Intelligent equipment management method and system for 5G Internet of Things

Abstract

The invention discloses an intelligent equipment management method and system for 5G Internet of Things, and relates to the technical field of equipment management, and the method comprises the steps: collecting environment video data and environment audio stream data in real time through 5G network connection and installation of an RGB monitoring camera and a microphone array, and carrying out the noise reduction processing through Gaussian filtering and spectral subtraction, and a network time protocol is configured to realize accurate perception of the environment. An RGB monitoring camera and a microphone array are connected through a 5G network, environment video data and environment audio stream data are collected in real time, noise reduction processing is carried out, environment features are extracted through Canny edge detection, frequency spectrum feature extraction is carried out on the audio data, the extracted features are normalized through a local sliding window, and the image quality is improved. And dynamically selecting a multi-modal response by combining context factors, establishing a management behavior classification framework of the equipment, selecting management behaviors through a decision tree, defining an environment state space and the management behaviors, and optimizing execution of the management behaviors of the equipment through a reward signal.

Description

Technical Field

[0001] The present invention relates to the technical field of device management, and in particular to an intelligent device management method and system for 5G Internet of Things. Background Art

[0002] With the continuous progress of artificial intelligence technology, the field of device management is undergoing a profound transformation, with the application scope continuously expanding, from traditional single-task execution to more complex and dynamic multi-task environments. In various human-computer interaction fields, device management systems need to have the ability to integrate multiple sensory inputs to achieve a higher level of intelligent decision-making and flexible response.

[0003] The existing technologies have the following deficiencies: In current device management, the use of a single type of sensor leads to an incomplete understanding of the environment; in real-time data processing, it is difficult to meet the response requirements in a rapidly changing environment, and noise is often not filtered out in a timely manner, thus affecting the accuracy of decision-making; the relative positions of sensors are not fixed enough, resulting in data asynchronization due to sensor position offset, thus affecting the accuracy of multi-modal data fusion; there is a lack of effective self-learning or adaptation mechanisms, making it difficult to perform dynamic adjustments under different environmental conditions.

[0004] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent device management method and system for 5G Internet of Things to solve the problems in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: An intelligent device management method for 5G Internet of Things, including the following steps: Step 1: Connect and install an RGB monitoring camera and a microphone array through a 5G network to collect environmental video data and environmental audio stream data in real time, and use Gaussian filtering and spectral subtraction for noise reduction processing, and configure the Network Time Protocol to achieve accurate perception of the environment; Step 2: Process static key frames into unified grayscale images and perform brightness, contrast adjustment and target detail enhancement, use Canny edge detection to extract the device environment edge features of static key frames, and extract the spectral features of the device environment audio stream data through short-time Fourier transform and apply Mel frequency conversion and cepstrum analysis; Step 3: Use a local sliding window to normalize the features extracted in Step 2, fuse them into multi-modal features according to weighted average, and dynamically select a multi-modal response environment according to the context factors of key time, device location and device status. Step 4: Establish a classification framework for the management behaviors of key devices, and dynamically select the management behaviors of the devices according to the multi-modal response environment by constructing a decision tree; Step 5: Define the state space set of the environment and all the management behaviors of the device, select the offspring individuals as the management behaviors of the device, set the reward signal for executing the management behaviors, and control the optimal management behavior for the successful execution of the management behaviors of the device by updating the environment state - management behavior value; Preferably, connect and install an RGB monitoring camera around the device through a 5G network to collect device environment video data in real time and extract static key frames, connect and install a microphone array as a multi-audio input point of the device through a 5G network to collect device environment audio stream data in real time and record the time stamp and the position of the multi-audio input point, support network time protocol configuration and connect to an NTP server to select the device master clock source as the time reference for the RGB monitoring camera and the microphone array, fix the relative positions of the RGB monitoring camera and the microphone array to maintain spatial synchronization, use Gaussian filtering to remove the noise of the static key frames for the device environment video data, and use spectral subtraction to remove the background noise for the device environment audio stream data.

[0007] Preferably, convert the static key frames into grayscale images with a fixed size of 224x224 pixels, use grayscale stretching to adjust the brightness and contrast of the static key frames, use sharpening and edge enhancement to highlight the target details in the device environment video data, extract the device environment color features of the static key frames through color space conversion, use Canny edge detection to extract the device environment edge features of the static key frames, separate the multi-audio input points of the microphone array based on beamforming, divide the device environment audio stream data into 20s frames, apply a Hamming window to each frame to reduce the edge effect and extract the spectral features of the device environment audio stream data through short-time Fourier transform, convert the spectral features to the Mel frequency domain through a Mel filter bank, use logarithmic processing to dynamically compress the range of the Mel frequency domain, and perform discrete cosine transform to obtain cepstral coefficients.

[0008] Preferably, set a local sliding window, and normalize the device environment color features, device environment edge features, and cepstral coefficients according to the local mean and local standard deviation. The specific formula is: .

[0009] Among them, represents the normalized device environment color features, device environment edge features, and cepstral coefficients, represents the device environment color features, device environment edge features, and cepstral coefficients currently being normalized, represents the i-th device environment color features, device environment edge features, and cepstral coefficients currently being normalized within the local sliding window, Indicates the number of device environment color features, device environment edge features, and cepstral coefficients contained within a local sliding window. The normalized device environment color features, device environment edge features, and cepstral coefficients are concatenated in chronological order and fused into multi-modal features according to weighted averaging. A multi-modal decision model is defined, and based on context factors such as critical time, device location, and device status, a multi-modal response to the device environment is dynamically selected.

[0010] Preferably, a classification framework for the management behaviors of key devices is established. By mapping the response behaviors of animals and plants to the environment, the management behaviors of the devices are classified and used as parental individuals. Using single-point crossover, a crossover point is randomly selected, and the two parental individuals are cut at the crossover point, and the genes of the parental individuals at the corresponding positions are exchanged to generate two new offspring individuals as the management behaviors of the devices. By repeating the iteration, a sufficient number of offspring individuals are generated, descriptive labels for the management behaviors of the devices are added, and a decision tree is constructed according to the multi-modal response environment to dynamically select the offspring individuals as the management behaviors of the devices.

[0011] Preferably, the device is set as an agent for executing management behaviors, the state space set of the environment and all management behaviors of the device are defined, and the offspring individuals selected as management behaviors are set with reward signals for executing management behaviors. A positive reward signal encourages the offspring individuals whose management behaviors of the device are successful, and a negative reward signal punishes the offspring individuals whose management behaviors of the device fail. The optimal strategy for controlling the management behaviors of the device is controlled by the environment state - management behavior value, and its specific update rule is as follows: 。

[0012] Among them, represents the environment state - management behavior value for executing the management behavior A of the device in the updated environment state S, represents the environment state - management behavior value for executing the management behavior A in the environment state S, represents a parameter between 0 and 1 indicating the centroid of the new environment state relative to the old environment state, represents the effect of a positive reward signal encouraging the successful management behavior A of the device after executing the management behavior A, represents a parameter between 0 and 1 measuring the importance of the reward signal, represents in the new environment state when executing the management behavior the environment state - management behavior value. The feedback of the management behavior of the device in the current new environment state is reflected by updating the list of environment state - management behavior values, and self-learning through the feedback prompts the control device to execute the management behavior in the current environment state.

[0013] A system of an intelligent device management system for 5G Internet of Things, including a data acquisition module, a data processing module, a fusion selection module, a classification selection module, and a control selection module; Data acquisition module: Connect through the 5G network and install RGB monitoring cameras and microphone arrays to collect environmental video data and environmental audio stream data in real time, and use Gaussian filtering and spectral subtraction for noise reduction processing, and configure the Network Time Protocol to achieve precise perception of the environment; Data processing module: Process static key frames into unified grayscale images and perform brightness, contrast adjustment and target detail enhancement, use Canny edge detection to extract the device environment edge features of static key frames, extract the spectral features of device environment audio stream data through short-time Fourier transform and apply Mel frequency conversion and cepstrum analysis; Fusion selection module: Use a local sliding window to normalize the features extracted in the second step, fuse them into multi-modal features according to weighted average, and dynamically select the multi-modal response environment according to the context factors of key time, device location and device status; Classification selection module: Establish a classification framework for the management behaviors of key devices, and construct a decision tree according to the multi-modal response environment to dynamically select the management behaviors of the devices; Control selection module: Define the state space set of the environment and all the management behaviors of the device, select the offspring individuals as the management behaviors of the device to set the reward signal for executing the management behavior, and control the best management behavior for the device to successfully execute the management behavior by updating the environment state-management behavior value.

[0014] In the above technical solution, the technical effects and advantages provided by the present invention: 1. By connecting an RGB surveillance camera and a microphone array through a 5G network, the system can collect environmental video and audio data in real time around the device, ensuring the timeliness of data transmission and enabling environmental monitoring to be carried out almost in real time. The extraction of static key frames by the RGB surveillance camera and the configuration of multiple audio input points of the microphone array enable the system to comprehensively capture environmental changes, ensuring the accuracy and integrity of the data. By supporting the configuration of the Network Time Protocol, it can effectively synchronize the time bases of various devices, further improving the efficiency and consistency of data processing. Gaussian filtering and spectral subtraction are used for noise removal of video data and audio streams, significantly improving the clarity of the signals. The gray conversion, brightness and contrast adjustment of static key frames, and the application of edge enhancement technology can effectively extract the color features and edge features of the device environment, enhancing the recognizability of target details. The processing of the audio stream through the application of the short-time Fourier transform and the Mel filter bank converts the spectral features into the Mel frequency domain, and cepstral coefficients are extracted through logarithmic processing and discrete cosine transform, ensuring high-quality analysis of audio data and enabling the system to respond to environmental changes in real time, providing a solid data basis for subsequent decision-making.

[0015] 2. By establishing a classification framework for the management behaviors of key devices and mapping the management behaviors of the devices to the response behaviors of animals and plants to the environment, intelligent classification and optimization of the management behaviors of the devices are achieved. The offspring individuals generated by using the single-point crossover algorithm can continuously evolve in multiple iterations to form management behaviors of devices with strong adaptability. They can dynamically adjust the management strategies of the devices according to environmental changes and historical data, thereby improving the management efficiency and response ability of the devices. During the decision-making process, optimization is carried out according to the environmental state and the reward signal of the management behavior, which can effectively reflect the effect of the management behavior of the device in the current environment and promote the selection of the optimal management strategy of the device in different environmental states. Through the fusion and weighted average of multi-modal features, the system can dynamically select the optimal management behavior according to the context factors of key time, device location, and device status, not only improving the intelligent level of device management but also enhancing the adaptability of the system to complex environments, and finally achieving efficient and intelligent device management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0017] Figure 1 It is a method flow chart of an intelligent device management method for a 5G Internet of Things according to the present invention.

[0018] Figure 2 This is the system structure diagram of an intelligent device management system for 5G Internet of Things according to the present invention. Detailed implementation manners

[0019] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that this disclosure will be more complete and thorough, and will fully convey the concept of the example embodiments to those skilled in the art.

[0020] The present invention provides an intelligent device management method for 5G Internet of Things as Figure 1 shown, including the following steps: Step 1: Connect and install an RGB monitoring camera and a microphone array through a 5G network to collect environmental video data and environmental audio stream data in real time, and use Gaussian filtering and spectral subtraction for noise reduction processing, and configure the Network Time Protocol to achieve accurate perception of the environment; Connect and install an RGB monitoring camera around the device through a 5G network to collect device environmental video data in real time and extract static key frames. Connect and install a microphone array as a multi-audio input point of the device through a 5G network to collect device environmental audio stream data in real time and record the time stamp and the position of the multi-audio input point. Support the configuration of the Network Time Protocol and connect to an NTP server to select the device master clock source as the time reference for the RGB monitoring camera and the microphone array. Fix the relative positions of the RGB monitoring camera and the microphone array to maintain spatial synchronization. Use Gaussian filtering to remove the noise of the static key frames for the device environmental video data, and use spectral subtraction to remove the background noise for the device environmental audio stream data.

[0021] Step 2: Process the static key frames into unified grayscale images and perform brightness, contrast adjustment and target detail enhancement. Use Canny edge detection to extract the device environmental edge features of the static key frames. Extract the spectral features of the device environmental audio stream data through short-time Fourier transform and apply Mel frequency conversion and cepstrum analysis; Convert the static key frame to grayscale and fix the size at 224x224 pixels. Adjust the brightness and contrast of the static key frame using grayscale stretching. Highlight the target details in the device environment video data using sharpening and edge enhancement. Extract the device environment color features of the static key frame through color space conversion. Extract the device environment edge features of the static key frame using Canny edge detection. Separate the multi-audio input points of the microphone array based on beamforming. Divide the device environment audio stream data into 20s frames. Apply a Hamming window to each frame to reduce edge effects and extract the spectral features of the device environment audio stream data through short-time Fourier transform. Convert the spectral features to the Mel frequency domain through a Mel filter bank. Use logarithmic processing to dynamically compress the range of the Mel frequency domain and obtain the cepstral coefficients through discrete cosine transform.

[0022] Step 3: Normalize the features extracted in Step 2 using a local sliding window, fuse them into multi-modal features according to weighted average, and dynamically select the multi-modal response environment based on context factors such as critical time, device location, and device status. Set a local sliding window, and normalize the device environment color features, device environment edge features, and cepstral coefficients according to the local mean and local standard deviation. The specific formula is: 。

[0023] Among them, represents the normalized device environment color features, device environment edge features, and cepstral coefficients. represents the device environment color features, device environment edge features, and cepstral coefficients currently being normalized. represents the i-th device environment color features, device environment edge features, and cepstral coefficients currently being normalized within the local sliding window. represents the number of device environment color features, device environment edge features, and cepstral coefficients included within the local sliding window. Concatenate the normalized device environment color features, device environment edge features, and cepstral coefficients in chronological order and fuse them into multi-modal features according to weighted average. Define a multi-modal decision model and dynamically select the multi-modal response device environment based on context factors such as critical time, device location, and device status.

[0024] Step 4: Establish a management behavior classification framework for key devices, and construct a decision tree based on the multi-modal response environment to dynamically select the management behavior of the device. Establish a classification framework for the management behaviors of key equipment. Classify the management behaviors of the equipment by mapping the response behaviors of animals and plants to the environment and use them as parental individuals. Randomly select a crossover point using single-point crossover, cut the two parental individuals at the crossover point, exchange the genes of the parental individuals at the corresponding positions, generate two new offspring individuals as the management behaviors of the equipment. Generate a sufficient number of offspring individuals through repeated iteration, add descriptive tags to the management behaviors of the equipment, and dynamically select the offspring individuals as the management behaviors of the equipment according to the multi-modal response environment.

[0025] Step 5: Define the state space set of the environment and all the management behaviors of the equipment. Select the offspring individuals as the management behaviors of the equipment and set the reward signal for executing the management behavior. Control the best management behavior for the successful execution of the management behavior of the equipment by updating the environment state-management behavior value; Set the equipment as an agent for executing the management behavior. Define the state space set of the environment and all the management behaviors of the equipment. Select the offspring individuals as the management behaviors of the equipment and set the reward signal for executing the management behavior. The positive reward signal encourages the offspring individuals for the successful management behavior of the equipment, and the negative reward signal punishes the offspring individuals for the failed management behavior of the equipment. Control the best strategy for the management behavior of the equipment through the environment state-management behavior value. The specific update rule is as follows: 。

[0026] Among them, represents the environment state-management behavior value for executing the management behavior A of the equipment in the environment state S after update, represents the environment state-management behavior value for executing the management behavior A of the equipment in the environment state S, represents a parameter between 0 and 1 indicating the centroid of the new environment state relative to the old environment state, represents the effect of the positive reward signal encouraging the successful management behavior A of the equipment after executing the management behavior A of the equipment, represents a parameter between 0 and 1 measuring the importance of the reward signal, represents in the new environment state under the execution of the management behavior of the equipment of the environment state-management behavior value. Reflect the feedback of the management behavior of the equipment in the current new environment state through the updated list of the environment state-management behavior value, and promote the control equipment to execute the management behavior of the current environment state through feedback self-learning.

[0027] In step one of the intelligent device management method for 5G Internet of Things in this embodiment, in a noisy environment, in the video frames captured by the RGB monitoring camera, the background noise and light changes cause image blurring. After applying Gaussian filtering, the noise of the static key frames is significantly reduced and the image clarity is improved. In the audio signals recorded by the microphone array in a noisy environment, a large amount of background noise is included. After spectral subtraction processing, the background noise in the audio signals is effectively removed and the voice signals are clearer.

[0028] In step two of the intelligent device management method for 5G Internet of Things in this embodiment, by converting the static key frames into unified grayscale images, the computational complexity is reduced. By adjusting the brightness and contrast, the features of the images become more obvious. By using sharpening and edge enhancement techniques, the details of the environmental objects are highlighted to provide clearer features. The Canny edge detection is used to effectively extract the edge features of the device environment, providing basic information for underlying visual understanding. The short-time Fourier transform is used to extract the spectral features of the audio stream data of the device environment, capturing the information in the time-frequency domain.

[0029] In step three of the intelligent device management method for 5G Internet of Things in this embodiment, normalization reduces the differences between different features, improves the compatibility of modalities, and accelerates the efficiency of feature fusion. The dynamic selection mechanism can respond more sensitively to environmental changes, improving the real-time performance and accuracy of decision-making. The multi-modal features after weighted average fusion combine the advantages of different information sources, making the decision-making more comprehensive. Assume that the current device environment color features are [150, 160, 155, 165, 170], the device environment edge features are [30, 35, 33, 40, 42], and the cepstral coefficients are [0.5, 0.55, 0.52, 0.58, 0.6]. Set the local sliding window size to 5. Then, after calculation by the normalization formula, the device environment color features are [−1.27, −0.634, −0.634, 0.634, 1.27], the device environment edge features are [−1.26, −0.21, −0.042, 0.84, 1.26], and the cepstral coefficients are [−1.86, 0.27, −0.15, 0.14, 1.27]. Assume that the weights set for the environmental color information features, environmental object edge features, and cepstral coefficients are 0.4, 0.3, and 0.3 respectively. Then the multi-modal features are [−1.444, −0.236, −0.312, 0.548, 1.267].

[0030] In step 4 of the intelligent device management method for 5G Internet of Things in this embodiment, the decision tree based on multi-modal data enables the device to more flexibly adapt to environmental changes, select appropriate management behaviors. The offspring individuals generated by the genetic algorithm have diverse behaviors, enhancing the adaptability of device control. Through repeated iteration, the behaviors of the offspring individuals can be gradually optimized, improving the efficiency of the device in complex environments. Through this process, the device can continuously optimize its decision-making ability and management behaviors based on observing and learning the behaviors of animals and plants, and better respond to the control of its environment.

[0031] In step 5 of the intelligent device management method for 5G Internet of Things in this embodiment, its behavior is continuously adjusted according to the feedback of the environment to find the most suitable behavior strategy in different environments. Assume that the environmental state - management behavior values of the device are as shown in the table.

[0032] Assume that the device obtains a positive reward signal +1 and transfers to a new environmental state S2. When the current device is in environmental state S1 and selects to execute management behavior A1, set α = 0.1 and γ = 0.9. , then its , through these updates, the device will be more inclined to execute management behavior A1 when facing environmental state S1, thereby gradually optimizing its management behavior. Over time and with the accumulation of more interactions, it can thus make the best management behavior control in different environments.

[0033] The present invention provides a system of an intelligent device management system for 5G Internet of Things as Figure 2 shown, including a data acquisition module, a data processing module, a fusion selection module, a classification selection module, and a control selection module; Data acquisition module: Connect through a 5G network and install an RGB monitoring camera and a microphone array to collect environmental video data and environmental audio stream data in real time, and perform noise reduction processing using Gaussian filtering and spectral subtraction. Configure the Network Time Protocol to achieve precise perception of the environment; Data processing module: Process static key frames into unified grayscale images and perform brightness, contrast adjustment, and target detail enhancement. Use Canny edge detection to extract the edge features of the device environment in static key frames, and extract the spectral features of the device environment audio stream data through short-time Fourier transform and apply Mel frequency conversion and cepstrum analysis; Fusion selection module: Use a local sliding window to normalize the features extracted in step 2, fuse them into multi-modal features according to weighted average, and dynamically select multi-modal responses to the environment according to context factors such as key time, device location, and device state; Classification Selection Module: Establish a classification framework for the management behaviors of key devices, and dynamically select the management behaviors of devices according to the multi-modal response environment by constructing a decision tree. Control Selection Module: Define the state space set of the environment and all the management behaviors of the device, select the offspring individuals as the management behaviors of the device, set the reward signals for executing the management behaviors, and control the best management behaviors for the successful execution of the management behaviors of the device by updating the environment state-management behavior values.

[0034] By connecting the RGB monitoring camera and the microphone array through the 5G network, the system can collect environmental video and audio data in real time around the device, ensuring the timeliness of data transmission and enabling environmental monitoring to be carried out almost in real time. The extraction of static key frames of the RGB monitoring camera and the configuration of multiple audio input points of the microphone array enable the system to comprehensively capture environmental changes, ensuring the accuracy and integrity of the data. By supporting the Network Time Protocol configuration, the time reference of each device can be effectively synchronized, further improving the efficiency and consistency of data processing. Gaussian filtering and spectral subtraction are used for noise removal of video data and audio streams, significantly improving the clarity of the signals. The gray conversion, brightness and contrast adjustment of static key frames, and the application of edge enhancement technology can effectively extract the color features and edge features of the device environment, enhancing the recognizability of target details. The processing of audio streams is carried out through the application of the short-time Fourier transform and the Mel filter bank, converting the spectral features into the Mel frequency domain, and extracting the cepstral coefficients through logarithmic processing and discrete cosine transform, ensuring the high-quality analysis of audio data, enabling the system to respond to environmental changes in real time, and providing a solid data basis for subsequent decision-making.

[0035] By establishing a classification framework for the management behaviors of key devices, mapping the management behaviors of the device to the response behaviors of animals and plants to the environment, the intelligent classification and optimization of the management behaviors of the device are realized. The offspring individuals generated by using the single-point crossover algorithm can continuously evolve in multiple iterations, forming highly adaptable management behaviors of the device. They can dynamically adjust the management strategy of the device according to environmental changes and historical data, thereby improving the management efficiency and response ability of the device. In the decision-making process, optimization is carried out according to the environmental state and the reward signals of the management behaviors, which can effectively reflect the effect of the management behaviors of the device in the current environment and promote the selection of the best management strategy of the device in different environmental states. Through the fusion and weighted average of multi-modal features, the system can dynamically select the optimal management behavior according to the context factors of key time, device location, and device state, not only improving the intelligent level of device management but also enhancing the adaptability of the system to complex environments, and ultimately realizing efficient and intelligent device management.

[0036] Only some exemplary embodiments of the present invention have been described above by way of illustration. Undoubtedly, for those of ordinary skill in the art, various different modifications can be made to the described embodiments without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An intelligent device management method for 5G Internet of Things, characterized in that: It includes the following steps: Step 1: Connect and install an RGB monitoring camera and a microphone array through a 5G network to collect environmental video data and environmental audio stream data in real time, and use Gaussian filtering and spectral subtraction for noise reduction processing. Configure the Network Time Protocol to achieve precise perception of the environment; Step 2: Process the static key frames into unified grayscale images and adjust the brightness, contrast, and enhance the target details. Use Canny edge detection to extract the device environment edge features of the static key frames. Extract the spectral features of the device environment audio stream data through short-time Fourier transform and apply Mel frequency conversion and cepstrum analysis; Step 3: Use a local sliding window to normalize the features extracted in Step 2, fuse them into multi-modal features according to weighted averaging, and dynamically select the multi-modal response environment according to the context factors of key time, device location, and device status; Step 4: Establish a management behavior classification framework for key devices, and construct a decision tree according to the multi-modal response environment to dynamically select the management behavior of the device; Step 5: Define the state space set of the environment and all management behaviors of the device. Select the offspring individuals as the management behaviors of the device and set the reward signals for executing the management behaviors. Control the best management behavior for the device to successfully execute the management behavior by updating the environment state-management behavior value.

2. The intelligent device management method for 5G Internet of Things according to claim 1, characterized in that: In Step 1, connect and install an RGB monitoring camera through a 5G network around the device to collect device environment video data in real time and extract static key frames. Connect and install a microphone array through a 5G network as the multi-audio input point of the device to collect device environment audio stream data in real time and record the time stamp and the location of the multi-audio input point. Support the configuration of the Network Time Protocol and connect to the NTP server to select the device's main clock source as the time reference for the RGB monitoring camera and the microphone array. Fix the relative positions of the RGB monitoring camera and the microphone array to maintain spatial synchronization. Use Gaussian filtering to remove the noise of the static key frames for the device environment video data, and use spectral subtraction to remove the background noise for the device environment audio stream data.

3. The intelligent device management method for 5G Internet of Things according to claim 1, characterized in that: In Step 2, convert the static key frames into grayscale images with a fixed size of 224x224 pixels. Use grayscale stretching to adjust the brightness and contrast of the static key frames. Use sharpening and edge enhancement to highlight the target details in the device environment video data. Extract the device environment color features of the static key frames through color space conversion. Use Canny edge detection to extract the device environment edge features of the static key frames. Based on beamforming, separate the multi-audio input points of the microphone array. Divide the device environment audio stream data into 20s frames. Apply a Hamming window to each frame to reduce the edge effect and extract the spectral features of the device environment audio stream data through short-time Fourier transform. Convert the spectral features to the Mel frequency domain through a Mel filter bank. Use logarithmic processing to dynamically compress the range of the Mel frequency domain, and perform discrete cosine transform to obtain cepstral coefficients.

4. An intelligent device management method for 5G Internet of Things according to claim 1, characterized in that: In step 3, the features extracted in step 2 are normalized using the local mean and local standard deviation of the local sliding window. The normalized device environment color features, device environment edge features, and cepstral coefficients are concatenated in chronological order and fused into multi-modal features according to weighted average. A multi-modal decision model is defined, and a multi-modal response device environment is dynamically selected according to the context factors of key time, device location, and device status.

5. The intelligent device management method for 5G Internet of Things according to claim 4, characterized in that: The specific formula for the normalization is as follows: ； Among them, represents the normalized device environment color feature, device environment edge feature, and cepstral coefficient, represents the device environment color feature, device environment edge feature, and cepstral coefficient currently being normalized, represents the i-th device environment color feature, device environment edge feature, and cepstral coefficient currently being normalized within the local sliding window, represents the number of device environment color features, device environment edge features, and cepstral coefficients included in the local sliding window. Represents the number of device environment color features, device environment edge features, and cepstral coefficients included in the local sliding window.

6. The intelligent device management method for 5G Internet of Things according to claim 1, characterized in that: In step 4, a classification framework for the management behaviors of key devices is established. The management behaviors of the devices are classified by mapping the reaction behaviors of animals and plants to the environment and used as parent individuals. A single-point crossover is used to randomly select a crossover point, and the two parent individuals are cut at the crossover point, and the genes of the parent individuals at the corresponding positions are exchanged to generate two new offspring individuals as the management behaviors of the devices. By repeating the iteration, a sufficient number of offspring individuals are generated, descriptive labels of the management behaviors of the devices are added, and a decision tree is constructed according to the multi-modal response environment to dynamically select the offspring individuals as the management behaviors of the devices.

7. An intelligent device management method for 5G Internet of Things according to claim 1, characterized in that: In step 5, the reward signals for the offspring individuals to execute the management behaviors are specifically as follows: positive reward signals encourage the offspring individuals with successful management behaviors of the devices, and negative reward signals punish the offspring individuals with failed management behaviors of the devices. The optimal strategy for controlling the management behaviors of the devices is controlled by the environment state - management behavior value. The feedback of the management behaviors of the devices in the current new environment state is reflected by updating the list of the environment state - management behavior values, and the control device is prompted to execute the management behaviors in the current environment state through feedback self-learning.

8. The intelligent device management method for 5G Internet of Things according to claim 7, characterized in that: The specific update rule for the optimal strategy of controlling the management behaviors of the devices by the environment state - management behavior value is as follows: ； Among them, represents the environmental state - management behavior value for performing the management behavior A of the device in the environmental state S after update, represents the environmental state - management behavior value for performing the management behavior A of the device in the environmental state S, represents a parameter between 0 and 1 indicating the centroid of the new environmental state relative to the old environmental state, represents the effect that after performing the management behavior A of the device, the positive reward signal encourages the success of the management behavior A of the device, represents a parameter between 0 and 1 measuring the importance of the reward signal, represents in the new environmental state for performing the management behavior of the device of the environmental state - management behavior value.

9. An intelligent device management system for 5G Internet of Things, which is used to implement the method for managing intelligent devices for 5G Internet of Things described in any one of the above claims 1-8, characterized in that, It includes a data acquisition module, a data processing module, a fusion selection module, a classification selection module, and a control selection module; Data acquisition module: Connect through the 5G network and install RGB monitoring cameras and microphone arrays to collect environmental video data and environmental audio stream data in real time, and perform noise reduction processing using Gaussian filtering and spectral subtraction. Configure the network time protocol to achieve precise perception of the environment; Data processing module: Process the static key frames into unified grayscale images and perform brightness, contrast adjustment, and target detail enhancement. Use Canny edge detection to extract the device environment edge features of the static key frames, and extract the spectral features of the device environment audio stream data through short-time Fourier transform and apply Mel frequency conversion and cepstral analysis; Fusion selection module: Normalize the features extracted in step 2 using the local sliding window, fuse them into multi-modal features according to weighted average, and dynamically select a multi-modal response environment according to the context factors of key time, device location, and device status; Classification selection module: Establish a classification framework for the management behaviors of key devices, and dynamically select the management behaviors of the devices by constructing a decision tree according to the multi-modal response environment; Control selection module: Define the state space set of the environment and all management behaviors of the device, select the offspring individuals that are the management behaviors of the device, set the reward signal for executing the management behaviors, and control the best management behavior for the device to successfully execute the management behaviors by updating the environment state - management behavior values.

Images