Sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization

Abstract

The invention discloses a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization, which comprises the following steps: S1, an environmental perception module calculates a standard deviation sigma of window data through original data collected by a sensor node and judges whether the standard deviation sigma exceeds a dynamic threshold value theta, so that the fluctuation condition of environmental data is monitored; s2, the dynamic sampling controller switches the sampling frequency according to different states of the fluctuation condition of the environment, and the energy consumption of the system is reduced while the data quality is ensured through down-sampling processing of content awareness and an intelligent batch transmission strategy. According to the sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization disclosed by the invention, the sampling frequency can be adaptively adjusted, and feature optimization is carried out on the sampled data through a content-aware CARAFE operator, so that the precision and energy efficiency of data acquisition are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of sensor sampling, and in particular relates to a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization. Background Art

[0002] In wireless sensor networks (WSN) and remote monitoring applications, a large number of unattended wireless sensor nodes are usually deployed in the monitoring area. These nodes realize real-time monitoring, perception and data collection of various monitoring objects through wireless ad hoc networks, and upload the information to the base station for storage and processing. Energy efficiency is one of the key indicators of wireless sensor networks. The energy consumption of sensor nodes is mainly concentrated in the CPU, sensor module and wireless transceiver. Among them, the energy consumption of wireless communication is relatively high, and data transmission accounts for about 80% of the energy consumption of the entire network.

[0003] Therefore, wireless sensor nodes need to adopt a reasonable sampling strategy to minimize the transmission of redundant data to reduce overall energy consumption. The traditional sensor sampling method is usually continuous sampling (periodic sampling), that is, the sensor node collects data at fixed time intervals. Although this method has a small sampling error and a simple control process, the sampling frequency is fixed, resulting in high power consumption and difficulty in meeting the energy requirements of wireless sensor nodes. Especially when the environmental parameters are relatively stable, too high a sampling frequency will increase unnecessary energy consumption. In addition, a fixed-frequency sampling system cannot flexibly respond to rapid changes in the environment and may miss key data when the environment fluctuates violently.

[0004] Therefore, further improvements are made to the above problems. Summary of the invention

[0005] The main purpose of the present invention is to provide a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization, which can adaptively adjust the sampling frequency and perform feature optimization on the sampled data through the content-aware CARAFE operator to improve the accuracy and energy efficiency of data acquisition. Under the synergistic effect of the environmental perception module, dynamic sampling controller, CARAFE downsampling module and data transmission and storage strategy, the energy consumption of wireless sensor nodes is effectively reduced, and it is suitable for low-power scenarios such as remote monitoring and wireless sensor networks.

[0006] To achieve the above objectives, the present invention provides a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization, comprising the following steps:

[0007] Step S1: The environment perception module calculates the standard deviation σ of the window data through the raw data collected by the sensor node and determines whether the standard deviation σ exceeds the dynamic threshold θ, thereby monitoring the fluctuation of the environment data;

[0008] Step S2: The dynamic sampling controller switches the sampling frequency according to different states of environmental fluctuations, and reduces the energy consumption of the system while ensuring data quality through content-aware downsampling processing and intelligent batch transmission strategy.

[0009] As a further preferred technical solution of the above technical solution, step S1 is specifically implemented as follows:

[0010] Step S1.1: The raw data collected by each sensor node of the environment perception module is allocated to a fixed time window, so as to calculate the standard deviation σ of the window data, and the formula is as follows:

[0011]

[0012] Among them, x i is the data value of the i-th sampling point, is the average value of the window data, N is the total number of sampling points in the time window, and the fluctuation of the standard deviation σ directly reflects the changes in the environment;

[0013] When σ exceeds the preset dynamic threshold θ, the system determines that the environment is in a fluctuating state; when σ is lower than θ, the system determines that the environment is in a stable state;

[0014] Step S1.2: (In order to improve the accuracy of environmental state judgment) the dynamic threshold θ is adaptively adjusted according to environmental changes, and the calculation formula is:

[0015] θ=α·σ histoty +β

[0016] Among them, σ history It represents the average standard deviation of historical data, and α and β are adjustment parameters used to control the baseline and sensitivity of the dynamic threshold (this adaptive dynamic threshold mechanism helps the system maintain a reasonable sampling frequency in different scenarios and improve the accuracy of environmental status monitoring).

[0017] As a further preferred technical solution of the above technical solution, step S2 is specifically implemented as the following steps:

[0018] Step S2.1: For a stable state, the dynamic sampling controller sets the sampling frequency to a low frequency (e.g., sampling once per hour) to save energy;

[0019] Step S2.2: For the fluctuating state, the dynamic sampling controller sets the sampling frequency to a high frequency (eg, sampling once per minute) to capture the rapidly changing environmental information.

[0020] As a further preferred technical solution of the above technical solution, in step S2, (in order to further optimize the efficiency and real-time performance of sampling frequency switching) the dynamic sampling controller predicts environmental fluctuations at future moments through a time series prediction model, thereby switching to a high-frequency mode in advance before environmental fluctuations occur. The time series prediction model is as follows:

[0021]

[0022] Among them, φ i is the coefficient of the autoregressive term, θ j is the coefficient of the moving average term, ∈ t-j is the white noise term, x t-i Represents past observation data. Parameters p, q, and t are the basic elements required for the definition and calculation of the model. Based on the time series prediction model, adaptive sampling frequency adjustment is achieved on the basis of predicting the environmental state (reducing the system overhead caused by frequent frequency switching while ensuring the effectiveness and timeliness of data collection).

[0023] As a further preferred technical solution of the above technical solution, for step S2, (in the high-frequency sampling mode, the amount of data collected by the sensor node increases, and direct transmission will significantly increase the communication energy consumption) the kernel prediction module and feature reconstruction module of the CARAFE operator are introduced to perform content-aware downsampling processing on the sampled data (the CARAFE operator can identify the change area in the data and perform weighted sampling on the samples to retain the key change information while removing redundant data. Unlike the traditional method using a 1×1 or 2×2 receptive field, the CARAFE operator has a larger receptive field and can consider the entire feature map, thereby better retaining the details and edge information of the image and reducing the jagged and blurring effects during the sampling process. The CARAFE operator supports the dynamic generation of adaptive kernels and can achieve content-aware processing according to different goals, while the traditional interpolation method cannot dynamically adjust the kernel size to match the data content), wherein:

[0024] In the kernel prediction module, the adaptive convolution kernel is generated according to the input features and is related to the position, thereby improving the ability to identify and retain the change information. The kernel prediction module generates a position-dependent adaptive convolution kernel according to the input features. The formula is:

[0025] K i,j =g(X i,j )

[0026] Among them, K i,j is the convolution kernel matrix, X i,j is the characteristic information of the original data, function g is the kernel prediction network, and the adaptive convolution kernel structure allows dynamic weighted optimization according to the characteristics of the environmental data;

[0027] In the weighted sampling step, the feature reconstruction module effectively extracts the data change area through resampling, and the CARAFE operator generates weights W according to the generated i,j , perform weighted convolution calculation on the sampled data, and use the convolution kernel K to perform content-aware resampling of the data area. The formula is:

[0028]

[0029] Among them, y i,j is the data after resampling, K is the convolution kernel, the size is (2k+1)×(2k+1), x i+m,j+n is the original data point, i and j represent the target position coordinates of the resampled data in the convolution operation, m and n are the relative position offsets of the convolution kernel, which are used to sample the original data points within the range of the convolution kernel. By weighted sampling in the salient area, the CARAFE operator (can) effectively remove unnecessary data redundancy, while reducing the amount of data and retaining the characteristic information of the environment (this method can reduce the amount of data transmission in high-frequency sampling mode while maintaining key change information, effectively improving the energy efficiency of data processing).

[0030] As a further preferred technical solution of the above technical solution, for step S2, the data volume problem caused by batch transmission is solved by an intelligent batch data transmission strategy, wherein:

[0031] In a stable state, the system temporarily stores the sampled data in a local storage unit and transmits it in batches when the environment enters a fluctuating state, thereby reducing the number of transmissions and communication power consumption. The system uses the following formula to quantify the total power consumption during the transmission process:

[0032]

[0033] Among them, T total Represents the total power consumption of the system, T sampling represents the power consumption of data acquisition, D represents the amount of data transmitted in batches, R represents the data transmission rate, and P transmit It represents the transmission power consumption. By reasonably controlling the size of the batch data volume D and the transmission frequency, and preferably using a low-power communication protocol (such as LoRa or BLE), the communication power consumption can be reduced and the service life of the sensor node can be extended while ensuring data integrity.

[0034] The beneficial effects of the present invention are:

[0035] The dynamic sampling algorithm of the present invention achieves significant reduction of system energy consumption while ensuring data quality through environmental perception, dynamic sampling adjustment, content-aware downsampling, and intelligent batch transmission. This solution based on environmental fluctuation adaptive sampling enables wireless sensor networks and remote monitoring applications to achieve efficient and low-power operation under battery power supply, and has broad application prospects and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the process of the present invention.

[0037] Figure 2 This is the overall framework of the CARAFE operator. DETAILED DESCRIPTION

[0038] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0039] In the preferred embodiments of the present invention, those skilled in the art should note that the sensors and the like involved in the present invention may be regarded as prior art.

[0040] Preferred embodiments.

[0041] like Figure 1-2 As shown, the present invention discloses a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization, comprising the following steps:

[0042] Step S1: The environment perception module calculates the standard deviation σ of the window data through the raw data collected by the sensor node and determines whether the standard deviation σ exceeds the dynamic threshold θ, thereby monitoring the fluctuation of the environment data;

[0043] Step S2: The dynamic sampling controller switches the sampling frequency according to different states of environmental fluctuations, and reduces the energy consumption of the system while ensuring data quality through content-aware downsampling processing and intelligent batch transmission strategy.

[0044] Specifically, step S1 is implemented as follows:

[0045] Step S1.1: The raw data collected by each sensor node of the environment perception module is allocated to a fixed time window, so as to calculate the standard deviation σ of the window data, and the formula is as follows:

[0046]

[0047] Among them, x i is the data value of the i-th sampling point, is the average value of the window data, N is the total number of sampling points in the time window, and the fluctuation of the standard deviation σ directly reflects the changes in the environment;

[0048] When σ exceeds the preset dynamic threshold θ, the system determines that the environment is in a fluctuating state; when σ is lower than θ, the system determines that the environment is in a stable state;

[0049] Step S1.2: (In order to improve the accuracy of environmental state judgment) the dynamic threshold θ is adaptively adjusted according to environmental changes, and the calculation formula is:

[0050] θ=α·σ history +β

[0051] Among them, σ history It represents the average standard deviation of historical data, and α and β are adjustment parameters used to control the baseline and sensitivity of the dynamic threshold (this adaptive dynamic threshold mechanism helps the system maintain a reasonable sampling frequency in different scenarios and improve the accuracy of environmental status monitoring).

[0052] More specifically, step S2 is implemented as follows:

[0053] Step S2.1: For a stable state, the dynamic sampling controller sets the sampling frequency to a low frequency (e.g., sampling once per hour) to save energy;

[0054] Step S2.2: For the fluctuating state, the dynamic sampling controller sets the sampling frequency to a high frequency (eg, sampling once per minute) to capture the rapidly changing environmental information.

[0055] Furthermore, in step S2, (in order to further optimize the efficiency and real-time performance of the sampling frequency switching) the dynamic sampling controller predicts the environmental fluctuations at future moments through a time series prediction model, so as to switch to the high-frequency mode in advance before the environmental fluctuations occur. The time series prediction model is as follows:

[0056]

[0057] Among them, φ i is the coefficient of the autoregressive term, θ j is the coefficient of the moving average term, ∈ t-j is the white noise term, x t-iRepresents past observation data. Parameters p, q, and t are the basic elements required for the definition and calculation of the model, which affect the accuracy and stability of the prediction. Based on the time series prediction model, adaptive sampling frequency adjustment is achieved on the basis of predicting the environmental state (reducing the system overhead caused by frequent frequency switching, while ensuring the effectiveness and timeliness of data collection). In the time series prediction model, the sensor node builds a model based on past sampling data to predict future environmental change trends. When the model predicts that the environment may fluctuate greatly, the system will automatically increase the sampling frequency to capture rapidly changing data in time; when the environment is expected to remain stable, the system will reduce the sampling frequency to reduce energy consumption. This adaptive adjustment logic not only improves the accuracy of sampling, but also effectively avoids the system overhead caused by frequent switching of sampling frequencies.

[0058] Furthermore, for step S2, (in high-frequency sampling mode, the amount of data collected by sensor nodes increases, and direct transmission will significantly increase communication energy consumption) the kernel prediction module and feature reorganization module of the CARAFE operator are introduced to perform content-aware downsampling processing on the sampled data (the CARAFE operator can identify the change area in the data and perform weighted sampling on the samples to retain key change information while removing redundant data. Unlike the traditional method using a 1×1 or 2×2 receptive field, the CARAFE operator has a larger receptive field and can consider the entire feature map, thereby better retaining the details and edge information of the image and reducing the jagged and blurring effects during the sampling process. The CARAFE operator supports dynamic generation of adaptive kernels and can achieve content-aware processing according to different goals, while the traditional interpolation method cannot dynamically adjust the kernel size to match the data content), where:

[0059] In the kernel prediction module, the adaptive convolution kernel is generated according to the input features and is related to the position, thereby improving the ability to identify and retain the change information. The kernel prediction module generates a position-dependent adaptive convolution kernel according to the input features. The formula is:

[0060] K i,j =g(X i,j )

[0061] Among them, K i,j is the convolution kernel matrix, X i,j is the characteristic information of the original data, function g is the kernel prediction network, and the adaptive convolution kernel structure allows dynamic weighted optimization according to the characteristics of the environmental data;

[0062] In the weighted sampling step, the feature reconstruction module effectively extracts the data change area through resampling, and the CARAFE operator generates weights W according to the generated i,j, perform weighted convolution calculation on the sampled data, and use the convolution kernel K to perform content-aware resampling of the data area. The formula is:

[0063]

[0064] Among them, y i,j is the data after resampling, K is the convolution kernel, the size is (2k+1)×(2k+1), x i+m,j+n is the original data point, i and j represent the target position coordinates of the resampled data in the convolution operation, m and n are the relative position offsets of the convolution kernel, which are used to sample the original data points within the range of the convolution kernel. By weighted sampling in the salient area, the CARAFE operator can effectively remove unnecessary data redundancy, while reducing the amount of data and retaining the characteristic information of the environment (this method can reduce the amount of data transmission in high-frequency sampling mode while maintaining key change information, effectively improving the energy efficiency of data processing). Weight W i,j The role of the CARAFE algorithm is to weight the convolution kernel at each position to highlight the more important feature areas. W is not reflected in the formula i,j The actual calculation process of is only mentioned as the result of generating the convolution kernel weights, and the specific application method is not shown in the formula. This weighting mechanism can more effectively concentrate sampling on data areas with significant changes and improve the accuracy of data processing.

[0065] Preferably, for step S2, an intelligent batch data transmission strategy is used to solve the data volume problem caused by batch transmission, wherein:

[0066] In a stable state, the system temporarily stores the sampled data in a local storage unit and transmits it in batches when the environment enters a fluctuating state, thereby reducing the number of transmissions and communication power consumption. The system uses the following formula to quantify the total power consumption during the transmission process:

[0067]

[0068] Among them, T total Represents the total power consumption of the system, T sampling represents the power consumption of data acquisition, D represents the amount of data transmitted in batches, R represents the data transmission rate, and P transmit It represents the transmission power consumption. By reasonably controlling the size of the batch data volume D and the transmission frequency, and preferably using a low-power communication protocol (such as LoRa or BLE), the communication power consumption can be reduced and the service life of the sensor node can be extended while ensuring data integrity.

[0069] The advantages of the present invention are:

[0070] The dynamic sampling method proposed in the present invention not only effectively reduces the energy consumption of sensor nodes, but also improves the efficiency and accuracy of data acquisition. Through the environmental fluctuation perception module, the algorithm can automatically adapt to different environmental changes, dynamically adjust the sampling frequency, and avoid unnecessary high-frequency sampling. At the same time, the content-aware CARAFE operator optimizes the features of the collected data to ensure that key information is not lost during low-frequency sampling, thereby improving the accuracy of data acquisition. Compared with traditional fixed sampling frequencies or simple adaptive sampling, the present invention achieves a higher energy efficiency ratio in a low-power environment, and provides a real-time and adaptable solution for wireless sensor networks and remote monitoring, which is suitable for scenarios with high requirements for energy consumption and data accuracy.

[0071] Through the dual judgment mechanism of standard deviation fluctuation and adaptive threshold, the system effectively distinguishes between the "fluctuation" and "stable" states of the environment, and achieves a dynamic balance between sampling frequency and energy consumption. Combined with the time series prediction model, the algorithm can predict environmental fluctuations before switching the sampling frequency in advance, thereby reducing the overhead of frequent switching. At the same time, through the core prediction and feature recombination modules of the CARAFE operator, the system performs weighted sampling on high-frequency sampled data, significantly compresses the data volume, and improves the energy efficiency of transmission and storage. Combining the above technologies, the invention is suitable for low-power remote monitoring and wireless sensor network scenarios, ensuring the validity of the data without increasing power consumption, and can meet the application requirements with high requirements for environmental status monitoring accuracy and energy efficiency.

[0072] It is worth mentioning that the technical features such as sensors involved in the patent application of this invention should be regarded as prior art. The specific structure, working principle and possible control method and spatial layout method of these technical features can be selected by conventional methods in the field, and should not be regarded as the inventive point of this patent. This patent will not be further elaborated.

[0073] For those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization, characterized in that: The following steps are involved: Step S1: The environment perception module calculates the standard deviation σ of the window data through the raw data collected by the sensor node and determines whether the standard deviation σ exceeds the dynamic threshold θ, thereby monitoring the fluctuation of the environment data; Step S2: The dynamic sampling controller switches the sampling frequency according to different states of environmental fluctuations, and reduces the energy consumption of the system while ensuring data quality through content-aware downsampling processing and intelligent batch transmission strategy.

2. According to claim 1, a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization is characterized in that: Step S1 is specifically implemented as follows: Step S1.1: The raw data collected by each sensor node of the environment perception module is allocated to a fixed time window, so as to calculate the standard deviation σ of the window data, and the formula is as follows: Among them, x i is the data value of the i-th sampling point, is the average value of the window data, N is the total number of sampling points in the time window, and the fluctuation of the standard deviation σ directly reflects the changes in the environment; When σ exceeds the preset dynamic threshold θ, the system determines that the environment is in a fluctuating state; when σ is lower than θ, the system determines that the environment is in a stable state; Step S1.2: The dynamic threshold θ is adaptively adjusted according to environmental changes, and the calculation formula is: θ=a·s history +b Among them, σ history It represents the average standard deviation of historical data. α and β are adjustment parameters used to control the baseline and sensitivity of the dynamic threshold.

3. According to claim 2, a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization is characterized in that: Step S2 is specifically implemented as the following steps: Step S2.1: For a stable state, the dynamic sampling controller sets the sampling frequency to a low frequency to save energy; Step S2.2: For the fluctuating state, the dynamic sampling controller sets the sampling frequency to a high frequency to capture the rapidly changing environmental information.

4. According to claim 3, a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization is characterized in that: In step S2, the dynamic sampling controller predicts the environmental fluctuations at future moments through a time series prediction model, so as to switch to the high-frequency mode in advance before the environmental fluctuations occur. The time series prediction model is as follows: Among them, φ i is the coefficient of the autoregressive term, θ j is the coefficient of the moving average term, ∈ t-j is the white noise term, x t-i Represents past observation data. Parameters p, q, and t are the basic elements required for the definition and calculation of the model. Based on the time series prediction model, adaptive sampling frequency adjustment is achieved based on the predicted environmental state.

5. According to claim 4, a sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization is characterized in that: For step S2, the kernel prediction module and feature reconstruction module of the CARAFE operator are introduced to perform content-aware downsampling processing on the sampled data, where: In the kernel prediction module, the adaptive convolution kernel is generated according to the input features and is related to the position, thereby improving the ability to identify and retain the change information. The kernel prediction module generates a position-dependent adaptive convolution kernel according to the input features. The formula is: K i，j =g(X i，j ) Among them, K i，j is the convolution kernel matrix, X i，j is the characteristic information of the original data, function g is the kernel prediction network, and the adaptive convolution kernel structure allows dynamic weighted optimization according to the characteristics of the environmental data; In the weighted sampling step, the feature reconstruction module effectively extracts the data change area through resampling, and the CARAFE operator generates weights W according to the generated i，j , perform weighted convolution calculation on the sampled data, and use the convolution kernel K to perform content-aware resampling of the data area. The formula is: Among them, y i , j is the data after resampling, K is the convolution kernel, the size is (2k+1)×(2k+1), x i+m，j+n is the original data point, i and j represent the target position coordinates of the resampled data in the convolution operation, m and n are the relative position offsets of the convolution kernel, which are used to sample the original data points within the range of the convolution kernel. By weighted sampling in the salient area, the CARAFE operator effectively removes unnecessary data redundancy and retains the characteristic information of the environment while reducing the amount of data.

6. The sensor dynamic sampling method based on environmental fluctuation perception and CARAFE optimization according to claim 5 is characterized in that: For step S2, the data volume problem caused by batch transmission is solved by an intelligent batch data transmission strategy, wherein: In a stable state, the system temporarily stores the sampled data in a local storage unit and transmits it in batches when the environment enters a fluctuating state, thereby reducing the number of transmissions and communication power consumption. The system uses the following formula to quantify the total power consumption during the transmission process: Among them, T total Represents the total power consumption of the system, T sampling represents the power consumption of data acquisition, D represents the amount of data transmitted in batches, R represents the data transmission rate, and P transmit It represents the transmission power consumption. By reasonably controlling the size of batch data D and the transmission frequency, and preferably using low-power communication protocols, the communication power consumption can be reduced and the service life of the sensor nodes can be extended while ensuring data integrity.

Images