Forest Fire Monitoring Method, Device and Equipment

Abstract

The forest fire monitoring method, device and equipment provided by this application relate to the field of computer technology. In this application, first, based on the environmental acquisition information corresponding to each sub-region, a regional relationship network is constructed; secondly, based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions, according to the association relationship represented by the regional relationship network, the environmental semantic vector of the environmental acquisition information corresponding to the target sub-region is semantically aggregated to form an aggregated environmental semantic vector; then, based on the image semantic vector corresponding to the forest remote sensing image, the aggregated environmental semantic vector is semantically enhanced to form an enhanced environmental semantic vector; finally, semantic analysis is performed based on the enhanced environmental semantic vector, and target fire monitoring data is output to reflect the probability of a fire occurring. Based on the above content, the problem of relatively low reliability of forest fire monitoring existing in the prior art can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular, to a forest fire monitoring method, device and equipment. Background Art

[0002] The monitoring of forest fires is generally achieved based on drone cruise monitoring, or based on manual patrol. However, both of these methods can only perform corresponding alarm and fire-fighting treatments after a fire occurs, which will cause certain losses. Therefore, in the prior art, fire prediction is carried out based on environmental information such as temperature and humidity. However, in the prior art, generally, threshold judgments are made on data in multiple dimensions. Thus, the problem of relatively low reliability of prediction is likely to occur. Summary of the Invention

[0003] In view of this, the purpose of the present application is to provide a forest fire monitoring method, device and equipment to improve the problem of relatively low reliability of forest fire monitoring existing in the prior art.

[0004] To achieve the above purpose, the present application adopts the following technical solutions:

[0005] A forest fire monitoring method includes:

[0006] Based on the environmental acquisition information corresponding to each sub-region in the target forest area, a regional relationship network corresponding to the target forest area is constructed, where the environmental acquisition information includes personnel activity information, temperature information, humidity information and wind speed information;

[0007] Based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions, according to the association relationship represented by the regional relationship network, the environmental semantic vectors of the environmental acquisition information corresponding to the target sub-region are semantically aggregated to form corresponding aggregated environmental semantic vectors;

[0008] Based on the image semantic vector of the forest remote sensing image corresponding to the target forest area, the aggregated environmental semantic vector is semantically enhanced to form a corresponding enhanced environmental semantic vector;

[0009] Based on the enhanced environmental semantic vector, semantic analysis is performed to output target fire monitoring data, where the target fire monitoring data is used to reflect the probability of a fire occurring in the target sub-region.

[0010] In a preferred selection of the present application, in the above forest fire monitoring method, the step of based on the image semantic vector of the forest remote sensing image corresponding to the target forest area, semantically enhancing the aggregated environmental semantic vector to form a corresponding enhanced environmental semantic vector includes:

[0011] Based on a pre-determined target color threshold, perform determination processing on the forest remote sensing image corresponding to the target forest area to obtain a corresponding target suspected dry area;

[0012] Perform binarization processing on the target suspected dry area in the forest remote sensing image, and perform value assignment processing on the pixel values of other areas to form a binarized remote sensing image;

[0013] Perform morphological operations on the binarized remote sensing image to form a region contour image;

[0014] Based on the region contour image and the forest remote sensing image, associate and mine the image semantic vector of the forest remote sensing image;

[0015] Based on the image semantic vector, perform semantic enhancement on the aggregated environmental semantic vector to form a corresponding enhanced environmental semantic vector.

[0016] In a preferred selection of the present application, in the above forest fire monitoring method, the step of performing morphological operations on the binarized remote sensing image to form a region contour image includes:

[0017] Perform first morphological dilation processing and second morphological dilation processing on the binarized remote sensing image respectively to obtain corresponding first dilated remote sensing image and second dilated remote sensing image, wherein the size of the structural element matrix corresponding to the first morphological dilation processing is smaller than the size of the structural element matrix corresponding to the second morphological dilation processing;

[0018] Perform first morphological erosion processing and second morphological erosion processing on the binarized remote sensing image respectively to obtain corresponding first eroded remote sensing image and second eroded remote sensing image, wherein the size of the structural element matrix corresponding to the first morphological erosion processing is equal to the size of the structural element matrix corresponding to the first morphological dilation processing, and the size of the structural element matrix corresponding to the second morphological erosion processing is equal to the size of the structural element matrix corresponding to the second morphological dilation processing;

[0019] Perform difference calculation on the first dilated remote sensing image and the first eroded remote sensing image to form a first region contour image, and perform difference calculation on the second dilated remote sensing image and the second eroded remote sensing image to form a second region contour image;

[0020] Based on the second region contour image, perform region contour update on the first region contour image to form a region contour image, wherein the region contour represented by the region contour image is located between the region contour of the second region contour image and the region contour of the first region contour image.

[0021] In a preferred option of the present application, in the above forest fire monitoring method, the step of associatively mining the image semantic vector of the forest remote sensing image based on the regional contour image and the forest remote sensing image includes:

[0022] Perform convolution on the regional contour image and the forest remote sensing image respectively to form corresponding regional contour convolution vectors and forest remote sensing convolution vectors;

[0023] Perform self-attention processing on the regional contour convolution vector and the forest remote sensing convolution vector respectively to form corresponding regional contour attention vectors and forest remote sensing attention vectors;

[0024] Perform non-linear mapping on the forest remote sensing attention vector to form a forest remote sensing mapping vector, where each vector parameter in the forest remote sensing mapping vector is greater than or equal to 0 and less than or equal to 1;

[0025] Multiply the vector parameters at the corresponding positions between the forest remote sensing mapping vector and the regional contour attention vector to form the image semantic vector of the forest remote sensing image.

[0026] In a preferred option of the present application, in the above forest fire monitoring method, the step of semantically enhancing the aggregated environmental semantic vector based on the image semantic vector to form a corresponding enhanced environmental semantic vector includes:

[0027] Determine a query vector based on the image semantic vector, and determine a key vector and a value vector based on the aggregated environmental semantic vector;

[0028] Based on the dot product between the query vector and the key vector, determine an attention distribution matrix, and based on the attention distribution matrix, perform weighted summation calculation on the value vector to form a corresponding enhanced environmental semantic vector.

[0029] In a preferred option of the present application, in the above forest fire monitoring method, the step of semantically aggregating the environmental semantic vector of the environmental acquisition information corresponding to other sub-regions to form a corresponding aggregated environmental semantic vector according to the association relationship represented by the regional relationship network includes:

[0030] Perform semantic space conversion on the environmental acquisition information corresponding to each sub-region respectively to form the environmental semantic vector corresponding to each sub-region;

[0031] Based on the degree of matching between each of the environmental semantic vectors and the fire environmental semantic vector, the most matching target sub-region is determined among the sub-regions, where the fire environmental semantic vector is formed by performing semantic space transformation on the historical environmental acquisition information before the occurred fire;

[0032] Based on the environmental semantic vectors corresponding to other sub-regions, according to the association relationship represented by the regional relationship network, semantic association aggregation is performed on the environmental semantic vector corresponding to the target sub-region to form a corresponding aggregated environmental semantic vector.

[0033] In a preferred selection of the present application, in the above forest fire monitoring method, the step of performing semantic association aggregation on the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the regional relationship network to form a corresponding aggregated environmental semantic vector includes:

[0034] Based on the environmental semantic vector corresponding to the target sub-region, attention processing is respectively performed on the environmental semantic vectors corresponding to each of the other sub-regions to form corresponding environmental attention vectors for each;

[0035] According to the association relationship represented by the regional relationship network, semantic association aggregation is performed on the environmental attention vector corresponding to each of the other sub-regions and the environmental semantic vector corresponding to the target sub-region, and the corresponding aggregated environmental semantic vector is output.

[0036] In a preferred selection of the present application, in the above forest fire monitoring method, the step of performing semantic association aggregation on the environmental attention vector corresponding to each of the other sub-regions and the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the regional relationship network and outputting the corresponding aggregated environmental semantic vector includes:

[0037] For each of the other sub-regions, according to the association relationship represented by the regional relationship network, in the regional relationship network, the shortest association path between the other sub-region and the target sub-region is determined, where there is an association relationship between every two adjacent sub-regions on the shortest association path;

[0038] Based on the length of the shortest association path between each of the other sub-regions and the target sub-region, linear mapping is respectively performed on the length of each of the shortest association paths to form an association relationship representation parameter between each of the other sub-regions and the target sub-region;

[0039] Taking the association relationship representation parameter as a weight parameter, weighted aggregation is performed on the environmental attention vectors corresponding to each of the other sub-regions to form an environmental aggregation vector;

[0040] Perform an addition operation on the environmental aggregation vector and the environmental semantic vector corresponding to the target sub-region to form a corresponding aggregated environmental semantic vector.

[0041] This application also provides a forest fire monitoring device, including:

[0042] A relationship network construction module, configured to construct a regional relationship network corresponding to the target forest area based on the environmental acquisition information corresponding to each sub-region in the target forest area, where the environmental acquisition information includes personnel activity information, temperature information, humidity information, and wind speed information;

[0043] A semantic aggregation module, configured to perform semantic aggregation on the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the regional relationship network based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions, to form a corresponding aggregated environmental semantic vector;

[0044] A semantic enhancement module, configured to perform semantic enhancement on the aggregated environmental semantic vector based on the image semantic vector of the forest remote sensing image corresponding to the target forest area, to form a corresponding enhanced environmental semantic vector;

[0045] A semantic analysis module, configured to perform semantic analysis based on the enhanced environmental semantic vector and output target fire monitoring data, where the target fire monitoring data is used to reflect the probability of a fire occurring in the target sub-region.

[0046] On the above basis, this application also provides an electronic device, including:

[0047] A memory, configured to store a computer program;

[0048] A processor connected to the memory, configured to execute the computer program stored in the memory to implement the above-mentioned forest fire monitoring method.

[0049] The forest fire monitoring method, device, and equipment provided by this application first construct a regional relationship network based on the environmental acquisition information corresponding to each sub-region. Secondly, based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions, the environmental semantic vector of the environmental acquisition information corresponding to the target sub-region is semantically aggregated according to the association relationship represented by the regional relationship network to form an aggregated environmental semantic vector. Then, based on the image semantic vector corresponding to the forest remote sensing image, the aggregated environmental semantic vector is semantically enhanced to form an enhanced environmental semantic vector. Finally, semantic analysis is performed based on the enhanced environmental semantic vector, and target fire monitoring data is output to reflect the probability of a fire occurring. Based on the above, on the one hand, since potential semantic information in the environmental acquisition information and the forest remote sensing image is captured to achieve fire prediction, the basis for analysis can be more abundant compared to conventional technical solutions that simply perform threshold comparisons. On the other hand, since a regional relationship network is constructed and the aggregation of potential semantic information in the environmental acquisition information is achieved based on the association relationship represented by the regional relationship network during the mining of potential semantic information, the reliability of the aggregation is higher, that is, the semantic representation accuracy of the potential semantic information corresponding to the obtained environmental acquisition information is higher, thereby further improving the reliability of fire prediction and further improving the problem of relatively low reliability of forest fire monitoring existing in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] To make the above objects, features, and advantages of this application more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes the following detailed description.

[0051] Figure 1 It is a structural block diagram of an electronic device provided by an embodiment of this application.

[0052] Figure 2 It is a schematic flowchart of a forest fire monitoring method provided by an embodiment of this application.

[0053] Figure 3 It is a schematic diagram of a regional relationship network provided by an embodiment of this application.

[0054] Figure 4 It is a schematic diagram of a first regional contour image provided by an embodiment of this application.

[0055] Figure 5 It is a schematic diagram of a second regional contour image provided by an embodiment of this application.

[0056] Figure 6 It is a schematic diagram of a regional contour image provided by an embodiment of this application.

[0057] Figure 7 It is a schematic block diagram of a forest fire monitoring device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part rather than all of the embodiments of this application. Generally, the components of the embodiments of this application described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0059] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application that is claimed, but merely represents selected embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts fall within the scope of protection of this application.

[0060] As Figure 1 shown, the embodiments of this application provide an electronic device. Among them, the electronic device may include a memory, a processor, and a forest fire monitoring device.

[0061] Specifically, the memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, the memory and the processor may be electrically connected through one or more communication buses or signal lines. The forest fire monitoring device includes at least one software function module stored in the memory in the form of software or firmware. The processor is used to execute the executable computer programs stored in the memory, such as the software function modules and computer programs included in the forest fire monitoring device, etc., to implement the forest fire monitoring method provided by the embodiments of this application.

[0062] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0063] Optionally, the processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0064] It can be understood that Figure 1 The structure shown is only schematic, and the electronic device may further include more or fewer components than those shown Figure 1 herein, or have a different configuration from that shown Figure 1 For example, it may further include a communication unit for information interaction with other devices (such as sensors, etc.).

[0065] In combination with Figure 2 , an embodiment of the present application further provides a forest fire monitoring method applicable to the above-mentioned electronic device. Among them, the method steps defined by the processes related to the forest fire monitoring method can be implemented by the electronic device.

[0066] Next, the specific process shown Figure 2 will be elaborated in detail.

[0067] Step S110: Based on the environmental acquisition information corresponding to each sub-region in the target forest area, construct a regional relationship network corresponding to the target forest area.

[0068] In an embodiment of the present application, the electronic device may construct a regional relationship network corresponding to the target forest area based on the environmental acquisition information corresponding to each sub-area in the target forest area. Wherein, the environmental acquisition information includes personnel activity information, temperature information, humidity information, and wind speed information. Exemplarily, the target forest area may refer to a complete forest area or a part of a complete forest area, which can be specifically configured according to actual requirements. In addition, the specific size of the sub-area is not limited either. For example, based on the need for accuracy, it can be determined based on the effective range of the corresponding sensor to collect environmental acquisition information, or, considering requirements such as cost, the target forest area can also be divided into equal areas to form multiple sub-areas (such as 10, 20, 100, etc.). In this way, corresponding sensors can be deployed at the center position of each sub-area to collect and form corresponding environmental acquisition information. Among them, the personnel activity information can be collected by an image sensor, the temperature information can be collected by a temperature sensor, the humidity information can be collected by a humidity sensor, and the wind speed information can be collected by a wind speed sensor. In addition, in the regional relationship network, two sub-areas with an association relationship can be connected together. In this way, in subsequent semantic aggregation, semantic information with an association relationship can be aggregated based on such a connection, thereby improving the reliability of semantic aggregation.

[0069] Step S120: Based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-areas, according to the association relationship represented by the regional relationship network, perform semantic aggregation on the environmental semantic vectors of the environmental acquisition information corresponding to the target sub-area to form corresponding aggregated environmental semantic vectors.

[0070] In an embodiment of the present application, after constructing the regional relationship network, the electronic device may perform semantic aggregation on the environmental semantic vectors of the environmental acquisition information corresponding to the target sub-area according to the association relationship represented by the regional relationship network based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-areas. That is to say, a target sub-area (such as having a higher importance in one or more dimensions) can be determined as the center or representative among the sub-areas included in the target forest area. In this way, the semantic information of other sub-areas can be aggregated into the semantic information of the target sub-area, thereby achieving targeted semantic aggregation, using the semantic information of the target sub-area as the main basis and the semantic information of other sub-areas as the auxiliary basis.

[0071] Step S130: Based on the image semantic vector of the forest remote sensing image corresponding to the target forest area, perform semantic enhancement on the aggregated environmental semantic vector to form a corresponding enhanced environmental semantic vector.

[0072] In the embodiment of the present application, after forming the aggregated environmental semantic vector, the electronic device may perform semantic enhancement on the aggregated environmental semantic vector based on the image semantic vector of the forest remote sensing image corresponding to the target forest area to form a corresponding enhanced environmental semantic vector. That is to say, since the semantic information corresponding to the environmental acquisition information of each sub-region is in the same semantic space, semantic aggregation can be performed first to obtain the aggregated environmental semantic vector. In this way, the global semantic information in the environmental dimension can be obtained. Based on this, after obtaining the image semantic vector of the forest remote sensing image, that is, after obtaining the global semantic information in the morphological dimension, the global semantic information of the two dimensions can be further fused. Among them, since the content represented by the global semantic information in the environmental dimension is richer and the content represented by the global semantic information in the morphological dimension is relatively single, the global semantic information in the morphological dimension can be fused into the semantic information in the environmental dimension to achieve the semantic enhancement of the image semantic vector to the aggregated environmental semantic vector.

[0073] Step S140, perform semantic analysis based on the enhanced environmental semantic vector and output target fire monitoring data.

[0074] In the embodiment of the present application, after obtaining the enhanced environmental semantic vector, the electronic device may perform semantic analysis based on the enhanced environmental semantic vector and output target fire monitoring data. The target fire monitoring data is used to reflect the probability of a fire occurring in the target sub-region. Exemplarily, a parameter, such as 0 - 1, can be obtained through linear regression of the enhanced environmental semantic vector to represent the probability of a fire occurring, where 0 indicates that no fire will occur and 1 indicates that a fire will occur. For example, in a specific case, when there is no human activity, the temperature is low, the humidity is high, the wind speed is low, and the morphology of the plants is in a normal growth state, the fire probability obtained may be 0. Another example, in another specific case, when there is frequent human activity, the temperature is high, the humidity is low, the wind speed is high, and the morphology of the plants shows many dry areas, the fire probability obtained may be 1.

[0075] Based on the above content, on the one hand, since the potential semantic information in the environmental acquisition information and the forest remote sensing image is captured to achieve fire prediction, the basis for analysis can be more abundant compared with the conventional technical solutions that simply perform threshold comparison. On the other hand, since a regional relationship network is constructed and the aggregation of potential semantic information is achieved based on the association relationship represented by the regional relationship network when mining the potential semantic information of the environmental acquisition information, the reliability of the aggregation is higher, that is, the semantic representation accuracy of the potential semantic information corresponding to the obtained environmental acquisition information is higher, thereby further improving the reliability of fire prediction and then improving the relatively low reliability of forest fire monitoring existing in the prior art.

[0076] In the first aspect, regarding step S110, it should be noted that the specific method of constructing the regional relationship network corresponding to the target forest area is not limited and can be selected according to actual needs.

[0077] For example, in an alternative implementation, on the one hand, two adjacent sub-regions can be determined to have an associated relationship. On the other hand, if there is a road in the target forest area, two sub-regions connected by the road can also be determined to have an associated relationship. And in some implementations, sub-regions with the same vegetation can also be determined to have an associated relationship. Based on this, after determining the associated relationship, the corresponding two sub-regions can be connected to form a regional relationship network, which can be as Figure 3 shown.

[0078] In the second aspect, regarding step S120, it should be noted that the specific method of semantic aggregation of the environmental semantic vectors of the environmental information collected for the target sub-region is not limited and can be selected accordingly according to actual needs.

[0079] For example, in an alternative implementation, a sub-region with the most frequent human activities can be first determined as the target sub-region, and then, the environmental semantic vectors of the environmental information collected for other sub-regions can be aggregated into the environmental semantic vector of the environmental information collected for the target sub-region, so as to form a corresponding aggregated environmental semantic vector.

[0080] Again, for example, in another alternative implementation, in order to ensure the reliability of semantic aggregation, step S120 described above can further include step S121, step S122, and step S123, where the specific content of each step is described as follows.

[0081] Step S121, perform semantic space conversion on the environmental information collected for each sub-region respectively to form the environmental semantic vector corresponding to each said sub-region.

[0082] In the embodiments of the present application, semantic space conversion can be performed on the environment acquisition information corresponding to each sub-region respectively to form an environment semantic vector corresponding to each of the sub-regions. For example, for sub-region a, the corresponding environment semantic vector 1 can be obtained. For sub-region b, the corresponding environment semantic vector 2 can be obtained. In addition, regarding the semantic space conversion, it should be noted that the personnel activity information can be the statistical number of people obtained by recognizing images in a period of time, and the temperature information, humidity information, and wind speed information can be the average values in a period of time. Based on this, the environment acquisition information corresponding to a sub-region can be "the statistical number of people is 10, the average temperature is 25 degrees Celsius, the average humidity is 56%, and the average wind speed is 2 meters per second". Based on this, word segmentation processing can be performed on this environment acquisition information to obtain corresponding words such as "personnel", "statistics", "quantity", "is", etc. Then, word embedding processing can be performed on each word respectively, which can be specifically implemented through models such as Word2Vec, so as to obtain the word embedding vectors corresponding to each word, such as:

[0083] "personnel": [0.25, -0.12, 0.33, 0.41, -0.55, 0.18, -0.44, 0.19, -0.31, 0.21,..., 0.10]; "statistics": [-0.32, 0.41, -0.22, 0.50, 0.34, -0.18, 0.60, -0.45, 0.12, 0.29,..., -0.08]; "quantity": [0.18, 0.14, 0.56, -0.10, 0.42, -0.36, 0.26, 0.15, 0.47, -0.21,..., 0.39]; "is": [-0.11, 0.33, 0.12, -0.05, 0.22, 0.38, -0.19, 0.50, 0.25, -0.14,..., 0.13].

[0084] Then, the word embedding vectors corresponding to each word can be concatenated (alternatively, the embedding vectors of the position encodings corresponding to each word can also be fused, and the specific method can refer to the relevant prior art and will not be elaborated here one by one), so as to form the corresponding environment semantic vector.

[0085] Step S122, based on the matching degree between each of the environment semantic vectors and the fire environment semantic vector, determine the most matching target sub-region among the sub-regions.

[0086] In the embodiment of the present application, after obtaining the environmental semantic vector, the most matching target sub-region can be determined in each sub-region based on the matching degree between each environmental semantic vector and the fire environmental semantic vector. Among them, the fire environmental semantic vector is formed by performing semantic space conversion on the historical environmental acquisition information before the occurred fire. That is to say, the historical environmental acquisition information of the target forest area (or other forest areas with certain similarities in environment, vegetation, etc. to the target forest area) before the fire (the time needs to have a certain relevance, that is, the interval time cannot be too long, such as one day, one week, etc.) can be subjected to semantic space conversion to obtain the fire environmental semantic vector. That is to say, the fire environmental semantic vector can represent the semantic information that may have occurred in the fire to a certain extent. Based on this, the sub-region most likely to have a fire can be determined through the above-mentioned matching and screening, and thus used as the key attention object, that is, as the target sub-region. In addition, the matching degree can refer to the cosine similarity between semantic vectors. In this way, the sub-region with the largest cosine similarity can be determined as the most matching sub-region, that is, as the target sub-region.

[0087] Step S123: Based on the environmental semantic vectors corresponding to other sub-regions, perform semantic association aggregation on the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the regional relationship network, so as to form a corresponding aggregated environmental semantic vector.

[0088] In the embodiment of the present application, after determining the target sub-region, based on the environmental semantic vectors corresponding to other sub-regions, perform semantic association aggregation on the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the regional relationship network, so as to form a corresponding aggregated environmental semantic vector.

[0089] It can be understood that in the above step S123, the specific manner of performing semantic association aggregation on the environmental semantic vector corresponding to the target sub-region is not limited. For example, in an alternative embodiment, in order to effectively capture the associated semantic information during the association aggregation process, so as to improve the representation ability of the obtained aggregated environmental semantic vector, the above step S123 can further include steps S123a and S123b, and the specific content of each step is as follows.

[0090] Step S123a: Based on the environmental semantic vector corresponding to the target sub-region, perform attention processing on the environmental semantic vector corresponding to each other sub-region respectively, so as to form a corresponding environmental attention vector for each one.

[0091] In an embodiment of the present application, based on the environmental semantic vector corresponding to the target sub-region, attention processing is respectively performed on the environmental semantic vectors corresponding to each of the other sub-regions to form corresponding environmental attention vectors for each of them. That is to say, for each of the other sub-regions, semantic information associated with the environmental semantic vector corresponding to the target sub-region can be extracted from the environmental semantic vector corresponding to this other sub-region, so as to obtain the corresponding environmental attention vector. Among them, when performing attention processing, the query vector is obtained by multiplying the environmental semantic vector corresponding to the target sub-region with the query matrix in the attention network, and the key vector and value vector are obtained by multiplying the environmental semantic vectors corresponding to the other sub-regions with the key matrix and value matrix in the attention network respectively.

[0092] Step S123b: According to the association relationship represented by the region relationship network, perform semantic association aggregation on the environmental attention vector corresponding to each of the other sub-regions and the environmental semantic vector corresponding to the target sub-region, and output the corresponding aggregated environmental semantic vector.

[0093] In an embodiment of the present application, after obtaining the corresponding environmental attention vector, semantic association aggregation can be performed on the environmental attention vector corresponding to each of the other sub-regions and the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the region relationship network, and the corresponding aggregated environmental semantic vector is output. That is to say, since the association degrees between the other sub-regions and the target sub-region are actually different, in the process of aggregation, in order to ensure the accuracy of aggregation, reliable aggregation needs to be achieved according to the corresponding association relationship.

[0094] It can be understood that in the above step S123b, the specific manner of performing semantic association aggregation on the environmental attention vector corresponding to each of the other sub-regions and the environmental semantic vector corresponding to the target sub-region is not limited. For example, in an alternative embodiment, in order to make full use of the association relationship between sub-regions to ensure the reliability of association aggregation, the above step S123b may further include the following content:

[0095] First, for each of the other sub-regions, according to the association relationship represented by the region relationship network, in the region relationship network, determine the shortest association path between this other sub-region and the target sub-region, where there is an association relationship between every two adjacent sub-regions on the shortest association path. For example, in Figure 3 the shown region relationship network, the shortest association path between sub-region a and sub-region e is "sub-region a - sub-region b - sub-region e" or "sub-region a - sub-region d - sub-region e";

[0096] Secondly, based on the lengths of the shortest association paths between each other sub-region and the target sub-region, the lengths of each of the shortest association paths can be linearly mapped respectively to form the association relationship characterization parameters between each other sub-region and the target sub-region. Exemplarily, the sum value of the reciprocals of the lengths of each of the shortest association paths can be determined, and then, for the length of each of the shortest association paths, the ratio between the reciprocal of this length and this sum value is calculated, so as to obtain the association relationship characterization parameter of the other sub-region corresponding to this length. That is to say, the larger the length, the smaller the association relationship characterization parameter, and the smaller the length, the larger the association relationship characterization parameter;

[0097] Then, the association relationship characterization parameter can be used as a weight parameter to perform weighted aggregation (summation) on the environmental attention vectors corresponding to each other sub-region to form an environmental aggregation vector, that is, environmental aggregation vector = environmental attention vector 1 * association relationship characterization parameter 1 + environmental attention vector 2 * association relationship characterization parameter 2 + environmental attention vector 3 * association relationship characterization parameter 3 +......;

[0098] Finally, an addition operation (or mean operation) can be performed on the environmental aggregation vector and the environmental semantic vector corresponding to the target sub-region to form the corresponding aggregated environmental semantic vector.

[0099] In the third aspect, it should be noted that for step S130, the specific manner of semantic enhancement for the aggregated environmental semantic vector is not limited and can be selected according to actual needs.

[0100] For example, in an alternative implementation manner, in order to achieve efficient semantic enhancement, the forest remote sensing image is subjected to convolutional processing to obtain the corresponding image semantic vector. Then, this image semantic vector and the aggregated environmental semantic vector can be concatenated or subjected to weighted summation operation, etc., so as to obtain the enhanced environmental semantic vector.

[0101] For another example, in another alternative implementation manner, in order to achieve more reliable semantic enhancement, during the process of semantic mining of the forest remote sensing image, some semantic information related to the occurrence of a fire can be focused on for mining, so that the obtained image semantic vector can fully and effectively represent the semantics related to the fire, thereby ensuring the accuracy of semantic enhancement and avoiding the interference of invalid semantic information on semantic enhancement. Based on this, the above step S130 can further include step S131, step S132, step S133, step S134 and step S135, where the specific content of each step is described as follows.

[0102] Step S131: Based on a pre-determined target color threshold, perform determination processing on the forest remote sensing image corresponding to the target forest area to obtain a corresponding target suspected dry area.

[0103] In an embodiment of the present application, based on a pre-determined target color threshold, determination processing can be performed on the forest remote sensing image corresponding to the target forest area to obtain a corresponding target suspected dry area. Exemplarily, through analysis, it can be known that the colors of dry trees usually tend to be brown, gray, brown, grayish-brown, charred black, etc., and these colors will vary with the dryness degree of the trees and the tree species. Based on this, in order to implement the determination processing of the suspected dry area, screening can be performed based on the corresponding target color threshold. For example, for the target color threshold, the red channel (R) is 20 - 220, the green channel (G) is 10 - 160, and the blue channel (B) is 10 - 150. That is, in the forest remote sensing image, each pixel point whose color channel values all belong to the interval corresponding to the target color threshold is screened out, so as to form a corresponding target suspected dry area.

[0104] Step S132: Perform binarization processing on the target suspected dry area in the forest remote sensing image, and perform value assignment processing on the pixel values of other areas to form a binarized remote sensing image.

[0105] In an embodiment of the present application, after determining the target suspected dry area, binarization processing can be performed on the target suspected dry area in the forest remote sensing image, and value assignment processing can be performed on the pixel values of other areas to form a binarized remote sensing image. For example, in the binarized remote sensing image, the pixel value of each pixel point in other areas is equal to 0.

[0106] Step S133: Perform morphological operations in the binarized remote sensing image to form a regional contour image.

[0107] In an embodiment of the present application, after obtaining the binarized remote sensing image, morphological operations can be performed in the binarized remote sensing image to form a regional contour image. That is to say, through morphological operations, the regional contour (actual dry area) therein can be further confirmed, so as to obtain a regional contour image with higher reliability.

[0108] Step S134: Based on the regional contour image and the forest remote sensing image, associate and mine the image semantic vector of the forest remote sensing image.

[0109] In an embodiment of the present application, after obtaining the regional contour image, an image semantic vector of the forest remote sensing image can be associated and mined based on the regional contour image and the forest remote sensing image. That is to say, since the regional contour image may only represent the contour of the dry area, there may actually be some other potential semantic information that plays a positive role in fire prediction. Therefore, it will be mined in combination with the forest remote sensing image to obtain an image semantic vector that can represent the dry area with emphasis and also represent other potential semantic information.

[0110] Step S135: Semantically enhance the aggregated environmental semantic vector based on the image semantic vector to form a corresponding enhanced environmental semantic vector.

[0111] In an embodiment of the present application, after obtaining the image semantic vector, the aggregated environmental semantic vector can be semantically enhanced based on the image semantic vector to form a corresponding enhanced environmental semantic vector.

[0112] It can be understood that in the above step S133, the specific manner of performing morphological operations on the binarized remote sensing image is not limited. For example, in an alternative embodiment, in order to improve the reliability of the determined regional contour image, the above step S133 may further include the following content:

[0113] First, the binarized remote sensing image can be respectively subjected to a first morphological dilation process and a second morphological dilation process to obtain corresponding first dilated remote sensing image and second dilated remote sensing image, where the size of the structural element matrix corresponding to the first morphological dilation process is smaller than the size of the structural element matrix corresponding to the second morphological dilation process. For example, when the structural element matrix is a rectangle, the size of the structural element matrix corresponding to the first morphological dilation process can be 3*3, and the size of the structural element matrix corresponding to the second morphological dilation process can be 5*5. Based on this, more details can be retained in the first dilated remote sensing image and the image expansion is smaller, while more details are lost in the second dilated remote sensing image and the image expansion is more obvious;

[0114] Secondly, the binary remote sensing image can be respectively subjected to a first morphological erosion process and a second morphological erosion process to obtain corresponding first eroded remote sensing image and second eroded remote sensing image. Among them, the size of the structural element matrix corresponding to the first morphological erosion process is equal to the size of the structural element matrix corresponding to the first morphological dilation process (such as both being 3*3), and the size of the structural element matrix corresponding to the second morphological erosion process is equal to the size of the structural element matrix corresponding to the second morphological dilation process (such as both being 5*5). In this way, more details can be retained in the first eroded remote sensing image, and the edge changes are small. In the second eroded remote sensing image, details and boundaries may be lost;

[0115] Then, the first dilated remote sensing image and the first eroded remote sensing image can be subjected to a difference calculation to form a first regional contour image (such as Figure 4 ), and the second dilated remote sensing image and the second eroded remote sensing image can be subjected to a difference calculation to form a second regional contour image (such as Figure 5 ); that is to say, it is necessary to perform a difference operation on the dilation result and the erosion result corresponding to the structural element matrix respectively. In this way, since the same structural element matrix is used for dilation and erosion, the obtained difference is the edge information of the image, thus realizing the effective extraction of the contour. However, since the structural element matrix corresponding to the first regional contour image is smaller than the structural element matrix corresponding to the second regional contour image, the contour of the first regional contour image will be smaller, and the contour of the second regional contour image will be larger;

[0116] Finally, the first regional contour image can be updated based on the second regional contour image to form a regional contour image, where the regional contour represented by the regional contour image is located between the regional contour of the second regional contour image and the regional contour of the first regional contour image; Exemplarily, the middle pixel points of the inner contour edge in the second regional contour image and the inner contour edge in the first regional contour image can be determined as the inner contour edge of the regional contour image, and the middle pixel points of the outer contour edge in the second regional contour image and the outer contour edge in the first regional contour image can be determined as the outer contour edge of the regional contour image, such as Figure 6 shown.

[0117] In addition, it should be noted that Figure 4 、 Figure 5 and Figure 6 The shapes of the contours in are the same, only the widths of the contours are different. In some cases, Figure 4 and Figure 5The shapes of the contours between them may not be exactly the same, but they still belong to relatively similar shapes, that is, the shapes are basically the same. Based on this, in order to update the regional contour of the first regional contour image based on the second regional contour image, the following solution can be adopted:

[0118] First, traverse each pixel point (or each sampled pixel point) on the outer edge of the contour in the first regional contour image. Then, determine the pixel point closest to this pixel point on the outer edge of the contour in the second regional contour image. After that, the intermediate pixel point (the intermediate value of the coordinates) of these two pixel points can be determined. Finally, the intermediate pixel points corresponding to each pixel point on the outer edge of the contour in the first regional contour image can be connected to form the outer edge of the contour of the regional contour image (the determination method of the inner edge of the contour is the same).

[0119] It can be understood that in step S134 above, the specific method of associatively mining the image semantic vector of the forest remote sensing image is not limited. For example, in an alternative implementation, the regional contour image and the forest remote sensing image can be associatively mined based on the attention mechanism to obtain the image semantic vector. Another example is that in another alternative implementation, step S134 above may include the following content:

[0120] First, the regional contour image and the forest remote sensing image can be convolved respectively to form corresponding regional contour convolution vectors and forest remote sensing convolution vectors. Among them, the regional contour image and the forest remote sensing image have the same size, and can be convolved through the same convolutional network layer;

[0121] Second, self-attention processing can be performed on the regional contour convolution vector and the forest remote sensing convolution vector respectively to form corresponding regional contour attention vectors and forest remote sensing attention vectors;

[0122] Then, a non-linear mapping can be performed on the forest remote sensing attention vector to form a forest remote sensing mapping vector, where each vector parameter in the forest remote sensing mapping vector is greater than or equal to 0 and less than or equal to 1. For example, a linear mapping can be performed on the forest remote sensing attention vector first, and then a non-linear mapping can be performed on the result of the linear mapping to obtain the forest remote sensing mapping vector. For example, y(x)=σ(ax + b), where ax + b represents the linear mapping, and σ() represents the non-linear mapping, such as being implemented through the Sigmoid activation function;

[0123] Finally, the vector parameters at the corresponding positions between the forest remote sensing mapping vector and the region contour attention vector can be multiplied (i.e., element-wise multiplication) to form the image semantic vector of the forest remote sensing image. In this way, based on the screening of the semantic information in the region contour attention vector by the forest remote sensing mapping vector, for example, through the vector parameters equal to 0 in the forest remote sensing mapping vector, the semantic information represented by the vector parameters at the corresponding positions in the region contour attention vector can be discarded; through the vector parameters equal to 1 in the forest remote sensing mapping vector, the semantic information represented by the vector parameters at the corresponding positions in the region contour attention vector can be completely retained; through the vector parameters greater than 0 and less than 1 in the forest remote sensing mapping vector, the semantic information represented by the vector parameters at the corresponding positions in the region contour attention vector can be partially retained, so as to realize the semantic association aggregation of the region contour image and the forest remote sensing image. Based on this, compared with the method of using attention for semantic association aggregation, the computational complexity of semantic association aggregation can be reduced to a certain extent, thereby improving the efficiency.

[0124] It can be understood that in the above step S135, the specific method of semantic enhancement of the aggregated environmental semantic vector is not limited. For example, in an alternative implementation, since the image semantic vector and the aggregated environmental semantic vector belong to semantic information in the image dimension and the text dimension respectively, considering that the attention mechanism has a better effect on cross-modal semantic information fusion, the attention mechanism is used to realize the semantic aggregation of the image semantic vector and the aggregated environmental semantic vector. The specific implementation process is as follows:

[0125] First, a query vector can be determined based on the image semantic vector (such as multiplying by a query matrix), and a key vector and a value vector can be determined based on the aggregated environmental semantic vector (such as multiplying by a key matrix and a value matrix respectively);

[0126] Second, an attention distribution matrix can be determined based on the dot product between the query vector and the key vector (for example, the dot product can be directly used as the corresponding attention distribution matrix, or the dot product can be further calculated, such as dividing by a target parameter to obtain the attention distribution matrix, and the target parameter can be related to the dimension of the key vector, such as equal to the 0.5th power of the dimension of the key vector), and based on the attention distribution matrix, a weighted sum calculation is performed on the value vector to form the corresponding enhanced environmental semantic vector.

[0127] Fourthly, regarding step S140, it should be noted that the specific manner of performing semantic analysis based on the enhanced environmental semantic vector is not limited and can be selected according to actual needs. For example, in an alternative embodiment, the enhanced environmental semantic vector can be subjected to a fully connected process to obtain a corresponding fully connected vector. The size of this fully connected vector can be 1*1, that is, it includes a vector parameter. Then, an identity mapping can be performed on this vector parameter, that is, f(x)=x, so as to obtain an output value, which can be used to represent the probability of a fire occurring in the target sub-region.

[0128] Combined with Figure 7 , the embodiments of the present application further provide a forest fire monitoring device applicable to the above-mentioned electronic device. Among them, the forest fire monitoring device may include a relationship network construction module, a semantic aggregation module, a semantic enhancement module, and a semantic analysis module.

[0129] Specifically, the relationship network construction module can be used to construct a regional relationship network corresponding to the target forest area based on the environmental acquisition information corresponding to each sub-region in the target forest area, where the environmental acquisition information includes personnel activity information, temperature information, humidity information, and wind speed information. In the embodiments of the present application, the relationship network construction module can be used to execute Figure 2 the step S110 shown in

[0130] For the relevant content of the relationship network construction module, reference can be made to the description of step S110 above. Figure 2 Specifically, the semantic aggregation module can be used to semantically aggregate the environmental semantic vector of the environmental acquisition information corresponding to the target sub-region according to the association relationship represented by the regional relationship network based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions, so as to form a corresponding aggregated environmental semantic vector. In the embodiments of the present application, the semantic aggregation module can be used to execute

[0131] the step S120 shown in Figure 2 For the relevant content of the semantic aggregation module, reference can be made to the description of step S120 above.

[0132] Specifically, the semantic analysis module can be used to perform semantic analysis based on the enhanced environmental semantic vector and output target fire monitoring data, where the target fire monitoring data is used to reflect the probability of a fire occurring in the target sub-region. In an embodiment of the present application, the semantic analysis module can be used to execute Figure 2 step S140 shown in the figure. For the relevant content of the semantic analysis module, reference can be made to the description of step S140 above.

[0133] In an embodiment of the present application, corresponding to the above forest fire monitoring method applied to the electronic device, a computer-readable storage medium is further provided. A computer program is stored in the computer-readable storage medium, and when the computer program runs, it executes each step of the forest fire monitoring method.

[0134] Among them, the steps executed when the foregoing computer program runs will not be elaborated one by one here, and reference can be made to the explanation of the forest fire monitoring method above.

[0135] In summary, for the forest fire monitoring method, device, and equipment provided in the present application, first, based on the environmental acquisition information corresponding to each sub-region, a regional relationship network is constructed; second, based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions, according to the association relationship represented by the regional relationship network, semantic aggregation is performed on the environmental semantic vector of the environmental acquisition information corresponding to the target sub-region to form an aggregated environmental semantic vector; then, based on the image semantic vector corresponding to the forest remote sensing image, semantic enhancement is performed on the aggregated environmental semantic vector to form an enhanced environmental semantic vector; finally, semantic analysis is performed based on the enhanced environmental semantic vector, and target fire monitoring data is output to reflect the probability of a fire occurring. Based on the above content, on the one hand, since potential semantic information in the environmental acquisition information and the forest remote sensing image will be captured to achieve fire prediction, the basis for analysis can be more abundant compared with the conventional technical solution of simply comparing thresholds. On the other hand, since a regional relationship network will be constructed, and when mining potential semantic information from the environmental acquisition information, the aggregation of potential semantic information will be achieved based on the association relationship represented by the regional relationship network, making the aggregation more reliable, that is, ensuring a higher accuracy of the semantic representation of the potential semantic information corresponding to the obtained environmental acquisition information, thereby further improving the reliability of fire prediction and further improving the problem of relatively low reliability of forest fire monitoring existing in the prior art.

[0136] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0137] In addition, each functional module in various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0138] If the function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes. It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements that are not explicitly listed, or further includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article, or device including the element.

[0139] The above are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A forest fire monitoring method, characterized in that, Including: Based on the environmental acquisition information corresponding to each sub-region in the target forest area, construct a regional relationship network corresponding to the target forest area, where the environmental acquisition information includes personnel activity information, temperature information, humidity information, and wind speed information; Based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions, according to the association relationship represented by the regional relationship network, perform semantic aggregation on the environmental semantic vectors of the environmental acquisition information corresponding to the target sub-region to form corresponding aggregated environmental semantic vectors. The target sub-region refers to the sub-region that is the center among the sub-regions, the sub-region with the most frequent personnel activities, or the sub-region most likely to have a fire. The determination method for the sub-region most likely to have a fire includes matching and screening based on the matching degree between the corresponding environmental semantic vector and the fire environmental semantic vector. This fire environmental semantic vector is formed by performing semantic space transformation on the historical environmental acquisition information before the occurred fire; Based on the image semantic vector of the forest remote sensing image corresponding to the target forest area, perform semantic enhancement on the aggregated environmental semantic vector to form a corresponding enhanced environmental semantic vector; Based on the enhanced environmental semantic vector, perform semantic analysis and output target fire monitoring data, where the target fire monitoring data is used to reflect the probability of a fire occurring in the target sub-region.

2. The forest fire monitoring method according to claim 1, wherein The step of performing semantic enhancement on the aggregated environmental semantic vector based on the image semantic vector of the forest remote sensing image corresponding to the target forest area to form a corresponding enhanced environmental semantic vector includes: Based on a pre-determined target color threshold, perform determination processing on the forest remote sensing image corresponding to the target forest area to obtain a corresponding target dry suspected area; Perform binarization processing on the target dry suspected area in the forest remote sensing image and assign pixel values to other areas to form a binarized remote sensing image; Perform morphological operations on the binarized remote sensing image to form a regional contour image; Based on the regional contour image and the forest remote sensing image, perform associated mining to obtain the image semantic vector of the forest remote sensing image; Based on the image semantic vector, perform semantic enhancement on the aggregated environmental semantic vector to form a corresponding enhanced environmental semantic vector.

3. The forest fire monitoring method according to claim 2, characterized in that The step of performing morphological operations on the binarized remote sensing image to form a regional contour image includes: Perform a first morphological dilation process and a second morphological dilation process on the binarized remote sensing image respectively to obtain corresponding first dilated remote sensing image and second dilated remote sensing image, where the size of the structural element matrix corresponding to the first morphological dilation process is smaller than the size of the structural element matrix corresponding to the second morphological dilation process; Perform first morphological erosion processing and second morphological erosion processing on the binary remote sensing image respectively to obtain corresponding first eroded remote sensing image and second eroded remote sensing image. Wherein, the size of the structural element matrix corresponding to the first morphological erosion processing is equal to the size of the structural element matrix corresponding to the first morphological dilation processing, and the size of the structural element matrix corresponding to the second morphological erosion processing is equal to the size of the structural element matrix corresponding to the second morphological dilation processing; Perform difference calculation on the first dilated remote sensing image and the first eroded remote sensing image to form a first regional contour image, and perform difference calculation on the second dilated remote sensing image and the second eroded remote sensing image to form a second regional contour image; Perform regional contour update on the first regional contour image based on the second regional contour image to form a regional contour image, wherein the regional contour represented by the regional contour image is located between the regional contour of the second regional contour image and the regional contour of the first regional contour image.

4. The forest fire monitoring method according to claim 2, wherein The step of associatively mining the image semantic vector of the forest remote sensing image based on the regional contour image and the forest remote sensing image includes: Perform convolution on the regional contour image and the forest remote sensing image respectively to form corresponding regional contour convolution vector and forest remote sensing convolution vector; Perform self-attention processing on the regional contour convolution vector and the forest remote sensing convolution vector respectively to form corresponding regional contour attention vector and forest remote sensing attention vector; Perform non-linear mapping on the forest remote sensing attention vector to form a forest remote sensing mapping vector, wherein each vector parameter in the forest remote sensing mapping vector is greater than or equal to 0 and less than or equal to 1; Multiply the vector parameters at the corresponding positions between the forest remote sensing mapping vector and the regional contour attention vector to form the image semantic vector of the forest remote sensing image.

5. The forest fire monitoring method according to claim 2, wherein, The step of semantically enhancing the aggregated environmental semantic vector based on the image semantic vector to form a corresponding enhanced environmental semantic vector includes: Determine a query vector based on the image semantic vector, and determine a key vector and a value vector based on the aggregated environmental semantic vector; Determine an attention distribution matrix based on the dot product between the query vector and the key vector, and perform weighted summation calculation on the value vector based on the attention distribution matrix to form a corresponding enhanced environmental semantic vector.

6. The forest fire monitoring method according to claim 1, characterized in that The step of semantically aggregating the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions to form a corresponding aggregated environmental semantic vector according to the association relationship represented by the regional relationship network includes: Perform semantic space transformation on the environmental acquisition information corresponding to each sub-region respectively to form the environmental semantic vector corresponding to each sub-region; Based on the matching degree between each of the environmental semantic vectors and the fire environmental semantic vector, the most matching target sub-region is determined among the sub-regions, where the fire environmental semantic vector is formed by performing semantic space transformation on the historical environmental acquisition information before the occurred fire; Based on the environmental semantic vectors corresponding to other sub-regions, semantic association aggregation is performed on the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the regional relationship network, so as to form a corresponding aggregated environmental semantic vector.

7. The forest fire monitoring method according to claim 6, wherein The step of performing semantic association aggregation on the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the regional relationship network based on the environmental semantic vectors corresponding to other sub-regions to form a corresponding aggregated environmental semantic vector includes: Based on the environmental semantic vector corresponding to the target sub-region, attention processing is respectively performed on the environmental semantic vectors corresponding to each of the other sub-regions to form corresponding environmental attention vectors for each; According to the association relationship represented by the regional relationship network, semantic association aggregation is performed on the environmental attention vector corresponding to each of the other sub-regions and the environmental semantic vector corresponding to the target sub-region, and a corresponding aggregated environmental semantic vector is output.

8. The forest fire monitoring method according to claim 7, characterized in that, The step of performing semantic association aggregation on the environmental attention vector corresponding to each of the other sub-regions and the environmental semantic vector corresponding to the target sub-region according to the association relationship represented by the regional relationship network and outputting a corresponding aggregated environmental semantic vector includes: For each of the other sub-regions, according to the association relationship represented by the regional relationship network, in the regional relationship network, the shortest association path between this other sub-region and the target sub-region is determined, where there is an association relationship between every two adjacent sub-regions on the shortest association path; Based on the length of the shortest association path between each of the other sub-regions and the target sub-region, linear mapping is respectively performed on the length of each of the shortest association paths to form an association relationship representation parameter between each of the other sub-regions and the target sub-region; Taking the association relationship representation parameter as a weight parameter, weighted aggregation is performed on the environmental attention vectors corresponding to each of the other sub-regions to form an environmental aggregation vector; An addition operation is performed on the environmental aggregation vector and the environmental semantic vector corresponding to the target sub-region to form a corresponding aggregated environmental semantic vector.

9. A forest fire monitoring device, characterized in that, including: A relationship network construction module, configured to construct a regional relationship network corresponding to the target forest area based on the environmental acquisition information corresponding to each sub-region in the target forest area, where the environmental acquisition information includes personnel activity information, temperature information, humidity information, and wind speed information; A semantic aggregation module, which is used to perform semantic aggregation on the environmental semantic vector of the environmental acquisition information corresponding to the target sub-region according to the association relationship represented by the regional relationship network based on the environmental semantic vectors of the environmental acquisition information corresponding to other sub-regions, so as to form a corresponding aggregated environmental semantic vector. The target sub-region refers to the sub-region that is the center among each sub-region, the sub-region with the most frequent human activities, or the sub-region most likely to have a fire. The determination method of the sub-region most likely to have a fire includes matching and screening based on the matching degree between the corresponding environmental semantic vector and the fire environmental semantic vector. This fire environmental semantic vector is formed by performing semantic space conversion on the historical environmental acquisition information before the occurred fire; A semantic enhancement module, which is used to perform semantic enhancement on the aggregated environmental semantic vector based on the image semantic vector of the forest remote sensing image corresponding to the target forest region to form a corresponding enhanced environmental semantic vector; A semantic analysis module, which is used to perform semantic analysis based on the enhanced environmental semantic vector and output target fire monitoring data. Among them, the target fire monitoring data is used to reflect the probability of a fire occurring in the target sub-region.

10. An electronic device, characterized in that, Comprising: A memory, which is used to store computer programs; A processor connected to the memory, which is used to execute the computer programs stored in the memory to implement the forest fire monitoring method according to any one of claims 1-8.

Images