control unit also regularly evaluates whether the channel configuration is optimal based on the performance indicators of the channel. If it is found that the configuration of a channel is unreasonable, such as the bandwidth upper limit is set too low and the delay threshold is set too high, it will automatically adjust the parameters. This parameter optimization mechanism ensures that the transmission channel can always remain in the best state, further improving the reliability and efficiency of data transmission; by dynamically adjusting the data allocation on each channel, the technical solution can ensure the reasonable allocation of transmission resources. It avoids waste and overuse of resources, so that each channel can be fully utilized; Due to the efficient use of transmission resources, this technical solution helps to reduce operating costs. It reduces repeated transmission and additional overhead caused by transmission delay and packet loss, thereby improving the overall economic benefits; This technical solution can process compressed and sorted multimodal data, including video, audio, text and other types of data. This multimodal data support enables the system to flexibly respond to different application scenarios and data requirements; By real-time monitoring of the usage and performance indicators of each transmission channel and dynamic adjustment based on these data, this technical solution can quickly adapt to changes in the network environment. This real-time monitoring and dynamic adjustment mechanism enhances the flexibility and adaptability of the system, enabling it to better cope with network fluctuations and abnormal situations; Through multi-channel parallel transmission and dynamic data allocation, this technical solution can significantly reduce the delay of data transmission and improve throughput. This is particularly important for application scenarios that require real-time transmission (such as video conferencing, online games, etc.), which can significantly improve user experience; This technical solution ensures the stability and reliability of data transmission through real-time monitoring and dynamic adjustment mechanisms. This helps reduce packet loss and errors during data transmission, improves the integrity and accuracy of data transmission, and thus improves service quality.

[0025] In one embodiment of the present invention, the S2 includes: S21. Extract features from the preprocessed image through a convolutional neural network (CNN) to identify key information such as objects, people, and behaviors in the image; and reduce the dimension and encode the extracted features; S22, analyzing the pre-processed sound signal through a recurrent neural network (RNN) to extract sound features such as speech content, intonation, and emotion; and normalizing the sound features; S23, taking environmental parameters such as temperature and humidity as auxiliary features, inputting them into the multimodal AI model together with information from other modalities, and performing feature engineering on the environmental parameters, such as discretization and standardization; S24, through the cross-modal matching algorithm, and based on the attention mechanism, the information such as images, sounds and environmental parameters are associated to identify the intrinsic connection between different modalities; S25. Use deep learning algorithms to comprehensively analyze the fused multimodal information, identify the operational behavior patterns in the kitchen, and cluster and analyze the identified behavior patterns.

[0026] The working principle of the above technical solution is to use convolutional neural network (CNN) to extract features from preprocessed images. Through structures such as convolutional layer, pooling layer and fully connected layer, CNN can automatically learn complex patterns and features in image data, such as edges, corners, textures, etc., so as to identify key information such as objects, people, behaviors, etc. in the image; the extracted features are often of high dimension, and direct use for subsequent processing will increase the computational complexity. Therefore, it is necessary to reduce the dimension and encode the extracted features to simplify the feature representation and reduce redundant information. Common dimensionality reduction methods include principal component analysis (PCA), linear discriminant analysis (LDA), etc., while encoding can be achieved through autoencoders and other methods; the preprocessed sound signal is analyzed using recurrent neural network (RNN) to extract sound features such as speech content, intonation, and emotions. RNN is good at processing sequence data and can capture the time dependency in sound signals, so as to accurately extract sound features; the extracted sound features may have different dimensions and value ranges. In order to unify the feature representation and eliminate the influence of dimensions, the sound features need to be normalized. Normalization can scale the feature value to a specific range (such as between 0 and 1) to make different features comparable; environmental parameters such as temperature and humidity are used as auxiliary features, which are of great significance for understanding the operating environment in the kitchen; feature engineering is performed on environmental parameters to improve their applicability for subsequent analysis. Common processing methods include discretization (dividing continuous values ​​into discrete intervals) and standardization (scaling feature values ​​to a distribution with a mean of 0 and a variance of 1). These processing can make environmental parameters more consistent with the input requirements of subsequent algorithms; through cross-modal matching algorithms, information such as images, sounds and environmental parameters are associated. These algorithms can learn the correspondence between different modalities, thereby identifying the intrinsic connection between different modalities; in the process of cross-modal matching, the introduction of attention mechanisms can make the algorithm pay more attention to important information and improve the accuracy of matching. The attention mechanism can achieve weighted processing of different information by calculating the attention weights between different modalities; and the deep learning algorithm is used to comprehensively analyze the fused multimodal information. Deep learning algorithms have powerful representation learning and generalization capabilities, and can process high-dimensional and complex data. Through deep learning algorithms, operational behavior patterns in the kitchen can be identified. These patterns can be specific action sequences or abstract behavioral features. Furthermore, cluster analysis can be performed on the identified behavioral patterns to discover potential violations or safety hazards. Cluster analysis can group similar behavioral patterns together, thereby helping people better understand the operations in the kitchen.

[0027] The effect of the above technical solution is: by integrating multimodal information such as images, sounds and environmental parameters, the technical solution can provide more comprehensive and accurate data input, thereby significantly improving the recognition accuracy. The complementarity of multimodal information enables the system to more accurately understand the operating behavior and environmental status in the kitchen; using convolutional neural networks (CNNs) to extract features from images, it can automatically learn and extract key information in the image, such as objects, people, behaviors, etc. At the same time, reducing the dimension and encoding the extracted features can simplify the feature representation, reduce redundant information, and improve the efficiency of subsequent processing; by analyzing the sound signal through a recurrent neural network (RNN), it is possible to extract sound features such as voice content, intonation, and emotions. These features are of great significance for understanding voice communication and the emotional state of people in the kitchen, and help to further improve the recognition accuracy; inputting environmental parameters such as temperature and humidity as auxiliary features into the multimodal AI large model can make the system more adaptable to different kitchen environments. These environmental parameters play an important role in understanding the operation behavior and environmental changes in the kitchen, and help improve the robustness and adaptability of the system; through the cross-modal matching algorithm and attention mechanism, the information such as images, sounds and environmental parameters are associated, and the intrinsic connection between different modes can be identified. This cross-modal information fusion method enables the system to better cope with the complex and changeable kitchen environment and improve the accuracy and stability of recognition; the deep learning algorithm is used to comprehensively analyze the fused multimodal information, which can identify the operation behavior patterns in the kitchen and cluster and analyze the identified behavior patterns. This intelligent processing method enables the system to automatically detect potential violations or safety hazards and improve the intelligent level of kitchen management; by clustering and analyzing the identified behavior patterns, the operation rules and potential problems in the kitchen can be further explored. This helps users better understand the operating status of the kitchen, discover and solve problems in a timely manner, and improve user experience; the technical solution uses a multimodal AI large model for information processing and analysis, which promotes the application and innovation of artificial intelligence technology in the field of kitchen management. This technological innovation helps improve the intelligence and automation level of kitchen management and promotes industrial upgrading and transformation. This technical solution integrates technologies from multiple fields such as computer vision, speech recognition, and environmental perception, and realizes the fusion and analysis of multimodal information. This cross-domain technological integration helps promote the coordinated development of related technologies and brings more technological innovation and application opportunities to the field of kitchen management.

[0028] In one embodiment of the present invention, the S24 includes: For time information, a synchronization algorithm based on timestamp matching aligns information from different modalities (such as image frames, sound clips, and environmental parameter records) in time; for image information, computer vision technology (such as target detection and tracking) is used to establish a position relationship diagram of objects and people in space; for sound signals, the location of the sound source is located through sound source localization technology; Based on the cross-attention mechanism, different attention weights are assigned by calculating the similarity between different modal features according to the importance of each modal information; the similarity is obtained by the following formula:

[0029] in, Represents the similarity between the i-th modal feature and the j-th modal feature; Represents the kth element of the i-th modal feature; represents the kth element of the jth modal feature; represents the attention weight assigned to the kth element, reflecting the importance of the element in the similarity calculation; n represents the dimension of the feature vector; After obtaining the attention weights, the multimodal features are fused through weighted summation. At the same time, deep learning techniques (such as Transformer or graph neural network) are used to learn the representation of the fused multimodal features. Use the knowledge graph to build a relationship model between multimodal information. Based on the built relationship model, use the path reasoning algorithm to identify the intrinsic connection between different modalities, including identifying which objects in the image are associated with specific sound events, and which changes in environmental parameters are related to specific behavior patterns. The cross-modal information is checked for consistency. Based on the results of the consistency check, the cross-modal matching algorithm and attention mechanism are optimized. At the same time, the feedback mechanism is used to continuously iterate and improve the model.

[0030] The working principle of the above technical solution is as follows: Time information alignment, based on the synchronization algorithm of timestamp matching, aligns information from different modalities (such as image frames, sound clips, and environmental parameter records) in time. By comparing the timestamps in each modal information, it is ensured that they can accurately correspond to the same time point or time period, thereby establishing a time synchronization relationship. Image information processing, through computer vision technology, such as target detection and tracking, using target detection algorithms to identify objects and people in the image, and using tracking algorithms to establish position relationship diagrams of these objects and people in space, and update their positions and states in real time. Sound signal processing, through sound source localization technology, using sound sensor arrays to receive sound signals, using sound source localization algorithms to calculate the location of the sound source, and associate it with the position relationship diagram in the image information. Multimodal feature fusion, through the cross-attention mechanism, according to the importance of each modal information, by calculating the similarity between different modal features to assign different attention weights, after obtaining the attention weights, the multimodal features are fused by weighted summation, and the representation of the fused multimodal features is learned using deep learning technology (such as Transformer or graph neural network) to extract higher-level feature information. The relationship model between multimodal information is constructed using knowledge graphs. Based on the constructed relationship model, the path reasoning algorithm is used to identify the intrinsic connection between different modalities, for example, to identify which objects in the image are associated with specific sound events, which changes in environmental parameters are related to specific behavior patterns, etc., to perform consistency checks on cross-modal information to ensure that the information between different modalities is coordinated and consistent. Based on the results of the consistency check, the cross-modal matching algorithm and attention mechanism are optimized, and the model is continuously iterated and improved using the feedback mechanism to improve the accuracy and robustness of the model.

[0031] The effects of the above technical solution are as follows: through a synchronization algorithm based on timestamp matching, information from different modalities (such as image frames, sound clips, and environmental parameter records) are accurately aligned in time, ensuring the synchronization of information and providing a reliable foundation for subsequent multimodal information fusion and analysis; using computer vision technology to establish a position relationship diagram of objects and personnel in space, and using sound source localization technology to locate the position of the sound source, the system can more accurately understand the spatial layout and dynamic changes in the scene, thereby improving the accuracy of information processing; by calculating the similarity between different modal features and assigning different attention weights, the system can pay more attention to important information, reduce the interference of noise and redundant information, and improve the effect of multimodal feature fusion; using deep learning technologies such as Transformer or graph neural networks to learn the representation of the fused multimodal features, it can automatically extract high-level feature information. information, further enhancing the effect of feature fusion and the generalization ability of the model; using the knowledge graph to build a relationship model between multimodal information, so that the system can better understand and explain the intrinsic connection between different modalities, and improve the ability to recognize information relationships; based on the constructed relationship model, a path reasoning algorithm is used to identify the intrinsic connection between different modalities, such as recognizing that objects in images are associated with specific sound events, and changes in environmental parameters are related to specific behavior patterns, etc., which provides strong support for subsequent decision-making and analysis; consistency verification is performed on cross-modal information to ensure that the information between different modalities is coordinated and consistent, thereby improving the accuracy and reliability of the model; based on the results of the consistency verification, the cross-modal matching algorithm and attention mechanism are optimized to improve the performance and efficiency of the model; the feedback mechanism is used to continuously iterate and improve the model, so that the model can adapt to the ever-changing environment and needs, and improve the robustness and adaptability of the system. The above formula calculates the similarity between each modal feature. The formula can accurately capture the correlation and interaction between different modalities, which helps the model to more accurately understand the intrinsic connection between different modalities when processing multimodal data, thereby improving the effect of information fusion. The attention weight in the formula reflects the importance of each element in the similarity calculation, so that the model can dynamically adjust the attention allocation according to the specific content of each modal feature. This flexibility helps the model to more efficiently utilize limited information resources when processing complex multimodal tasks, thereby improving the performance and efficiency of the model. Since the formula takes into account each element of the feature vector and performs weighted processing through the attention weight, the model is more robust to small changes in the input data.

[0032] In one embodiment of the present invention, S3 includes: S31. Use deep learning models to analyze operators’ actions in real time, identify possible violations, classify violations, and take different warning and intervention measures according to the severity; S32. By analyzing the operator's facial expressions and voice characteristics, the operator's emotional state is judged; when abnormal emotions are detected, early warning information is issued in a timely manner, and corresponding intervention measures are taken; S33. Real-time monitoring of the operating status of kitchen equipment, such as stove power, refrigerator temperature, etc.; when equipment fails or is abnormal, an early warning message is immediately issued and the corresponding emergency handling procedures are initiated; S34. Continuously monitor the temperature, humidity and other environmental parameters in the kitchen; when the environmental parameters exceed the normal range, issue warning information in time and take corresponding adjustment measures; S35. Design an early warning mechanism based on monitoring results and risk levels, and determine the method of sending early warning information (such as sound and light alarms, text messages, emails, etc.) and the recipients (such as managers, operators, etc.); formulate corresponding intervention measures based on the type and severity of the early warning information; intervention measures include suspending equipment operation, cutting off power supply, initiating emergency evacuation procedures, etc.

[0033] The working principle of the above technical solution is as follows: the deep learning model captures the operator's actions through devices such as cameras and performs real-time analysis. The model has been trained with a large amount of data and can identify the normal operations and violations of the operator. Once a violation is identified, the model will grade it according to the severity of the violation; for example, in a kitchen environment, the deep learning model identifies that an operator directly touches the food without wearing protective gloves, which is a minor violation. The system immediately issues a sound warning and displays a prompt message "Please wear protective gloves"; if the model identifies that the operator has irregular actions when using knives, such as excessive waving or not keeping a safe distance, this is a moderate violation. The system not only issues a sound warning, but also notifies the management personnel through SMS or email, and recommends suspending the operator's work for safety training. The system captures the operator's facial expressions and voice characteristics through cameras and microphones, and uses emotion recognition algorithms to judge their emotional state. When abnormal emotions (such as anger, anxiety, etc.) are detected, the system will issue a warning message in time. For example, in the kitchen work, an operator feels tired and anxious due to long hours of work, and his facial expressions and voice characteristics show obvious abnormalities. After the system recognizes this emotional state, it immediately issues an early warning message, and prompts the operator to take a rest through voice, and notifies the management staff to pay attention. The system uses sensors to monitor the operating status of the equipment in the kitchen in real time, such as the stove firepower, refrigerator temperature, etc. Once the equipment fails or is abnormal, the system will immediately issue an early warning message and start the corresponding emergency handling procedures. For example, the stove in the kitchen fails due to long-term use, and the firepower increases abnormally. After the system recognizes this abnormality, it immediately issues a sound warning and automatically cuts off the power supply of the stove to prevent fire accidents. At the same time, the system notifies the management staff to troubleshoot and repair the fault. The system continuously monitors environmental parameters such as temperature and humidity in the kitchen through sensors to ensure that these parameters remain within the normal range. Once the environmental parameters exceed the normal range, the system will promptly issue an early warning message and take corresponding adjustment measures. For example, if the humidity in the kitchen is too high, it may cause the food to deteriorate or the equipment to get damp. After the system recognizes this abnormality, it immediately issues an early warning message and starts the dehumidification equipment to reduce the humidity in the kitchen. At the same time, the system notifies the management staff to pay attention to ensure the safety of food and equipment. According to the monitoring results and risk levels, the system designs an early warning mechanism and determines the way to send the early warning information (such as sound and light alarms, text messages, emails, etc.) and the recipients (such as managers, operators, etc.). According to the type and severity of the early warning information, the system formulates corresponding intervention measures, such as suspending equipment operation, cutting off power, and starting emergency evacuation procedures. For example, when the system detects that there is a fire risk in the kitchen, it immediately activates the sound and light alarm and notifies the managers and operators via text messages and emails. At the same time, the system automatically cuts off the power supply of all equipment in the kitchen and starts the emergency evacuation procedure to ensure the safe evacuation of personnel.

[0034] The effects of the above technical solution are: through real-time analysis of the operator's actions by the deep learning model, possible violations can be accurately identified, and graded warning and intervention measures can be taken according to the severity. This helps to correct the operator's wrong behavior in time, prevent accidents, and thus improve the safety management level of the kitchen; by analyzing the operator's facial expressions and voice characteristics, the emotional state can be judged, and warning information can be issued in time when abnormal emotions are detected. This helps managers to pay attention to the psychological state of the operator in time, avoid safety hazards caused by emotional fluctuations, and improve the overall working atmosphere and safety of the kitchen; the system can monitor the operating status of the equipment in the kitchen in real time, such as stove firepower, refrigerator temperature, etc. Once a fault or abnormality is found, it will immediately issue a warning message and start the emergency handling procedure. This helps to eliminate equipment failures in time, avoid equipment damage and accidents, and ensure the safe and stable operation of kitchen equipment; continuously monitor environmental parameters such as temperature and humidity in the kitchen to ensure that these parameters remain within the normal range. When the environmental parameters exceed the normal range, the system can issue a warning message in time and take corresponding adjustment measures. This helps to maintain the comfort and safety of the kitchen environment and provide a good working environment for operators; according to the monitoring results and risk levels, a scientific and reasonable early warning mechanism is designed, and the sending method and recipients of the early warning information are determined. This helps to ensure that the early warning information can be conveyed to relevant personnel in a timely and accurate manner, and improve the speed and efficiency of emergency response; according to the type and severity of the early warning information, corresponding intervention measures are formulated, such as suspending equipment operation, cutting off power supply, and initiating emergency evacuation procedures. These measures can respond to risks quickly and effectively in emergency situations and ensure the safety of personnel and equipment; This technical solution integrates a variety of advanced technologies such as deep learning, emotion recognition, and sensor monitoring to achieve comprehensive monitoring and management of kitchen environment, equipment, and personnel. This helps to promote the development of kitchen intelligence and automation and improve the efficiency and accuracy of kitchen management; By collecting and analyzing a large amount of data, the system can generate valuable analysis reports and statistical information to provide decision-making support for managers. This helps managers to formulate management strategies and optimize work processes more scientifically, and improve the overall operational efficiency of the kitchen.

[0035] In one embodiment of the present invention, the S31 includes: The operator's action details are captured through high-precision cameras and sensors, including limb position, movement trajectory, speed, etc. The captured action data is pre-processed to extract key features such as action mode, movement rhythm, strength, etc., and construct feature vectors; based on the extracted feature vectors, a deep learning model (such as convolutional neural network CNN or long short-term memory network LSTM) is constructed and trained; Input the operator's real-time action data into the trained deep learning model for online analysis to quickly identify possible violations; Classify violations into different levels, such as minor, moderate, or severe, based on the type, duration, and frequency of the identified violations; and record the identified violations and their classification information in a database; Design corresponding early warning mechanisms based on the classification results of violations, such as voice prompts or flashing lights for minor violations, SMS or email reminders for moderate violations, and immediate sound and light alarms for serious violations; formulate corresponding intervention measures for different levels of violations; verbal warnings or on-site guidance for minor violations; safety training or restricted operating permissions for moderate violations; and immediate suspension of equipment operation, power cut-off, and initiation of emergency evacuation procedures for serious violations; Dynamically adjust the early warning mechanism and intervention measures according to the effectiveness of early warning and intervention measures, collect feedback from operators, and continuously optimize the deep learning model and early warning intervention process; According to the designed early warning mechanism, the early warning information is sent to the corresponding recipients, such as managers, operators, etc. After receiving the early warning information, the corresponding intervention measures are initiated, and the execution process of the early warning and intervention measures is monitored in real time to evaluate its execution effect and compliance.

[0036] The working principle of the above technical solution is as follows: use high-precision cameras and sensors to capture the details of the operator's movements, including key information such as limb position, movement trajectory, speed, etc.; pre-process the captured movement data, such as denoising, filtering, normalization, etc., to improve data quality and the accuracy of subsequent analysis; extract key features, such as movement pattern, movement rhythm, strength, etc., and construct feature vectors. These feature vectors can characterize the movement characteristics of the operator and provide a basis for the subsequent training of deep learning models; select suitable deep learning models according to the needs of movement recognition, such as convolutional neural networks (CNN) or long short-term memory networks (LSTM). CNN is suitable for processing image and video data and can extract spatial features; LSTM is suitable for processing time series data and can capture temporal dependencies; use a large amount of compliance and violation behavior data to train the deep learning model so that it can accurately identify the operator's movement pattern and distinguish between compliance and violation behaviors. During the training process, the recognition performance of the model is optimized by continuously adjusting the model parameters; the operator's real-time movement data is input into the trained deep learning model for online analysis. The model can quickly identify possible violations and output the identification results; Based on the type, duration, frequency and other factors of the identified violations, the violations are graded, such as minor violations, moderate violations, and serious violations. The grading results can reflect the severity of the violations and provide a basis for the formulation of subsequent warning and intervention measures; According to the grading results of the violations, the corresponding warning mechanism is designed. For example, voice prompts or flashing lights are used for minor violations, SMS or email reminders are sent for moderate violations, and sound and light alarms are immediately triggered for serious violations. The early warning mechanism can ensure that managers and operators obtain violation information in a timely manner and take corresponding measures; According to different levels of violations, corresponding intervention measures are formulated. Minor violations can be given verbal warnings or on-site guidance, moderate violations require safety training or restricted operating permissions, and serious violations require immediate suspension of equipment operation, power cuts, and emergency evacuation procedures. Intervention measures can effectively curb the spread of violations and ensure the safety of personnel and equipment; According to the execution effect of early warning and intervention measures, the early warning mechanism and intervention measures are dynamically adjusted. By collecting feedback from operators and on-site data, the effectiveness of early warning and intervention measures is evaluated, and corresponding adjustments and optimizations are made; the deep learning model is continuously optimized to improve the recognition accuracy and generalization ability of the model. The performance of the model is continuously improved by introducing new data samples and algorithm improvements; according to the designed early warning mechanism, the early warning information is sent to the corresponding recipients, such as managers, operators, etc. Early warning information can ensure that relevant personnel obtain violation information in a timely manner and take corresponding measures; after receiving the early warning information, the corresponding intervention measures are initiated. At the same time, the execution process of early warning and intervention measures is monitored in real time to evaluate their execution effect and compliance.Through monitoring and evaluation, we ensure the effective implementation of early warning and intervention measures to ensure the safe operation of the kitchen.

[0037] The effect of the above technical solution is: etc. key information. After preprocessing, these data can extract key features, such as action mode, movement rhythm, strength, etc., to provide strong support for subsequent analysis; using the trained deep learning model, the real-time action data of the operator is analyzed online, which can quickly identify possible violations. This real-time analysis capability helps to find problems in time and reduce safety hazards; the deep learning model is trained with a large amount of data, and can accurately identify the action mode of the operator and distinguish between compliance and violation. This greatly improves the recognition accuracy of violations and reduces false alarms and missed reports; violations are graded according to the type, duration, frequency and other factors of the identified violations. This graded processing method helps managers take corresponding early warning and intervention measures for violations of different levels, and improves the pertinence and effectiveness of management. At the same time, the identified violations and their graded information are recorded in the database for subsequent analysis and tracing; according to the graded results of the violations, the corresponding early warning mechanism is designed. Minor violations use voice prompts or flashing lights, moderate violations send text messages or email reminders, and serious violations immediately trigger sound and light alarms. This flexible and diverse early warning method can ensure that early warning information is conveyed to relevant personnel in a timely and accurate manner, and improve the speed of emergency response; formulate corresponding intervention measures for different levels of violations. Minor violations can be treated with verbal warnings or on-site guidance, moderate violations require safety training or restricted operating permissions, and serious violations require immediate suspension of equipment operation, power cut-off, and emergency evacuation procedures. These intervention measures can take corresponding treatment measures for different levels of violations and effectively curb the spread of violations; According to the execution effect of early warning and intervention measures, the early warning mechanism and intervention measures are dynamically adjusted. This dynamic adjustment mechanism can ensure that the early warning and intervention measures are always consistent with the actual situation and improve the flexibility and adaptability of management; collect feedback from operators and continuously optimize the deep learning model and early warning intervention process. This continuous optimization mechanism can continuously improve the technical solution and improve its practicality and reliability; monitor the execution process of early warning and intervention measures in real time and evaluate their execution effect and compliance. This real-time monitoring mechanism can ensure that the early warning and intervention measures are effectively implemented and improve the efficiency and safety of management; through the implementation of the above technical solutions, violations can be discovered and handled in a timely manner, safety hazards can be reduced, and the overall safety level can be improved. This is of great significance for ensuring the safety of personnel and equipment.

[0038] In one embodiment of the present invention, the S4 includes: S41. The system provides a variety of warning information templates for users to choose or customize; managers or operators set the type, content and sending method of warning information according to their own preferences and needs; S42. Establish a response strategy library, which contains intervention measures and emergency plans corresponding to different warning information; intelligently predict possible consequences based on historical data and current conditions, and take corresponding preventive measures in advance; S43. When receiving warning information, the system automatically selects the best response strategy from the response strategy library and executes it; during the execution process, the system continuously monitors changes in the situation and adjusts the response strategy as needed.

[0039] The working principle of the above technical solution is as follows: the system provides a variety of warning information templates: these templates cover common warning scenarios, such as natural disasters, equipment failures, market fluctuations, etc., and each template contains the corresponding information structure and format; managers or operators can select ready-made warning information templates in the system according to their actual needs, or create warning information templates that meet their specific needs through simple editing and configuration; set the type, content and sending method of warning information: for example, you can choose SMS, email, APP push and other methods to send warning information, and set the specific content of the information, such as warning level, scope of impact, recommended measures, etc. For example: suppose a user is using the system to monitor the operating status of his production line. When a key equipment on the production line fails, the system can automatically send an early warning message to the manager's mobile phone, and the information content includes the equipment name, fault type, fault time and recommended maintenance measures. Establish a response strategy library, which contains intervention measures and emergency plans for different types of warning information. These strategies and plans are based on historical data and expert experience, and are designed to provide effective response measures. The system uses machine learning and data analysis technology to mine and analyze historical data to predict the possible consequences and impact of the current warning information. Based on the prediction results, the system can automatically or suggest managers to take preventive measures to reduce potential risks and losses. When the system receives warning information, it will automatically select the best response strategy from the response strategy library and execute it based on the type and content of the warning information. This includes operations such as launching emergency plans, sending notifications, and adjusting equipment parameters. In the process of executing the response strategy, the system will continuously monitor changes in relevant indicators and data to promptly discover and respond to new problems. If it is detected that the situation has changed or the current strategy is not effective, the system will automatically or suggest managers to adjust the response strategy to ensure the effectiveness and pertinence of the response measures.

[0040] The effect of the above technical solution is: the system provides a variety of warning information templates, which cover the format and content of warning information for different scenarios and needs. This allows managers or operators to quickly select appropriate templates for setting according to their preferences and needs, improving the flexibility and ease of use of warning information; in addition to the preset templates, the system also supports user-defined warning information. This means that managers or operators can customize warning information that meets their specific needs according to actual conditions, including the type, content, and sending method of the warning. This customization function makes the warning information more accurate and personalized, which helps to better deal with various potential risks; the system has established a response strategy library, which contains intervention measures and emergency plans corresponding to different warning information. This enables the system to quickly call the corresponding response strategy from the library when receiving the warning information, improving the speed and efficiency of the response; the system can intelligently predict possible consequences based on historical data and current conditions, and take corresponding preventive measures in advance. This predictive ability helps managers or operators to prepare in advance and reduce potential risks and losses; when receiving the warning information, the system can automatically select the best response strategy from the response strategy library and execute it. This automated processing reduces the reliance on manual intervention and improves the efficiency and accuracy of emergency response. In the process of executing the response strategy, the system can continuously monitor changes in the situation and adjust the response strategy as needed. This dynamic adjustment capability enables the system to respond flexibly according to the actual situation, improving the flexibility and adaptability of emergency response. Through the above technical solutions, the system can issue early warning information more quickly and accurately, and take effective response measures. This helps to timely discover and deal with potential safety risks and reduce the probability of safety accidents. The system can reasonably allocate and dispatch resources, such as personnel and equipment, based on early warning information and response strategies. This way of optimizing resource allocation improves the efficiency of resource use and reduces safety management costs. By using the S4 system, managers can have a more comprehensive understanding of the company's safety status, more accurately assess the risk level, and thus formulate more effective safety management measures. This helps to improve the overall safety management level of the company.

[0041] In one embodiment of the present invention, S5 includes: S51. Continuously collect kitchen safety monitoring data, and clean and organize the collected data; S52. Use big data analysis technology to conduct in-depth mining of monitoring data, discover potential safety hazards and improvement points, and visualize the analysis results; S53. Based on the collected data and analysis results, the multimodal AI large model is trained and optimized based on the reinforcement learning algorithm; the optimized model is verified and tested, and the model is fine-tuned and improved according to the test results.

[0042] The working principle of the above technical solution is as follows: the system first continuously collects safety monitoring data in the kitchen through sensors, cameras and other equipment. These data include multimodal information (such as images, sounds, temperature, humidity, etc.), early warning records (such as smoke alarms, gas leak alarms, etc.) and intervention measures (such as automatic power cut-off, exhaust system startup, etc.); the collected data may be redundant, erroneous or inconsistent, so it needs to be cleaned and sorted. Data cleaning includes steps such as removing duplicate data, correcting erroneous data, and filling missing data to ensure the accuracy and completeness of the data. Data sorting is to classify, sort and format the cleaned data for subsequent analysis and processing; the cleaned and sorted data is input into the big data analysis system. The system uses big data analysis techniques (such as cluster analysis, correlation analysis, time series analysis, etc.) to deeply mine the monitoring data to discover potential safety hazards and improvement points; the analysis results are presented to users through visual display technology (such as charts, curves, heat maps, etc.). These visualizations help users understand the data and analysis results more intuitively, so as to make more informed decisions. Through big data analysis, the system can identify possible safety risk areas, high-risk time periods, and common types of safety hazards in the kitchen. This information provides important data support for subsequent model training and optimization. Based on the collected data and analysis results, the system begins to train the multimodal AI large model based on the reinforcement learning algorithm. Reinforcement learning is a machine learning algorithm that learns how to take actions to maximize a certain reward signal through interaction with the environment. In this scenario, the reward signal can be a goal such as reducing safety hazards and improving kitchen safety. During the training process, the system continuously provides the model with kitchen safety monitoring data as input, and adjusts the model's parameters based on the comparison between the model's prediction results and the actual situation. Through multiple iterations of training, the model gradually learns how to predict potential safety hazards based on the input data and take corresponding intervention measures. After the training is completed, the system verifies and tests the optimized model. The purpose of verification and testing is to ensure that the model has sufficient accuracy and reliability in practical applications. If the test results are not ideal, the system will fine-tune and improve the model to improve its performance.

[0043] The effects of the above technical solutions are: continuously collecting kitchen safety monitoring data, covering multimodal information (such as images, sounds, temperature, etc.), early warning records, and intervention measures, etc., to ensure the comprehensiveness and diversity of the data. This helps to understand the safety status of the kitchen more comprehensively and provide a solid data foundation for subsequent analysis and decision-making; cleaning and sorting the collected data, removing redundant, erroneous or inconsistent data, and ensuring the accuracy and completeness of the data. This helps to improve the accuracy and reliability of data analysis and provide high-quality data input for subsequent big data analysis and model training; using big data analysis technology to conduct in-depth mining of monitoring data can discover potential safety hazards and improvement points. This deep mining capability helps to identify potential safety risks in advance and provide strong support for timely preventive measures; visualizing the analysis results makes safety hazards and improvement points more intuitive and easy to understand. This helps managers quickly understand the data and analysis results, so as to make more informed decisions; training and optimizing the multimodal AI large model based on the reinforcement learning algorithm enables the model to predict potential safety hazards based on the input data and take corresponding intervention measures. This training method helps to improve the accuracy and adaptability of the model, enabling it to better cope with complex kitchen safety monitoring scenarios; the optimized model is verified and tested to ensure that the model has sufficient accuracy and reliability in practical applications. This rigor helps to avoid false positives or omissions in the model in practical applications, thereby improving the overall safety monitoring effect; through the above technical solutions, kitchen safety management has achieved automation and intelligence in the entire process from data collection and analysis to model training and optimization. This greatly improves the efficiency and accuracy of safety management and reduces reliance on manual intervention; by continuously collecting and analyzing data, fine-tuning and improving the model, the accuracy and effectiveness of safety monitoring can be continuously improved. This ability to continuously improve helps to adapt to the ever-changing safety environment and needs.

[0044] One embodiment of the present invention, as Figure 2 As shown, a kitchen safety monitoring system based on a multimodal AI large model, the system includes: Data acquisition module: collects multimodal information such as text, images, and sounds in the kitchen in real time through the sensor network installed in the kitchen (such as cameras, microphones, temperature sensors, edge computing nodes, and central processing units, etc.); and pre-processes the collected multimodal information; Feature extraction module: Based on the multimodal AI big model, the pre-processed data is fused and processed to extract the key features of each modal information; and through the deep learning algorithm, the information of different modalities is associated and matched to form a comprehensive understanding of the operation behavior in the kitchen; Behavior monitoring module: Based on the fused multimodal information, it monitors the operation behaviors in the kitchen in real time, analyzes the operator's actions, expressions, sounds and other behaviors, and judges whether there are any violations or safety hazards based on the analysis results; and monitors the equipment status and environmental parameters in the kitchen in real time to discover and warn of potential faults or dangers; Mechanism trigger module: When a violation or safety hazard is detected, the system immediately triggers the early warning mechanism and sends early warning information to management personnel or operators through sound and light alarms, text messages, emails, etc. According to the severity of the early warning information, the system automatically takes corresponding intervention measures, such as suspending equipment operation, cutting off power supply, etc. Optimization and upgrade module: By continuously collecting and analyzing kitchen safety monitoring data, the multimodal AI large model is continuously optimized and upgraded, and the system functions are expanded and improved based on user feedback and needs.

[0045] The working principle of the above technical solution is as follows: cameras, microphones, temperature sensors and other sensors, as well as edge computing nodes and central processing units are arranged in the kitchen to form a complete sensor network; these sensors are responsible for capturing multimodal information in the kitchen in real time, such as text (such as cooking guides, food labels, etc.), images (such as operator's actions, kitchen environment, etc.), sounds (such as operator's voice, operating sound of kitchen equipment, etc.); the collected multimodal information is preliminarily processed, such as denoising, enhancement, compression, etc., to improve the efficiency and accuracy of subsequent processing; the image information is processed by clarity enhancement, color correction, etc.; the sound information is processed by denoising, volume adjustment, etc.; the temperature and other sensor data are calibrated and formatted; based on The multimodal AI large model fuses the pre-processed information of different modalities to form a unified information representation; uses the attention mechanism, multimodal fusion layer and other technologies in the deep learning algorithm to effectively integrate information of different modalities such as images, sounds, and texts; extracts key features from the fused multimodal information, which can reflect the essential attributes of the operation behavior in the kitchen; for example, extracts the action features and expression features of the operator from the image; extracts the voice features of the operator and the operating sound features of the kitchen equipment from the sound; through the association learning, matching learning and other technologies in the deep learning algorithm, the information of different modalities is associated and matched to form a comprehensive understanding of the operation behavior in the kitchen; for example, the operator's action The characteristics are associated with the operating sound characteristics of the kitchen equipment to determine whether the operator's behavior complies with safety regulations; based on the fused multimodal information, the operating behavior in the kitchen is monitored in real time; by analyzing the operator's movements, expressions, sounds and other behavioral characteristics, it is determined whether there are any violations or safety hazards; at the same time, the equipment status and environmental parameters in the kitchen are monitored in real time; by analyzing the equipment operating status data (such as temperature, pressure, current, etc.) and environmental parameter data (such as temperature, humidity, smoke concentration, etc.), potential faults or dangers are discovered and warned; when violations or safety hazards are detected, the system immediately triggers the warning mechanism; warning information is sent to management personnel or operators through various methods such as sound and light alarms, text messages, and emails. , reminding them to take corresponding measures; according to the severity of the warning information, the system automatically takes corresponding intervention measures; for example, when the unsafe behavior of the operator is detected, the system can suspend the operation of related equipment; when the fire hazard in the kitchen is detected, the system can automatically cut off the power supply, etc.; by continuously collecting and analyzing kitchen safety monitoring data, the multimodal AI large model is continuously optimized and upgraded; using online learning, incremental learning and other technologies in machine learning, the parameters and weights of the model are continuously updated to improve the accuracy and generalization ability of the model; according to user feedback and needs, the system functions are expanded and improved; for example, new sensor types are added to capture more modal information; the early warning mechanism is optimized to improve the accuracy and timeliness of the early warning;Add remote monitoring and control functions to facilitate managers to remotely manage the kitchen. ;

[0046] The effects of the above technical solution are as follows: by integrating information collected by multiple sensors such as cameras, microphones, temperature sensors, etc., comprehensive coverage of multimodal information such as text, images, and sounds in the kitchen is achieved; the fusion processing of multimodal information can extract key features from each modal information, and associate and match them through deep learning algorithms, so as to more accurately understand the operation behavior in the kitchen; the system can monitor the operator's actions, expressions, sounds and other behaviors in real time, as well as changes in the equipment status and environmental parameters in the kitchen; once a violation or safety hazard is detected, the system will immediately trigger the early warning mechanism and send a warning message to the management personnel or operators, thereby improving the timeliness and accuracy of monitoring; based on the deep learning algorithm, the system can intelligently identify the operator's violation or safety hazard, such as not wearing protective equipment such as chef hats and masks, or abnormal situations such as equipment failure; the system can also automatically take corresponding intervention measures according to the severity of the warning information, such as suspending equipment operation, cutting off power, etc., thereby avoiding potential safety risks; through the automated processing process, the system can automatically record and analyze kitchen safety monitoring data, and continuously optimize and upgrade the multimodal AI large model; it not only reduces the burden on management personnel, but also improves the efficiency and accuracy of safety management ; The system can comprehensively monitor the operation behavior and equipment status in the kitchen and record relevant data and events; this provides managers with rich data support to facilitate their subsequent analysis and decision-making; through the recorded data and events, managers can trace back to the time and place of any violations or safety hazards; it helps them find the root cause of the problem and take effective measures to rectify and prevent it; the method can monitor in real time whether the behavior of operators meets the standard, so as to prompt them to work in accordance with the prescribed operating procedures; it helps to reduce safety risks and quality problems caused by improper operation; through the assistance and reminders of the system, managers can have a more comprehensive understanding of the safety status of the kitchen and the behavior of operators; it helps them to improve their safety management awareness and ability and better perform their management responsibilities; the method integrates advanced technologies such as multimodal AI large models and deep learning algorithms, and provides new technical means and solutions for kitchen safety management; this helps to promote the innovation and development of kitchen safety management technology and improve the overall safety level; through real-time monitoring and early warning mechanisms, the method realizes dynamic management and control of kitchen safety; it helps to break the traditional static management model and promote kitchen safety management to develop in a more intelligent and dynamic direction.

[0047] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A kitchen safety monitoring method based on a multimodal AI large model, characterized in that: The method comprises: S1. Collect multimodal information in the kitchen in real time through a sensor network installed in the kitchen; and pre-process the collected multimodal information; S2. Based on the multimodal AI big model, the pre-processed data is fused and processed to extract the key features of each modal information; and through the deep learning algorithm, the information of different modalities is associated and matched to form a comprehensive understanding of the operation behavior in the kitchen; S3. Based on the fused multimodal information, the operation behavior in the kitchen is monitored in real time, and the operator's actions, expressions, sounds and other behaviors are analyzed. Based on the analysis results, it is determined whether there are any violations or safety hazards; the equipment status and environmental parameters in the kitchen are monitored in real time to discover and warn of potential failures or dangers; S4. When any violation or safety hazard is detected, the system immediately triggers the early warning mechanism and sends early warning information to management personnel or operators in various ways. The system automatically takes corresponding intervention measures according to the severity of the early warning information. S5. By continuously collecting and analyzing kitchen safety monitoring data, the multimodal AI large model is continuously optimized and upgraded, and the system functions are expanded and improved based on user feedback and needs.

2. The kitchen safety monitoring method based on the multimodal AI large model according to claim 1 is characterized in that: Said S1 comprises: S11, collecting multimodal information in the kitchen in real time through sensor equipment installed in the kitchen, and sending the collected multimodal information to the corresponding edge computing node for preliminary preprocessing; S12, the edge computing node transmits the preliminarily preprocessed multimodal data to the central control unit through a multi-channel transmission protocol; S13. The central control unit preprocesses the preliminarily preprocessed data received from the edge computing node.

3. The kitchen safety monitoring method based on the multimodal AI large model according to claim 2 is characterized in that: The S12 comprises: S121, the edge computing node evaluates the size, urgency and type of the preliminarily preprocessed multimodal data based on preset standards, and dynamically allocates an optimal transmission channel for each data type based on the evaluation results; S122. Before data transmission, the edge computing node compresses the multimodal data using a data compression algorithm and sets a priority based on the urgency and importance of the data. S123, the edge computing node sends the compressed and sorted data to multiple transmission channels in parallel through a multi-channel transmission protocol. At the same time, the central control unit monitors the usage of each channel and dynamically adjusts data allocation; S124, based on the intelligent scheduling algorithm, dynamically adjust the transmission strategy according to the real-time network status, and based on the error recovery mechanism, immediately start the retransmission or data repair process when a data transmission error is detected; S125. The central control unit regularly collects performance indicators of each transmission channel, and performs in-depth analysis based on the performance indicators to identify potential bottlenecks and redundancies; and automatically adjusts channel configuration according to the analysis results.

4. The kitchen safety monitoring method based on the multimodal AI large model according to claim 3 is characterized in that: The S121 includes: The edge computing node analyzes the size and complexity of each data type and classifies the data into three levels: high, medium, and low according to the urgency and importance of the data; A transmission channel characteristic database is maintained through the edge computing node. Based on the data feature analysis results and transmission channel characteristics, the edge computing node runs an optimal channel selection algorithm; According to the selected transmission channel, the edge computing node formulates a specific transmission strategy for each data type. During the data transmission process, the edge computing node monitors the usage and performance indicators of each transmission channel in real time. If a channel is found to be congested or the performance is degraded, the dynamic adjustment mechanism is immediately triggered; Based on the real-time monitoring data, the edge computing node fine-tunes the transmission strategy and updates the transmission channel characteristic database through the performance data.

5. The kitchen safety monitoring method based on the multimodal AI large model according to claim 3 is characterized in that: The S123 includes: The edge computing node initializes and allocates transmission channels based on the currently available transmission channel resources; According to the characteristics of the transmission channel, the edge computing node divides the compressed and sorted multimodal data into packets of appropriate size and adds metadata information to each packet; The edge computing node sends the segmented data packets to multiple transmission channels in parallel through a multi-channel transmission protocol, and monitors the usage of each transmission channel in real time through the central control unit; Based on real-time monitoring data, the central control unit dynamically adjusts the data allocation on each channel; the central control unit also regularly evaluates whether the channel configuration is optimal based on the channel's performance indicators; if it is found that the configuration of a channel does not meet the preset standards, the central control unit will automatically adjust the parameters.

6. The kitchen safety monitoring method based on the multimodal AI large model according to claim 1 is characterized in that: The S2 comprises: S21, extracting features from the preprocessed image through a convolutional neural network to identify key information in the image; and performing dimensionality reduction and encoding on the extracted features; S22, analyzing the preprocessed sound signal through a recurrent neural network to extract sound features; and normalizing the sound features; S23, using environmental parameters as auxiliary features, inputting them into the multimodal AI large model together with information from other modalities, and performing feature engineering on the environmental parameters; S24, through the cross-modal matching algorithm, and based on the attention mechanism, the image, sound and environmental parameters are associated to identify the intrinsic connection between different modalities; S25. Use deep learning algorithms to comprehensively analyze the fused multimodal information, identify the operational behavior patterns in the kitchen, and cluster and analyze the identified behavior patterns.

7. The kitchen safety monitoring method based on the multimodal AI large model according to claim 1 is characterized in that: The S3 includes: S31. Use deep learning models to analyze operators’ actions in real time, identify existing violations, classify violations, and take different early warning and intervention measures according to the severity; S32. By analyzing the operator's facial expressions and voice characteristics, the operator's emotional state is judged; when abnormal emotions are detected, early warning information is issued in a timely manner, and corresponding intervention measures are taken; S33. Real-time monitoring of the operating status of equipment in the kitchen; when equipment fails or is abnormal, an early warning message is immediately issued and the corresponding emergency treatment procedures are initiated; S34. Continuously monitor the environmental parameters in the kitchen; when the environmental parameters exceed the normal range, issue early warning information in a timely manner and take corresponding adjustment measures; S35. Design an early warning mechanism based on monitoring results and risk levels, and determine the method for sending early warning information and the recipients; formulate corresponding intervention measures based on the type and severity of the early warning information.

8. The kitchen safety monitoring method based on the multimodal AI large model according to claim 1 is characterized in that: The S4 comprises: S41. The system provides a variety of warning information templates for users to choose or customize; managers or operators set the type, content and sending method of warning information according to their own preferences and needs; S42. Establish a response strategy library, which contains intervention measures and emergency plans corresponding to different warning information; intelligently predict possible consequences based on historical data and current conditions, and take corresponding preventive measures in advance; S43. When receiving warning information, the system automatically selects the best response strategy from the response strategy library and executes it; during the execution process, the system continuously monitors changes in the situation and adjusts the response strategy as needed.

9. The kitchen safety monitoring method based on the multimodal AI large model according to claim 1 is characterized in that: The S5 comprises: S51. Continuously collect kitchen safety monitoring data, and clean and organize the collected data; S52. Use big data analysis technology to conduct in-depth mining of monitoring data, discover potential safety hazards and improvement points, and visualize the analysis results; S53. Based on the collected data and analysis results, the multimodal AI large model is trained and optimized based on the reinforcement learning algorithm; the optimized model is verified and tested, and the model is fine-tuned and improved according to the test results.

10. A kitchen safety monitoring system based on a multimodal AI large model, characterized in that: The system comprises: Data acquisition module: collects multimodal information in the kitchen in real time through the sensor network installed in the kitchen; and pre-processes the collected multimodal information; Feature extraction module: Based on the multimodal AI big model, the pre-processed data is fused and processed to extract the key features of each modal information; and through the deep learning algorithm, the information of different modalities is associated and matched to form a comprehensive understanding of the operation behavior in the kitchen; Behavior monitoring module: Based on the fused multimodal information, it monitors the operation behaviors in the kitchen in real time, analyzes the operator's actions, expressions, sounds and other behaviors, and judges whether there are any violations or safety hazards based on the analysis results; and monitors the equipment status and environmental parameters in the kitchen in real time to discover and warn of potential faults or dangers; Mechanism trigger module: When a violation or safety hazard is detected, the system immediately triggers the early warning mechanism and sends early warning information to management personnel or operators in a variety of ways; and according to the severity of the early warning information, the system automatically takes corresponding intervention measures; Optimization and upgrade module: By continuously collecting and analyzing kitchen safety monitoring data, the multimodal AI large model is continuously optimized and upgraded, and the system functions are expanded and improved based on user feedback and needs.

Images