Forest monitoring method and system based on multi-dimensional data analysis

Abstract

This application provides a forest monitoring method and system based on multidimensional data analysis, relating to the field of data processing technology. In this application, firstly, current forest images and recent meteorological data for the target forest are acquired; secondly, texture features related to healthy growth are extracted from the current forest image to obtain forest image texture features; then, using a forest health identification model, semantic encoding is guided by the latent semantic information of the forest image texture features and the target meteorological data during the semantic encoding process of the current forest image, forming a forest health coding vector; finally, the forest health identification model is used to semantically decode the forest health coding vector to form target forest health status data. Based on the above, the relatively low reliability of forest health status in existing technologies can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a forest monitoring method and system based on multidimensional data analysis. Background Technology

[0002] With the impact of factors such as climate change, environmental pollution, and human intervention, global forest ecosystems are facing increasing challenges. Monitoring and assessing forest health is crucial for the timely detection of forest ecological problems, ensuring ecological security, and implementing effective forest management measures. However, existing forest health monitoring methods still have many shortcomings. Currently, forest health monitoring methods mainly rely on field surveys or remote sensing technology. While these methods can provide some information about forest health status, they generally suffer from the following problems: First, traditional field surveys are limited by time, space, and personnel, require significant human and material resources, and are subject to a degree of subjectivity, resulting in relatively low reliability. Second, remote sensing image analysis mainly relies on the visible information of images, making it difficult to effectively capture the deeper semantic information related to forest health status, thus also reducing the reliability of the analysis.

[0003] Furthermore, with the rapid development of big data technology, data analysis-based methods have been gradually introduced into the field of forest health monitoring. Although some studies have attempted to combine multi-dimensional data to predict forest health status, these methods typically rely on manual feature extraction and traditional statistical models, making it difficult to effectively mine the potential information in the data, and they also have certain errors and limitations. In other words, under current technology, the monitoring of forest health status suffers from relatively low reliability. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a forest monitoring method and system based on multidimensional data analysis, so as to improve the problem of relatively low reliability of forest health status in the prior art.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] A forest monitoring method based on multidimensional data analysis includes:

[0007] Acquire current forest images of the target forest and target meteorological data for the most recent time period;

[0008] Extract texture features related to healthy growth from the current forest image to obtain the forest image texture features of the current forest image;

[0009] Using a forest health identification model, during the semantic encoding of the current forest image, semantic encoding is guided by the latent semantic information of the forest image texture features and the target meteorological data to form a forest health encoding vector. The forest health identification model is a neural network model formed by learning sample data and label data.

[0010] Using the forest health identification model, the forest health encoding vector is semantically decoded to form target forest health status data.

[0011] In a preferred embodiment of this application, in the aforementioned forest monitoring method based on multidimensional data analysis, the step of using a forest health identification model to guide semantic encoding of the current forest image through the latent semantic information of the forest image texture features and the target meteorological data to form a forest health coding vector includes:

[0012] Using the semantic space mapping unit included in the forest health identification model, the current forest image, the forest image texture features, and the target meteorological data are respectively mapped into the semantic space to form a forest image mapping vector, a texture feature mapping vector, and a meteorological data mapping vector. The forest image texture features are characterized by the distribution of the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest image.

[0013] Using the first semantic coding unit included in the forest health identification model, during the process of semantically coding the forest image mapping vector, the texture feature mapping vector is used to guide the semantic coding, forming a forest image guidance vector.

[0014] Using the second semantic coding unit included in the forest health identification model, during the process of semantic coding of the forest image guidance vector, the meteorological data mapping vector is used for semantic coding guidance to form a forest health coding vector.

[0015] In a preferred embodiment of this application, in the aforementioned forest monitoring method based on multidimensional data analysis, the step of using the first semantic coding unit included in the forest health identification model to semantically code the forest image mapping vector, and guiding the semantic coding through the texture feature mapping vector to form a forest image guidance vector, includes:

[0016] Using the deep mining subunit in the first semantic coding unit of the forest health identification model, the forest image mapping vector is semantically encoded to obtain forest image depth vectors of multiple depths;

[0017] Using the encoding guidance subunit in the first semantic encoding unit, the forest image depth vector at each depth is encoded and guided by the texture feature mapping vector to obtain forest image saliency vectors at multiple depths. The forest image depth vector at a later depth is obtained based on the semantic encoding of the forest image saliency vector at a previous depth. The forest image saliency vector is used to characterize the saliency semantic information in the forest image depth vector.

[0018] Based on the forest image saliency vector of the last depth, the forest image guiding vector is obtained.

[0019] In a preferred embodiment of this application, in the aforementioned forest monitoring method based on multidimensional data analysis, the step of semantically encoding the forest image mapping vector using the depth mining subunit in the first semantic coding unit of the forest health identification model to obtain forest image depth vectors of multiple depths includes:

[0020] The forest image mapping vector is loaded into the deep mining subunit of the first semantic coding unit included in the forest health identification model;

[0021] In the semantic encoding of the first depth, the forest image mapping vector is subjected to convolution mining and pooling compression to form the forest image depth vector of the first depth;

[0022] In the second-depth semantic encoding, the forest image saliency vector of the first depth is subjected to convolution mining and pooling compression to form the forest image depth vector of the second depth. The forest image saliency vector of the first depth is obtained based on the encoding guidance of the forest image depth vector of the first depth.

[0023] In a preferred embodiment of this application, in the aforementioned forest monitoring method based on multidimensional data analysis, the step of using the encoding guidance subunit in the first semantic coding unit to encode the forest image depth vector for each depth through the texture feature mapping vector to obtain forest image saliency vectors for multiple depths includes:

[0024] The texture feature mapping vector and the forest image depth vector are loaded into the coding guidance subunit in the first semantic coding unit;

[0025] In the first depth of the encoding guidance, the correlation parameter distribution between the texture feature mapping vector and the representation vector of the first depth and the forest image depth vector of the first depth is analyzed. Based on the correlation parameter distribution, salient semantic information is mined from the forest image depth vector of the first depth to obtain the salient vector of the forest image of the first depth.

[0026] In the second-depth encoding guidance, the distribution of correlation parameters between the texture feature mapping vector and the representation vector of the second depth and the forest image depth vector of the second depth is analyzed. Then, the significance index of the correlation parameter distribution is evaluated. When the result of the significance index evaluation meets the preset conditions, based on the correlation parameter distribution, significant semantic information is mined from the forest image depth vector of the second depth to obtain the forest image significance vector of the second depth. Alternatively, when the result of the significance index evaluation does not meet the preset mining conditions, the forest image depth vector of the second depth is skip-connected to obtain the forest image significance vector of the second depth. The result of the significance index evaluation is used to reflect the distribution differences of each parameter in the correlation parameter distribution.

[0027] In a preferred embodiment of this application, in the aforementioned forest monitoring method based on multidimensional data analysis, the step of using the second semantic coding unit included in the forest health identification model to semantically code the forest image guidance vector, and guiding the semantic coding through the meteorological data mapping vector to form a forest health coding vector, includes:

[0028] The forest image guiding vector and the meteorological data mapping vector are loaded into the second semantic coding unit included in the forest health identification model;

[0029] The forest image guiding vector and the meteorological data mapping vector are spliced ​​together to form a forest global vector. Then, the forest global vector is subjected to multi-head attention processing to form multiple forest attention vectors. The forest global vector is then linearly mapped to form a forest global linear vector, wherein the size of the forest global linear vector matches the number of forest attention vectors.

[0030] Based on the parameters in the global linear vector of the forest, the multiple forest attention vectors are weighted and summed to form a forest health coding vector.

[0031] In a preferred embodiment of this application, in the aforementioned forest monitoring method based on multidimensional data analysis, the step of extracting texture features related to healthy growth from the current forest image to obtain the forest image texture features of the current forest image includes:

[0032] The current forest image is converted to grayscale to form a current forest grayscale image, and the red channel of the current forest image is extracted to form a current forest red channel image;

[0033] Co-occurrence features are extracted from the current forest grayscale image to form a forest grayscale co-occurrence distribution, wherein each parameter in the forest grayscale co-occurrence distribution is used to characterize the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest grayscale image;

[0034] Co-occurrence features are extracted from the current forest red channel map to form a forest red co-occurrence distribution, wherein each parameter in the forest red co-occurrence distribution is used to characterize the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest red channel map;

[0035] Based on the forest gray-scale co-occurrence distribution and the forest red-scale co-occurrence distribution, the forest image texture features of the current forest image are determined.

[0036] In a preferred embodiment of this application, in the aforementioned forest monitoring method based on multidimensional data analysis, the step of semantically decoding the forest health coding vector using the forest health identification model to form target forest health status data includes:

[0037] The forest health encoding vector is fully connected to form a forest health fully connected vector.

[0038] The forest health fully connected vector is activated and output to form target forest health status data, which is used to characterize the health of the target forest.

[0039] In a preferred embodiment of this application, the forest monitoring method based on multidimensional data analysis further includes:

[0040] Acquire sample forest images, sample meteorological data, and corresponding health status labels;

[0041] Extract texture features related to healthy growth from the sample forest image to obtain the sample image texture features of the sample forest image;

[0042] Using the primary forest health identification model, during the semantic encoding of the sample forest images, semantic encoding is guided by the latent semantic information of the sample image texture features and the sample meteorological data to form a sample health encoding vector. The primary forest health identification model is a neural network model to be learned and trained.

[0043] Using the original forest health identification model, the sample health encoding vector is semantically decoded to form sample forest health status data;

[0044] Based on the error between the sample forest health status data and the health status label, the model parameters of the original forest health identification model are updated until the error converges, thus forming the forest health identification model.

[0045] Based on the above, this application also provides a forest monitoring system based on multidimensional data analysis, comprising:

[0046] Memory, used to store computer programs;

[0047] A processor connected to the memory is used to execute the computer program stored in the memory to implement the above-described forest monitoring method based on multidimensional data analysis.

[0048] The forest monitoring method and system based on multidimensional data analysis provided in this application first acquires the current forest image of the target forest and the target meteorological data for the most recent time period; second, it extracts texture features related to healthy growth from the current forest image to obtain the forest image texture features; then, using a forest health identification model, during the semantic encoding process of the current forest image, it guides semantic encoding through the potential semantic information of the forest image texture features and the target meteorological data to form a forest health coding vector; finally, it uses the forest health identification model to perform semantic decoding on the forest health coding vector to form the target forest health status data. Based on the above, on the one hand, semantic encoding through neural network models can capture some potential semantic information, making the basis for forming target forest health status data more sufficient. On the other hand, during the encoding process, semantic encoding is guided by the potential semantic information of forest image texture features and target meteorological data, which are directly related to the forest's growth status. Therefore, the basis for the corresponding semantic encoding guidance is reliable, ensuring the accuracy of the semantic representation of the forest's growth status by the formed forest health encoding vector. This further improves the reliability of the formed target forest health status data, thereby addressing the problem of relatively low reliability of forest health status in existing technologies. Attached Figure Description

[0049] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0050] Figure 1 A structural block diagram of a forest monitoring system based on multidimensional data analysis provided in an embodiment of this application.

[0051] Figure 2This is a flowchart illustrating the forest monitoring method based on multidimensional data analysis provided in an embodiment of this application.

[0052] Figure 3 This is a schematic diagram of the forest grayscale symbiotic distribution provided in an embodiment of this application.

[0053] Figure 4 This is a first schematic diagram illustrating semantic encoding and guidance provided in an embodiment of this application.

[0054] Figure 5 This is a second schematic diagram illustrating semantic encoding and guidance provided in an embodiment of this application. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0056] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0057] like Figure 1 As shown, this application provides a forest monitoring system based on multidimensional data analysis. The forest monitoring system based on multidimensional data analysis may include a memory, a processor, and a forest monitoring device based on multidimensional data analysis.

[0058] Specifically, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The forest monitoring device based on multidimensional data analysis includes at least one software functional module stored in the memory in the form of software or firmware. The processor is used to execute executable computer programs stored in the memory, such as the software functional modules and computer programs included in the forest monitoring device based on multidimensional data analysis, to implement the forest monitoring method based on multidimensional data analysis provided in this application embodiment.

[0059] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0060] 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, or discrete hardware components.

[0061] Optionally, the forest monitoring device based on multidimensional data analysis may include:

[0062] The data acquisition module is used to acquire current forest images of the target forest and target meteorological data for the most recent time period;

[0063] The feature extraction module is used to extract texture features related to healthy growth from the current forest image to obtain the forest image texture features of the current forest image;

[0064] The encoding guidance module is used to utilize the forest health identification model to guide semantic encoding during the semantic encoding of the current forest image by using the potential semantic information of the forest image texture features and the target meteorological data respectively, thereby forming a forest health encoding vector. The forest health identification model is a neural network model formed by learning sample data and label data.

[0065] The semantic decoding module is used to perform semantic decoding on the forest health encoding vector using the forest health identification model to form target forest health status data.

[0066] Understandable. Figure 1 The structure shown is for illustrative purposes only; the forest monitoring system based on multidimensional data analysis may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices (such as image acquisition devices, meteorological databases, etc.).

[0067] Combination Figure 2 This application also provides a forest monitoring method based on multidimensional data analysis, applicable to the aforementioned forest monitoring system based on multidimensional data analysis. The method steps defined in the relevant process of the forest monitoring method based on multidimensional data analysis can be implemented by the forest monitoring system based on multidimensional data analysis. The following will describe... Figure 2 The process shown will be explained.

[0068] Step S110: Obtain the current forest image of the target forest and the target meteorological data for the most recent time period.

[0069] In this embodiment of the application, the forest monitoring system based on multidimensional data analysis can acquire current forest images of the target forest and target meteorological data for the most recent time period. It should be noted that the end time of the most recent time period and the time when the current forest image was captured can be the same or different, as long as the interval between the two is less than a set threshold. This set threshold can be a day, a week, etc., and the time period can be a month, a quarter, a year, etc.

[0070] Step S120: Extract texture features related to healthy growth from the current forest image to obtain the forest image texture features of the current forest image.

[0071] In this embodiment, after acquiring the current forest image, the forest monitoring system based on multidimensional data analysis can extract texture features related to healthy growth from the current forest image to obtain the forest image texture features. It should be noted that extracting image texture features can reveal information such as the structure and surface properties of different regions in the image. For forest health analysis, texture analysis can help capture attributes such as the uniformity and roughness of the forest canopy, which plays an important role in judging the growth status of the forest.

[0072] Step S130: Using the forest health identification model, during the semantic encoding process of the current forest image, semantic encoding is guided by the potential semantic information of the forest image texture features and the target meteorological data to form a forest health encoding vector.

[0073] In this embodiment, after obtaining the forest image texture features and the target meteorological data, the forest monitoring system based on multidimensional data analysis can utilize a forest health identification model. During the semantic encoding of the current forest image, the system guides semantic encoding through the latent semantic information inherent in the forest image texture features and the target meteorological data, forming a forest health encoding vector. In other words, during the semantic encoding of the current forest image, the system is guided by latent semantic information directly related to the growth state, making it more likely to capture semantic information related to the growth state. The forest health identification model is a neural network model formed by learning sample data (such as sample forest images and sample meteorological data) and label data (such as labels representing forest health status, such as whether it is healthy or the degree of health). Furthermore, the learning and training process of the neural network model can refer to relevant existing technologies.

[0074] Step S140: Using the forest health identification model, the forest health encoding vector is semantically decoded to form target forest health status data.

[0075] In this embodiment, after obtaining the forest health coding vector, the forest monitoring system based on multidimensional data analysis can utilize the forest health identification model to perform semantic decoding on the forest health coding vector to form target forest health status data. It should be noted that the forest health identification model may include an encoding part and a decoding part. The encoding part can be used to execute the aforementioned step S130 to achieve corresponding semantic encoding and semantic encoding guidance, and the decoding part can be used to execute the aforementioned step S140 to achieve corresponding semantic decoding.

[0076] Based on the above, on the one hand, semantic encoding through neural network models can capture some potential semantic information, making the basis for forming target forest health status data more sufficient. On the other hand, during the encoding process, semantic encoding is guided by the potential semantic information of forest image texture features and target meteorological data, which are directly related to the forest's growth status. Therefore, the basis for the corresponding semantic encoding guidance is reliable, ensuring the accuracy of the semantic representation of the forest's growth status by the formed forest health encoding vector. This further improves the reliability of the formed target forest health status data, thereby addressing the problem of relatively low reliability of forest health status in existing technologies.

[0077] Firstly, regarding step S110, it should be noted that the specific method for obtaining the current forest image and target meteorological data is not limited and can be selected according to actual needs.

[0078] For example, in an alternative implementation, to ensure the timeliness of forest health monitoring, images of the target forest at the current time (e.g., images captured by image acquisition equipment) and target meteorological data (e.g., data related to sunlight, precipitation, and temperature) over a period of time up to the current moment can be acquired in real time. Alternatively, in another alternative implementation, historical data stored in a database can be retrieved as the current forest image and target meteorological data, allowing for the tracing of forest health status at historical moments.

[0079] Secondly, regarding step S120, it should be noted that the specific method for extracting texture features related to healthy growth from the current forest image is not limited and can be selected according to actual needs.

[0080] For example, in an alternative implementation, in order to improve the sufficiency of texture feature extraction so that the obtained forest image texture features can fully and effectively reflect the features related to healthy growth in the current forest image, the above step S120 may further include steps S121, S122, S123 and S124, as described below.

[0081] Step S121: Perform grayscale processing on the current forest image to form a current forest grayscale image; and extract the red channel from the current forest image to form a current forest red channel image.

[0082] In this embodiment, the current forest image can be grayscale processed (referring to relevant existing technologies, such as converting the current forest image belonging to the RGB three channels into a grayscale image) to form a current forest grayscale image. Furthermore, the red channel of the current forest image can be extracted (e.g., extracting the R channel data from the current forest image belonging to the RGB three channels) to form a current forest red channel image. It should be noted that there is a close relationship between red light reflectance and chlorophyll content, which makes it possible to have a potential mapping relationship for the characterization of photosynthetic capacity and plant health status. Therefore, by extracting red channel information, it is convenient to capture potential semantic information related to plant health status in subsequent steps.

[0083] Step S122: Extract symbiotic features from the current forest grayscale map to form a forest grayscale symbiotic distribution.

[0084] In this embodiment, after obtaining the current forest grayscale image, co-occurrence features can be extracted from the current forest grayscale image to form a forest grayscale co-occurrence distribution. Each parameter in the forest grayscale co-occurrence distribution characterizes the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest grayscale image. For example, the forest grayscale co-occurrence distribution can be represented by a grayscale co-occurrence matrix (GLCM), such as... Figure 3 As shown, P(i,j) represents the co-occurrence probability (or frequency) of pixel pairs with gray values ​​i and j in an image at a specific direction and distance. For example, the co-occurrence of gray values ​​of adjacent pixels in the horizontal direction, where i belongs to 0-255 and j belongs to 0-255. Specifically, P(i,j) represents the probability value calculated by dividing the number of times a pixel with gray value i appears together with its neighboring pixel with gray value j (according to a predetermined distance and direction) by the total number of times all pixel pairs appear.

[0085] Step S123: Extract symbiotic features from the current forest red channel map to form a forest red symbiotic distribution.

[0086] In this embodiment, after obtaining the current forest red channel map, co-occurrence features can be extracted from the current forest red channel map (as described above) to form a forest red co-occurrence distribution. Each parameter in the forest red co-occurrence distribution characterizes the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest red channel map.

[0087] Step S124: Based on the forest gray-scale co-occurrence distribution and the forest red co-occurrence distribution, determine the forest image texture features of the current forest image.

[0088] In this embodiment, after obtaining the forest gray-scale co-occurrence distribution and the forest red-scale co-occurrence distribution, the forest image texture features of the current forest image can be determined based on these distributions. For example, parameters such as contrast, homogeneity, energy, and correlation (refer to relevant prior art) can be extracted from the forest gray-scale co-occurrence distribution and the forest red-scale co-occurrence distribution to characterize the forest image texture features. Alternatively, the forest gray-scale co-occurrence distribution and the forest red-scale co-occurrence distribution can be directly used as the forest image texture features. In this way, in subsequent processing, the powerful processing capabilities of the neural network model can capture more potential semantic information related to growth, improving the reliability of monitoring.

[0089] Thirdly, regarding step S130, it should be noted that the specific method for forming the forest health coding vector is not restricted and can be selected according to actual needs.

[0090] For example, in an alternative implementation, in order to ensure the semantic representation accuracy of the formed forest health coding vector, cascaded semantic coding guidance can be performed during the semantic coding process so that the three potential semantic information can be fully integrated. Based on this, the above step S130 can further include steps S131, S132 and S133, as described below.

[0091] Step S131: Using the semantic space mapping unit included in the forest health identification model, the current forest image, the forest image texture features, and the target meteorological data are mapped into the semantic space respectively, forming a forest image mapping vector, a texture feature mapping vector, and a meteorological data mapping vector.

[0092] In the embodiments of this application, combined with Figure 4 The semantic space mapping unit included in the forest health identification model can be used to map the current forest image, the forest image texture features, and the target meteorological data into the semantic space, forming a forest image mapping vector, a texture feature mapping vector, and a meteorological data mapping vector, respectively. The forest image texture features are characterized by the distribution of the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest image, such as... Figure 3The matrix shown, when including the forest gray-level co-occurrence distribution and the forest red co-occurrence distribution, can be concatenated with the forest gray-level co-occurrence distribution and then semantic space mapping can be performed on the concatenated matrix. Further, semantic space mapping can be implemented using convolutional units. For example, the semantic space mapping unit can include a first convolutional unit, a second convolutional unit, and a third convolutional unit. The first convolutional unit can be used to perform convolution processing on the current forest image to obtain a forest image mapping vector; the second convolutional unit can be used to perform convolution processing on the forest image texture features to obtain a texture feature mapping vector; and the third convolutional unit can be used to perform convolution processing on the target meteorological data to obtain a meteorological data mapping vector. Furthermore, it should be noted that the convolution processing of the first convolutional unit can be three-dimensional convolution or two-dimensional convolution (i.e., unfolding the data of each channel in the current forest image before convolution), the convolution processing of the second convolutional unit can be two-dimensional convolution, and the convolution processing of the third convolutional unit can be one-dimensional convolution (e.g., including only one dimension of time-series data such as temperature, precipitation, and illumination) or two-dimensional convolution (including at least two dimensions of time-series data such as temperature, precipitation, and illumination). It should also be further noted that the first, second, and third convolutional units can all include convolutional layers, pooling layers, and fully connected layers. Through corresponding convolution, pooling, and fully connected processing, the resulting forest image mapping vector, texture feature mapping vector, and meteorological data mapping vector have the same size.

[0093] Step S132: Using the first semantic coding unit included in the forest health identification model, during the process of semantic coding of the forest image mapping vector, semantic coding guidance is performed through the texture feature mapping vector to form a forest image guidance vector.

[0094] In this embodiment, after obtaining the forest image mapping vector and the texture feature mapping vector, the first semantic encoding unit included in the forest health recognition model can be used to guide the semantic encoding of the forest image mapping vector through the texture feature mapping vector, forming a forest image guidance vector. That is, since both the forest image mapping vector and the texture feature mapping vector are actually latent semantic information mined from the current forest image, the forest image mapping vector tends to represent global information in the current forest image, while the texture feature mapping vector tends to represent local information related to healthy growth in the current forest image. However, because they are in similar semantic spaces, semantic encoding guidance can be performed first, resulting in relatively higher accuracy. This allows the forest image guidance vector to represent global information while also emphasizing local information related to healthy growth.

[0095] Step S133: Using the second semantic coding unit included in the forest health identification model, during the process of semantic coding of the forest image guidance vector, semantic coding guidance is performed through the meteorological data mapping vector to form a forest health coding vector.

[0096] In this embodiment, after obtaining the forest image guidance vector and the meteorological data mapping vector, the second semantic coding unit included in the forest health identification model can be used to guide semantic coding through the meteorological data mapping vector during the semantic coding process of the forest image guidance vector, thus forming a forest health coding vector. Based on this, since the semantic coding guidance from the first semantic coding unit allows the semantic information in the forest image guidance vector to be related to healthy growth to a certain extent, further semantic coding guidance when the meteorological data mapping vector can also represent semantic information related to healthy growth can achieve better fusion of semantic information from two dimensions. This means guiding the extraction of more semantic information related to healthy growth from the forest image guidance vector, allowing the formed forest health coding vector to focus more on representing semantic information related to healthy growth.

[0097] It is understood that the specific method of semantic encoding guidance through the texture feature mapping vector in step S132 above is not limited. For example, in an alternative implementation, in order to enable the formed forest image guidance vector to represent some high-level, abstract semantic information related to healthy growth while focusing on representing semantic information related to healthy growth, so as to improve the semantic representation capability, step S132 above may further include steps S132a, S132b and S132c, as described below.

[0098] Step S132a: Using the depth mining subunit in the first semantic coding unit of the forest health identification model, the forest image mapping vector is semantically encoded to obtain forest image depth vectors of multiple depths.

[0099] In this embodiment, the depth mining subunit within the first semantic coding unit of the forest health identification model can be used to semantically encode the forest image mapping vector, obtaining forest image depth vectors at multiple depths. In other words, by mining the potential semantic information present in the forest image mapping vector at multiple different depths, the richness of semantic information can be significantly improved, thereby further enhancing the semantic representation capability.

[0100] Step S132b: Using the encoding guidance subunit in the first semantic coding unit, the forest image depth vector at each depth is encoded and guided by the texture feature mapping vector to obtain forest image saliency vectors at multiple depths.

[0101] In this embodiment, after obtaining the forest image depth vectors of multiple depths, the encoding guidance subunit in the first semantic encoding unit can be used to guide the encoding of each depth's forest image depth vector through the texture feature mapping vector, thereby obtaining forest image saliency vectors of multiple depths. The forest image depth vector of a later depth is obtained based on the semantic encoding of the forest image saliency vector of the previous depth, and the forest image saliency vector is used to represent the salient semantic information in the forest image depth vectors. Furthermore, during the encoding guidance process, by mining and capturing salient semantic information, the ability of the resulting forest image saliency vectors to represent important semantic information can be further improved.

[0102] Step S132c: Based on the forest image saliency vector of the last depth, obtain the forest image guiding vector.

[0103] In this embodiment, after obtaining forest image saliency vectors at multiple depths, a forest image guiding vector can be obtained based on the forest image saliency vector at the last depth. For example, the forest image saliency vector at the last depth can be used as the forest image guiding vector.

[0104] It is understood that the specific method of semantic encoding of the forest image mapping vector in step S132a above is not limited. For example, in an alternative implementation, in order to fully mine more complex high-level semantic information in the process of semantic encoding, step S132a above may further include steps a1, a2 and a3, as described below.

[0105] Step a1: Load the forest image mapping vector into the deep mining subunit of the first semantic coding unit included in the forest health identification model.

[0106] In this embodiment of the application, the forest image mapping vector can be loaded into the deep mining subunit of the first semantic coding unit included in the forest health identification model.

[0107] Step a2: In the semantic encoding of the first depth, the forest image mapping vector is subjected to convolution mining and pooling compression to form the forest image depth vector of the first depth.

[0108] In this embodiment of the application, after loading the forest image mapping vector into the depth mining subunit, corresponding depth mining can be performed in the depth mining subunit, combined with... Figure 5 In the first depth semantic encoding, the forest image mapping vector can be subjected to convolution mining and pooling compression (to reduce the vector size and achieve downsampling while capturing important high-level, abstract semantic information) to form the first depth forest image depth vector.

[0109] In step a3, in the semantic encoding of the second depth, the forest image saliency vector of the first depth is subjected to convolution mining and pooling compression to form the forest image depth vector of the second depth.

[0110] In this embodiment, after obtaining the forest image depth vector of the first depth, the saliency vector of the forest image of the first depth can be subjected to convolution mining and pooling compression in the semantic encoding of the second depth to form the forest image depth vector of the second depth. The saliency vector of the forest image of the first depth is obtained based on the encoding guidance of the forest image depth vector of the first depth. That is, after obtaining the forest image depth vector of the first depth, the saliency semantic information can be mined from it through the encoding guidance subunit to obtain the forest image saliency vector of the first depth. Then, the semantic encoding of the second depth is performed to obtain the forest image depth vector of the second depth.

[0111] It should be noted that in the semantic encoding of the third depth, the forest image saliency vector of the second depth can be subjected to convolution mining and pooling compression to form the forest image depth vector of the third depth. In the semantic encoding of the fourth depth, the forest image saliency vector of the third depth can be subjected to convolution mining and pooling compression to form the forest image depth vector of the fourth depth. This process continues, and so on, to obtain the forest image depth vector for each depth. The specific depth can be configured according to actual needs. For example, a shallow depth may be difficult to effectively capture complex semantic relationships; a large depth may lead to excessive computation and semantic distortion.

[0112] It is understood that in step S132b above, the specific method of encoding the forest image depth vector for each depth through the texture feature mapping vector is not limited. For example, in an alternative implementation, in order to fully mine higher-level semantic information with greater accuracy during the encoding guidance process, step S132b above may further include steps b1, b2 and b3, as described below.

[0113] Step b1: Load the texture feature mapping vector and the forest image depth vector into the coding guidance subunit in the first semantic coding unit.

[0114] In this embodiment of the application, the texture feature mapping vector and the forest image depth vector can be loaded into the coding guidance subunit in the first semantic coding unit, so that subsequent coding guidance can be performed in the coding guidance subunit.

[0115] Step b2: In the first depth encoding guidance, the correlation parameter distribution between the texture feature mapping vector and the representation vector of the first depth and the forest image depth vector of the first depth is analyzed. Based on the correlation parameter distribution, salient semantic information is mined from the forest image depth vector of the first depth to obtain the forest image salient vector of the first depth.

[0116] In the embodiments of this application, further combined with Figure 5Corresponding to the aforementioned multiple depths of semantic encoding, there are also multiple depths of encoding guidance, allowing encoding guidance to be performed separately at multiple depths, thereby further improving the accuracy of encoding guidance. Specifically, in the encoding guidance of the first depth, the distribution of correlation parameters between the texture feature mapping vector and the representation vector of the first depth (for example, the texture feature mapping vector can be downsampled to form the representation vector of the first depth, the size of which can be the same as the size of the forest image depth vector of the first depth) and the forest image depth vector of the first depth can be analyzed. Based on this distribution of correlation parameters, salient semantic information is mined from the forest image depth vector of the first depth to obtain the salient vector of the forest image of the first depth. For example, the distribution of correlation parameters can be the distribution of attention parameters, and mining salient semantic information can refer to performing a weighted summation operation on the forest image depth vector of the first depth based on the attention parameter distribution to obtain the salient vector of the forest image of the first depth. That is to say, the semantic information with correlation between two vectors can be regarded as important semantic information, that is, salient semantic information.

[0117] Step b3: In the second depth encoding guidance, the distribution of correlation parameters between the texture feature mapping vector and the representation vector of the second depth forest image and the second depth forest image depth vector is analyzed. The significance index of the correlation parameter distribution is evaluated. When the result of the significance index evaluation meets the preset conditions, significant semantic information is mined from the forest image depth vector of the second depth based on the correlation parameter distribution to obtain the forest image significance vector of the second depth. Alternatively, when the result of the significance index evaluation does not meet the preset mining conditions, the forest image depth vector of the second depth forest image is skipped to obtain the forest image significance vector of the second depth forest image.

[0118] In this embodiment, after obtaining the forest image saliency vector at the first depth, the forest image saliency vector at the first depth can be subjected to convolution mining and pooling compression in the semantic encoding at the second depth to form the forest image depth vector at the second depth. Then, in the encoding guidance at the second depth, the distribution of correlation parameters between the texture feature mapping vector and the representation vector at the second depth (for example, the representation vector at the first depth can be downsampled to form the representation vector at the second depth, and the size of the representation vector can be the same as the size of the forest image depth vector at the second depth) and the forest image depth vector at the second depth is analyzed. The saliency index of the correlation parameter distribution is evaluated, and when the result of the saliency index evaluation meets the preset conditions, saliency semantic information is mined from the forest image depth vector at the second depth based on the correlation parameter distribution to obtain the forest image saliency vector at the second depth. Alternatively, when the result of the saliency index evaluation does not meet the preset mining conditions, the forest image depth vector at the second depth is skipped to obtain the forest image saliency vector at the second depth, such as directly using the forest image depth vector at the second depth as the forest image saliency vector at the second depth. The results of the significance index evaluation are used to reflect the distribution differences of each parameter in the distribution of the associated parameters (e.g., it can be characterized by the dispersion of each parameter; the greater the dispersion, the greater the distribution difference between the parameters, meaning that significant and non-significant information can be distinguished. Therefore, it can be used to meet preset mining conditions, such as determining that the dispersion is greater than a set threshold; conversely, the smaller the dispersion, the smaller the distribution difference between the parameters, meaning that significant and non-significant information cannot be distinguished. Therefore, it can be used to determine that the preset mining conditions are not met, such as determining that the dispersion is not greater than a set threshold. The set threshold can be configured according to actual needs, such as values ​​like 0.5, 0.7, and 1). It should be noted that by setting the significance index evaluation, a balance can be achieved between mining significant semantic information and computational efficiency. That is, when the effect of significance mining is poor, it can be directly output to improve computational efficiency and reduce computational costs; when the effect of significance mining is good, significant semantic information can be mined through weighted summation calculation to improve the reliability of semantic mining.

[0119] Furthermore, after obtaining the forest image saliency vector at the second depth, convolution mining and pooling compression can be performed on the second depth forest image saliency vector during the semantic encoding at the third depth to form the third depth forest image depth vector. Then, in the encoding guidance at the third depth, the distribution of correlation parameters between the texture feature mapping vector and the representation vector at the third depth (for example, the representation vector at the second depth can be downsampled to form the representation vector at the third depth, and the size of the representation vector can be the same as the size of the third depth forest image depth vector) and the third depth forest image depth vector is analyzed. A saliency index is then evaluated on this correlation parameter distribution. If the saliency index evaluation result meets preset conditions, saliency-related semantic information is mined from the third depth forest image depth vector based on this correlation parameter distribution to obtain the third depth forest image saliency vector. Alternatively, if the saliency index evaluation result does not meet the preset mining conditions, the third depth forest image depth vector is skip-connected to obtain the third depth forest image saliency vector. This process can be repeated to obtain the forest image saliency vector at each depth.

[0120] It is understood that the specific method of forming the forest health coding vector in step S133 above is not limited. For example, in an alternative implementation, in order to enable the meteorological data mapping vector to fully guide the semantic coding of the forest image guiding vector, that is, to achieve full fusion of semantic information in two dimensions, thereby improving the semantic representation capability of the formed forest health coding vector, step S133 above may further include steps S133a, S133b and S133c, as described below.

[0121] Step S133a: Load the forest image guiding vector and the meteorological data mapping vector into the second semantic coding unit included in the forest health identification model.

[0122] In this embodiment of the application, the forest image guiding vector and the meteorological data mapping vector can be loaded into the second semantic coding unit included in the forest health identification model, so that subsequent processing can be performed in the second semantic coding unit.

[0123] Step S133b: Concatenate the forest image guiding vector and the meteorological data mapping vector to form a forest global vector; perform multi-head attention processing on the forest global vector to form multiple forest attention vectors; and perform linear mapping on the forest global vector to form a forest global linear vector.

[0124] In this embodiment, the forest image guiding vector and the meteorological data mapping vector can be concatenated in the second semantic encoding unit to form a forest global vector. Furthermore, multi-head attention processing is performed on the forest global vector to form multiple forest attention vectors, and a linear mapping is applied to the forest global vector to form a forest global linear vector. The size of the forest global linear vector matches the number of forest attention vectors; for example, the number of parameters included in the forest global linear vector is equal to the number of forest attention vectors, and also equal to the number of attention heads in the multi-head attention processing. That is, the multiple forest attention vectors obtained through multiple attention processing can each focus on the attention and representation of different important semantic information. However, different semantic information may have different degrees of importance in actual application reasoning. Therefore, a linear mapping, such as convolution and fully connected processing, can be applied to the forest global vector to form a weight distribution that can give different degrees of attention to the semantic information focused on by each attention, i.e., the forest global linear vector.

[0125] Step S133c: Based on the parameters in the global linear vector of the forest, perform a weighted summation calculation on the multiple forest attention vectors to form a forest health coding vector.

[0126] In this embodiment, after obtaining the global linear vector of the forest and the plurality of forest attention vectors, a weighted summation can be performed on the plurality of forest attention vectors based on the parameters in the global linear vector of the forest to form a forest health coding vector. For example, if the global linear vector of the forest is (a1, a2, a3, a4), and the plurality of forest attention vectors are B1, B2, B3, and B4, then the forest health coding vector = a1*B1 + a2*B2 + a3*B3 + a4*B4. This can further improve the semantic representation accuracy of the formed forest health coding vector.

[0127] Fourthly, regarding step S140, it should be noted that the specific method for semantically decoding the forest health encoding vector is not limited and can be selected according to actual needs.

[0128] For example, in an alternative implementation, in order to improve the efficiency of semantic decoding, step S140 above may further include steps S141 and S142, as described below.

[0129] Step S141: Perform a fully connected processing on the forest health encoding vector to form a fully connected forest health vector.

[0130] In this embodiment, the forest health encoding vector can be fully connected to form a forest health fully connected vector. The size of this fully connected vector can be configured differently based on various monitoring needs. For example, when multiple classification outputs (such as "healthy," "healthy," "unhealthy," "very unhealthy," etc.) are required, the size of the fully connected vector can be determined based on the number of specific output types, where one parameter in the fully connected vector corresponds to one output type. Alternatively, when linear regression is required to determine the degree of health, such as when the output is a value between 0 and 1, the size of the fully connected vector can be 1*1, including only one parameter.

[0131] Step S142: Activate and output the fully connected vector of forest health to form target forest health status data.

[0132] In this embodiment, after obtaining the forest health fully connected vector, the forest health fully connected vector can be activated and output to form target forest health status data. The target forest health status data is used to characterize the health of the target forest. It should be noted that when multi-class classification output is required, activation output can be achieved using classification functions such as softmax to obtain the target forest health status data; when linear regression output is required, activation output can be achieved using activation functions such as sigmoid to obtain the target forest health status data.

[0133] Fifthly, regarding the forest monitoring method based on multidimensional data analysis, it should also be noted that, in an alternative implementation, it may further include a step of training to form the forest health identification model. Specifically, the forest monitoring method based on multidimensional data analysis may further include steps S150, S160, S170, S180, and S190, as described below.

[0134] Step S150: Obtain sample forest images, sample meteorological data, and corresponding health status labels.

[0135] In this embodiment of the application, sample forest images (such as the target forest image in step S110), sample meteorological data (such as the target meteorological data in step S110) and corresponding health status labels (i.e., characterizing the health status of the forest corresponding to the sample forest image and the sample meteorological data, which can be identified based on other neural network models or formed by manual annotation) can be obtained.

[0136] Step S160: Extract texture features related to healthy growth from the sample forest image to obtain the sample image texture features of the sample forest image.

[0137] In this embodiment of the application, texture features related to healthy growth can be extracted from the sample forest image to obtain the sample image texture features of the sample forest image. Please refer to the previous explanation of step S120.

[0138] Step S170: Using the primary forest health identification model, during the semantic encoding process of the sample forest image, semantic encoding is guided by the potential semantic information of the sample image texture features and the sample meteorological data to form a sample health encoding vector.

[0139] In this embodiment, a primary forest health identification model can be used to guide semantic encoding of the sample forest images by leveraging the latent semantic information inherent in the texture features of the sample images and the meteorological data, thereby forming a sample health encoding vector. Refer to the preceding explanation of step S130. The primary forest health identification model is a neural network model to be trained.

[0140] Step S180: Using the original forest health identification model, the sample health encoding vector is semantically decoded to form sample forest health status data.

[0141] In this embodiment of the application, the original forest health identification model can be used to semantically decode the sample health encoding vector to form sample forest health status data. Please refer to the relevant explanation of step S140 above.

[0142] Step S190: Based on the error between the sample forest health status data and the health status label, update the model parameters of the original forest health identification model until the error converges, thus forming the forest health identification model.

[0143] In this embodiment of the application, the model parameters of the original forest health identification model can be updated based on the error between the sample forest health status data and the health status label (the error can be calculated using any relevant existing error or loss calculation method) until the error converges (e.g., the error is less than the error threshold, or the decrease in error is less than the magnitude threshold, or the number of updates exceeds the number threshold, etc.), thus forming the forest health identification model.

[0144] In summary, the forest monitoring method and system based on multidimensional data analysis provided in this application firstly acquires the current forest image of the target forest and the target meteorological data for the most recent time period; secondly, it extracts texture features related to healthy growth from the current forest image to obtain the forest image texture features; then, using a forest health identification model, during the semantic encoding process of the current forest image, it guides semantic encoding through the potential semantic information of the forest image texture features and the target meteorological data to form a forest health coding vector; finally, it uses the forest health identification model to semantically decode the forest health coding vector to form the target forest health status data. Based on the above, on the one hand, semantic encoding through neural network models can capture some potential semantic information, making the basis for forming target forest health status data more sufficient. On the other hand, during the encoding process, semantic encoding is guided by the potential semantic information of forest image texture features and target meteorological data, which are directly related to the forest's growth status. Therefore, the basis for the corresponding semantic encoding guidance is reliable, ensuring the accuracy of the semantic representation of the forest's growth status by the formed forest health encoding vector. This further improves the reliability of the formed target forest health status data, thereby addressing the problem of relatively low reliability of forest health status in existing technologies.

[0145] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a 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 than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0146] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0147] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0148] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A forest monitoring method based on multidimensional data analysis, characterized in that, include: Acquire current forest images of the target forest and target meteorological data for the most recent time period; Extract texture features related to healthy growth from the current forest image to obtain the forest image texture features of the current forest image; Using the semantic space mapping unit included in the forest health identification model, the current forest image, the forest image texture features, and the target meteorological data are mapped into the semantic space, forming a forest image mapping vector, a texture feature mapping vector, and a meteorological data mapping vector, respectively. The forest image texture features are characterized by the distribution of the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest image. Using the depth mining subunit in the first semantic encoding unit of the forest health identification model, the forest image mapping vector is semantically encoded to obtain multiple depths of forest image depth vectors. The texture feature mapping vector and the forest image depth vector are loaded into the encoding guidance subunit in the first semantic encoding unit. In the encoding guidance of the first depth, the distribution of correlation parameters between the texture feature mapping vector and the representation vector of the first depth and the forest image depth vector of the first depth are analyzed. Based on this correlation parameter distribution, salient semantic information is mined from the forest image depth vector of the first depth to obtain the forest image saliency vector of the first depth. In the encoding guidance of the second depth... The method analyzes the distribution of correlation parameters between the texture feature mapping vector at the second depth and the forest image depth vector at the second depth. It then evaluates the significance of this correlation parameter distribution. If the evaluation result meets a preset condition, significant semantic information is extracted from the forest image depth vector at the second depth based on this correlation parameter distribution, resulting in a significant vector for the forest image at the second depth. If the evaluation result does not meet the preset mining condition, the forest image depth vector at the second depth is directly output to obtain a significant vector for the forest image at the second depth. The evaluation result reflects the distribution differences of each parameter in the correlation parameter distribution. Based on the significant vector of the forest image at the last depth, a forest image guidance vector is obtained. Using the second semantic encoding unit included in the forest health identification model, semantic encoding guidance is performed through the meteorological data mapping vector during the semantic encoding process of the forest image guidance vector, forming a forest health encoding vector. The forest health identification model is a neural network model formed by learning sample data and label data. Using the forest health identification model, the forest health encoding vector is semantically decoded to form target forest health status data.

2. The forest monitoring method based on multidimensional data analysis according to claim 1, characterized in that, The step of semantically encoding the forest image mapping vector using the depth mining subunit in the first semantic coding unit of the forest health identification model to obtain forest image depth vectors of multiple depths includes: The forest image mapping vector is loaded into the deep mining subunit of the first semantic coding unit included in the forest health identification model; In the semantic encoding of the first depth, the forest image mapping vector is subjected to convolution mining and pooling compression to form the forest image depth vector of the first depth; In the second-depth semantic encoding, the forest image saliency vector of the first depth is subjected to convolution mining and pooling compression to form the forest image depth vector of the second depth. The forest image saliency vector of the first depth is obtained based on the encoding guidance of the forest image depth vector of the first depth.

3. The forest monitoring method based on multidimensional data analysis according to claim 1, characterized in that, The step of using the second semantic coding unit included in the forest health identification model to semantically code the forest image guidance vector, and guiding the semantic coding through the meteorological data mapping vector to form a forest health coding vector, includes: The forest image guiding vector and the meteorological data mapping vector are loaded into the second semantic coding unit included in the forest health identification model; The forest image guiding vector and the meteorological data mapping vector are spliced ​​together to form a forest global vector. Then, the forest global vector is subjected to multi-head attention processing to form multiple forest attention vectors. The forest global vector is then linearly mapped to form a forest global linear vector, wherein the size of the forest global linear vector matches the number of forest attention vectors. Based on the parameters in the global linear vector of the forest, the multiple forest attention vectors are weighted and summed to form a forest health coding vector.

4. The forest monitoring method based on multidimensional data analysis according to any one of claims 1-3, characterized in that, The step of extracting texture features related to healthy growth from the current forest image to obtain the forest image texture features of the current forest image includes: The current forest image is converted to grayscale to form a current forest grayscale image, and the red channel of the current forest image is extracted to form a current forest red channel image; Co-occurrence features are extracted from the current forest grayscale image to form a forest grayscale co-occurrence distribution, wherein each parameter in the forest grayscale co-occurrence distribution is used to characterize the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest grayscale image; Co-occurrence features are extracted from the current forest red channel map to form a forest red co-occurrence distribution, wherein each parameter in the forest red co-occurrence distribution is used to characterize the probability of co-occurrence of pixel value combinations corresponding to two pixels in the current forest red channel map; Based on the forest gray-scale co-occurrence distribution and the forest red-scale co-occurrence distribution, the forest image texture features of the current forest image are determined.

5. The forest monitoring method based on multidimensional data analysis according to any one of claims 1-3, characterized in that, The step of using the forest health identification model to semantically decode the forest health encoding vector to form target forest health status data includes: The forest health encoding vector is fully connected to form a forest health fully connected vector. The forest health fully connected vector is activated and output to form target forest health status data, which is used to characterize the health of the target forest.

6. The forest monitoring method based on multidimensional data analysis according to any one of claims 1-3, characterized in that, The forest monitoring method based on multidimensional data analysis also includes: Acquire sample forest images, sample meteorological data, and corresponding health status labels; Extract texture features related to healthy growth from the sample forest image to obtain the sample image texture features of the sample forest image; Using the primary forest health identification model, during the semantic encoding process of the sample forest images, semantic encoding is guided by the latent semantic information of the sample image texture features and the sample meteorological data to form a sample health encoding vector. The primary forest health identification model is a neural network model to be learned and trained. Using the original forest health identification model, the sample health encoding vector is semantically decoded to form sample forest health status data; Based on the error between the sample forest health status data and the health status label, the model parameters of the original forest health identification model are updated until the error converges, thus forming the forest health identification model.

7. A forest monitoring system based on multidimensional data analysis, characterized in that, include: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the forest monitoring method based on multidimensional data analysis as described in any one of claims 1-6.

Images