An online data completion method for large-scale sparse mobile IoT perception

Abstract

The present invention discloses an online data completion method for large-scale sparse mobile Internet of Things perception, comprising: S1, dividing a target perception area into a plurality of sub-areas; S2, randomly assigning a perception importance score matrix of the first cycle to perform a cold start; S3, obtaining a pre-filled perception matrix for the next cycle; S4, after the next cycle arrives, selecting sub-areas that need to be perceived in the current cycle according to the perception importance score matrix of the current cycle, and collecting and uploading real data of the sub-areas that need to be perceived to obtain a sparse perception matrix for the current cycle; S5, using the sparse perception matrix of the current cycle to obtain a complete perception matrix for the corresponding cycle; S6, based on the pre-filled perception matrix and the complete perception matrix of the corresponding cycle, obtaining a perception importance score matrix for the target area in the next cycle; S7, adding the complete perception matrix of the corresponding cycle to a historical completion data set, entering a loop for the next cycle, and returning to S3.

Description

Technical Field

[0001] The present invention belongs to the field of mobile computing and mobile Internet of Things perception technology, and in particular relates to an online data completion method for large-scale sparse mobile Internet of Things perception. Background Art

[0002] With the widespread adoption of portable devices, mobile IoT sensing has become a mainstream data collection method. It collects data through sensors (such as GPS, accelerometers, and cameras) embedded in participants and their devices, and is widely used in fields such as environmental monitoring, traffic scheduling, and disease control. For example, precision agriculture relies on extensive real-time environmental data, such as soil moisture, temperature, and crop growth. Mobile IoT sensing leverages participants and their mobile devices to provide stable data support for crop monitoring and precision fertilization, helping farmers improve agricultural production efficiency without the need for specialized sensors.

[0003] However, mobile IoT sensing faces challenges such as high recruitment costs and high participant mobility, resulting in insufficient data coverage. This has led to the rise of sparse mobile IoT sensing. This approach recruits a small number of participants to collect data from a limited area and uses data completion algorithms to infer values for unperceived areas, thereby reducing costs while ensuring perception quality. For example, in precision agriculture, data can be selectively collected from a limited area and then inferred from a data completion algorithm to generate global agricultural data.

[0004] Traditional data completion methods often use sparse matrix modeling, dividing the target area into multiple sub-areas, accumulating data from several periods to form a matrix, and then using methods such as matrix completion, deep decomposition, and compressed sensing to fill in missing values. Matrix completion infers missing values by exploiting the low-rank nature of the data; deep decomposition uses nonlinear relationships to approximate real data; and compressed sensing reduces sampling costs by modeling signal sparsity. These methods are effective under certain conditions, but when faced with sparse mobile IoT sensing tasks, the matrix size increases dramatically, significantly increasing computational costs and resource requirements, and potentially reducing completion accuracy.

[0005] Existing methods also suffer from insufficient real-time performance. Many data technologies require the accumulation of multiple cycles of data before completing them in a unified manner, making it difficult to quickly complete data collection after a single cycle. This delay limits their performance in applications with high real-time requirements, such as traffic scheduling and environmental monitoring.

[0006] Furthermore, the importance of different subregions fluctuates with environmental changes. A single-minded focus on high-importance regions can lead to neglect of less important ones, missing critical information and limiting the system's dynamic adaptability. Therefore, dynamically assessing the importance of subregions and rationally selecting sensing regions is crucial. This not only optimizes data collection efficiency but also enhances the system's responsiveness to changing environments.

[0007] In summary, it is necessary to invent a data completion method that can cope with large-scale sparse perception tasks, avoid data accumulation over multiple cycles, obtain complete and high-precision perception results online, and dynamically adjust the data collection strategy through importance evaluation and sub-region selection to improve the efficiency and accuracy of overall data completion. Summary of the Invention

[0008] To solve the above technical problems, the present invention proposes an online data completion method for large-scale sparse mobile Internet of Things perception, which realizes the efficient completion of large-scale sparse task data and significantly improves the accuracy and timeliness of the completion.

[0009] To achieve the above objectives, the present invention provides an online data completion method for large-scale sparse mobile Internet of Things perception, comprising:

[0010] S1. Divide the target perception area into several sub-areas to form a spatial grid representation of the perception area, and divide the time intervals of equal length into perception cycles;

[0011] S2, cold start the first cycle’s perceived importance score matrix by random assignment;

[0012] S3. Based on the complete perception matrix in the historical completion dataset, the ConvLSTM model is used to obtain the pre-filled perception matrix of each sub-region in the next cycle;

[0013] S4. When the next cycle arrives, select the sub-areas that need to be sensed in the current cycle based on the perception importance score matrix of the current cycle, and recruit participants to collect and upload real data of the sub-areas that need to be sensed to obtain the sparse perception matrix of the current cycle;

[0014] S5. Using the sparse perception matrix of the current cycle, the pre-filled perception matrix is corrected online through the BERT-based model to obtain the complete perception matrix of the corresponding cycle;

[0015] S6. Obtaining a perception importance score matrix of the target area in the next cycle based on the pre-filled perception matrix and the complete perception matrix of the corresponding cycle;

[0016] S7. Add the complete perception matrix of the corresponding period to the historical completion data set, enter the loop of the next period, and return to S3.

[0017] Optionally, obtaining the sparse sensing matrix of the current cycle includes:

[0018] Obtain the importance score of each sub-region according to the perceived importance score matrix of the current cycle;

[0019] Based on the importance score of each sub-region, combined with the improved multi-armed bandit model, an ∈-greedy strategy is introduced to select the sub-region that needs to be sensed in the current cycle;

[0020] According to the sub-areas that need to be sensed in the selected current period, a list of sub-areas that need to be sensed in the current period is obtained;

[0021] According to the list of sub-areas that need to be perceived in the current cycle, participants are recruited to collect and upload real data of the sub-areas that need to be perceived to obtain the sparse perception matrix of the current cycle.

[0022] Optionally, based on the importance score of each sub-region, combined with the improved multi-armed bandit model, an ∈-greedy strategy is introduced to select the sub-regions that need to be sensed in the current cycle, including:

[0023]

[0024] Where P[i, j] is the probability of selecting the subregion at position (i, j), ∈ is the exploration ratio, L is the total number of perception regions, and I[i, j] is the importance score of each subregion.

[0025] Optionally, the list of sub-areas to be sensed in the current cycle is obtained, including:

[0026] S d ={(i1,j1),(i2,j2),...,(i d ,j d )},

[0027] Among them, S d is the list of sub-areas to be sensed in the current cycle, (i1, j1), (i2, j2), ..., (i d ,j d ) is the selected position, and d is the preset number of sub-areas to be sensed.

[0028] Optionally, obtaining a complete sensing matrix for a corresponding period includes:

[0029] Combine the sparse perception matrix of the current cycle with the pre-filled perception matrix to obtain the model input matrix;

[0030] Constructing a spatiotemporal embedding layer to embed each value in the model input matrix with spatial information, time information of the current cycle, value, and value type information, and flattening the matrix to obtain a model input sequence;

[0031] Build a BERT model, input the model input sequence into the BERT model, and map the output result to 1D through a fully connected layer to obtain a data correction matrix;

[0032] A complete sensing matrix of a corresponding period is obtained based on the sparse sensing matrix, the pre-filled sensing matrix and the data correction matrix.

[0033] Optionally, combining the sparse sensing matrix of the current cycle with the pre-filled sensing matrix includes:

[0034]

[0035] Among them, X input is the model input matrix, X input [i,j] is the value at position (i,j) of the two-dimensional input matrix, is the sparse perception matrix, is the pre-filled perception matrix.

[0036] Optionally, obtaining a complete sensing matrix for a corresponding period based on the sparse sensing matrix, the pre-filled sensing matrix, and the data correction matrix includes:

[0037]

[0038] Among them, X′ K+1 is the complete perception matrix, To pre-fill the perception matrix, is the data correction matrix, is the sparse perception matrix.

[0039] Optionally, obtaining a perception importance score matrix of a target area in a next period based on the pre-filled perception matrix and the complete perception matrix of the corresponding period includes:

[0040] According to the data correction matrix, the pre-filling importance of each sub-region in the current period is obtained by calculation;

[0041] Obtaining, by calculation, a correction importance of each sub-region in a current period according to the pre-filled sensing matrix, the data correction matrix, and the sparse sensing matrix;

[0042] Calculating the information gain of each sub-region according to the pre-filled perception matrix and the complete perception matrix;

[0043] The importance score of each region is updated according to the pre-filled importance of each sub-region in the current cycle, the revised importance of each sub-region in the current cycle, and the information gain of each sub-region.

[0044] Technical effects of the present invention: The present invention discloses an online data completion method for large-scale sparse mobile Internet of Things perception, which realizes efficient completion of large-scale sparse task data and significantly improves the accuracy and timeliness of completion; the present invention can quickly correct prediction results online, timely reflect environmental changes, and optimize resource allocation and utilization by comprehensively evaluating the importance of sub-regions, ensuring data collection in key areas; in addition, the present invention has good adaptability to dynamic environments, can flexibly adjust perception strategies, balance exploration and utilization, avoid information blind spots, thereby improving the comprehensiveness and effectiveness of overall data completion. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0046] Figure 1 The figure is a flow chart of an online data completion method for large-scale sparse mobile Internet of Things perception according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0048] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0049] like Figure 1 As shown, this embodiment provides an online data completion method for large-scale sparse mobile Internet of Things perception, including:

[0050] S1. Divide the target perception area into several sub-areas to form a spatial grid representation of the perception area, and divide the time intervals of equal length into perception cycles;

[0051] S2, cold start the first cycle’s perceived importance score matrix by random assignment;

[0052] S3. Based on the complete perception matrix in the historical completion dataset, the ConvLSTM model is used to obtain the pre-filled perception matrix of each sub-region in the next cycle;

[0053] S4. When the next cycle arrives, select the sub-areas that need to be sensed in the current cycle based on the perception importance score matrix of the current cycle, and recruit participants to collect and upload real data of the sub-areas that need to be sensed to obtain the sparse perception matrix of the current cycle;

[0054] S5. Using the sparse perception matrix of the current cycle, the pre-filled perception matrix is corrected online through the BERT-based model to obtain the complete perception matrix of the corresponding cycle;

[0055] S6. Obtaining a perception importance score matrix of the target area in the next cycle based on the pre-filled perception matrix and the complete perception matrix of the corresponding cycle;

[0056] S7. Add the complete perception matrix of the corresponding period to the historical completion data set, enter the loop of the next period, and return to S3.

[0057] Furthermore, obtaining the sparse perception matrix of the current cycle includes:

[0058] Obtain the importance score of each sub-region according to the perceived importance score matrix of the current cycle;

[0059] Based on the importance score of each sub-region, combined with the improved multi-armed bandit model, an ∈-greedy strategy is introduced to select the sub-region that needs to be sensed in the current cycle;

[0060] According to the sub-areas that need to be sensed in the selected current period, a list of sub-areas that need to be sensed in the current period is obtained;

[0061] According to the list of sub-areas that need to be perceived in the current cycle, participants are recruited to collect and upload real data of the sub-areas that need to be perceived to obtain the sparse perception matrix of the current cycle.

[0062] Furthermore, based on the importance score of each sub-region, combined with the improved multi-armed bandit model, an ∈-greedy strategy is introduced to select the sub-regions that need to be sensed in the current cycle, including:

[0063]

[0064] Where P[i, j] is the probability of selecting the subregion at position (i, j), ∈ is the exploration ratio, L is the total number of perception regions, and I[i, j] is the importance score of each subregion.

[0065] Get the list of sub-areas that need to be sensed in the current cycle, including:

[0066] S d ={(i1,j1),(i2,j2),...,(i d ,j d )},

[0067] Among them, S d is the list of sub-areas to be sensed in the current cycle, (i1,j1), (i2,j2),..., (i d ,j d ) is the selected position, and d is the preset number of sub-areas to be sensed.

[0068] Furthermore, obtaining a complete perception matrix for the corresponding period includes:

[0069] Combine the sparse perception matrix of the current cycle with the pre-filled perception matrix to obtain the model input matrix;

[0070] Constructing a spatiotemporal embedding layer to embed each value in the model input matrix with spatial information, time information of the current cycle, value, and value type information, and flattening the matrix to obtain a model input sequence;

[0071] Build a BERT model, input the model input sequence into the BERT model, and map the output result to 1D through a fully connected layer to obtain a data correction matrix;

[0072] A complete sensing matrix of a corresponding period is obtained based on the sparse sensing matrix, the pre-filled sensing matrix and the data correction matrix.

[0073] Furthermore, combining the sparse sensing matrix of the current cycle with the pre-filled sensing matrix includes:

[0074]

[0075] Among them, X input is the model input matrix, X input [i, j] is the value at position (i, j) of the two-dimensional input matrix, is the sparse perception matrix, is the pre-filled perception matrix.

[0076] Obtaining a complete sensing matrix for a corresponding period based on the sparse sensing matrix, the pre-filled sensing matrix, and the data correction matrix includes:

[0077]

[0078] Among them, X′ K+1 is the complete perception matrix, To pre-fill the perception matrix, is the data correction matrix, is the sparse perception matrix.

[0079] Furthermore, based on the pre-filled perception matrix and the complete perception matrix of the corresponding period, obtaining the perception importance score matrix of the target area in the next period includes:

[0080] According to the data correction matrix, the pre-filling importance of each sub-region in the current period is obtained by calculation;

[0081] Obtaining, by calculation, a correction importance of each sub-region in a current period according to the pre-filled sensing matrix, the data correction matrix, and the sparse sensing matrix;

[0082] Calculating the information gain of each sub-region according to the pre-filled perception matrix and the complete perception matrix;

[0083] The importance score of each region is updated according to the pre-filled importance of each sub-region in the current cycle, the revised importance of each sub-region in the current cycle, and the information gain of each sub-region.

[0084] As a technical means that can be added, based on the data correction matrix, the pre-filling importance of each sub-region in the current cycle is obtained by calculation, including:

[0085]

[0086] in, The pre-fill importance of each sub-region for the current period, Correction matrix for the data.

[0087] As a technical means that may be added, obtaining the correction importance of each sub-region in the current period by calculation based on the pre-filled sensing matrix, the data correction matrix, and the sparse sensing matrix includes:

[0088]

[0089] Among them, the correction importance of each sub-area in the current period is To pre-fill the perception matrix, is the data correction matrix, is the sparse perception matrix.

[0090] As a technical means that may be added, calculating the information gain of each sub-region according to the pre-filled perception matrix and the complete perception matrix includes:

[0091] The information entropy formula is: Where P(x) is the distribution of perception data in the sub-region;

[0092] The information gain of each sub-region is:

[0093] As a technical measure that can be added, the importance score of each region is updated according to the pre-filled importance of each sub-region in the current period, the revised importance of each sub-region in the current period, and the information gain of each sub-region, including:

[0094]

[0095] Among them, α, β, and γ are weight coefficients that can be adjusted according to the specific application scenario. This formula combines prediction accuracy, completion accuracy, and information gain to provide a comprehensive sub-region importance assessment to improve the overall data completion accuracy.

[0096] The solution of the present invention has shown significant technical effects in many aspects, including the following advantages:

[0097] (1) High accuracy in data completion for large-scale sparse tasks

[0098] The present invention achieves efficient completion of sparse data by combining historical data and advanced deep learning models (such as ConvLSTM and BERT), significantly improving the accuracy of data completion for large-scale sparse tasks. First, by using the data from the first few cycles in the historical completion dataset, the ConvLSTM model can capture the complex patterns and trends in the time series, making the preliminary perception matrix have higher accuracy. Secondly, after the data collection of each cycle is completed, online correction is performed through real data, and the system can dynamically adjust the prediction results based on the latest information. In addition, the sub-region importance evaluation part and the perception region selection part further enhance the effect of data completion. By comprehensively evaluating the pre-filled importance, corrected importance and information gain of each sub-region, the system can accurately identify key areas, thereby optimizing resource allocation and improving the accuracy of data completion in dynamically changing large-scale perception systems. This multi-level mechanism ensures that the system can provide reliable perception information in a large-scale data environment, significantly improving the accuracy of prediction, especially in a sparse data environment, effectively reducing the impact of missing data and ensuring the comprehensiveness and accuracy of overall data completion.

[0099] (2) Realized online acquisition of data completion results

[0100] The present invention focuses on completing the sparse data of the current cycle, and adopts an online correction method so that the data completion results can be quickly obtained after the end of the cycle. By quickly collecting real data and making immediate corrections within each cycle, the system can promptly reflect environmental changes and ensure the timeliness of perception data. Specifically, after the end of each cycle, the present invention uses the newly obtained sparse perception matrix and corrects it through a BERT-based model to quickly generate a complete perception matrix. This ability to obtain completion results online greatly improves the response speed and flexibility of the system in a dynamic environment, allowing users to obtain accurate data in the shortest time and adapt to rapidly changing needs.

[0101] (3) Optimize resource allocation and utilization

[0102] The present invention effectively optimizes the allocation and utilization of resources by comprehensively evaluating the pre-filling importance, correction importance and information gain of each sub-region. First, the pre-filling importance helps to identify key areas that have a greater impact on the accuracy of pre-filling, so that the accuracy of pre-filling is further improved. Secondly, the correction importance analysis ensures that important areas are preferentially perceived during the correction stage, so that the data in this area can further improve the accuracy of the completion results in the subsequent data correction steps. The information gain measures the degree of improvement in the overall data completion after collecting data from a certain sub-region, which promotes the improvement of the accuracy of the data completion results as a whole. This comprehensive evaluation mechanism enables the system to accurately identify key sub-regions and reasonably allocate resources, thereby ensuring the best data completion effect with limited resources.

[0103] (4) Dynamic environmental adaptability

[0104] The present invention has good adaptability to dynamic environments and can adjust perception strategies in a timely manner to cope with environmental changes. By analyzing the importance of each sub-area, the system not only focuses on data collection in areas of high importance, but also maintains exploration of areas of low importance. This balanced strategy of exploration and utilization ensures that potentially important information is not missed in the long-term perception process, and avoids information blind spots caused by excessive concentration on high-importance areas. As the perception environment changes dynamically, the present invention can flexibly adjust perception strategies and respond to changing needs in a timely manner, thereby ensuring the comprehensiveness and accuracy of data collection and improving the flexibility and effectiveness of the system in practical applications.

[0105] In summary, the present invention focuses on completing the sparse data of the current cycle, rather than accumulating sparse data of multiple cycles to construct a sparse matrix, so that the data completion result can be obtained after the end of the cycle; the present invention first uses the historical complete data to pre-fill the perception data of the next cycle through the ConvLSTM model, and after the sparse perception data of the next cycle arrives, it embeds time, space and value type features, and uses the BERT model to correct the pre-filled data with sparse real data to obtain complete perception data; the present invention comprehensively evaluates the pre-filling importance, correction importance and information gain of each sub-region, and forms a new importance score through weighted combination for the next A reference is provided for the sub-region selection of one cycle. This multi-dimensional importance evaluation mechanism ensures that the system can accurately identify key areas and reasonably allocate resources, thereby making the data pre-filling step and the data correction step more effective, and further improving the accuracy of the overall data completion; the present invention models the sub-region selection problem as an improved multi-armed bandit model and introduces an ∈-greedy strategy. It can dynamically select the sub-region to be perceived according to the perception importance score in each cycle, ensuring priority perception of high-importance areas while maintaining exploration of low-importance areas, thereby achieving an effective balance between exploration and utilization, and optimizing the dynamic adaptability of data completion.

[0106] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An online data completion method for large-scale sparse mobile Internet of Things perception, characterized by: include: S1. Divide the target perception area into several sub-areas to form a spatial grid representation of the perception area, and divide the time intervals of equal length into perception cycles; S2, cold start the first cycle’s perceived importance score matrix by random assignment; S3. Based on the complete perception matrix in the historical completion dataset, the ConvLSTM model is used to obtain the pre-filled perception matrix of each sub-region in the next cycle; S4. When the next cycle arrives, select the sub-areas that need to be sensed in the current cycle based on the perception importance score matrix of the current cycle, and recruit participants to collect and upload real data of the sub-areas that need to be sensed to obtain the sparse perception matrix of the current cycle; S5. Using the sparse perception matrix of the current cycle, the pre-filled perception matrix is corrected online through the BERT-based model to obtain the complete perception matrix of the corresponding cycle; S6. Obtaining a perception importance score matrix of the target area in the next cycle based on the pre-filled perception matrix and the complete perception matrix of the corresponding cycle; S7. Add the complete perception matrix of the corresponding period to the historical completion data set, enter the loop of the next period, and return to S3.

2. The online data completion method for large-scale sparse mobile Internet of Things perception according to claim 1 is characterized in that: Obtaining the sparse perception matrix of the current cycle includes: Obtain the importance score of each sub-region according to the perceived importance score matrix of the current cycle; Based on the importance score of each sub-region, combined with the improved multi-armed bandit model, an ∈-greedy strategy is introduced to select the sub-region that needs to be sensed in the current cycle; According to the sub-areas that need to be sensed in the selected current period, a list of sub-areas that need to be sensed in the current period is obtained; According to the list of sub-areas that need to be perceived in the current cycle, participants are recruited to collect and upload real data of the sub-areas that need to be perceived to obtain the sparse perception matrix of the current cycle.

3. The online data completion method for large-scale sparse mobile Internet of Things perception according to claim 2 is characterized in that: Based on the importance score of each sub-region, combined with the improved multi-armed bandit model, the ∈-greedy strategy is introduced to select the sub-regions that need to be sensed in the current cycle, including: Where P[i, j] is the probability of selecting the subregion at position (i, j), ∈ is the exploration ratio, L is the total number of perception regions, and I[i, j] is the importance score of each subregion.

4. The online data completion method for large-scale sparse mobile Internet of Things perception according to claim 2 is characterized in that: Get the list of sub-areas that need to be sensed in the current cycle, including: S d ={(i1,j1),(i2,j2),...,(i d ,j d )}, Among them, S d is the list of sub-areas to be sensed in the current cycle, (i1, j1), (i2, j2), ..., (i d ,j d ) is the selected position, and d is the preset number of sub-areas to be sensed.

5. The online data completion method for large-scale sparse mobile Internet of Things perception according to claim 1 is characterized in that: Obtaining the complete perception matrix for the corresponding period includes: Combine the sparse perception matrix of the current cycle with the pre-filled perception matrix to obtain the model input matrix; Constructing a spatiotemporal embedding layer to embed each value in the model input matrix with spatial information, time information of the current cycle, value, and value type information, and flattening the matrix to obtain a model input sequence; Build a BERT model, input the model input sequence into the BERT model, and map the output result to 1D through a fully connected layer to obtain a data correction matrix; A complete sensing matrix of a corresponding period is obtained based on the sparse sensing matrix, the pre-filled sensing matrix and the data correction matrix.

6. The online data completion method for large-scale sparse mobile Internet of Things perception according to claim 5 is characterized in that: Combining the sparse perception matrix of the current cycle with the pre-filled perception matrix includes: Among them, X input is the model input matrix, X input [i, j] is the value at position (i, j) of the two-dimensional input matrix, is the sparse perception matrix, is the pre-filled perception matrix.

7. The online data completion method for large-scale sparse mobile Internet of Things perception according to claim 5 is characterized in that: Obtaining a complete sensing matrix for a corresponding period based on the sparse sensing matrix, the pre-filled sensing matrix, and the data correction matrix includes: Among them, X′ K+1 is the complete perception matrix, To pre-fill the perception matrix, is the data correction matrix, is the sparse perception matrix.

8. The online data completion method for large-scale sparse mobile Internet of Things perception according to claim 5 is characterized in that: Based on the pre-filled perception matrix and the complete perception matrix of the corresponding cycle, the perception importance score matrix of the target area in the next cycle is obtained, which includes: According to the data correction matrix, the pre-filling importance of each sub-region in the current period is obtained by calculation; Obtaining, by calculation, a correction importance of each sub-region in a current period according to the pre-filled sensing matrix, the data correction matrix, and the sparse sensing matrix; Calculating the information gain of each sub-region according to the pre-filled perception matrix and the complete perception matrix; The importance score of each region is updated according to the pre-filled importance of each sub-region in the current cycle, the revised importance of each sub-region in the current cycle, and the information gain of each sub-region.

Images