Titanium alloy smart home temperature control method and system

Through the improved DeepLab algorithm and LSTM network, combined with titanium alloy heat sink, real-time temperature regulation of smart home temperature control system is achieved, solving the problems of inaccurate temperature regulation and energy waste in the existing system, and improving the accuracy of temperature regulation and efficiency.

CN119006977APending Publication Date: 2024-11-22SHENZHEN NAO JIANGLAN INVESTMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411074859.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing smart home temperature control system is difficult to accurately control indoor temperatures, and cannot predict temperature demands based on historical data and real-time human activities, resulting in inaccurate temperature regulation and waste of energy.

Method used

Using the improved DeepLab algorithm and LSTM long and short-term memory network, the human body's position and activity state are identified through real-time video analysis, feature vectors are generated and time series analysis is performed, and the working state of the titanium alloy heat sink is dynamically adjusted to achieve temperature adjustment.

Benefits of technology

Accurate adjustment of indoor temperature is achieved, user experience and energy efficiency is improved, and the accuracy and efficiency of temperature adjustment is improved by adaptively adjusting the hollow rate and selecting heat sinks for variable materials at different stages.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006977A_ABST
    Figure CN119006977A_ABST
Patent Text Reader

Abstract

The invention discloses a titanium alloy smart home temperature control method and system. The method comprises the steps that S1, a data acquisition unit acquires indoor data in real time; s2, the server performs semantic segmentation on the indoor video frame image by using an improved DeepLab algorithm, and integrates features extracted from a segmented human body region and corresponding time and temperature data into a feature vector Ft; s3, inputting the feature vector into a trained LSTM (Long Short-Term Memory) network for time sequence analysis to obtain an expected user target temperature range; and S4, the control unit compares the monitored environment temperature with the expected user target temperature range, and then the working state of the titanium alloy cooling fin is adjusted. According to the method, the deep learning algorithm DeepLab is used for carrying out real-time video analysis, information such as the position, the activity state and the number of residents is recognized, and the temperature requirements of the residents are predicted in real time in combination with the long-short-term memory network, so that the working state of the titanium alloy cooling fins is dynamically adjusted, and the optimal temperature adjusting effect is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of smart home technology, and in particular to a titanium alloy smart home temperature control method and system. Background Art

[0002] In modern living environments, the efficiency and intelligence of temperature control systems are important factors in improving living comfort and energy efficiency. Traditional home temperature control systems rely on simple thermodynamic principles, such as using air conditioners and heating equipment to manually adjust indoor temperature. These systems are usually slow to respond, consume a lot of energy, and it is difficult to accurately control the temperature level to personal preference. Therefore, with the development of smart home technology, there is an urgent need for an intelligent temperature control solution that can respond to environmental changes in real time and adjust the temperature in a personalized way.

[0003] In recent years, titanium alloys have been widely used in various industrial fields due to their excellent mechanical properties and corrosion resistance. In the smart home temperature control system, titanium alloys are used as heat sink materials to achieve more efficient and rapid heat transfer by taking advantage of their high thermal conductivity and good thermal stability. In addition, the lightweight and high-strength characteristics of titanium alloy heat sinks also make the design of temperature control systems more flexible and durable. The development of titanium alloy smart home temperature control methods combines the latest achievements in materials science, information technology and communication technology, and provides a practical technical solution for achieving a more comfortable, energy-saving and intelligent living environment. Through the integrated application of these technologies, future home temperature control systems will be more humane, efficient and environmentally friendly.

[0004] In order to further improve the intelligence and energy efficiency of the temperature control system, researchers began to explore the combination of high-tech materials such as phase change materials and advanced information technology. Phase change materials can absorb or release a large amount of heat energy at a fixed temperature, and adjust the ambient temperature through physical state changes (such as from solid to liquid), so that the indoor temperature is maintained in a relatively stable and comfortable range. In terms of information technology, the smart home system integrates a large number of sensors and data processing units, enabling the system to monitor and analyze indoor and outdoor temperature, humidity, human activity and other data in real time.

[0005] However, the existing temperature adjustment is inaccurate and it is difficult to accurately control the temperature level of personal preference, and it is not possible to achieve accurate and energy-saving temperature control. For example, the existing technology basically does not consider the automatic judgment of the appropriate temperature for the current active group based on historical training results, and the temperature cannot be adjusted in a targeted manner with different number of people and exercise types, resulting in inaccurate prediction results; and the existing heat sink structure is single and cannot achieve precise temperature control, resulting in increasing temperature adjustment errors instead of decreasing, which not only affects the user experience but also leads to a significant decrease in energy saving efficiency; and the existing DeepLab algorithm has low accuracy in the process of extracting and identifying the moving human body area, and cannot automatically adjust according to the on-site situation. A new solution is urgently needed for high-accuracy judgment and automated processing methods to improve processing efficiency and comfort. Summary of the invention

[0006] In view of the above problems mentioned in the prior art, the present invention provides a titanium alloy smart home temperature control method and system. In the method, first, a data acquisition unit acquires indoor data in real time, including temperature sensors arranged at multiple positions indoors to monitor ambient temperature data and acquire indoor video frame image data in real time, and transmits the acquired indoor data to a server; secondly, the server uses an improved DeepLab algorithm to perform semantic segmentation on the indoor video frame image to obtain the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F t ; Again, the eigenvector F t The data is input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range; finally, the control unit compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result. This application uses deep learning algorithms such as DeepLab for real-time video analysis, and the system can identify information such as the location, activity status, and number of residents. Combined with the long short-term memory network (LSTM), these data can be converted into a prediction model to predict the temperature requirements of the residents in real time, thereby dynamically adjusting the working state of the titanium alloy heat sink to achieve the optimal temperature regulation effect.

[0007] The present application provides a titanium alloy smart home temperature control method, comprising the steps of:

[0008] S1: The data acquisition unit collects indoor data in real time, including temperature sensors arranged at multiple locations indoors to monitor ambient temperature data and collect indoor video frame image data in real time, and transmits the collected indoor data to the server;

[0009] S2: The server uses the improved DeepLab algorithm to perform semantic segmentation on the indoor video frame images, obtains the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F t ;

[0010] S21: performing semantic segmentation on each frame of the indoor video frame image;

[0011] S22: Extract features from the semantic segmentation results, including human position, action type, and indoor number features; and combine the extracted features with the corresponding time and temperature data to form a feature vector F t ;

[0012] S3: The feature vector F t Input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range;

[0013] S31: Initialize the parameters of the LSTM long short-term memory network;

[0014] S32: Combine the features of consecutive frames into time series data X LSYM ;X LSTM =[F1,F2,...F T ], where T represents the total number of frames in the video sequence;

[0015] S33: Convert the time series data X LSTM Input to the LSTM long short-term memory network for forward propagation; the LSTM long short-term memory network outputs the expected user target temperature range;

[0016] S4: The control unit compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result; including: when the monitored ambient temperature is higher than the maximum value of the expected user target temperature range, the control unit increases the surface area of ​​the titanium alloy heat sink, increases the fan speed, and enhances the coolant flow of the heat sink; when the monitored ambient temperature is lower than the minimum value of the expected user target temperature range, the control unit reduces the surface area of ​​the heat sink, reduces the fan speed, and reduces the coolant flow.

[0017] Preferably, the collected indoor data is transmitted to the server, including: using a 5G network or a Wi-Fi network to transmit the data in real time to a remote server data center, and storing the video data in a cloud storage server.

[0018] Preferably, the step S21: performing semantic segmentation on each frame of indoor video frame image includes S211: using a convolutional neural network to extract features of the input indoor video frame image to obtain F CNN ; S212: Use dilated convolution for FCNN Convolution operation is performed to maintain spatial resolution and capture multi-scale information, and F dilated ;

[0019]

[0020] Among them, F CNN The feature map extracted by the convolutional neural network is used as the input of the dilated convolution, F dilated is the feature map output by the dilated convolution; W is the convolution kernel, whose size is (2a+1)*(2b+1); d is the dilation rate, which is used to control the size of the dilated convolution; x, y are the horizontal and vertical coordinates in the feature map, respectively; i, j are the horizontal and vertical coordinates of the position in the convolution kernel, respectively, traversing the convolution kernel space range; a, b represent the parameter values ​​of the convolution kernel space range, respectively; d0 represents the dilation rate when the target size is equal to the benchmark size; N is the number of human body regions; s k is the size of the kth human body region; s0 is the set reference value; the larger the human body region, the larger the void ratio d;

[0021] S213: Use spatial pyramid pooling to F dilated Capture information of different scales and obtain multi-scale features F spp ; S214: Multi-scale feature F through upsampling and deconvolution spp The original spatial resolution is restored and a category label is assigned to each pixel, outputting the semantic segmentation result F t .

[0022] Preferably, in S213: spatial pyramid pooling is used to pool F dilated Capture information of different scales and obtain multi-scale features F spp , including: first creating pyramid levels; secondly performing the maximum pooling operation on the input feature map at each level; and finally merging the pooling results of all levels into a long vector to obtain the multi-scale feature F spp .

[0023] Preferably, the S214: performing upsampling and deconvolution on the multi-scale feature F spp The original spatial resolution is restored and a category label is assigned to each pixel, outputting the semantic segmentation result F t , including: first, using a linear interpolation algorithm to upsample and increase the image size, and then performing deconvolution, which is the inverse operation of convolution, to increase the spatial resolution of the feature map.

[0024] Preferably, the step S22: extracting features from the semantic segmentation results, including human body position, action type, and indoor number features; and combining the extracted features with the corresponding time and temperature data to form a feature vector F t, where the human body position includes the area of ​​the human body region and the bounding box coordinates; the action types include sitting, walking, running, yoga, dancing, playing table tennis, playing badminton, and playing basketball; the indoor population characteristics include the number of people and the change in the number of people.

[0025] Preferably, the working state of the titanium alloy heat sink is adjusted based on the comparison result, and the control unit automatically selects the first titanium alloy heat sink or the second titanium alloy heat sink or the third titanium alloy heat sink or the fourth titanium alloy heat sink according to the deviation between the ambient temperature and the target temperature range and the size of the set threshold, wherein the first titanium alloy heat sink contains the phase change material capric acid, which has a phase change at 29.8°C, and the phase change latent heat is 190J / g; the second titanium alloy heat sink contains the phase change material lauric acid, which has a phase change at about 43°C, and the phase change latent heat is 178J / g; the third titanium alloy heat sink contains the phase change material sodium sulfate decahydrate, which has a phase change at 32°C, and the phase change latent heat is about 250J / g; the fourth titanium alloy heat sink contains the phase change material p-hydroxybenzoic acid, which has a phase change at 25.0°C, and the phase change latent heat is 189J / g; the phase change material is encapsulated in microcapsules, the microcapsules are evenly distributed in the titanium alloy matrix, and the shell of the microcapsules is composed of polyphenylene sulfide material.

[0026] The present application also provides a titanium alloy smart home temperature control system, comprising:

[0027] A data acquisition unit is used to collect indoor data in real time, including temperature sensors arranged at multiple locations indoors to monitor ambient temperature data and collect indoor video frame image data in real time, and transmit the collected indoor data to a server;

[0028] In the feature vector generation module, the server uses the improved DeepLab algorithm to perform semantic segmentation on the indoor video frame images, obtains the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F t ;

[0029] A semantic segmentation module performs semantic segmentation on each frame of indoor video;

[0030] The feature extraction module extracts features from the semantic segmentation results, including human position, action type, and indoor number of people; and combines the extracted features with the corresponding time and temperature data to form a feature vector F t ;

[0031] The module for calculating the expected user target temperature range converts the feature vector F t Input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range;

[0032] Initialization module, initializes the parameters of LSTM long short-term memory network;

[0033] The time series data generation module combines the features of consecutive frames into time series data X LSTM ;X LSTM =[F1,F2,...F T ], where T represents the total number of frames in the video sequence;

[0034] LSTM long short-term memory network processing module, time series data X LSTM Input to the LSTM long short-term memory network for forward propagation; the LSTM long short-term memory network outputs the expected user target temperature range;

[0035] The judgment and adjustment module comprises a control unit which compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result; including: when the monitored ambient temperature is higher than the maximum value of the expected user target temperature range, the control unit increases the surface area of ​​the titanium alloy heat sink, increases the fan speed, and enhances the coolant flow of the heat sink; when the monitored ambient temperature is lower than the minimum value of the expected user target temperature range, the control unit reduces the surface area of ​​the heat sink, reduces the fan speed, and reduces the coolant flow.

[0036] Preferably, the collected indoor data is transmitted to the server, including: using a 5G network or a Wi-Fi network to transmit the data in real time to a remote server data center, and storing the video data in a cloud storage server.

[0037] Preferably, the working state of the titanium alloy heat sink is adjusted based on the comparison result, and the control unit automatically selects the first titanium alloy heat sink or the second titanium alloy heat sink or the third titanium alloy heat sink or the fourth titanium alloy heat sink according to the deviation between the ambient temperature and the target temperature range and the size of the set threshold, wherein the first titanium alloy heat sink contains the phase change material capric acid, which has a phase change at 29.8°C, and the phase change latent heat is 190J / g; the second titanium alloy heat sink contains the phase change material lauric acid, which has a phase change at about 43°C, and the phase change latent heat is 178J / g; the third titanium alloy heat sink contains the phase change material sodium sulfate decahydrate, which has a phase change at 32°C, and the phase change latent heat is about 250J / g; the fourth titanium alloy heat sink contains the phase change material p-hydroxybenzoic acid, which has a phase change at 25.0°C, and the phase change latent heat is 189J / g; the phase change material is encapsulated in microcapsules, the microcapsules are evenly distributed in the titanium alloy matrix, and the shell of the microcapsules is composed of polyphenylene sulfide material.

[0038] The present invention provides a titanium alloy smart home temperature control method and system, which can achieve the following beneficial technical effects:

[0039] 1. In the present invention, the data acquisition unit first collects indoor data in real time; secondly, the server uses the improved DeepLab algorithm to perform semantic segmentation on the indoor video frame image, obtains the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F t ; Again, the eigenvector F t The data is input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range; finally, the control unit compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result. This application uses deep learning algorithms such as DeepLab for real-time video analysis, and the system can identify information such as the location, activity status, and number of residents. Combined with the long short-term memory network (LSTM), these data can be converted into a prediction model to predict the temperature requirements of the residents in real time, thereby dynamically adjusting the working state of the titanium alloy heat sink to achieve the optimal temperature regulation effect.

[0040] 2. The present invention adopts the improved DeepLab algorithm in the process of human body area extraction, introduces the human body area into the calculation of the dilation rate, and dynamically adjusts the dilation rate according to the human body area. The dilation rate d is used to control the size of the dilation in the dilation convolution. The larger the human body area, the larger the dilation rate d. The adaptive adjustment of the dilation rate is realized, which greatly enhances the accurate extraction of the human body area and the accurate judgment of the motion situation.

[0041] 3. The control unit of the present invention automatically selects the first titanium alloy heat sink, the second titanium alloy heat sink, the third titanium alloy heat sink, or the fourth titanium alloy heat sink according to the deviation between the ambient temperature and the target temperature range and the size of the set threshold. The step-change material is encapsulated in microcapsules, and the microcapsules are evenly distributed in the titanium alloy matrix. The shell of the microcapsule is composed of polyphenylene sulfide material. The titanium alloy heat sink is selected according to different deviation conditions, and the temperature is adjusted according to the characteristics of the step-change material, which greatly improves the temperature adjustment accuracy and efficiency and effectively reduces energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 It is a schematic diagram of the steps of a titanium alloy smart home temperature control method of the present invention;

[0044] Figure 2It is a schematic diagram of a titanium alloy smart home temperature control system of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Embodiment 1:

[0047] In view of the above problems mentioned in the prior art, in order to solve the above technical problems, as shown in the attached Figure 1 As shown: This application provides a titanium alloy smart home temperature control method, including the steps of:

[0048] S1: The data acquisition unit collects indoor data in real time, including temperature sensors arranged at multiple locations indoors to monitor ambient temperature data and collect indoor video frame image data in real time, and transmits the collected indoor data to the server; in one embodiment, at least one temperature sensor is installed in each room (such as a gym, badminton hall, indoor basketball court, living room, kitchen, master bedroom, children's room and study), and each sensor can measure the temperature once a minute and send data to the central server in real time. In the main activity areas, such as the gym, badminton hall, indoor basketball court, living room, kitchen and corridor, a high-resolution video camera is installed. The camera is installed near the ceiling to cover a wide range of viewing angles to ensure that human activities in the room can be clearly captured. The camera collects video frame images in real time and transmits the video stream to the server in real time. All sensors and cameras are connected to the central server of the home via a Wi-Fi network. This arrangement ensures low data transmission latency and does not rely on external Internet connections. The server is equipped with a high-speed data processing unit that can receive, store and analyze large amounts of data from various devices. The server uses an improved DeepLab algorithm to process the video frame images received from the camera and performs semantic segmentation to identify and track human activities and positions in the room. At the same time, the server integrates the ambient temperature data from various temperature sensors and combines it with the video data analysis results to form a feature vector. The feature vector is input into the trained LSTM long short-term memory network to predict the optimal temperature setting for each room or area. According to the temperature range output by the LSTM, the server dynamically adjusts the working status of the heat sink in each room, such as adjusting the heat sink surface area, fan speed and coolant flow rate, to achieve precise temperature control.

[0049] S2: The server uses the improved DeepLab algorithm to perform semantic segmentation on the indoor video frame images, obtains the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F4;

[0050] S21: performing semantic segmentation on each frame of the indoor video frame image;

[0051] S22: Extract features from the semantic segmentation results, including human position, action type, and indoor number features; and combine the extracted features with the corresponding time and temperature data to form a feature vector F t ; In one embodiment, the camera continuously monitors each room and transmits video data to a central server in real time. The temperature sensor measures the room temperature once a minute and sends the data to the server. The video frames received by the server are first processed by the DeepLab algorithm, which is specially optimized to adapt to indoor environments and dynamic conditions. The DeepLab algorithm performs semantic segmentation on each frame of the image and accurately identifies the human body area in the image. In the video frame of the living room, the algorithm can distinguish between a person sitting on the sofa and a child playing on the floor. From the results of the semantic segmentation, the system extracts key features. The human body position includes determining the specific location of the human body in the room, such as "near the window" or "near the door". The action type includes analyzing the human body's actions, such as "sitting", "walking" or "exercise". The number of people in the room includes counting the total number of people in the room for adjusting the temperature setting. These features are integrated with the temperature data at the corresponding time point to form a detailed feature vector. The feature vector is input into a trained LSTM network, which learns to predict the optimal temperature setting in different situations based on past data. The LSTM network analyzes the patterns in the feature vector and predicts the ideal temperature range for each room. Based on the output of the LSTM network, the server controls the temperature control equipment in each room to make corresponding adjustments. If it detects an increase in activity in a room (a party), the system increases the output of the air conditioner to accommodate the heat generated by more people. For less used rooms, such as an empty study, the system will reduce temperature control activity to save energy.

[0052] DeepLab is an efficient deep learning model mainly used for image semantic segmentation. It can understand the category to which each pixel in the image belongs, such as human body, object, background, etc. The DeepLab algorithm was developed by researchers from Google. Its main feature is that it combines deep convolutional neural networks (DCNNs) with conditional random fields (CRFs). This combination improves the segmentation accuracy of image edges, especially the boundaries of objects in complex backgrounds.

[0053] One of the core features of DeepLab’s algorithm is the dilated convolution, which allows the network to increase the receptive field (i.e., the range of view) while maintaining the image size without pooling or downsampling. This allows the model to grasp detailed information in the image more accurately, especially in edge areas. Dilated convolution is achieved by adjusting the dilation rate in the convolution kernel, which determines the spacing between weights inside the convolution kernel. ASPP In order to capture multi-scale information, DeepLab introduced the ASPP module, which uses dilated convolutions with different dilation rates for parallel processing, so that information at multiple scales can be considered simultaneously. This is particularly effective for processing objects of different sizes and complex image structures. Fully connected CRFs After the output of the deep network, DeepLab uses fully connected CRFs to further optimize the segmentation results. CRFs improve and refine segmentation boundaries by modeling subtle interactions between pixels, especially more accurately processing details near object boundaries.

[0054] S3: The feature vector F t Input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range;

[0055] S31: Initialize the parameters of the LSTM long short-term memory network;

[0056] S32: Combine the features of consecutive frames into time series data X LSTM ;X LSTM =[F1,F2,...F T ], where T represents the total number of frames in the video sequence;

[0057] S33: Convert the time series data X LSTM Input to the LSTM long short-term memory network for forward propagation; the LSTM long short-term memory network outputs the expected user target temperature range;

[0058] In one embodiment, the camera continuously monitors the activities in the room, capturing one frame of video image every minute. The image is processed by the DeepLab algorithm to identify the human body area in the image and extract features, such as human body position, action type and number of people in the room. These features are combined with the corresponding temperature data to form a feature vector. Each vector represents the indoor situation at a time point. The continuously collected feature vectors are arranged in chronological order to form a time series data set. If data is collected once a minute, there will be 60 feature vectors in an hour. These vectors are combined together to form a larger data structure for input to the LSTM network. Before the system is deployed, the LSTM network is trained on similar historical data to learn how to predict indoor temperature changes based on the input feature vectors. The trained model will be used to analyze real-time data. The model predicts the next temperature demand based on the data trend in the last few hours. The LSTM network processes the input time series data through the forward propagation process and outputs the predicted target temperature range. According to the prediction results output by the LSTM, the system automatically adjusts the temperature control equipment, such as adjusting the working status of the heat sink or the setting of the air conditioner. If the predicted target temperature is higher than the current temperature, the system increases the heating intensity; if the predicted temperature is lower than the current temperature, the system starts cooling measures.

[0059] Long Short-Term Memory (LSTM) is a special type of Recurrent Neural Network (RNN) that is particularly suitable for processing and predicting long-term dependencies in time series data. LSTM regulates the flow of information by introducing three main gating mechanisms - forget gate, input gate and output gate, which enables it to avoid the gradient vanishing or gradient exploding problems faced by traditional RNN when processing long sequence data. The forget gate determines which information should be forgotten or discarded. It looks at the current input and the previous hidden state through a sigmoid function and outputs a value between 0 and 1 for each value in the cell state. Values ​​close to 0 indicate forgetting, while values ​​close to 1 indicate keeping. The input gate determines which new information is stored in the cell state. First, a sigmoid layer determines which values ​​we are going to update, and then a tanh layer creates a new candidate vector, which is added to the state. The cell state is the core of the LSTM, which flows through the network and is passed to the next time step after being slightly modified. It is updated by forgetting old information through the forget gate and adding new information through the input gate. The output gate determines the value of the next hidden state, which contains information about the previous time step and is used for prediction or decision making. The output is based on the current cell state but is filtered through a sigmoid layer.

[0060] In one embodiment, there is a smart home system whose task is to dynamically adjust the temperature of the room based on the activities in the home and the changes in the external temperature. Using the LSTM network to predict the user's target temperature range can be divided into the following steps: First, we need to collect relevant data, including: Indoor temperature: the real-time temperature of each room. Human activity data: collected through indoor cameras, including the number of people, location, activity type, etc. These data will be integrated into feature vectors as inputs to the LSTM. The LSTM model is trained using historical data, with feature vectors as inputs and the ideal temperature manually set by the user (collected from the historical settings of the smart temperature control device) as outputs. In this way, the model learns the temperature that residents tend to set under different indoor activities and external climate conditions. When the system is running, the feature vectors collected in real time are continuously input into the trained LSTM model. The model processes the input data and predicts the ideal indoor temperature range under the current conditions.

[0061] S4: The control unit compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result; including: when the monitored ambient temperature is higher than the maximum value of the expected user target temperature range, the control unit increases the surface area of ​​the titanium alloy heat sink, increases the fan speed, and increases the coolant flow of the heat sink; when the monitored ambient temperature is lower than the minimum value of the expected user target temperature range, the heat sink surface area is reduced, the fan speed is reduced, and the coolant flow is reduced. In one embodiment, the temperature sensor in each room monitors and records the ambient temperature in real time, and the video camera monitors the activities in the room, such as the number of people and the type of activity. These data are sent to the central control unit for analysis and decision-making. The central control unit first compares the real-time monitored ambient temperature with the target temperature range predicted by the LSTM model. If the actual temperature exceeds the maximum value of the target temperature range, the control unit sends an instruction to the relevant heat sink, then increases the fan speed by increasing the speed of the fan on the heat sink to speed up the exhaust speed of the hot air, increases the surface area of ​​the heat sink, and increases the effective surface area of ​​the heat sink through the mechanical expansion device to improve the heat dissipation capacity. Increasing the coolant flow rate quickly reduces the temperature of the heat sink by increasing the flow rate of the liquid in the cooling system. If the actual temperature is lower than the minimum value of the target temperature range, the control unit adjusts the heat sink to reduce heat dissipation: Reducing the fan speed reduces the cooling effect by reducing the fan speed to increase the room temperature. Reducing the heat sink surface area reduces the surface area of ​​the heat sink through mechanical devices, and reducing the coolant flow rate slows down the heat dissipation process by reducing the flow rate of the coolant. Due to a party with many people in the living room, the room temperature rose to 27°C, while the system set the comfortable temperature range to 22°C to 25°C. After the control unit detected that the temperature exceeded the maximum value, it automatically increased the fan speed of the living room heat sink, expanded the heat sink surface area, and increased the flow of coolant. As a result, the temperature of the living room fell back to the comfortable range in a short time.

[0062] In some embodiments, the collected indoor data is transmitted to the server, including: using a 5G network or a Wi-Fi network to transmit the data in real time to a remote server data center, and storing the video data in a cloud storage server.

[0063] In some embodiments, the step S21: performing semantic segmentation on each frame of indoor video frame image includes S211: using a convolutional neural network to extract features of the input indoor video frame image to obtain F CNN ; S212: Use dilated convolution for F CNN Convolution operation is performed to maintain spatial resolution and capture multi-scale information, and F dilated ;

[0064]

[0065] Among them, F CNN The feature map extracted by the convolutional neural network is used as the input of the dilated convolution, F dilated is the feature map output by the dilated convolution; W is the convolution kernel, whose size is (2a+1)*(2b+1); d is the dilation rate, which is used to control the size of the dilated convolution; x, y are the horizontal and vertical coordinates in the feature map, respectively; i, j are the horizontal and vertical coordinates of the position in the convolution kernel, respectively, traversing the convolution kernel space range; a, b represent the parameter values ​​of the convolution kernel space range, respectively; d0 represents the dilation rate when the target size is equal to the benchmark size; N is the number of human body regions; s k is the size of the kth human body region; s0 is the set reference value; the larger the human body region, the larger the void ratio d;

[0066] S213: Use spatial pyramid pooling to F dilated Capture information of different scales and obtain multi-scale features F spp ; S214: Multi-scale feature F through upsampling and deconvolution spp The original spatial resolution is restored and a category label is assigned to each pixel, outputting the semantic segmentation result F t .

[0067] In one embodiment, in the video monitoring and processing of the smart home system, the ability of semantic segmentation is enhanced by using dilated convolution to effectively improve the effect. The dilated convolution adds "holes" (i.e., intervals) to the convolution kernel, so that the network can expand the receptive field without increasing the computational burden, thereby capturing a wider range of contextual information. This is particularly useful when processing indoor monitoring images, because the human body and its activities can be more accurately identified and analyzed, regardless of whether the target object is close or far from the camera. A smart home system is installed in a large living space, and the system aims to automatically adjust the room environment, such as temperature and light, through video monitoring. The camera in the system continuously monitors and transmits video to the central processor in real time. Each frame of video captured by the camera is first subjected to preliminary feature extraction by a model based on FCNN (full convolutional neural network). These features include basic visual elements such as edges, textures, and colors. Next, these feature maps are input into a network using dilated convolution. In this step, the convolution kernel expands its scope by increasing the dilation rate (d), allowing the network to observe contextual information in a larger area. The dynamic adjustment of the dilation rate is based on the size of the detected human body area. A larger human body area indicates that the target is farther away and a larger dilation rate is required to capture enough context; a smaller human body area indicates that the target is closer and a smaller dilation rate is sufficient. Through spatial pyramid pooling, the system can capture features from multiple scales to ensure that the behavior of the human body at different distances to the camera can be accurately identified. Finally, using upsampling and deconvolution techniques, the system restores these features to the original resolution and classifies each pixel to complete high-precision semantic segmentation. The application of dilated convolution significantly improves the accuracy and efficiency of semantic segmentation, especially in indoor environments, and can better handle objects of different scales. The technical effects include capturing more contextual information through a larger receptive field, improving the segmentation accuracy of various objects (especially human bodies) indoors, and dynamically adjusting the dilation rate so that the system can adapt to the different sizes of people in the picture and optimize the processing effect. Compared with traditional methods for expanding the receptive field (such as multi-level pooling or repeated convolution), dilated convolution does not require additional parameters and calculations, which improves efficiency.

[0068] In some embodiments, the S213: using spatial pyramid pooling to dilated Capture information of different scales and obtain multi-scale features F spp , including: first creating pyramid levels; secondly performing the maximum pooling operation on the input feature map at each level; and finally merging the pooling results of all levels into a long vector to obtain the multi-scale feature F spp. Spatial pyramid pooling is a computer vision technique used to capture multi-scale information from images. It allows neural networks to process inputs of varying sizes while extracting spatial hierarchical features, which is very useful for tasks such as image recognition and semantic segmentation. In smart home temperature control systems, SPP can enhance the system's understanding of human bodies and other objects in indoor scenes and improve the accuracy of temperature control. In one embodiment, a smart home system that adjusts the temperature of a room by monitoring human activities in the room, the camera in the system continuously captures indoor video images and processes them through a central processing unit. Each video frame captured by the camera is first processed by a model based on a deep convolutional neural network to generate a feature map (FCNN output), which contains the basic visual information of the image. In the step of creating the pyramid level, the feature map is divided into different regional levels, usually several grids of varying sizes, 1x1, 2x2, 4x4. In this way, the grids of each level will cover different parts of the image, from coarse to fine, and the maximum pooling operation is performed within the grids of each level. Max pooling will select the most significant feature (i.e., the maximum value) in each grid, which can retain important information in the image while reducing the dimension. Finally, the pooling results of all levels are merged into a long vector (multi-scale feature Fs). This vector contains image features extracted from different spatial scales, providing rich information for subsequent processing. This long vector (Fs) is then used for further image analysis tasks, such as time series analysis through LSTM networks to predict changes in indoor temperature demand. Such multi-scale feature processing allows the system to more accurately understand the indoor environment and occupant behavior, and provide more accurate temperature control.

[0069] In some embodiments, the S214: performing upsampling and deconvolution on the multi-scale features F spp The original spatial resolution is restored and a category label is assigned to each pixel, outputting the semantic segmentation result F t, including: first, using a linear interpolation algorithm to upsample and increase the image size, and then performing deconvolution, which is the inverse operation of convolution, to increase the spatial resolution of the feature map. In one embodiment, the system first uses a convolutional neural network (CNN) and spatial pyramid pooling (SPP) to extract features from the video frame captured by the camera. These features capture detailed information of the image at different levels of scale, and use a linear interpolation algorithm to upsample the multi-scale features to increase the image size to gradually approach the resolution of the original video frame. This step is achieved by inserting new pixels between existing pixels. The interpolation method calculates the value of the new pixel based on the value of the adjacent pixels. Next, the deconvolution operation is used to further refine the feature map. Deconvolution expands the feature map through a process opposite to convolution, which is used to learn how to restore from a compressed feature map to a more detailed spatial resolution. After the above process, each pixel has a detailed feature representation. Finally, a classification layer processes these features and assigns a category label (such as "human body", "furniture", "wall", etc.) to each pixel, thereby generating a final semantic segmentation map. In one embodiment, in a family living room, the system needs to identify and distinguish between human bodies, furniture, and other objects in order to intelligently adjust the air conditioner and lights. When family members gather to watch TV in the evening, the system analyzes the semantic segmentation map of the living room, detects multiple human tags and automatically lowers the air-conditioning temperature, while brightening the lights in the TV area to provide a more comfortable viewing environment.

[0070] In some embodiments, S22: extracting features from the semantic segmentation results, including human body position, action type, and indoor number features; and combining the extracted features with the corresponding time and temperature data to form a feature vector F t, where the human body position includes the area and bounding box coordinates of the human body region; the action types include sitting, walking, running, yoga, dancing, playing table tennis, playing badminton, and playing basketball; the indoor number characteristics include the number of people and the change in the number of people. In one embodiment, multiple cameras are installed in a home environment, covering the main activity areas such as the living room, kitchen, and bedroom. These cameras continuously capture video data, which are analyzed by the central processing unit, and the video frames are processed using the improved DeepLab model to identify and mark the human body and other important objects in each frame. The output semantic segmentation map provides basic data for subsequent feature extraction. From the human body area identified in the semantic segmentation map, the bounding box coordinates and area of ​​each human body are extracted. These data help the system understand the specific position of each person in the room and the size of the space occupied. By analyzing the movement and posture changes of the human body, the type of activity being carried out is identified, such as sitting, walking, running, yoga, dancing, etc. This requires time series analysis of human body movements in consecutive frames. Count the number of people in each room and monitor their changes. This information is crucial for adjusting the temperature settings of the room, especially when there are multiple people. At the same time, the ambient temperature data provided by the temperature sensor is collected and recorded, and the above extracted features are combined with the temperature data at the corresponding time point to form a detailed feature vector.

[0071] In some embodiments, the working state of the titanium alloy heat sink is adjusted based on the comparison result, and the control unit automatically selects the first titanium alloy heat sink or the second titanium alloy heat sink or the third titanium alloy heat sink or the fourth titanium alloy heat sink according to the deviation between the ambient temperature and the target temperature range and the size of the set threshold, wherein the first titanium alloy heat sink contains the phase change material capric acid, which has a phase change at 29.8°C, and the phase change latent heat is 190J / g; the second titanium alloy heat sink contains the phase change material lauric acid, which has a phase change at about 43°C, and the phase change latent heat is 178J / g; the third titanium alloy heat sink contains the phase change material sodium sulfate decahydrate, which has a phase change at 32°C, and the phase change latent heat is about 250J / g; the fourth titanium alloy heat sink contains the phase change material p-hydroxybenzoic acid, which has a phase change at 25.0°C, and the phase change latent heat is 189J / g; the phase change material is encapsulated in microcapsules, the microcapsules are evenly distributed in the titanium alloy matrix, and the shell of the microcapsules is composed of polyphenylene sulfide material. In one embodiment, an intelligent temperature control system is installed in a home environment, which includes multiple titanium alloy heat sinks, each of which is equipped with different phase change materials. These heat sinks are installed in key locations of the home, such as the living room, bedroom, kitchen, etc. The first titanium alloy heat sink: contains capric acid as a phase change material, which changes phase at 29.8°C and is suitable for warmer environments. The second titanium alloy heat sink: contains lauric acid, with a phase change point of about 43°C, which is suitable for high temperature environments. The third titanium alloy heat sink: contains sodium sulfate decahydrate, with a phase change point at 32°C, suitable for medium temperatures. The fourth titanium alloy heat sink: contains parahydroxybenzoic acid, with a phase change point at 25.0°C, suitable for cooler environments. The phase change material of each heat sink is encapsulated in microcapsules, which are evenly distributed in the titanium alloy matrix, ensuring the effective absorption and release of thermal energy. The shell of the microcapsule is made of polyphenylene sulfide material, which ensures the stability and long-term durability of the structure. The temperature sensor in the system continuously monitors the real-time temperature of each room and compares it with the target temperature range set by the user. According to the deviation between the monitored temperature and the target temperature, the control unit automatically selects the most suitable heat sink to adjust the room temperature. If the temperature of a room is much higher than the target temperature, and the room is equipped with a first and a second titanium alloy heat sink, the control unit will activate the second heat sink containing lauric acid because its phase change temperature is more suitable for high temperature environments. In one embodiment, on a hot summer day, the temperature in the living room quickly rises to 30°C, while the target temperature is set to 24°C. The system automatically detects the temperature deviation and determines that stronger cooling measures need to be enabled. At this time, the system activates the third heat sink containing sodium sulfate decahydrate because its phase change temperature is closest to the current temperature, which can effectively absorb excess heat and quickly adjust the indoor temperature back to a comfortable range.

[0072] In one embodiment, the target temperature range is 22°C to 26°C, and the system can set different thresholds to trigger the switching of the heat sink: If the indoor temperature is lower than the target minimum temperature of 22°C by more than 2°C (i.e., 20°C and below), the fourth heat sink (parahydroxybenzoic acid) is activated because its phase change temperature is closest to the low temperature requirement. If the indoor temperature is higher than the target maximum temperature of 26°C by more than 3°C (i.e., 29°C and above), the first heat sink (decanoic acid) is activated because its phase change temperature is closest to the high temperature requirement. If the indoor temperature is between 27°C and 31°C, the third heat sink (sodium sulfate decahydrate) is activated, which is suitable for cooling in this temperature range. If the indoor temperature is between 23°C and 25°C, the second heat sink (lauric acid) is activated. Although its phase change temperature is higher, its latent heat can slowly adjust the indoor temperature to avoid the temperature from dropping too quickly. On a hot summer afternoon, the indoor temperature rises to 30°C. The control unit detects that the deviation of the indoor temperature from the target temperature range exceeds the high temperature threshold, so the third heat sink (sodium sulfate decahydrate) is automatically activated. Since its phase change temperature is 32°C, it can effectively absorb excess heat energy and quickly adjust the indoor temperature back to a level closer to the target temperature.

[0073] The present application also provides a titanium alloy smart home temperature control system, such as Figure 2As shown, it includes: the central control unit as the brain of the system, processing all sensor data, controlling the operation of the heat sink, and interacting with the user interface, including a high-performance microprocessor or microcontroller, memory, and necessary communication interfaces. The temperature sensor monitors the real-time temperature of each room. The digital temperature sensor, such as DS18B20, can provide accurate temperature readings and is connected to the central control unit via a data line. The video camera monitors indoor activities and provides real-time video streams for analyzing the activities and locations of the occupants. The IP camera supports Wi-Fi connection and can directly send data streams to the central control unit. The heat sink device adjusts the indoor temperature according to the command of the control unit. The titanium alloy heat sink is embedded with phase-variable material microcapsules and is equipped with an electric mechanical control system for adjusting the heat sink surface area and fan speed. The user interface allows the user to set temperature preferences, view the current temperature and system status, and the touch screen display or smartphone application is connected to the central control unit via Wi-Fi or wired network. The communication device ensures seamless data transmission between all devices. Wi-Fi module, Bluetooth module or Zigbee module ensures that each device can be interconnected. The central control unit is connected to the home network via Wi-Fi or Ethernet, and communicates with the heat sink and temperature sensors via RS-485 or wireless modules. The temperature sensors are distributed in various rooms of the home and connected back to the central control unit via wired (such as RS-485) or wireless (such as Zigbee). The video camera is directly connected to the central control unit via the Wi-Fi network and transmits video data in real time. The heat sink device adjusts the operation through the electromechanical interface and is connected to the central control unit through the control line to receive adjustment instructions.

[0074] A data acquisition unit is used to collect indoor data in real time, including temperature sensors arranged at multiple locations indoors to monitor ambient temperature data and collect indoor video frame image data in real time, and transmit the collected indoor data to a server;

[0075] In the feature vector generation module, the server uses the improved DeepLab algorithm to perform semantic segmentation on the indoor video frame images, obtains the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F t ;

[0076] A semantic segmentation module performs semantic segmentation on each frame of indoor video;

[0077] The feature extraction module extracts features from the semantic segmentation results, including human position, action type, and indoor number of people; and combines the extracted features with the corresponding time and temperature data to form a feature vector F t ;

[0078] The module for calculating the expected user target temperature range converts the feature vector F t Input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range;

[0079] Initialization module, initializes the parameters of LSTM long short-term memory network;

[0080] The time series data generation module combines the features of consecutive frames into time series data X LSTM ;X LSTM =[F1,F2,...F t ], where T represents the total number of frames in the video sequence;

[0081] LSTM long short-term memory network processing module, time series data X LSTM Input to the LSTM long short-term memory network for forward propagation; the LSTM long short-term memory network outputs the expected user target temperature range;

[0082] The judgment and adjustment module comprises a control unit which compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result; including: when the monitored ambient temperature is higher than the maximum value of the expected user target temperature range, the control unit increases the surface area of ​​the titanium alloy heat sink, increases the fan speed, and enhances the coolant flow of the heat sink; when the monitored ambient temperature is lower than the minimum value of the expected user target temperature range, the control unit reduces the surface area of ​​the heat sink, reduces the fan speed, and reduces the coolant flow.

[0083] In some embodiments, the collected indoor data is transmitted to the server, including: using a 5G network or a Wi-Fi network to transmit the data in real time to a remote server data center, and storing the video data in a cloud storage server.

[0084] In some embodiments, the working state of the titanium alloy heat sink is adjusted based on the comparison result, and the control unit automatically selects the first titanium alloy heat sink or the second titanium alloy heat sink or the third titanium alloy heat sink or the fourth titanium alloy heat sink according to the deviation between the ambient temperature and the target temperature range and the size of the set threshold, wherein the first titanium alloy heat sink contains the phase change material capric acid, which has a phase change at 29.8°C, and the phase change latent heat is 190J / g; the second titanium alloy heat sink contains the phase change material lauric acid, which has a phase change at about 43°C, and the phase change latent heat is 178J / g; the third titanium alloy heat sink contains the phase change material sodium sulfate decahydrate, which has a phase change at 32°C, and the phase change latent heat is about 250J / g; the fourth titanium alloy heat sink contains the phase change material p-hydroxybenzoic acid, which has a phase change at 25.0°C, and the phase change latent heat is 189J / g; the phase change material is encapsulated in microcapsules, the microcapsules are evenly distributed in the titanium alloy matrix, and the shell of the microcapsules is composed of polyphenylene sulfide material.

[0085] The present invention provides a titanium alloy smart home temperature control method and system, which can achieve the following beneficial technical effects:

[0086] 1. In the present invention, the data acquisition unit first collects indoor data in real time; secondly, the server uses the improved DeepLab algorithm to perform semantic segmentation on the indoor video frame image, obtains the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F t ; Again, the eigenvector F t The data is input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range; finally, the control unit compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result. This application uses deep learning algorithms such as DeepLab for real-time video analysis, and the system can identify information such as the location, activity status, and number of residents. Combined with the long short-term memory network (LSTM), these data can be converted into a prediction model to predict the temperature requirements of the residents in real time, thereby dynamically adjusting the working state of the titanium alloy heat sink to achieve the optimal temperature regulation effect.

[0087] 2. The present invention adopts the improved DeepLab algorithm in the process of human body area extraction, introduces the human body area into the calculation of the dilation rate, and dynamically adjusts the dilation rate according to the human body area. The dilation rate d is used to control the size of the dilation in the dilation convolution. The larger the human body area, the larger the dilation rate d. The adaptive adjustment of the dilation rate is realized, which greatly enhances the accurate extraction of the human body area and the accurate judgment of the motion situation.

[0088] 3. The control unit of the present invention automatically selects the first titanium alloy heat sink, the second titanium alloy heat sink, the third titanium alloy heat sink, or the fourth titanium alloy heat sink according to the deviation between the ambient temperature and the target temperature range and the size of the set threshold. The step-change material is encapsulated in microcapsules, and the microcapsules are evenly distributed in the titanium alloy matrix. The shell of the microcapsule is composed of polyphenylene sulfide material. The titanium alloy heat sink is selected according to different deviation conditions, and the temperature is adjusted according to the characteristics of the step-change material, which greatly improves the temperature adjustment accuracy and efficiency and effectively reduces energy consumption.

[0089] A titanium alloy smart home temperature control method and system are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention. At the same time, for general technicians in this field, according to the ideas and methods of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A titanium alloy smart home temperature control method, characterized in that: Includes steps: S1: The data acquisition unit collects indoor data in real time, including temperature sensors arranged at multiple locations indoors to monitor ambient temperature data and collect indoor video frame image data in real time, and transmits the collected indoor data to the server; S2: The server uses the improved DeepLab algorithm to perform semantic segmentation on the indoor video frame images, obtains the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F t ; S21: performing semantic segmentation on each frame of the indoor video frame image; S22: Extract features from the semantic segmentation results, including human position, action type, and indoor number of people; The extracted features are combined with the corresponding time and temperature data to form a feature vector F t ; S3: The feature vector F t Input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range; S31: Initialize the parameters of the LSTM long short-term memory network; S32: Combine the features of consecutive frames into time series data X LSTM ;X LSTM =[F1,F2,...F T ], where T represents the total number of frames in the video sequence; S33: Convert the time series data X LSTM Input to the LSTM long short-term memory network for forward propagation; the LSTM long short-term memory network outputs the expected user target temperature range; S4: The control unit compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result; including: when the monitored ambient temperature is higher than the maximum value of the expected user target temperature range, the control unit increases the surface area of ​​the titanium alloy heat sink, increases the fan speed, and enhances the coolant flow of the heat sink; when the monitored ambient temperature is lower than the minimum value of the expected user target temperature range, the control unit reduces the surface area of ​​the heat sink, reduces the fan speed, and reduces the coolant flow.

2. A titanium alloy smart home temperature control method as claimed in claim 1, characterized in that: The collected indoor data is transmitted to the server, including: using a 5G network or a Wi-Fi network to transmit the data in real time to a remote server data center, and storing the video data in a cloud storage server.

3. A titanium alloy smart home temperature control method as claimed in claim 2, characterized in that: S21: performing semantic segmentation on each frame of indoor video frame image, including S211: using a convolutional neural network to extract features of the input indoor video frame image to obtain F CNN ; S212: Use dilated convolution for F CNN Convolution operation is performed to maintain spatial resolution and capture multi-scale information, and F dilated ; Among them, F CNN The feature map extracted by the convolutional neural network is used as the input of the dilated convolution, F dilated is the feature map output by the dilated convolution; W is the convolution kernel, whose size is (2a+1)*(2b+1); d is the dilation rate, which is used to control the size of the dilated convolution; x, y are the horizontal and vertical coordinates in the feature map, respectively; i, j are the horizontal and vertical coordinates of the position in the convolution kernel, respectively, traversing the convolution kernel space range; a, b represent the parameter values ​​of the convolution kernel space range, respectively; d0 represents the dilation rate when the target size is equal to the benchmark size; N is the number of human body regions; s k is the size of the kth human body region; s0 is the set reference value; the larger the human body region, the larger the void ratio d; S213: Use spatial pyramid pooling to F dilated Capture information of different scales and obtain multi-scale features F spp ; S214: Multi-scale feature F through upsampling and deconvolution spp The original spatial resolution is restored and a category label is assigned to each pixel, outputting the semantic segmentation result F t .

4. A titanium alloy smart home temperature control method as claimed in claim 3, characterized in that: S213: Using spatial pyramid pooling to F dilated Capture information of different scales and obtain multi-scale features F spp , including: first creating pyramid levels; secondly performing the maximum pooling operation on the input feature map at each level; and finally merging the pooling results of all levels into a long vector to obtain the multi-scale feature F spp .

5. A titanium alloy smart home temperature control method as claimed in claim 3, characterized in that: S214: Multi-scale feature F through upsampling and deconvolution spp The original spatial resolution is restored and a category label is assigned to each pixel, outputting the semantic segmentation result F t , including: first, using a linear interpolation algorithm to upsample and increase the image size, and then performing deconvolution, which is the inverse operation of convolution, to increase the spatial resolution of the feature map.

6. A titanium alloy smart home temperature control method as claimed in claim 3, characterized in that: S22: extracting features from the semantic segmentation result, including human body position, action type, and indoor number features; The extracted features are combined with the corresponding time and temperature data to form a feature vector F t , where the human body position includes the area of ​​the human body region and the bounding box coordinates; the action types include sitting, walking, running, yoga, dancing, playing table tennis, playing badminton, and playing basketball; the indoor population characteristics include the number of people and the change in the number of people.

7. A titanium alloy smart home temperature control method as claimed in claim 3, characterized in that: The method further includes adjusting the working state of the titanium alloy heat sink based on the comparison result, and the control unit automatically selects the first titanium alloy heat sink or the second titanium alloy heat sink or the third titanium alloy heat sink or the fourth titanium alloy heat sink according to the deviation between the ambient temperature and the target temperature range and the size of the set threshold, wherein the first titanium alloy heat sink contains the phase change material decanoic acid, which has a phase change at 29.8°C, and the phase change latent heat is 190J / g; the second titanium alloy heat sink contains the phase change material lauric acid, which has a phase change at about 43°C, and the phase change latent heat is 178J / g; the third titanium alloy heat sink contains the phase change material sodium sulfate decahydrate, which has a phase change at 32°C, and the phase change latent heat is about 250J / g; the fourth titanium alloy heat sink contains the phase change material p-hydroxybenzoic acid, which has a phase change at 25.0°C, and the phase change latent heat is 189J / g; the phase change material is encapsulated in microcapsules, the microcapsules are evenly distributed in the titanium alloy matrix, and the shell of the microcapsules is composed of polyphenylene sulfide material.

8. A titanium alloy smart home temperature control system, characterized in that: include: A data acquisition unit is used to collect indoor data in real time, including temperature sensors arranged at multiple locations indoors to monitor ambient temperature data and collect indoor video frame image data in real time, and transmit the collected indoor data to a server; In the feature vector generation module, the server uses the improved DeepLab algorithm to perform semantic segmentation on the indoor video frame images, obtains the human body area in each frame of the video image, and extracts features from the human body area. The extracted features are connected end to end with the corresponding time and temperature data to form a feature vector F t ; A semantic segmentation module performs semantic segmentation on each frame of indoor video; The feature extraction module extracts features from the semantic segmentation results, including human position, action type, and indoor number of people; The extracted features are combined with the corresponding time and temperature data to form a feature vector F t ; The module for calculating the expected user target temperature range converts the feature vector F t Input into the trained LSTM long short-term memory network for time series analysis to obtain the expected user target temperature range; Initialization module, initializes the parameters of LSTM long short-term memory network; The time series data generation module combines the features of consecutive frames into time series data X LSTM ;X LSTM =[F1,F2,...F T ], where T represents the total number of frames in the video sequence; LSTM long short-term memory network processing module, time series data X LSTM Input to the LSTM long short-term memory network for forward propagation; the LSTM long short-term memory network outputs the expected user target temperature range; The judgment and adjustment module comprises a control unit which compares the monitored ambient temperature with the expected user target temperature range, and adjusts the working state of the titanium alloy heat sink based on the comparison result; including: when the monitored ambient temperature is higher than the maximum value of the expected user target temperature range, the control unit increases the surface area of ​​the titanium alloy heat sink, increases the fan speed, and enhances the coolant flow of the heat sink; when the monitored ambient temperature is lower than the minimum value of the expected user target temperature range, the control unit reduces the surface area of ​​the heat sink, reduces the fan speed, and reduces the coolant flow.

9. A titanium alloy smart home temperature control system as claimed in claim 8, characterized in that: The collected indoor data is transmitted to the server, including: using a 5G network or a Wi-Fi network to transmit the data in real time to a remote server data center, and storing the video data in a cloud storage server.

10. A titanium alloy smart home temperature control system as claimed in claim 8, characterized in that: The method further includes adjusting the working state of the titanium alloy heat sink based on the comparison result, and automatically selecting the first titanium alloy heat sink or the second titanium alloy heat sink or the third titanium alloy heat sink or the fourth titanium alloy heat sink according to the deviation between the ambient temperature and the target temperature range and the size of the set threshold value by the control unit, wherein the first titanium alloy heat sink contains the phase change material capric acid, which has a phase change at 29.8°C, and the phase change latent heat is 190J / g; the second titanium alloy heat sink contains the phase change material lauric acid, which has a phase change at about 43°C, and the phase change latent heat is 178J / g; the third titanium alloy heat sink contains the phase change material sodium sulfate decahydrate, which has a phase change at 32°C, and the phase change latent heat is about 250J / g; the fourth titanium alloy heat sink contains the phase change material p-hydroxybenzoic acid, which has a phase change at 25.0°C, and the phase change latent heat is 189J / g; the phase change material is encapsulated in microcapsules, the microcapsules are evenly distributed in the titanium alloy matrix, and the shell of the microcapsules is composed of polyphenylene sulfide material.