Target situation awareness method and system based on low-altitude Internet of Things

Abstract

The invention relates to the technical field of low-altitude Internet of Things, and particularly provides a target situation awareness method and system based on the low-altitude Internet of Things, and the method comprises the steps: firstly obtaining a target awareness data set which is cooperatively collected by multiple nodes in the low-altitude Internet of Things and comprises three-dimensional position data and environment disturbance data, and then carrying out the feature extraction of the target awareness data set, the method comprises the following steps: generating spatial distribution characteristics and continuous change characteristics of a target object, inputting the spatial distribution characteristics and the continuous change characteristics into a pre-trained situation awareness model to generate situation association characteristics, and obtaining current situation attributes and situation development tendency information of the target object through a situation classification layer; and finally, according to the current situation attribute and the situation development tendency information, generating a collaborative awareness instruction containing a target positioning identifier, and sending the collaborative awareness instruction to a low-altitude Internet of Things management terminal, thereby comprehensively and accurately sensing the low-altitude target situation, providing an effective collaborative awareness instruction for the low-altitude Internet of Things, and improving the safety and management efficiency of the low-altitude field.

Description

Technical Field

[0001] The present invention relates to the technical field of low-altitude intelligent networking, and in particular, to a method and system for target situation awareness based on low-altitude intelligent networking. Background Art

[0002] In the low-altitude field, with the increasing number of targets such as unmanned aerial vehicles and low-altitude aircraft, the complexity and uncertainty of the low-altitude environment have increased significantly. As a new type of network architecture, the low-altitude intelligent network provides a data basis for the situation awareness of low-altitude targets by collecting target perception data through multi-node collaboration. However, there are many deficiencies in the existing low-altitude target situation awareness technologies.

[0003] On the one hand, existing technologies often only focus on the position information of targets and ignore the important role of environmental disturbance data in target situations. In fact, environmental factors such as wind speed, wind direction, and air pressure have a significant impact on the flight state of low-altitude targets, and it is difficult to accurately grasp the true situation of targets only relying on position information. On the other hand, in terms of data processing, existing technologies lack effective feature extraction of target perception data and cannot fully mine the feature information of target objects in the spatial and temporal dimensions, resulting in insufficient accuracy and comprehensiveness of situation awareness. In addition, the existing situation awareness models have limitations in generating situation correlation features and situation classification, making it difficult to accurately judge the current situation attributes and situation development tendencies of target objects, and unable to provide effective collaborative perception instructions for the low-altitude intelligent network management terminal, thus affecting the timeliness and accuracy of linkage response operations. Summary of the Invention

[0004] In view of the problems mentioned above, in combination with the first aspect of the present invention, the present invention provides a method for target situation awareness based on low-altitude intelligent networking, and the method includes: Obtain a set of target perception data collected by multi-node collaboration in the low-altitude intelligent network, where the set of target perception data includes three-dimensional position data and environmental disturbance data collected by different nodes in a continuous time period; Perform feature extraction processing on the set of target perception data to generate the spatial distribution feature and the continuous change feature of the target object, where the spatial distribution feature represents the position correlation relationship of the target object in the low-altitude area, and the continuous change feature represents the state change law of the target object in a continuous time period; Input the spatial distribution feature and the continuous change feature into a pre-trained situation awareness model to generate the situation correlation feature of the target object; Input the situation correlation feature into the situation classification layer of the situation awareness model to obtain the current situation attribute and situation development tendency information of the target object; Generate a collaborative perception instruction containing a target positioning identifier according to the current situation attribute and the situation development tendency information, and send the collaborative perception instruction to the low-altitude intelligent network management terminal to trigger a linkage response operation.

[0005] In another aspect, the present invention further provides a target situation perception system based on a low-altitude intelligent network, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0006] Based on the above aspects, the present invention obtains a target perception data set containing three-dimensional position data and environmental disturbance data collected by multi-node collaboration, comprehensively considers various factors affecting the target situation, performs feature extraction processing on the target perception data set, generates the spatial distribution feature and continuous change feature of the target object, can accurately describe the state of the target object from two dimensions of space and time, inputs the spatial distribution feature and continuous change feature into a pre-trained situation perception model to generate situation correlation features, and further obtains the current situation attribute and situation development tendency information of the target object through a situation classification layer, realizing the accurate judgment and prediction of the target situation. Generate a collaborative perception instruction containing a target positioning identifier based on this information, and send it to the low-altitude intelligent network management terminal to trigger a linkage response operation, so that the low-altitude intelligent network can make corresponding adjustments and response measures in a timely manner according to the real-time situation of the target, improving the safety and management efficiency in the low-altitude field, and enhancing the intelligent level and collaborative combat ability of the low-altitude intelligent network. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a schematic execution flow diagram of a target situation perception method based on a low-altitude intelligent network provided by an embodiment of the present invention.

[0008] Figure 2 is a schematic diagram of exemplary hardware and software components of a target situation perception system based on a low-altitude intelligent network provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0009] The present invention will be specifically described below with reference to the accompanying drawings of the specification. Figure 1 is a schematic flow diagram of a target situation perception method based on a low-altitude intelligent network provided by an embodiment of the present invention. The target situation perception method based on a low-altitude intelligent network will be introduced in detail below.

[0010] Step S110: Obtain a target perception data set collected by multi-node collaboration in a low-altitude intelligent network. The target perception data set contains three-dimensional position data and environmental disturbance data collected by different nodes within a continuous time period.

[0011] In this embodiment, in the application scenario of the low-altitude intelligent Internet of Things, in order to accurately perceive the situation of the target, the target perception data set is first obtained. The target perception data set covers the three-dimensional position data and environmental disturbance data collected by different nodes within a continuous time period. For example, in a low-altitude area, there are multiple target objects, and the motion states of these target objects will be affected by the surrounding environment. Different nodes can comprehensively and continuously record the relevant information of the target objects through collaborative acquisition.

[0012] Step S111: Determine multiple sensing nodes in the low-altitude intelligent Internet of Things that cover the target monitoring area. The sensing nodes include positioning sensing devices deployed on low-altitude aircraft and environmental monitoring devices deployed on ground stations.

[0013] In the low-altitude intelligent Internet of Things, in order to achieve full coverage of the target monitoring area, it is necessary to reasonably arrange the sensing nodes. Among them, the positioning sensing devices deployed on low-altitude aircraft have the characteristics of mobility and flexibility, and can follow the movement of the target object in real time and collect its position information. The environmental monitoring devices deployed on ground stations can stably monitor the environmental conditions around the target object. For example, in a vast low-altitude area, there are multiple low-altitude aircraft flying at different heights and in different areas. Each aircraft is equipped with a positioning sensing device, and at the same time, multiple environmental monitoring stations are set up at different positions on the ground. These sensing nodes together constitute a data acquisition network covering the target monitoring area.

[0014] Step S112: Collect the three-dimensional position data of the target object through the positioning sensing device within a continuous time period. The three-dimensional position data includes the spatial coordinate points of the target object in each time period.

[0015] Then, use the positioning sensing device to collect the three-dimensional position data of the target object within a continuous time period. The positioning sensing device usually relies on advanced positioning technologies such as the Global Positioning System (GPS) or other high-precision positioning algorithms to accurately obtain the position of the target object in three-dimensional space. Within a continuous time period, the positioning sensing device will continuously record the spatial coordinate points of the target object, forming a continuous position data sequence. For example, within a certain period of time, the positioning sensing device will collect the three-dimensional position data of the target object at regular time intervals. These three-dimensional position data can be represented as a series of coordinate points (x1, y1, z1), (x2, y2, z2), etc. Through these coordinate points, the movement trajectory of the target object can be clearly depicted.

[0016] Step S113: Collect the environmental disturbance data around the target object through the environmental monitoring device within the same continuous time period. The environmental disturbance data includes air flow direction information and wind speed fluctuation information.

[0017] Meanwhile, the environmental monitoring device collects environmental disturbance data around the target object within the same continuous time period. The environmental monitoring device is usually equipped with a variety of sensors, such as a wind direction sensor and a wind speed sensor, which can accurately measure the air flow direction and wind speed. During the continuous time period, the environmental monitoring device continuously records the air flow direction information and wind speed fluctuation information, which are of great significance for analyzing the motion state and situation of the target object. For example, when the target object is in an area with strong air flow, its motion trajectory may be significantly affected, and the motion law of the target object can be better understood by analyzing the environmental disturbance data.

[0018] Step S114: Synchronize and align the three-dimensional position data and the environmental disturbance data according to the acquisition timestamps to generate a perception data unit with a time correspondence relationship.

[0019] After that, it is necessary to synchronize and align the collected three-dimensional position data and environmental disturbance data according to the acquisition timestamps. Since the three-dimensional position data and environmental disturbance data are collected by different devices, there may be a certain difference in their acquisition times. To ensure the accuracy and effectiveness of the data, these data need to be aligned according to the acquisition timestamps so that each set of three-dimensional position data can correspond to the corresponding environmental disturbance data. For example, the positioning sensing device collects the three-dimensional position data (x1, y1, z1) of the target object at time t1, and the environmental monitoring device collects the air flow direction information A1 and wind speed fluctuation information V1 at time t1. By aligning these data according to the timestamps, a perception data unit (t1, (x1, y1, z1), A1, V1) with a time correspondence relationship can be generated.

[0020] Step S115: Classify and integrate the perception data units according to the node identifiers to generate perception data subsets corresponding to different nodes, where the perception data subsets include the three-dimensional position data and environmental disturbance data collected by the corresponding perception nodes.

[0021] In this embodiment, each perception node has a unique node identifier, and the perception data units can be classified through this node identifier, so that the data belonging to the same node are integrated together. For example, for the perception data units collected by the positioning sensing device deployed on the low-altitude aircraft A, they are grouped into one category to form a perception data subset corresponding to this perception node; for the perception data units collected by the environmental monitoring device deployed at the ground station B, they are also grouped into one category to form another perception data subset corresponding to a perception node. Thus, the perception data subsets corresponding to different perception nodes include the three-dimensional position data and environmental disturbance data collected by the perception nodes.

[0022] Step S116: Perform cross-node information fusion processing on all subsets of perception data to generate a target perception data set containing multi-node collaboration information.

[0023] In the low-altitude intelligent network, the data collected by different nodes is complementary. Through cross-node information fusion, the advantages of these data can be fully utilized to generate a target perception data set containing multi-node collaboration information. For example, fusing the three-dimensional position data collected by the positioning and sensing devices on low-altitude aircraft and the environmental disturbance information collected by ground station environmental monitoring devices can provide a more comprehensive understanding of the motion state of the target object and the impact of the surrounding environment. During the fusion process, various fusion algorithms, such as data association methods, can be used to ensure that the fused data can accurately reflect the true situation of the target object.

[0024] Step S120: Perform feature extraction processing on the target perception data set to generate the spatial distribution features and continuous change features of the target object. The spatial distribution features represent the position association relationship of the target object in the low-altitude area, and the continuous change features represent the state change law of the target object in a continuous time period.

[0025] After obtaining the target perception data set, this step needs to perform feature extraction processing on it to extract the spatial distribution features and continuous change features of the target object. The spatial distribution features can reflect the position association relationship of the target object in the low-altitude area, such as the distance and relative position between target objects; the continuous change features reflect the state change law of the target object in a continuous time period, such as speed change and direction change.

[0026] Step S121: Perform spatial relationship analysis processing on the three-dimensional position data in the target perception data set to calculate the relative distance parameter and azimuth angle parameter between different target objects. The relative distance parameter represents the straight-line distance of the target object in three-dimensional space, and the azimuth angle parameter represents the direction angle of the target object relative to the reference point.

[0027] First, perform spatial relationship analysis processing on the three-dimensional position data in the target perception data set. In three-dimensional space, the position relationship between different target objects can be described by the relative distance parameter and azimuth angle parameter. To calculate these parameters, it is necessary to extract the three-dimensional position data of multiple target objects in the same time period from the target perception data set.

[0028] Step S1211: Extract the three-dimensional position data of multiple target objects in the same time period from the target perception data set. The three-dimensional position data includes the spatial coordinate points of each target object.

[0029] Next, extract the three-dimensional position data of multiple target objects within the same time period from the target perception data set. Since the target perception data set is collected within consecutive time periods, in order to accurately analyze the spatial relationships between target objects, it is necessary to select data from the same time period. For example, at time t, there are multiple target objects moving in the low-altitude area, and the positioning and sensing device records the three-dimensional position data of the above-mentioned target objects. Each three-dimensional position data contains the spatial coordinate points (x, y, z) of the target object.

[0030] Step S1212: Select the spatial coordinate point of one of the target objects as the reference point.

[0031] In this embodiment, the selection of the reference point can be random or determined according to actual needs. For example, select the coordinate point of a representative target object as the reference point to better analyze the positional relationship of other target objects relative to this reference point.

[0032] Step S1213: For each of the other target objects, extract its spatial coordinate point, and calculate the three-dimensional Euclidean distance between this spatial coordinate point and the reference point, and use the calculation result as the relative distance parameter between this target object and the reference point.

[0033] For each of the other target objects except the target object corresponding to the reference point, it is necessary to extract its spatial coordinate point and calculate the three-dimensional Euclidean distance between this coordinate point and the reference point. The three-dimensional Euclidean distance is a common method for measuring the straight-line distance between two points in three-dimensional space. For example, let the spatial coordinates of the reference point be (x0, y0, z0), and the spatial coordinates of another target object be (x1, y1, z1). Through the corresponding calculation method, the three-dimensional Euclidean distance between these two points can be obtained, and this distance is the relative distance parameter between this target object and the reference point.

[0034] Step S1214: Calculate the horizontal projection coordinates of the spatial coordinate point of this target object relative to the reference point, and the horizontal projection coordinates represent the position of the target object on the horizontal plane.

[0035] In this embodiment, the horizontal projection coordinates are the coordinates obtained by projecting the three-dimensional coordinates of the target object onto the horizontal plane, and they can represent the position of the target object on the horizontal plane. For example, by ignoring the z coordinate in the three-dimensional coordinates (x, y, z) of the target object, the projection coordinates (x, y) on the horizontal plane are obtained, and these projection coordinates are the horizontal projection coordinates of the target object relative to the reference point.

[0036] Step S1215: Calculate the azimuth angle between the horizontal projection coordinates, and use the calculation result as the azimuth angle parameter of this target object relative to the reference point.

[0037] Then, calculate the azimuth angle between the horizontal projection coordinates. The azimuth angle represents the included angle of the direction of the target object relative to the reference point. By performing corresponding calculations on the horizontal projection coordinates, the azimuth angle parameter of the target object relative to the reference point can be obtained. For example, using methods such as trigonometric functions, the azimuth angle is calculated based on the differences in the horizontal projection coordinates, and this azimuth angle can accurately describe the direction of the target object relative to the reference point.

[0038] Step S1216: Summarize and organize the relative distance parameters and azimuth angle parameters of all target objects to generate a spatial relationship analysis result including the spatial relationships of multiple target objects.

[0039] In this embodiment, the relative distance parameters and azimuth angle parameters of each target object relative to the reference point can be collected and integrated to form a spatial relationship analysis result including the spatial relationships of multiple target objects. This spatial relationship analysis result can clearly show the positional relationships of different target objects in the three-dimensional space.

[0040] Step S122: Construct a spatial distribution description structure based on the relative distance parameters and the azimuth angle parameters, where the spatial distribution description structure includes a set of position coordinates of the target object and a position association matrix.

[0041] After calculating the relative distance parameters and azimuth angle parameters between different target objects, this step needs to construct a spatial distribution description structure based on the relative distance parameters and azimuth angle parameters. The spatial distribution description structure can more comprehensively represent the positional association relationships of the target objects in the low-altitude area. Among them, the set of position coordinates includes the three-dimensional position coordinates of all target objects, and these coordinate points determine the specific positions of the target objects in the three-dimensional space; the position association matrix reflects the relative positional relationships between the target objects, and the distance, azimuth and other information between the target objects can be represented by the numerical values of the matrix elements. For example, the elements in the position association matrix can be assigned values according to the relative distance parameters and azimuth angle parameters, so that the matrix can accurately describe the positional associations between the target objects.

[0042] Step S123: Perform a time period difference analysis process on the three-dimensional position data in the target perception data set, and calculate the displacement amount and the displacement direction change amount of the target object in adjacent time periods. The displacement amount represents the moving distance of the target object per unit time, and the displacement direction change amount represents the adjustment amplitude of the moving direction of the target object.

[0043] In this embodiment, by comparing the three-dimensional position data of the target object in adjacent time periods, the displacement amount and the displacement direction change amount of the target object can be calculated. The displacement amount reflects the moving distance of the target object per unit time, which can be obtained by calculating the difference between the position coordinates of adjacent time periods; the displacement direction change amount reflects the adjustment amplitude of the moving direction of the target object. For example, it can be determined by calculating the included angle between the displacement vectors of adjacent time periods. For example, between time period t1 and time period t2, the position of the target object moves from (x1, y1, z1) to (x2, y2, z2), and the displacement amount and the displacement direction change amount of the target object during this period can be obtained through corresponding calculation methods.

[0044] Step S124: Perform time period trend extraction processing on the environmental disturbance data to extract the change period of the air flow direction information in consecutive time periods and the fluctuation frequency of the wind speed fluctuation information.

[0045] In this embodiment, the air flow direction information and the wind speed fluctuation information in the environmental disturbance data change with time. By extracting the change period and the fluctuation frequency of these information in consecutive time periods, the change law of environmental factors can be understood. For example, for the air flow direction information, its periodic change within a period of time can be analyzed to determine its change period; for the wind speed fluctuation information, its fluctuation frequency can be counted to master the change of the wind speed. The change laws of these environmental factors will affect the motion state of the target object. Therefore, it is very important to extract these information for analyzing the situation of the target object.

[0046] Step S125: Perform feature fusion processing on the displacement amount, the displacement direction change amount, the change period, and the fluctuation frequency to generate a continuous change description structure, and the continuous change description structure includes the state change rate of the target object and the environmental impact correlation degree.

[0047] In this embodiment, the displacement amount, the displacement direction change amount, the change period, and the fluctuation frequency respectively reflect the state change of the target object and the influence of environmental factors from different perspectives. Through fusion processing, a more comprehensive continuous change description structure can be generated. In the fusion process, a weighted fusion method can be adopted, and different weights are assigned according to the importance of each feature. For example, for the displacement amount and the displacement direction change amount, relatively high weights can be given because they directly reflect the change of the motion state of the target object; for the change period and the fluctuation frequency, appropriate weights can be given according to the influence degree of environmental factors on the target object. The generated continuous change description structure after fusion includes the state change rate of the target object and the environmental impact correlation degree. The state change rate reflects how fast the state of the target object changes in consecutive time periods, and the environmental impact correlation degree represents the influence degree of environmental factors on the state change of the target object.

[0048] Step S126: Use the spatial distribution description structure and the continuous change description structure as the spatial distribution feature and the continuous change feature respectively.

[0049] Finally, use the spatial distribution description structure and the continuous change description structure as the spatial distribution feature and the continuous change feature of the target object respectively. The spatial distribution feature can intuitively display the position correlation relationship of the target object in the low-altitude area; the continuous change feature reflects the state change law of the target object in a continuous time period and the influence of environmental factors, which helps to predict the future situation of the target object. Through the spatial distribution feature and the continuous change feature, the situation information of the target object can be understood more comprehensively.

[0050] Step S130: Input the spatial distribution feature and the continuous change feature into a pre-trained situation awareness model to generate the situation correlation feature of the target object.

[0051] After generating the spatial distribution feature and the continuous change feature of the target object, this step needs to input these features into a pre-trained situation awareness model to generate the situation correlation feature of the target object. The situation correlation feature can more deeply reflect the situation information of the target object. It combines the spatial distribution feature and the continuous change feature. Through the processing and analysis of the situation awareness model, the potential situation relationship of the target object can be mined.

[0052] Step S131: Input the spatial distribution feature into the first encoding layer of the situation awareness model, and perform dimension unification processing on the position coordinate set and the position correlation matrix in the spatial distribution feature through the first encoding layer to generate a spatial encoding feature with a standard dimension representation.

[0053] In this embodiment, the main function of the first encoding layer is to perform dimension unification processing on the position coordinate set and the position correlation matrix in the spatial distribution feature. Since the original dimensions of the position coordinate set and the position correlation matrix may be inconsistent, in order to facilitate the processing and analysis of the model, they need to be unified to the standard dimension. For example, by expanding or compressing the dimension of the position coordinate set to match the dimension of the position correlation matrix; performing normalization processing on the position correlation matrix to make the value range and dimension of its elements meet the requirements of the model. After the dimension unification processing, a spatial encoding feature with a standard dimension representation is generated, and this spatial encoding feature can be more effectively utilized by the model.

[0054] Step S132: Input the continuous change feature into the second encoding layer of the situation awareness model, and perform time sequence alignment processing on the state change rate and the environmental impact correlation degree in the continuous change feature through the second encoding layer to generate a change encoding feature with a time correspondence relationship.

[0055] In this embodiment, the second encoding layer needs to perform a temporal alignment process on the rate of state change and the correlation degree of environmental impact in the continuously varying features. Since the rate of state change and the correlation degree of environmental impact are collected within a continuous time period, their time series may be different. To ensure the accuracy and consistency of the data, they need to be aligned in chronological order so that the rate of state change and the correlation degree of environmental impact have a corresponding relationship at the same time point. For example, by methods such as interpolation and resampling, the time series of the rate of state change and the correlation degree of environmental impact are adjusted to be synchronized in time. After the temporal alignment process, a change encoding feature with a time correspondence relationship is generated, and this change encoding feature can better reflect the state change and environmental impact of the target object within a continuous time period.

[0056] Step S133: Input the spatial encoding feature and the change encoding feature into the weight assignment unit of the correlation analysis module of the situation awareness model to dynamically calculate the correlation weight between the spatial encoding feature and the change encoding feature. The correlation weight is used to represent the contribution degree of the spatial encoding feature and the change encoding feature to the situation correlation feature.

[0057] In this embodiment, the role of the weight assignment unit is to dynamically calculate the correlation weight between the spatial encoding feature and the change encoding feature. The correlation weight represents the contribution degree of the spatial encoding feature and the change encoding feature to the situation correlation feature, and different weight values reflect the importance of these two features in the generation process of the situation correlation feature. For example, in some cases, the spatial encoding feature may contribute more to the situation correlation feature, and the weight assignment unit will accordingly assign a higher weight to it; while in other cases, the change encoding feature may be more critical, and the weight assignment unit will assign a higher weight to it. The weight assignment unit determines the appropriate correlation weight by analyzing and processing the input spatial encoding feature and change encoding feature.

[0058] Step S1331: Input the spatial encoding feature and the change encoding feature into the weight assignment unit, and the weight assignment unit includes a fully connected layer and an activation function layer.

[0059] In this embodiment, the weight allocation unit consists of a fully connected layer and an activation function layer. The fully connected layer can perform a linear transformation on the input features, mapping the input features to a new feature space; the activation function layer then performs a non-linear transformation on the output of the fully connected layer to make it have a stronger expressive ability. For example, the fully connected layer will perform a linear combination of the spatial encoding features and the variation encoding features according to a preset weight matrix, perform a weighted sum on each dimension of the input features, and obtain a new feature representation. Then, the activation function layer will process the output of the fully connected layer. Common activation functions include the Sigmoid function, the ReLU function, etc. Through the action of the activation function, the value range and distribution of the features more conform to the requirements of the model.

[0060] Step S1332: Perform a linear transformation process on the spatial encoding features and the variation encoding features through the fully connected layer to generate a feature importance evaluation vector.

[0061] When the fully connected layer performs a linear transformation on the input spatial encoding features and variation encoding features, it can perform a weighted sum on each dimension of the features according to the preset weight parameters. Assume that the spatial encoding feature is represented as A, the variation encoding feature is represented as B, and the weight matrices of the fully connected layer are W_A and W_B respectively. For each dimension of the spatial encoding feature A, it can be multiplied by the corresponding element in the weight matrix W_A, and then all the products are added; the same operation is performed on the variation encoding feature B. Through the above linear combination, a new vector is obtained, and this vector contains the preliminary evaluation information of the importance of each dimension of the spatial encoding feature and the variation encoding feature, which is defined as the feature importance evaluation vector. For example, the spatial encoding feature A has multiple dimensions (a_1, a_2,..., a_n), and the weight matrix W_A also has multiple elements (w_a1, w_a2,..., w_an) accordingly. Then the result after the linear transformation is a_1*w_a1 + a_2*w_a2 +... + a_n*w_an. A similar calculation is performed on the variation encoding feature B, and finally the two results are combined into the feature importance evaluation vector.

[0062] Step S1333: Perform a normalization process on the feature importance evaluation vector through the activation function layer to generate a first weight value corresponding to the spatial encoding feature and a second weight value corresponding to the variation encoding feature, and the sum of the first weight value and the second weight value is 1.

[0063] After the activation function layer receives the feature importance evaluation vector, it can perform normalization processing on it. The purpose of normalization is to make the generated weight values within a reasonable range and ensure that the sum of the first weight value corresponding to the spatial coding feature and the second weight value corresponding to the change coding feature is 1. For example, if the Sigmoid activation function is used, it can map each element in the feature importance evaluation vector to the range between 0 and 1. Then, through the set calculation method, the result after being processed by the Sigmoid function is adjusted so that the sum of the finally obtained first weight value and the second weight value is equal to 1. Specifically, assuming that the value related to the spatial coding feature obtained after being processed by the Sigmoid function is S_A, and the value related to the change coding feature is S_B, then the first weight value can be calculated by S_A / (S_A + S_B), and the second weight value can be calculated by S_B / (S_A + S_B), which ensures that the sum of the two weight values is 1.

[0064] Step S1334: Use the first weight value as the associated weight of the spatial coding feature, and use the second weight value as the associated weight of the change coding feature; the weight allocation unit adjusts the contribution ratio of the spatial coding feature and the change coding feature in the fusion process according to the first weight value and the second weight value.

[0065] Use the calculated first weight value as the associated weight of the spatial coding feature, and the second weight value as the associated weight of the change coding feature. In the subsequent feature fusion process, the weight allocation unit will adjust the contribution ratio of the spatial coding feature and the change coding feature according to these two weight values. For example, if the first weight value is larger, it means that the spatial coding feature is more important in the generation process of the situation association feature. Then, a higher weight will be given to the spatial coding feature during fusion, making its impact on the final result greater; conversely, if the second weight value is larger, the contribution of the change coding feature will be greater. Through the above method, the roles of the two features in the fusion process can be dynamically adjusted according to different situations to generate more accurate situation association features.

[0066] Step S134: Perform weighted fusion processing on the spatial coding feature and the change coding feature based on the associated weights to generate a fusion association feature.

[0067] After obtaining the correlation weights of the spatial encoding features and the change encoding features, it is necessary to perform weighted fusion processing on the spatial encoding features and the change encoding features. The way of weighted fusion needs to be determined according to the nature of the features and the actual requirements, whether to use weighted addition or weighted concatenation. If the spatial encoding features and the change encoding features have a certain complementarity in dimension and meaning and can be numerically combined in the same dimension, then the weighted addition method can be used. For example, multiply the value of each dimension of the spatial encoding features by its corresponding correlation weight (the first weight value), multiply the value of each dimension of the change encoding features by its corresponding correlation weight (the second weight value), and then add the values of the corresponding dimensions to obtain the fused feature value. If the dimensions and meanings of the two features are quite different and not suitable for addition in the same dimension, then the weighted concatenation method is more reasonable. That is, the weighted spatial encoding features and the change encoding features are concatenated together in a set order to form the fused correlation features. Through the weighted fusion processing, the information of the spatial encoding features and the change encoding features can be effectively integrated, highlighting the role of important features.

[0068] Step S135: Perform information enhancement processing on the fused correlation features through the feature enhancement unit of the correlation analysis module to generate the situation correlation features of the target object.

[0069] The feature enhancement unit of the correlation analysis module will perform information enhancement processing on the fused correlation features. The feature enhancement unit can use a variety of methods to achieve information enhancement, such as using convolution operations, pooling operations, etc. The convolution operation can extract local features of the fused correlation features. By sliding different convolution kernels over the features, local feature information of different scales and directions can be extracted. The pooling operation can perform dimensionality reduction on the features, while retaining important feature information and reducing data redundancy. For example, using the maximum pooling operation, the maximum value in the local area of the feature can be selected as the representative value of this area, which can highlight the important information in the feature. Through the processing of the feature enhancement unit, the potential information in the fused correlation features can be further mined, the expression ability of the features can be enhanced, making the generated situation correlation features of the target object more accurate and rich, and better reflecting the real situation information of the target object.

[0070] Step S140: Input the situation correlation features into the situation classification layer of the situation awareness model to obtain the current situation attributes and situation development tendency information of the target object.

[0071] After generating the situation correlation features of the target object, this step needs to input this feature into the situation classification layer of the situation awareness model to obtain the current situation attributes and situation development tendency information of the target object. The role of the situation classification layer is to analyze and judge the situation correlation features, determine the situation type in which the target object is currently located, and predict its future development tendency.

[0072] Step S141: Input the situation correlation feature into the situation classification layer of the situation awareness model. Through the situation classification layer, perform attribute matching processing on the situation correlation feature to generate a candidate set of situation attributes of the target object and the matching confidence degrees corresponding to each situation type identifier in the candidate set of situation attributes. The candidate set of situation attributes contains multiple candidate situation type identifiers.

[0073] After inputting the situation correlation feature into the situation classification layer, the situation classification layer will perform attribute matching processing on it. The situation classification layer pre-stores multiple situation type identifiers and the feature patterns corresponding to each situation type. By comparing and matching the input situation correlation feature with these pre-stored feature patterns, the situation type most similar to the situation correlation feature is found to form a candidate set of situation attributes of the target object. At the same time, calculate the matching confidence degree for each candidate situation type identifier. The matching confidence degree represents the matching degree between the situation type and the input situation correlation feature. For example, using the method of similarity calculation, calculate the similarity between the situation correlation feature and each pre-stored feature pattern, and use the similarity value as the matching confidence degree. If the feature pattern of a certain situation type has a higher similarity with the situation correlation feature, then the corresponding matching confidence degree will be larger, indicating that this situation type is more likely to be the situation type in which the target object is currently located.

[0074] Step S142: Select the situation type identifier with the highest matching confidence degree as the current situation attribute of the target object.

[0075] After obtaining the candidate set of situation attributes and the matching confidence degrees corresponding to each situation type identifier, select the situation type identifier with the highest matching confidence degree as the current situation attribute of the target object. Because the highest matching confidence degree indicates that this situation type is the most matched with the input situation correlation feature and can most accurately reflect the current situation of the target object. For example, in the candidate set of situation attributes, there are several different situation type identifiers, namely type A, type B, type C, etc., and their corresponding matching confidence degrees are R_A, R_B, R_C, etc. By comparing the magnitudes of these matching confidence degrees, select the situation type identifier corresponding to the largest matching confidence degree as the current situation attribute of the target object.

[0076] Step S143: Input the situation correlation feature into the tendency prediction layer of the situation awareness model. Through the tendency prediction layer, perform historical information backtracking processing on the situation correlation feature, extract the situation change pattern of the target object in the historical period, and perform tendency prediction processing on the current situation attribute in the future period based on the situation change pattern, generating situation development tendency information including a direction indication and a rate indication. The direction indication represents the possible evolution direction of the target situation, and the rate indication represents the speed of the target situation evolution.

[0077] Input the situation correlation feature into the tendency prediction layer of the situation awareness model. The tendency prediction layer will first perform historical information backtracking processing on the situation correlation feature. Since the situation correlation feature contains the situation information of the target object over a period of time, by analyzing and mining this information, the situation change pattern of the target object in the historical period can be extracted. For example, the tendency prediction layer will perform time series analysis on the situation correlation feature in the historical period, observe information such as the change trend and periodicity of the feature, so as to determine the situation evolution law of the target object in history. Then, based on the extracted situation change pattern, perform tendency prediction processing on the current situation attribute of the target object in the future period. The tendency prediction layer will combine the current situation correlation feature and the historical situation change pattern to predict the possible evolution direction and the speed of evolution of the target object's situation in the future period. The direction indication represents the direction in which the target situation may develop, such as the situation may strengthen, weaken or change, etc.; the rate indication represents the speed of the target situation evolution, such as rapid evolution, slow evolution, etc.

[0078] Step S1431: Extract the correlation feature sequence of the historical period according to the situation correlation feature. The correlation feature sequence contains the situation correlation features of the target object in multiple historical periods.

[0079] First, extract the correlation feature sequence of the historical period according to the situation correlation feature. The situation correlation feature is collected and generated over a period of time, which contains the situation information of the target object in different historical periods. By dividing and extracting the situation correlation feature in the time dimension, the situation correlation features of the target object in multiple historical periods can be obtained, and these features are arranged in chronological order to form a correlation feature sequence. For example, assume that the situation correlation feature is collected over a relatively long period of time. Divide this period into multiple sub-periods, each sub-period corresponds to a situation correlation feature, and arrange the situation correlation features of these sub-periods in sequence to obtain the correlation feature sequence. This correlation feature sequence can reflect the situation change of the target object in the historical period.

[0080] Step S1432: Input the correlation feature sequence into the tendency prediction layer. The tendency prediction layer includes a long short-term memory network structure and a pattern extraction structure.

[0081] Input the associated feature sequence into the tendency prediction layer. The tendency prediction layer consists of a long short-term memory network structure and a pattern extraction structure. The long short-term memory (LSTM) network structure is a special recurrent neural network structure that can process sequence data and effectively solve the problems of gradient vanishing and gradient explosion in traditional recurrent neural networks. The LSTM structure has memory cells that can retain long-term information when processing sequence data, which is very effective for extracting the situation change information in the historical period. The pattern extraction structure is responsible for further analyzing and processing the output of the LSTM structure to extract the situation change pattern of the target object in the historical period.

[0082] Step S1433: Perform temporal information storage processing on the associated feature sequence through the long short-term memory network structure to generate a memory state vector containing historical situation information.

[0083] The long short-term memory network structure will perform temporal information storage processing on the input associated feature sequence. The LSTM structure consists of an input gate, a forget gate, an output gate, and memory cells. When processing the associated feature sequence, the input gate determines which new information can enter the memory cells, the forget gate determines which old information needs to be forgotten, and the output gate determines which information is output from the memory cells. Through the control of these gates, the LSTM structure can store the situation information of the historical period in the memory cells during the process of processing sequence data. As the associated feature sequence is gradually processed, the LSTM structure continuously updates the state of the memory cells and finally generates a memory state vector containing historical situation information. This memory state vector contains the important situation information of the target object in the historical period.

[0084] Step S1434: Perform feature decomposition processing on the memory state vector through the pattern extraction structure to separate the stable situation pattern and the fluctuating situation pattern of the target object in the historical period. The stable situation pattern represents the situation characteristics that the target object maintains for a long time, and the fluctuating situation pattern represents the situation characteristics of the short-term change of the target object.

[0085] The pattern extraction structure performs eigen-decomposition processing on the memory state vector. Since the memory state vector contains various situation information of the target object in the historical period, including both situation features that remain relatively stable in the long term and situation features that change in the short term. The pattern extraction structure will adopt appropriate methods, such as principal component analysis, clustering analysis, etc., to decompose the memory state vector. Principal component analysis can perform dimensionality reduction on the features in the memory state vector and extract the main feature components, which can reflect the stable situation pattern of the target object. Clustering analysis can classify the features in the memory state vector and cluster similar features into one category, thereby separating the stable situation pattern and the fluctuating situation pattern. The stable situation pattern represents the situation features that the target object maintains in a relatively long time, such as the basic movement law of the target object, the long-term situation state, etc.; the fluctuating situation pattern represents the situation changes that the target object undergoes in the short term, such as sudden situation fluctuations, temporary situation adjustments, etc. By separating the stable situation pattern and the fluctuating situation pattern, it is possible to more clearly understand the situation changes of the target object in the historical period.

[0086] Step S1435: Use the stable situation pattern and the fluctuating situation pattern as the situation change patterns of the target object in the historical period.

[0087] Use the separated stable situation pattern and fluctuating situation pattern as the situation change patterns of the target object in the historical period. These two patterns can comprehensively reflect the situation evolution law of the target object in the historical period. The stable situation pattern reflects the long-term situation features of the target object; the fluctuating situation pattern reflects the situation changes of the target object in the short term. By comprehensively considering the stable situation pattern and the fluctuating situation pattern, the tendency prediction layer can more accurately perform tendency prediction processing on the current situation attribute of the target object in the future period and generate more reliable situation development tendency information.

[0088] Step S144: Use the current situation attribute and the situation development tendency information as the target situation determination result.

[0089] Finally, use the current situation attribute and the situation development tendency information of the target object as the target situation determination result. This target situation determination result includes the situation type where the target object is currently located and the possible future development tendency. For example, in the application scenario of the low-altitude intelligent Internet of Things, according to the target situation determination result, corresponding measures can be taken in a timely manner, such as adjusting the monitoring strategy, issuing early warnings, etc., to ensure the effective management and control of the target object.

[0090] Step S150: Generate a collaborative perception instruction including the target positioning identifier according to the current situation attribute and the situation development tendency information, and send the collaborative perception instruction to the low-altitude intelligent Internet of Things management terminal to trigger a linkage response operation.

[0091] After obtaining the current situation attributes and the information on the tendency of situation development of the target object, this step is required to generate a collaborative perception instruction containing the target positioning identifier based on this information, and send this instruction to the low-altitude intelligent network management terminal to trigger a linkage response operation. The generation of the collaborative perception instruction needs to comprehensively consider the current situation and future development tendency of the target object, and at the same time include an accurate target positioning identifier, so that the management terminal can accurately monitor and respond to the target object according to the instruction.

[0092] Step S151: Analyze the preset response rule library corresponding to the current situation attributes, and extract the response level identifier and the collaborative policy code associated with the current situation attributes. The response level identifier is used to represent the urgency of the response to be triggered, and the collaborative policy code is used to represent the type of nodes to be linked.

[0093] First, analyze the preset response rule library corresponding to the current situation attributes. The preset response rule library is established in advance, which contains the response rules and strategies corresponding to different situation attributes. By matching and analyzing the current situation attributes, the response level identifier and the collaborative policy code associated with this situation attribute can be extracted from the preset response rule library. The response level identifier represents the urgency of the response to be triggered. For example, it may be divided into different levels, such as high urgency, medium urgency, low urgency, etc. Different levels of urgency correspond to different response measures and resource allocation. The collaborative policy code is used to represent the type of nodes to be linked. For example, it is necessary to link the positioning and sensing devices on the low-altitude aircraft, or the environmental monitoring devices at the ground stations, or both need to be linked, etc. By extracting the response level identifier and the collaborative policy code, it can provide clear response strategies and linkage information for the subsequent generation of collaborative perception instructions.

[0094] Step S152: According to the three-dimensional position data in the target perception data set, calculate the position centroid coordinates and the boundary of the position coverage area of the target object in the current period. The position centroid coordinates represent the central position of the target object, and the boundary of the position coverage area represents the spatial distribution range of the target object.

[0095] Based on the three-dimensional position data in the target perception data set, calculate the position centroid coordinates and the boundary of the position coverage area of the target object in the current period. The position centroid coordinates can be obtained by performing a weighted average calculation on all the three-dimensional position data of the target object. Suppose the target object has multiple three-dimensional position coordinates (x_1, y_1, z_1), (x_2, y_2, z_2),..., (x_n, y_n, z_n) in the current period. The x, y, and z values of each coordinate can be weighted and summed respectively, and then divided by the number of coordinates n to obtain the position centroid coordinates (x_c, y_c, z_c). There are various methods for calculating the boundary of the position coverage area. For example, a bounding box can be determined by finding the maximum and minimum values of all the position coordinates, and the boundary of this bounding box can approximately represent the boundary of the position coverage area of the target object. Or more complex methods can be used, such as the convex hull algorithm, by finding the convex hull of all the position coordinates to obtain a more accurate boundary of the position coverage area. The position centroid coordinates represent the central position of the target object, and the boundary of the position coverage area represents the spatial distribution range of the target object.

[0096] Step S153: Perform coordinate system unification processing on the position centroid coordinates and the boundary of the position coverage area to generate a standardized positioning identifier that is consistent with the geographical coordinate system of the low-altitude intelligent network. The standardized positioning identifier includes the central coordinate point of the target object and a sequence of boundary coordinate points.

[0097] Perform coordinate system unification processing on the calculated position centroid coordinates and the boundary of the position coverage area. Since the three-dimensional position data in the target perception data set may be collected based on different coordinate systems, and the low-altitude intelligent network has its own geographical coordinate system, in order to ensure the accuracy and consistency of the positioning identifier, it is necessary to convert the position centroid coordinates and the boundary of the position coverage area to a coordinate system that is consistent with the geographical coordinate system of the low-altitude intelligent network. Coordinate system unification processing can be achieved through a series of coordinate transformation operations, such as translation, rotation, and scaling. First, determine the transformation parameters between the two coordinate systems, which can be calculated through known control points. Control points are points with accurate coordinates in both coordinate systems. Through the coordinate correspondence of these control points, the translation parameters, rotation parameters, and scaling parameters can be calculated. Then, each coordinate point of the position centroid coordinates and the boundary of the position coverage area is transformed according to these transformation parameters to obtain the new coordinates in the geographical coordinate system of the low-altitude intelligent network. After coordinate system unification processing, a standardized positioning identifier that is consistent with the geographical coordinate system of the low-altitude intelligent network is generated. This standardized positioning identifier includes the central coordinate point of the target object (i.e., the coordinates after the transformation of the position centroid coordinates) and a sequence of boundary coordinate points (i.e., the sequence of coordinates after the transformation of the boundary of the position coverage area). This standardized positioning identifier can accurately represent the position and spatial distribution range of the target object in the geographical coordinate system of the low-altitude intelligent network.

[0098] Step S154: Combine the direction indication and rate indication in the situation development tendency information to perform position prediction processing on the standardized positioning identifier for a future time period, and generate a predicted positioning identifier sequence of the target object in subsequent time periods.

[0099] Combine the direction indication and rate indication in the situation development tendency information to perform position prediction processing on the standardized positioning identifier. The direction indication in the situation development tendency information represents the possible evolution direction of the target situation, and the rate indication represents the speed of the target situation evolution. Based on this information, the moving direction and moving speed of the target object in the future time period can be inferred. For the central coordinate point and the boundary coordinate point sequence in the standardized positioning identifier, determine the possible moving direction of each coordinate point in the future time period according to the direction indication, and determine the possible moving distance of each coordinate point per unit time according to the rate indication. For example, assume that the central coordinate point is (x0, y0, z0), the direction indication indicates that the target object will move in a specific direction, and the rate indication indicates that its moving speed is a certain value. Then in a future time period t, the new coordinates (x1, y1, z1) of the central coordinate point after this time period can be calculated according to the direction and speed. The same processing is performed for each point in the boundary coordinate point sequence. By continuously predicting the position for each time period, a predicted positioning identifier sequence of the target object in subsequent time periods is generated. This predicted positioning identifier sequence includes the predicted central coordinate points and the predicted boundary coordinate point sequence of the target object in multiple future time periods.

[0100] Step S155: Integrate the response level identifier, cooperation strategy code, standardized positioning identifier, and predicted positioning identifier sequence to generate a cooperation awareness instruction including a positioning identifier chain with timestamp alignment, and the coordinate dimension of each identifier node in the positioning identifier chain is consistent with the coordinate dimension of the standardized positioning identifier.

[0101] Integrate the response level identifier, collaborative policy encoding, standardized positioning identifier, and predicted positioning identifier sequence for information processing. The purpose of information integration is to combine these different types of information into a complete collaborative perception instruction so that the low-altitude intelligent network management terminal can accurately perform linkage response operations according to this instruction. First, add a timestamp to each identifier node in the predicted positioning identifier sequence, where the timestamp represents the future time period corresponding to the identifier node. Then, integrate the response level identifier, collaborative policy encoding, standardized positioning identifier, and the predicted positioning identifier sequence with timestamps added in a set format. During the integration process, ensure that the coordinate dimensions of each identifier node in the positioning identifier chain are consistent with the coordinate dimensions of the standardized positioning identifier, which can guarantee the consistency and accuracy of information. For example, the response level identifier and collaborative policy encoding can be used as the header information of the instruction, the standardized positioning identifier as the positioning information at the current moment, and the predicted positioning identifier sequence arranged in chronological order as the positioning information for future time periods, forming a collaborative perception instruction containing a positioning identifier chain with timestamp alignment. This collaborative perception instruction can clearly convey information such as the current position of the target object, its possible future positions, the response strategies to be taken, and the types of linkage nodes.

[0102] Step S156: Send the collaborative perception instruction to the management terminal so that the management terminal schedules the corresponding nodes to perform target monitoring or response operations according to the collaborative perception instruction.

[0103] Send the generated collaborative perception instruction to the low-altitude intelligent network management terminal. In the low-altitude intelligent network, the management terminal is responsible for the management and scheduling of the entire network. When the management terminal receives the collaborative perception instruction, it can perform corresponding processing according to the information in the collaborative perception instruction. First, the management terminal parses the collaborative perception instruction and extracts information such as the response level identifier, collaborative strategy code, standardized positioning identifier, and predicted positioning identifier sequence. Then, it determines the urgency of the response to be triggered according to the response level identifier and the type of nodes to be linked according to the collaborative strategy code. For example, if the response level identifier is high urgency, the management terminal will quickly allocate more resources and nodes to participate in the target monitoring and response operations; if the collaborative strategy code indicates that the positioning sensing device on the low-altitude aircraft and the environmental monitoring device at the ground station need to be linked, the management terminal will send corresponding instructions to these nodes. Next, the management terminal will schedule the corresponding nodes to perform accurate monitoring and response operations on the target object according to the standardized positioning identifier and the predicted positioning identifier sequence. At the current moment, the node will position and monitor the target object according to the standardized positioning identifier; for the future period, the node will make preparations in advance according to the predicted positioning identifier sequence to timely track the position change of the target object. In the above way, the low-altitude intelligent network management terminal can effectively schedule the corresponding nodes to execute target monitoring or response operations according to the collaborative perception instruction, realizing the situation awareness and effective management of low-altitude targets.

[0104] Furthermore, in order to enable the situation awareness model to accurately classify and predict the situation of the target object, it is necessary to train the situation awareness model. The training process of the situation awareness model is a complex process, involving the adjustment and optimization of multiple modules and levels.

[0105] First, collect a large amount of sample data, which should contain various situation information of the target object. The sample data can be selected from the historical target perception data set or generated through simulation experiments. The sample data needs to contain the spatial distribution characteristics, continuous change characteristics, and corresponding true situation attributes and situation development tendency information of the target object. For example, collect the movement data of different target objects in different environments, calculate their spatial distribution characteristics and continuous change characteristics, and determine their true situation attributes (such as normal flight, abnormal flight, etc.) and situation development tendency information (such as situation enhancement, situation weakening, etc.) through manual annotation or other reliable methods. Then, divide the sample data into a training set, a validation set, and a test set. The training set is used for the parameter learning of the model, the validation set is used to adjust the hyperparameters of the model during the training process, and the test set is used to evaluate the performance of the trained model. The division ratio can be adjusted according to the actual situation. Generally speaking, the training set accounts for a relatively large proportion, while the validation set and the test set account for a relatively small proportion.

[0106] On this basis, a situation awareness model is constructed, which includes multiple necessary modules and levels. The situation awareness model mainly consists of a first encoding layer, a second encoding layer, a correlation analysis module, a situation classification layer, a tendency prediction layer, etc. The first encoding layer is used to uniformly process the spatial distribution features. It can adopt structures such as a fully connected layer to convert the set of position coordinates and the position correlation matrix of the input spatial distribution features into spatial encoding features with a standard dimension representation. The second encoding layer is used to perform temporal alignment processing on the continuously changing features. Similarly, it can adopt structures such as a fully connected layer or a recurrent neural network layer to convert the state change rate and the environmental impact correlation degree of the input continuously changing features into change encoding features with a time correspondence relationship. The correlation analysis module includes a weight assignment unit and a feature enhancement unit. The weight assignment unit is used to dynamically calculate the correlation weights between the spatial encoding features and the change encoding features. It can be composed of a fully connected layer and an activation function layer. The feature enhancement unit is used to perform information enhancement processing on the fused correlation features and can adopt structures such as a convolutional layer and a pooling layer. The situation classification layer is used to perform attribute matching processing on the situation correlation features to determine the current situation attributes of the target object. It can adopt structures such as a fully connected layer and a Softmax activation function. The tendency prediction layer is used to perform historical information backtracking processing and future tendency prediction processing on the situation correlation features. It can include structures such as a long short-term memory network structure and a pattern extraction structure. Each module and level are connected according to the set connection relationships. For example, the outputs of the first encoding layer and the second encoding layer serve as the inputs of the correlation analysis module, and the output of the correlation analysis module serves as the inputs of the situation classification layer and the tendency prediction layer.

[0107] Then, input the training set data into the constructed situation awareness model for training. During the training process, a suitable loss function is used to measure the difference between the output result of the model and the true label. For the situation classification layer, the cross-entropy loss function can be used, which can measure the difference between the situation attributes predicted by the model and the true situation attributes, and prompt the prediction result of the model to be closer to the true situation. For the tendency prediction layer, the mean squared error loss function can be used, which can measure the difference between the situation development tendency information predicted by the model (such as direction indication and rate indication) and the true situation development tendency information, enabling the model to more accurately predict the future situation changes of the target object. Then, use an optimization algorithm to update the parameters of the model. Common optimization algorithms include the stochastic gradient descent method, the Adam optimization algorithm, etc. The optimization algorithm will continuously adjust the parameters of the model according to the gradient information of the loss function, making the value of the loss function gradually decrease, that is, the difference between the output result of the model and the true label gradually shrinks. During the training process, some training parameters also need to be set, such as the learning rate, batch size, and number of training epochs. The learning rate controls the step size of parameter update, the batch size represents the number of samples input into the model each time, and the number of training epochs represents the number of times the entire training set data is trained by the model. Appropriate setting of training parameters is crucial for the training effect of the model and requires continuous experiments and adjustments to determine the optimal parameter combination.

[0108] During the training process, use the validation set data to validate the model regularly. The purpose of validation is to evaluate the performance of the model on unseen data and prevent the model from overfitting. Overfitting refers to the situation where the model performs well on the training set but poorly on the validation set or test set. Based on the evaluation results of the validation set, the hyperparameters of the model, such as the learning rate, batch size, etc., can be adjusted to optimize the performance of the model. For example, if it is found that the value of the loss function of the model on the validation set no longer decreases or starts to increase, it indicates that overfitting may have occurred. At this time, methods such as reducing the learning rate or increasing the regularization term can be used to alleviate the overfitting problem. At the same time, according to the evaluation results of the validation set, the model parameters that perform best during the training process can also be selected as the final model parameters.

[0109] After the model training is completed, the test set data is used to perform the final test on the model. The test set data is data that the model has never seen during the training and validation processes, and can more realistically evaluate the generalization ability of the model. Through the evaluation results of the test set, performance indicators such as the accuracy rate, recall rate, and F1 value of the model can be obtained. These indicators can comprehensively reflect the accuracy and reliability of the model in classifying and predicting the target object situation. If the test performance of the model meets the requirements, the trained model can be applied to the actual low-altitude intelligent network target situation awareness task; if the test performance of the model does not meet the requirements, it is necessary to readjust the model structure, training parameters, or increase the sample data, etc., and perform training and testing again until the performance of the model reaches a satisfactory level.

[0110] Figure 2 FIG. shows a schematic diagram of exemplary hardware and software components of a target situation awareness system 100 based on a low-altitude intelligent network that can implement the inventive concept provided by some embodiments of the present invention. For example, a processor 120 can be used on the target situation awareness system 100 based on a low-altitude intelligent network and is used to execute the functions in the present invention.

[0111] The target situation awareness system 100 based on a low-altitude intelligent network can be a general-purpose server or a special-purpose server, both of which can be used to implement the target situation awareness method based on a low-altitude intelligent network of the present invention. Although only one server is shown in the present invention, for convenience, the functions described in the present invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0112] For example, the target situation awareness system 100 based on a low-altitude intelligent network can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the target situation awareness system 100 based on a low-altitude intelligent network can also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The target situation awareness system 100 based on a low-altitude intelligent network also includes an I / O interface 150 between the computer and other input / output devices.

[0113] For ease of description, only one processor is described in the target situation awareness system 100 based on the low-altitude intelligent Internet of Things. However, it should be noted that the target situation awareness system 100 based on the low-altitude intelligent Internet of Things in the present invention may also include multiple processors. Therefore, the steps performed by one processor described in the present invention may also be jointly performed or separately performed by multiple processors. For example, if the processor of the target situation awareness system 100 based on the low-altitude intelligent Internet of Things performs step A and step B, it should be understood that step A and step B may also be jointly performed by two different processors or separately performed in one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0114] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned target situation awareness method based on the low-altitude intelligent Internet of Things is implemented.

[0115] It should be noted that, in order to simplify the description of the present invention disclosure and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. A target situation awareness method based on a low-altitude intelligent Internet of Things, characterized in that, The method includes: Obtaining a set of target perception data collected by multi-node collaboration in a low-altitude intelligent network, where the set of target perception data includes three-dimensional position data and environmental disturbance data collected by different nodes within a continuous time period; Performing feature extraction processing on the set of target perception data to generate spatial distribution features and continuous change features of the target object, where the spatial distribution features represent the position association relationship of the target object in the low-altitude area, and the continuous change features represent the state change law of the target object within a continuous time period; Inputting the spatial distribution features and the continuous change features into a pre-trained situation awareness model to generate situation association features of the target object; Inputting the situation association features into the situation classification layer of the situation awareness model to obtain the current situation attribute and situation development tendency information of the target object; Generating a collaborative perception instruction including a target positioning identifier according to the current situation attribute and the situation development tendency information, and sending the collaborative perception instruction to a low-altitude intelligent network management terminal to trigger a linkage response operation.

2. The target situation awareness method based on the low-altitude intelligent Internet of Things according to claim 1, characterized in that The obtaining of the set of target perception data collected by multi-node collaboration in the low-altitude intelligent network includes: Determining multiple perception nodes covering the target monitoring area in the low-altitude intelligent network, where the perception nodes include positioning sensing devices deployed on low-altitude aircraft and environmental monitoring devices deployed on ground stations; Collecting three-dimensional position data of the target object within a continuous time period through the positioning sensing device, where the three-dimensional position data includes spatial coordinate points of the target object in each time period; Collecting environmental disturbance data around the target object within the same continuous time period through the environmental monitoring device, where the environmental disturbance data includes air flow direction information and wind speed fluctuation information; Performing synchronous alignment processing on the three-dimensional position data and the environmental disturbance data according to the collection timestamp to generate perception data units with time correspondence relationships; Classifying and integrating the perception data units according to the node identifier to generate perception data subsets corresponding to different nodes, where the perception data subsets include three-dimensional position data and environmental disturbance data collected by the corresponding perception nodes; Performing cross-node information fusion processing on all perception data subsets to generate a set of target perception data including multi-node collaboration information.

3. The target situation awareness method based on the low-altitude intelligent Internet of Things according to claim 2, wherein, The performing of feature extraction processing on the set of target perception data to generate spatial distribution features and continuous change features of the target object includes: Performing spatial relationship analysis processing on the three-dimensional position data in the set of target perception data, and calculating relative distance parameters and azimuth angle parameters between different target objects, where the relative distance parameters represent the straight-line distance of the target object in three-dimensional space, and the azimuth angle parameters represent the direction angle of the target object relative to a reference point; Constructing a spatial distribution description structure based on the relative distance parameters and the azimuth angle parameters, where the spatial distribution description structure includes a position coordinate set and a position association matrix of the target object; Perform time period difference analysis processing on the three-dimensional position data in the target perception data set, calculate the displacement amount and the displacement direction change amount of the target object in adjacent time periods, where the displacement amount represents the moving distance of the target object per unit time, and the displacement direction change amount represents the adjustment amplitude of the moving direction of the target object; Perform time period trend extraction processing on the environmental disturbance data, and extract the change period of the air flow direction information and the fluctuation frequency of the wind speed fluctuation information in consecutive time periods; Perform feature fusion processing on the displacement amount, the displacement direction change amount, the change period, and the fluctuation frequency to generate a continuous change description structure, where the continuous change description structure includes the state change rate of the target object and the environmental impact correlation degree; Use the spatial distribution description structure and the continuous change description structure as the spatial distribution feature and the continuous change feature respectively.

4. The target situation awareness method based on the low-altitude intelligent Internet of Things according to claim 3, wherein Perform spatial relationship analysis processing on the three-dimensional position data in the target perception data set, and calculate the relative distance parameter and the azimuth angle parameter between different target objects, including: Extract the three-dimensional position data of multiple target objects in the same time period from the target perception data set, where the three-dimensional position data includes the spatial coordinate points of each target object; Select the spatial coordinate point of one of the target objects as the reference point; For each of the other target objects, extract its spatial coordinate point, and calculate the three-dimensional Euclidean distance between this spatial coordinate point and the reference point, and use the calculation result as the relative distance parameter between this target object and the reference point; Calculate the horizontal projection coordinate of the spatial coordinate point of this target object relative to the reference point, where the horizontal projection coordinate represents the position of the target object on the horizontal plane; Calculate the azimuth angle between the horizontal projection coordinates, and use the calculation result as the azimuth angle parameter of this target object relative to the reference point; Summarize and organize the relative distance parameters and azimuth angle parameters of all target objects to generate a spatial relationship analysis result including the spatial relationship of multiple target objects.

5. The target situation awareness method based on the low-altitude intelligent Internet of Things according to claim 1, wherein, Input the spatial distribution feature and the continuous change feature into a pre-trained situation awareness model to generate a situation correlation feature of the target object, including: Input the spatial distribution feature into the first encoding layer of the situation awareness model, and perform dimension unification processing on the position coordinate set and the position association matrix in the spatial distribution feature through the first encoding layer to generate a spatial encoding feature with a standard dimension representation; Input the continuous change feature into the second encoding layer of the situation awareness model, and perform time series alignment processing on the state change rate and the environmental impact correlation degree in the continuous change feature through the second encoding layer to generate a change encoding feature with a time correspondence relationship; Input the spatial encoding feature and the change encoding feature into the weight assignment unit of the association analysis module of the situation awareness model to dynamically calculate the association weights of the spatial encoding feature and the change encoding feature, where the association weights are used to represent the contribution degrees of the spatial encoding feature and the change encoding feature to the situation correlation feature; Perform weighted fusion processing on the spatial encoding feature and the change encoding feature based on the association weights to generate a fusion association feature; The feature enhancement unit of the correlation analysis module performs information enhancement processing on the fused correlation features to generate the situation correlation features of the target object.

6. The target situation awareness method based on the low-altitude intelligent Internet of Things according to claim 5, characterized in that, The dynamic calculation of the correlation weights of the spatial coding features and the change coding features by the weight assignment unit of the correlation analysis module of the situation awareness model when inputting the spatial coding features and the change coding features includes: Input the spatial coding features and the change coding features into the weight assignment unit, and the weight assignment unit includes a fully connected layer and an activation function layer; Perform linear transformation processing on the spatial coding features and the change coding features through the fully connected layer to generate a feature importance evaluation vector; Perform normalization processing on the feature importance evaluation vector through the activation function layer to generate a first weight value corresponding to the spatial coding features and a second weight value corresponding to the change coding features, and the sum of the first weight value and the second weight value is 1; Use the first weight value as the correlation weight of the spatial coding features, and use the second weight value as the correlation weight of the change coding features; the weight assignment unit adjusts the contribution ratio of the spatial coding features and the change coding features in the fusion process according to the first weight value and the second weight value.

7. The method for target situation awareness based on low-altitude intelligent Internet of Things according to claim 1, characterized in that, The input of the situation correlation features into the situation classification layer of the situation awareness model to obtain the current situation attribute and the situation development tendency information of the target object includes: Input the situation correlation features into the situation classification layer of the situation awareness model, and perform attribute matching processing on the situation correlation features through the situation classification layer to generate a situation attribute candidate set of the target object and the matching confidence corresponding to each situation type identifier in the situation attribute candidate set, and the situation attribute candidate set contains multiple candidate situation type identifiers; Select the situation type identifier with the highest matching confidence as the current situation attribute of the target object; Input the situation correlation features into the tendency prediction layer of the situation awareness model, and perform historical information backtracking processing on the situation correlation features through the tendency prediction layer to extract the situation change pattern of the target object in the historical period, and perform tendency prediction processing on the current situation attribute in the future period based on the situation change pattern to generate situation development tendency information including a direction indication and a rate indication, where the direction indication represents the possible evolution direction of the target situation, and the rate indication represents the speed of the target situation evolution; Use the current situation attribute and the situation development tendency information as the target situation determination result.

8. The method for target situation awareness based on low-altitude intelligent Internet of Things according to claim 7, wherein The input of the situation correlation features into the tendency prediction layer of the situation awareness model, and the extraction of the situation change pattern of the target object in the historical period by performing historical information backtracking processing on the situation correlation features through the tendency prediction layer includes: Extract the correlation feature sequence of the historical period according to the situation correlation features, and the correlation feature sequence contains the situation correlation features of the target object in multiple historical periods; Input the correlation feature sequence into the tendency prediction layer, and the tendency prediction layer includes a long short-term memory network structure and a pattern extraction structure; Perform temporal information storage processing on the associated feature sequence through the long short-term memory network structure to generate a memory state vector containing historical situation information; Perform feature decomposition processing on the memory state vector through the pattern extraction structure to separate the stable situation pattern and the fluctuating situation pattern of the target object in the historical period. The stable situation pattern represents the situation characteristics that the target object maintains for a long time, and the fluctuating situation pattern represents the situation characteristics of the short-term change of the target object; Use the stable situation pattern and the fluctuating situation pattern as the situation change pattern of the target object in the historical period.

9. The target situation awareness method based on the low-altitude intelligent Internet of Things according to claim 1, wherein, The generating a collaborative perception instruction containing a target positioning identifier according to the current situation attribute and the situation development tendency information, and sending the collaborative perception instruction to the low-altitude intelligent network management terminal to trigger a linkage response operation includes: Parse the preset response rule library corresponding to the current situation attribute, and extract the response level identifier and the collaborative strategy code associated with the current situation attribute. The response level identifier is used to represent the urgency of the response to be triggered, and the collaborative strategy code is used to represent the type of nodes to be linked; According to the three-dimensional position data in the target perception data set, calculate the position centroid coordinates and the position coverage area boundary of the target object in the current period. The position centroid coordinates represent the central position of the target object, and the position coverage area boundary represents the spatial distribution range of the target object; Perform coordinate system unification processing on the position centroid coordinates and the position coverage area boundary to generate a standardized positioning identifier consistent with the geographical coordinate system of the low-altitude intelligent network. The standardized positioning identifier includes the central coordinate point and the boundary coordinate point sequence of the target object; Combine the direction indication and the rate indication in the situation development tendency information to perform future period position prediction processing on the standardized positioning identifier to generate a predicted positioning identifier sequence of the target object in the subsequent period; Perform information integration processing on the response level identifier, the collaborative strategy code, the standardized positioning identifier, and the predicted positioning identifier sequence to generate a collaborative perception instruction containing a positioning identifier chain with timestamp alignment. The coordinate dimension of each identifier node in the positioning identifier chain is consistent with the coordinate dimension of the standardized positioning identifier; Send the collaborative perception instruction to the management terminal so that the management terminal schedules the corresponding nodes to perform target monitoring or response operations according to the collaborative perception instruction.

10. An object situation awareness system based on a low-altitude intelligent Internet of Things, characterized in that, It includes a processor and a memory. The memory is connected to the processor. The memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the method for target situation perception based on a low-altitude intelligent network according to any one of claims 1-9 above.

Images