Water and soil loss automatic monitoring and early warning method based on transfer learning

Abstract

The invention provides a water and soil loss automatic monitoring and early warning method based on transfer learning, and relates to the technical field of artificial intelligence. The method comprises the following steps: firstly, mining a project monitoring data vector and a project prediction data vector according to water and soil loss monitoring data; secondly, performing A-stage interference applying operation on a target remote sensing image of the target region based on the target interference vector to form a region remote sensing image vector; then, on the basis of a project monitoring data vector and a project prediction data vector, performing A-stage interference removal operation on the regional remote sensing image vector to form a multi-modal water and soil loss vector; and finally, performing water and soil loss prediction operation based on the multi-modal water and soil loss vector to form water and soil loss prediction data, and performing water and soil loss early warning operation based on the water and soil loss prediction data. On the basis of the content, the problem that in the prior art, the reliability of automatic monitoring and early warning of water and soil loss is relatively low can be solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an automated soil and water loss monitoring and early warning method based on transfer learning. Background Art

[0002] Soil erosion is a widespread environmental problem worldwide, particularly in mountainous, hilly, and arid regions, where it severely impacts agricultural production, the ecological environment, and water resource management. Soil erosion not only reduces soil fertility but can also trigger a range of serious natural disasters, such as floods, mudslides, and landslides. Therefore, early monitoring and early warning of soil erosion have become crucial research topics in environmental management. Existing soil erosion monitoring methods typically rely on manual measurements, remote sensing data analysis, and the application of empirical formulas. While these methods can provide a certain degree of information on soil erosion, they suffer from long monitoring cycles, complex data processing, and poor real-time performance. For example, remote sensing technology can be affected by factors such as cloud cover and climate change when monitoring large areas, resulting in inaccurate image information and thus undermining the reliability of monitoring results. Existing monitoring methods often only provide a single data dimension, failing to integrate and efficiently analyze multimodal data. Consequently, the reliability of automated soil erosion monitoring and early warning systems in existing technologies remains relatively low. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide an automatic monitoring and early warning method for soil and water loss based on transfer learning, so as to improve the problem of relatively low reliability of automatic monitoring and early warning of soil and water loss in the prior art.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] An automated soil and water loss monitoring and early warning method based on transfer learning, comprising:

[0006] Determining soil and water loss monitoring data for a target area, and mining, based on the soil and water loss monitoring data, project monitoring data vectors for a first number of measured items in the soil and water loss monitoring data, and mining project prediction data vectors for a second number of predicted items, wherein the first number of measured items includes at least one of terrain, soil, and vegetation, and the second number of predicted items includes at least one of predicted precipitation, predicted wind speed, and predicted wind direction;

[0007] Determining a target interference vector, and, based on the target interference vector, performing A stages of interference application operations on the target remote sensing image of the target area to form a regional remote sensing image vector;

[0008] Based on the project monitoring data vectors of the first number of measured projects and the project prediction data vectors of the second number of predicted projects, performing A stages of interference removal operations on the regional remote sensing image vector to form a multimodal soil and water loss vector corresponding to the a-th stage of interference removal operations;

[0009] A soil and water loss prediction operation is performed based on the multimodal soil and water loss vector corresponding to the interference removal operation in the ath stage to form soil and water loss prediction data corresponding to the target area; and, based on the soil and water loss prediction data, a soil and water loss early warning operation is performed on the target area.

[0010] In a preferred embodiment of the present invention, in the above-mentioned automated soil and water loss monitoring and early warning method based on transfer learning, the step of performing A stages of interference removal operations on the regional remote sensing image vector based on the project monitoring data vectors of the first number of measured projects and the project prediction data vectors of the second number of predicted projects to form a multimodal soil and water loss vector corresponding to the interference removal operation of the ath stage includes:

[0011] Performing A stages of interference removal operations on the regional remote sensing image vector to form a multimodal soil and water loss vector corresponding to the interference removal operation of the a-th stage, wherein, for the interference removal operation of the a-th stage among the A-stage interference removal operations, if a is equal to 1, then the input of the interference removal operation of the a-th stage is the regional remote sensing image vector; if a is greater than 1 and less than or equal to A, then the input of the interference removal operation of the a-th stage is the multimodal soil and water loss vector of the interference removal operation of the a-1-th stage;

[0012] In the interference removal operation of the a-th stage, B interference removal units are used to perform an interference removal sub-operation, wherein, for the b-th interference removal unit among the B interference removal units, if b is equal to 1, then the input of the b-th interference removal unit is the input of the interference removal operation of the a-th stage; if b is greater than 1 and less than or equal to B, then the input of the b-th interference removal unit includes the multimodal soil and water loss vector of the b-1-th interference removal unit; if b is equal to B, then the multimodal soil and water loss vector of the interference removal operation of the a-th stage is the multimodal soil and water loss vector of the b-th interference removal unit;

[0013] For the b-th interference removal unit, an interference removal sub-operation is performed on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1-th interference removal unit and the project monitoring data vectors of the first number of measured projects to form a first local soil and water loss vector; and an interference removal sub-operation is performed on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1-th interference removal unit and the project prediction data vectors of the second number of prediction projects to form a second local soil and water loss vector;

[0014] The first local soil and water loss vector and the second local soil and water loss vector are aggregated to form a multimodal soil and water loss vector of the bth interference removal unit.

[0015] In a preferred embodiment of the present invention, in the above-mentioned automated soil and water loss monitoring and early warning method based on transfer learning, the steps of performing an interference removal sub-operation on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1th interference removal unit and the project monitoring data vectors of the first number of measured projects to form a first local soil and water loss vector, and performing an interference removal sub-operation on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1th interference removal unit and the project prediction data vectors of the second number of prediction projects to form a second local soil and water loss vector include:

[0016] For the b-th interference removal unit, using the b-th interference removal unit, compressing the multimodal soil and water loss vector of the b-1-th interference removal unit to form a corresponding compressed soil and water loss vector;

[0017] Performing internal association mining on the semantic information in the compressed soil and water loss vector to form a corresponding associated soil and water loss vector;

[0018] performing an interference removal sub-operation based on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector;

[0019] An interference removal sub-operation is performed respectively according to the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects to form second local soil and water loss vectors corresponding to the second number of prediction projects.

[0020] In a preferred embodiment of the present invention, in the above-mentioned automated soil and water loss monitoring and early warning method based on transfer learning, the step of performing an interference removal sub-operation based on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector includes:

[0021] External association mining of semantic information is performed on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector.

[0022] In a preferred embodiment of the present invention, in the above-mentioned automated soil and water loss monitoring and early warning method based on transfer learning, the step of performing external association mining of semantic information on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector includes:

[0023] Performing external correlation mining of semantic information on the associated soil and water loss vector and the project monitoring data vector of each of the first number of measured projects, to form a project external correlation vector corresponding to the project monitoring data vector of each measured project;

[0024] The project external correlation vectors corresponding to the project monitoring data vectors of the first number of measured projects are aggregated to form corresponding first local soil and water loss vectors.

[0025] In a preferred embodiment of the present invention, in the above-mentioned automated soil and water loss monitoring and early warning method based on transfer learning, the step of performing external association mining of semantic information on the associated soil and water loss vector and the project monitoring data vector of each of the first number of measured projects to form a project external association vector corresponding to the project monitoring data vector of each measured project includes:

[0026] For each of the project monitoring data vectors of the first number of measured projects, mapping the project monitoring data vector according to a first mapping parameter formed by learning the sample data to form a corresponding first mapping parameter distribution and a second mapping parameter distribution, and mapping the associated soil and water loss vector to form a corresponding third mapping parameter distribution;

[0027] performing a dot product operation on the first mapping parameter distribution and the third mapping parameter distribution to form a first associated parameter distribution corresponding to the project monitoring data vector;

[0028] Each first correlation parameter in the first correlation parameter distribution corresponding to the project monitoring data vector is used as a weighting coefficient, and a weighted sum operation is performed on the second mapping parameter distribution to form a project external correlation vector corresponding to the project monitoring data vector.

[0029] In a preferred embodiment of the present invention, in the above-mentioned automated soil and water loss monitoring and early warning method based on transfer learning, the step of performing interference removal sub-operations based on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects to form second local soil and water loss vectors corresponding to the second number of prediction projects includes:

[0030] External association mining of semantic information is performed on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects respectively, to form a second local soil and water loss vector corresponding to each of the second number of prediction projects.

[0031] In a preferred embodiment of the present invention, in the above-mentioned automated soil and water loss monitoring and early warning method based on transfer learning, the step of performing external association mining of semantic information on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects to form a second local soil and water loss vector corresponding to each of the second number of prediction projects includes:

[0032] For each of the project prediction data vectors corresponding to the second number of prediction projects, mapping the project prediction data vector according to the second mapping parameter formed by learning the sample data to form a corresponding fourth mapping parameter distribution and a fifth mapping parameter distribution, and mapping the associated soil and water loss vector to form a corresponding sixth mapping parameter distribution;

[0033] performing a dot product operation on the fourth mapping parameter distribution and the sixth mapping parameter distribution to form a second associated parameter distribution corresponding to the project prediction data vector;

[0034] Each first associated parameter in the second associated parameter distribution corresponding to the project prediction data vector is used as a weighting coefficient, and a weighted sum operation is performed on the fifth mapping parameter distribution to form a second local soil and water loss vector corresponding to the project prediction data vector.

[0035] In a preferred embodiment of the present invention, in the above-mentioned method for automated monitoring and early warning of soil and water loss based on transfer learning, the step of aggregating the first local soil and water loss vector and the second local soil and water loss vector to form a multimodal soil and water loss vector of the bth interference removal unit includes:

[0036] For each of the second local soil and water loss vectors, mapping the second local soil and water loss vector according to a third mapping parameter formed by learning the sample data to form a corresponding seventh mapping parameter distribution, and mapping the first local soil and water loss vector to form a corresponding eighth mapping parameter distribution, wherein the second number of project prediction data vectors for the prediction projects corresponds to a second number of second local soil and water loss vectors;

[0037] calculating a distribution similarity between the seventh mapping parameter distribution and the eighth mapping parameter distribution, and using the distribution similarity as a corresponding weighting coefficient to perform a weighted sum operation on each of the second local soil and water loss vectors to form a target local soil and water loss vector;

[0038] The first local soil and water loss vector and the target local soil and water loss vector are aggregated to form a multimodal soil and water loss vector of the bth interference removal unit.

[0039] In a preferred embodiment of the present invention, in the above-mentioned automatic monitoring and early warning method for soil and water loss based on transfer learning, the automatic monitoring and early warning method for soil and water loss further comprises:

[0040] Determine the soil and water loss monitoring samples in the sample area;

[0041] Using the candidate prediction network, mining project monitoring data vectors of a first number of sample measured projects in the soil and water loss monitoring samples based on the soil and water loss monitoring samples, and mining project prediction data vectors of a second number of sample prediction projects;

[0042] Determining a sample interference vector using the candidate prediction network, and performing A stages of interference application operations on the sample remote sensing image of the sample area based on the sample interference vector to form a sample area remote sensing image vector;

[0043] Using the candidate prediction network, based on the project monitoring data vectors of the first number of sample measured projects and the project prediction data vectors of the second number of sample prediction projects, perform A stages of interference removal operations on the remote sensing image vectors of the sample area to form a sample multimodal soil and water loss vector corresponding to the a-th stage of interference removal operations;

[0044] Using the candidate prediction network, a soil and water loss prediction operation is performed based on the sample multimodal soil and water loss vector corresponding to the interference removal operation in the a-th stage to form a soil and water loss prediction sample corresponding to the sample area, and based on the error between the soil and water loss prediction sample and the actual soil and water loss data of the sample area, the network parameters of the candidate prediction network are updated to form a target prediction network, wherein the target prediction network is used to perform the soil and water loss prediction operation to form the soil and water loss prediction data corresponding to the target area.

[0045] The present invention provides an automated soil and water loss monitoring and early warning method based on transfer learning. First, based on soil and water loss monitoring data, a project monitoring data vector and a project prediction data vector are mined; second, based on the target interference vector, a target remote sensing image of the target area is subjected to A-stage interference application operations to form a regional remote sensing image vector; then, based on the project monitoring data vector and the project prediction data vector, the regional remote sensing image vector is subjected to A-stage interference removal operations to form a multimodal soil and water loss vector; finally, a soil and water loss prediction operation is performed based on the multimodal soil and water loss vector to form soil and water loss prediction data, and a soil and water loss early warning operation is performed based on the soil and water loss prediction data. Based on the above content, on the one hand, because the semantic information of the measured project and the predicted project is combined with the semantic information in the target remote sensing image, prediction and early warning can be performed based on the obtained multimodal soil and water loss vector. In this way, when the multimodal soil and water loss vector has a relatively rich semantic information representation capability, the reliability of the prediction and early warning can be guaranteed. On the other hand, the application of interference information can break the integrity of the target remote sensing image itself, so that when fusing other information (i.e., performing interference removal operations based on other information), more reliance is placed on relevant information to restore the target remote sensing image. In this way, it is not just about directly operating on the target remote sensing image, but also about integrating various information to improve the understanding and processing capabilities of the target. In this way, the quality of information fusion can be improved, thereby obtaining more accurate and rich inference results. Therefore, the problem of relatively low reliability of automated monitoring and early warning of soil and water loss in existing technologies can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings.

[0047] Figure 1 This is a structural block diagram of an electronic device provided by an embodiment of the present invention.

[0048] Figure 2 A flow chart of an automated soil and water loss monitoring and early warning method based on transfer learning provided in an embodiment of the present invention.

[0049] Figure 3 A schematic diagram of interference removal operations in stages A provided by an embodiment of the present invention.

[0050] Figure 4 A schematic diagram of an interference removal sub-operation provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0052] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0053] like Figure 1 As shown, an embodiment of the present invention provides an electronic device, wherein the electronic device may include a memory and a processor.

[0054] 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 may be electrically connected via one or more communication buses or signal lines. The processor is configured to execute an executable computer program stored in the memory to implement the automated soil and water loss monitoring and early warning method based on transfer learning provided in an embodiment of the present invention.

[0055] 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.

[0056] Furthermore, 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.

[0057] I understand. Figure 1 The structure shown is only for illustration, and the electronic device may also include Figure 1 More or fewer components than shown, or with Figure 1 The different configurations shown, for example, may also include a communication unit for exchanging information with other devices (such as image acquisition devices, etc.).

[0058] Combine Figure 2 The embodiment of the present invention also provides a method for automatic monitoring and early warning of soil and water loss based on transfer learning that can be applied to the above electronic device. The method steps defined in the process related to the automatic monitoring and early warning method of soil and water loss based on transfer learning can be implemented by the electronic device. Figure 2 The specific process shown is explained in detail.

[0059] Step S110 , determining soil and water loss monitoring data of a target area, and mining project monitoring data vectors of a first number of measured items in the soil and water loss monitoring data based on the soil and water loss monitoring data, and mining project prediction data vectors of a second number of predicted items.

[0060] In an embodiment of the present invention, the electronic device can determine the soil and water loss monitoring data of the target area, and mine the project monitoring data vectors of a first number of measured items in the soil and water loss monitoring data based on the soil and water loss monitoring data, and mine the project prediction data vectors of a second number of predicted items. The first number of measured items includes at least one of topography, soil and vegetation, and the second number of predicted items includes at least one of predicted precipitation, predicted wind speed and predicted wind direction. That is, the first number of measured items may only include at least part of topography, soil and vegetation, or may also include other items when including at least part of topography, soil and vegetation. Similarly, the second number of predicted items may also only include at least part of predicted precipitation, predicted wind speed and predicted wind direction, or may also include other items when including at least part of predicted precipitation, predicted wind speed and predicted wind direction. In addition, the target area may refer to an area that needs to be monitored and warned. The specific size of the area is not limited and can be selected according to actual needs. In addition, the project monitoring data vector and the project prediction data vector may be a type of data that performs semantic mining on the data of the corresponding project and is represented in the form of a vector. For example, the soil and water loss monitoring data may be "within 1 hour after the heavy rain, the precipitation reaches 100 mm, the slope is 30 degrees, the soil is sandy soil, the vegetation coverage rate is low, about 30%,...", where "the soil is sandy soil" belongs to a measured item, namely soil, and "within 1 hour after the heavy rain, the precipitation reaches 100 mm" belongs to a predicted item, namely predicted precipitation. In this way, "within 1 hour after the heavy rain, the precipitation reaches 100 mm" can be processed by word segmentation and embedding (such as through the word embedding model in the corresponding neural network, such as the target prediction network described later), and the following can be obtained:

[0061] exist:

[0062] [0.12,-0.07,0.22,-0.15,0.43,-0.19,0.31,0.25,-0.34,0.06,-0.28,0.18,0.44,-0.22,-0.15,0.26,0.19,-0.31,......,0.09];

[0063] rainstorm:

[0064] [0.45,0.39,-0.21,0.52,-0.12,0.04,0.18,-0.09,0.34,0.22,-0.27,0.39,-0.13,0.18,0.26,-0.07,0.16,0.11,......,0.37];

[0065] back:

[0066] [0.11,0.17,-0.02,0.21,0.08,0.13,0.09,0.12,-0.03,-0.09,0.24,-0.01,0.28,0.06,-0.19,0.31,0.29,-0.16,......,0.14];

[0067] of:

[0068] [0.02,0.19,0.07,-0.13,0.03,0.21,0.09,0.16,0.14,-0.01,0.17,0.12,-0.02,0.11,0.03,0.16,0.25,0.08,......,0.06]; 1:

[0070] [0.71,0.89,-0.71,0.54,-0.13,0.14,0.12,-0.19,0.35,0.21,-0.37,0.38,-0.14,0.19,0.27,-0.08,0.17,0.12,......,0.69];

[0071] Hour:

[0072] [0.38,0.43,0.29,0.35,-0.09,0.18,-0.08,0.19,0.23,-0.04,0.12,0.25,-0.21,0.31,0.14,0.22,0.05,0.11,......,0.26];

[0073] Inside:

[0074] [0.21,0.13,0.14,0.08,0.27,-0.18,0.12,-0.04,0.10,0.19,-0.01,0.14,0.22,-0.12,0.16,0.29,0.21,0.11,......,0.07];

[0075] Precipitation:

[0076] [0.52,0.43,0.29,0.62,-0.09,0.33,0.21,-0.10,0.41,0.27,-0.28,0.46,0.29,-0.18,0.13,0.28,0.16,0.34,......,0.24];

[0077] achieve:

[0078] [0.34,0.36,0.43,-0.02,0.29,0.12,0.15,0.11,0.34,0.41,-0.06,0.37,0.24,-0.09,0.29,0.21,0.18,0.17,......,0.33]; 100:

[0080] [0.32,0.33,0.49,0.72,-0.99,0.32,0.21,-0.11,0.41,0.25,-0.27,0.45,0.21,-0.17,0.18,0.28,0.26,0.56,......,0.78];

[0081] mm:

[0082] [0.39,0.22,0.47,0.38,-0.15,0.18,0.12,0.26,-0.04,0.27,0.33,-0.11,0.31,0.15,0.14,0.19,0.28,-0.08,......,0.16].

[0083] Based on this, the embedding vectors of each word can be concatenated and averaged to achieve fusion, thereby forming the corresponding project prediction data vector. In other words, the fusion result can be further processed, such as convolution, self-attention, etc., to obtain the project prediction data vector; or, in other embodiments, the fusion result can be fused with other information, for example, the embedding vectors of the positions of each word in the text can be fused, etc.

[0084] Step S120 : determining a target interference vector, and performing A stages of interference application operations on the target remote sensing image of the target area based on the target interference vector to form a regional remote sensing image vector.

[0085] In an embodiment of the present invention, the electronic device can determine a target interference vector, and, based on the target interference vector, perform A stages of interference application operations on the target remote sensing image of the target area to form a regional remote sensing image vector. The target interference vector can be a random vector, and the random vector can be a normal distribution with a mean of 0. In this way, the target remote sensing image can be convolved to form a corresponding convolution vector, or the convolution vector can be self-attention processed to obtain a self-attention vector. In this way, the convolution vector or the self-attention vector can be used as a semantic representation vector of the target remote sensing image, and then the target interference vector can be applied to the semantic representation vector to form a regional remote sensing image vector with interference information. Specifically, in the first stage of the interference application operation, the target interference vector and the semantic representation vector can be weighted and summed to obtain the interference application vector corresponding to the interference application operation of the first stage. In the second stage of the interference application operation, the interference application vector corresponding to the interference application operation of the first stage and the target interference vector can be weighted and summed to obtain the interference application vector corresponding to the interference application operation of the second stage. This cycle can be repeated to obtain the interference application vector corresponding to the interference application operation of the last stage, that is, as the regional remote sensing image vector, which is calculated as follows:

[0086] X t =α t *X t-1 +(1-α t )*e;

[0087] Here, t represents the current stage, t-1 represents the previous stage, X represents the interference application vector (X0 represents the semantic representation vector), α represents the weighting coefficient, which can be formed in the corresponding neural network, and e represents the target interference vector. In this way, as the stages progress, the interference information gradually increases.

[0088] Step S130: Based on the project monitoring data vectors of the first number of measured projects and the project prediction data vectors of the second number of predicted projects, the regional remote sensing image vector is subjected to A stages of interference removal operations to form a multimodal soil and water loss vector corresponding to the interference removal operation of the ath stage.

[0089] In an embodiment of the present invention, the electronic device can perform A stages of interference removal operations on the regional remote sensing image vector based on the project monitoring data vectors of the first number of measured projects and the project prediction data vectors of the second number of predicted projects, to form a multimodal soil and water loss vector corresponding to the interference removal operation of the ath stage, that is, a multimodal soil and water loss vector corresponding to the removal operation of the last stage. That is to say, by introducing the relevant project monitoring data vectors of the first number of measured projects and the project prediction data vectors of the second number of predicted projects, the interference information in the regional remote sensing image vector is suppressed or removed (by gradually removing it in A stages, the accuracy and reliability of the processing can be improved to a certain extent), so that the relevant semantic information in the two parts of the vector can be mined, so that the representation accuracy of the mined semantic information can be higher, that is, the reliability of the multimodal soil and water loss vector is guaranteed. In addition, the specific value of A is not limited and can be selected according to actual needs, for example, values such as 3, 4, 5, 6, 7, etc.

[0090] Step S140: performing a soil and water loss prediction operation based on the multimodal soil and water loss vector corresponding to the interference removal operation in the a-th stage to form soil and water loss prediction data corresponding to the target area; and performing a soil and water loss early warning operation on the target area based on the soil and water loss prediction data.

[0091] In an embodiment of the present invention, the electronic device can perform a soil and water loss prediction operation based on the multimodal soil and water loss vector corresponding to the interference removal operation of the a-th stage, form soil and water loss prediction data corresponding to the target area, and perform a soil and water loss early warning operation on the target area based on the soil and water loss prediction data. For example, the soil and water loss prediction operation can be performed based on the soil and water loss prediction unit in the corresponding neural network. Specifically, the soil and water loss prediction unit can include a fully connected subunit and an output subunit. In this way, the fully connected subunit can perform a fully connected processing on the multimodal soil and water loss vector corresponding to the interference removal operation of the a-th stage to obtain a corresponding fully connected result. Then, the output subunit can process the fully connected result and output the corresponding soil and water loss prediction data. The soil and water loss prediction data can be a classification result such as severe soil and water loss or not severe soil and water loss. In this way, the output subunit can include a classification function such as softmax. The soil and water loss prediction data may also be the intensity of soil and water loss such as 0.5 tons / hectare. In this way, the output subunit may include a corresponding linear regression function (such as Sigmoid activation function + linear transformation), etc., which may be selected according to actual needs. In addition, after obtaining the soil and water loss prediction data, corresponding soil and water loss early warning operations may be performed based on the soil and water loss prediction data. For example, when soil and water loss is serious, corresponding warning information may be issued to notify corresponding personnel to handle the situation, or the corresponding equipment may be directly controlled, such as the opening of flood discharge equipment. Alternatively, when the intensity of soil and water loss exceeds a preset value, corresponding warning information may be issued to notify corresponding personnel to handle the situation, etc., which may also be selected according to actual needs.

[0092] Based on the above, on the one hand, since the semantic information of the measured and predicted items is combined with the semantic information in the target remote sensing image, prediction and early warning can be performed based on the obtained multimodal soil and water loss vector. In this way, when the multimodal soil and water loss vector has a relatively rich semantic information representation capability, the reliability of the prediction and early warning can be guaranteed. On the other hand, the application of interference information can break the integrity of the target remote sensing image itself, so that when fusing other information (i.e., performing interference removal operations based on other information), more reliance is placed on relevant information to restore the target remote sensing image. In this way, it is not just a direct operation on the target remote sensing image, but also the integration of multiple aspects of information to improve the understanding and processing capabilities of the target. In this way, the quality of information fusion can be improved, thereby obtaining more accurate and rich inference results. Therefore, the problem of relatively low reliability of automated soil and water loss monitoring and early warning in the existing technology can be improved.

[0093] Based on the above content, the specific implementation method of the A-stage interference removal operation in step S130 is not limited and can be selected according to actual conditions.

[0094] For example, in an alternative embodiment, in order to reliably perform the interference operation so that the obtained multimodal soil and water loss vector can reliably represent the effective semantic information in the project monitoring data vector, the target prediction data vector, and the regional remote sensing image vector, the above-mentioned step S130 may further include step S131, and the step S131 may specifically include the following contents:

[0095] The regional remote sensing image vector may be subjected to A stages of interference removal operations to form a multimodal soil and water loss vector corresponding to the a-th stage of interference removal operations;

[0096] Among them, for the interference removal operation of the ath stage in the A-stage interference removal operation, if a is equal to 1, then the input of the interference removal operation of the ath stage is the regional remote sensing image vector; if a is greater than 1 and less than or equal to A, then the input of the interference removal operation of the ath stage is the multimodal soil and water loss vector of the interference removal operation of the a-1th stage; that is, the input of the interference removal operation of the first stage is the quality of the regional remote sensing image, and the input of the interference removal operation of each stage after the first stage is the multimodal soil and water loss vector of the interference removal operation of the previous stage. For details, please refer to Figure 3 The content shown;

[0097] Furthermore, in the interference removal operation of the a-th stage, B interference removal units are used to perform an interference removal sub-operation, wherein, for the b-th interference removal unit among the B interference removal units, if b is equal to 1, then the input of the b-th interference removal unit is the input of the interference removal operation of the a-th stage; if b is greater than 1 and less than or equal to B, then the input of the b-th interference removal unit includes the multimodal soil and water loss vector of the b-1-th interference removal unit; if b is equal to B, then the multimodal soil and water loss vector of the interference removal operation of the a-th stage is the multimodal soil and water loss vector of the b-th interference removal unit; that is, in the interference removal operation of each stage, B interference removal units are used to perform an interference removal sub-operation; in addition, the input of the first interference removal unit is the input of the interference removal operation of the a-th stage, the input of each interference removal unit after the first is the multimodal soil and water loss vector of the previous interference removal unit, and the multimodal soil and water loss vector of the last interference removal unit is used as the multimodal soil and water loss vector of the interference removal operation of the corresponding stage.

[0098] It is understood that in the above step S131, the specific manner of performing the interference removal sub-operation using the bth interference removal unit is not limited and can be selected according to actual circumstances. For example, in an alternative embodiment, to ensure the reliability of the interference removal sub-operation, the above step S131 may further include the following steps S131a and S131b.

[0099] Step S131a, for the b-th interference removal unit, based on the multimodal soil and water loss vector of the b-1-th interference removal unit and the project monitoring data vectors of the first number of measured projects, an interference removal sub-operation is performed on the regional remote sensing image vector to form a first local soil and water loss vector, and based on the multimodal soil and water loss vector of the b-1-th interference removal unit and the project prediction data vectors of the second number of prediction projects, an interference removal sub-operation is performed on the regional remote sensing image vector to form a second local soil and water loss vector.

[0100] In an embodiment of the present invention, for the b-th interference removal unit, an interference removal sub-operation can be performed on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1-th interference removal unit and the project monitoring data vectors of the first number of measured projects to form a first local soil and water loss vector, and an interference removal sub-operation can be performed on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1-th interference removal unit and the project prediction data vectors of the second number of predicted projects to form a second local soil and water loss vector. In other words, the interference removal sub-operation can be performed based on the data of the two dimensions of the project monitoring data vector and the project prediction data vector, respectively, to form the first local soil and water loss vector and the second local soil and water loss vector, respectively. In this way, the accuracy of the operation can be improved to a certain extent. In addition, for the first interference removal unit in the first stage, an interference removal sub-operation can be performed on the regional remote sensing image vector based on the regional remote sensing image vector and the project monitoring data vectors of the first number of measured projects to form a first local soil and water loss vector, and an interference removal sub-operation can be performed on the regional remote sensing image vector based on the regional remote sensing image vector and the project prediction data vectors of the second number of prediction projects to form a second local soil and water loss vector.

[0101] Step S131b: Aggregate the first local soil and water loss vector and the second local soil and water loss vector to form a multimodal soil and water loss vector of the bth interference removal unit.

[0102] In this embodiment of the present invention, the first local soil and water loss vector and the second local soil and water loss vector may be aggregated to form a multimodal soil and water loss vector for the bth interference removal unit. That is, because the semantic information in the first local soil and water loss vector and the second local soil and water loss vector are both effective, the first local soil and water loss vector and the second local soil and water loss vector may be aggregated to obtain a semantically richer and more reliable multimodal soil and water loss vector.

[0103] It is understandable that in the above step S131a, the specific method of performing the interference removal sub-operation on the regional remote sensing image vector based on the two-dimensional data is not limited and can be selected according to the actual situation. For example, in an alternative embodiment, in order to ensure the reliability of the interference removal sub-operation, the interference information can be fully removed or suppressed by mining the association between the data to obtain a local soil and water loss vector with better semantic representation ability. Figure 4 The above-mentioned step S131a may include the following steps a1, a2, a3 and a4.

[0104] Step a1: For the b-th interference removal unit, use the b-th interference removal unit to compress the multimodal soil and water loss vector of the b-1-th interference removal unit to form a corresponding compressed soil and water loss vector.

[0105] In an embodiment of the present invention, for the bth interference removal unit, the Bth interference removal unit is used to compress the multimodal soil and water loss vector of the b-1th interference removal unit to form a corresponding compressed soil and water loss vector. In other words, the multimodal soil and water loss vector of the previous interference removal unit can be compressed to form a corresponding compressed soil and water loss vector. Furthermore, for the first interference removal unit in the first stage, a compression operation can be performed on the regional remote sensing image vector to form a corresponding compressed soil and water loss vector. For the first interference removal unit in each stage after the first stage, a compression operation can be performed on the multimodal soil and water loss vector of the last interference removal unit in the previous stage to form a corresponding compressed soil and water loss vector. Furthermore, the specific method of the compression operation is not limited. For example, convolution and / or pooling processing can be performed. Furthermore, when performing convolution and pooling simultaneously, the corresponding stride in at least one of the processes needs to be greater than 1, such as 2, 3, or 4, to enable parameter compression.

[0106] Step a2: performing internal association mining on the semantic information in the compressed soil and water loss vector to form a corresponding associated soil and water loss vector.

[0107] In an embodiment of the present invention, the semantic information in the compressed soil and water loss vector can be subjected to internal association mining to form a corresponding associated soil and water loss vector. For example, in order to achieve internal association mining, the compressed soil and water loss vector can be subjected to self-attention processing, and the obtained self-attention vector is used as the corresponding associated soil and water loss vector. Alternatively, in other embodiments, the compressed soil and water loss vector can be subjected to two different pooling methods so that different important information can be paid attention to. Then, the two obtained pooling vectors are subjected to external association mining (the specific processing process is described below), thereby obtaining the corresponding associated soil and water loss vector.

[0108] Step a3: performing an interference removal sub-operation based on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector.

[0109] In an embodiment of the present invention, an interference removal sub-operation can be performed based on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector. In other words, the project monitoring data vectors of the first number of measured projects can be integrated into the associated soil and water loss vector. Specifically, by mining the associated information therein, interference information without an associated relationship is removed, thereby completing the interference removal sub-operation.

[0110] Step a4: performing interference removal sub-operations based on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects, respectively, to form second local soil and water loss vectors corresponding to the second number of prediction projects.

[0111] In an embodiment of the present invention, an interference removal sub-operation can be performed based on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction items, respectively, to form second local soil and water loss vectors corresponding to the second number of prediction items. In other words, the project prediction data vectors for the second number of prediction items can be fused into the associated soil and water loss vector, i.e., by mining the associated information therein, interference information without an associated relationship is removed, thereby completing the interference removal sub-operation.

[0112] It is understood that in step a3 above, the specific manner of performing the interference removal sub-operation based on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects is not limited and can be selected according to actual circumstances. For example, in an alternative embodiment, to ensure the reliability of the interference removal operation and ensure that the obtained first local soil and water loss vector has a high degree of reliability, step a3 above may include the following:

[0113] The associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects are subjected to external association mining of semantic information to form a corresponding first local soil and water loss vector. In this way, interference information that has no association relationship can be effectively removed through association mining.

[0114] It is understood that, in the above content, the specific method of performing external association mining of semantic information between the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects is not limited and can be selected according to actual circumstances. For example, in an alternative embodiment, in order to improve the reliability of external association mining, the following contents may be included:

[0115] First, external association mining of semantic information can be performed on the associated soil and water loss vector and the project monitoring data vector of each measured project in the first number of measured projects, respectively, to form a project external association vector corresponding to the project monitoring data vector of each measured project; for example, external association mining of semantic information can be performed based on the project monitoring data vector of the first measured project and the associated soil and water loss vector, to form a project external association vector corresponding to the project monitoring data vector of the first measured project; for another example, external association mining of semantic information can be performed based on the project monitoring data vector of the second measured project and the associated soil and water loss vector, to form a project external association vector corresponding to the project monitoring data vector of the second measured project;

[0116] Secondly, the project external association vectors corresponding to the project monitoring data vectors of the first number of measured projects can be aggregated to form a corresponding first local soil and water loss vector; for example, the project external association vectors corresponding to the project monitoring data vectors of the first number of measured projects can be spliced or fused by mean calculation to obtain the first local soil and water loss vector, so that the semantics of the first local soil and water loss vector can be richer, and by performing external association mining of semantic information separately, the accuracy of the external association mining can also be guaranteed, avoiding problems such as mutual interference between project monitoring data vectors of different measured projects.

[0117] It is understandable that, in the above content, the specific method of performing external association mining of semantic information on the associated soil and water loss vector and the project monitoring data vector of each of the first number of measured projects is not limited and can be selected according to actual conditions. For example, in an alternative embodiment, in order to fully remove interference information in the associated soil and water loss vector through external association mining of semantic information, that is, to achieve full fusion of semantic information between the associated soil and water loss vector and the project monitoring data vector, the following content may be further included:

[0118] First, for the project monitoring data vector of each measured project in the first number of measured projects, the project monitoring data vector is mapped according to the first mapping parameter formed by learning the sample data to form a corresponding first mapping parameter distribution and a second mapping parameter distribution, and the associated soil and water loss vector is mapped to form a corresponding third mapping parameter distribution; for example, the first mapping parameter may include three parameter sub-matrices, and the project monitoring data vector is multiplied with the first parameter sub-matrix and the second parameter sub-matrix respectively (which may be the multiplication of the parameters at the corresponding positions) to obtain the first mapping parameter distribution and the second mapping parameter distribution; and the associated soil and water loss vector is multiplied with the third parameter sub-matrix to obtain the third mapping parameter distribution; in addition, the first mapping parameter may be a network parameter of the target prediction network as described later, so that it is formed by updating in the process of the corresponding candidate prediction network learning the sample data, and the second mapping parameter and the third mapping parameter described later can also be formed in the corresponding process;

[0119] Secondly, a dot product operation can be performed on the first mapping parameter distribution and the third mapping parameter distribution to form a first associated parameter distribution corresponding to the project monitoring data vector; for example, the first mapping parameter distribution can be transposed and then dot-producted with the third mapping parameter distribution to obtain the corresponding first associated parameter distribution. In this way, the association relationship between the two vectors can be represented by the first associated parameter distribution;

[0120] Then, each first association parameter in the first association parameter distribution corresponding to the project monitoring data vector can be used as a weighting coefficient, and the second mapping parameter distribution can be weighted summed to form the project external association vector corresponding to the project monitoring data vector. Based on this, mining based on association relationships can be achieved, that is, mining of association information can be achieved.

[0121] It is understood that in step a4 above, the specific manner of performing the interference removal sub-operation based on the associated soil and water loss vector and the project prediction data vectors of the second number of prediction projects is not limited and can be selected according to actual circumstances. For example, in an alternative embodiment, to ensure the reliability of the interference removal operation and ensure that the obtained second local soil and water loss vector has a high degree of reliability, step a4 above may include the following:

[0122] External association mining of semantic information is performed on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction items respectively to form a second local soil and water loss vector corresponding to each prediction item in the second number of prediction items; for example, external association mining of semantic information can be performed on the associated soil and water loss vector and the project prediction data vector corresponding to the first prediction item to form a second local soil and water loss vector corresponding to the first prediction item; for another example, external association mining of semantic information can be performed on the associated soil and water loss vector and the project prediction data vector corresponding to the second prediction item to form a second local soil and water loss vector corresponding to the second prediction item.

[0123] It is understandable that, in the above content, the specific method of performing external association mining of semantic information on the associated soil and water loss vector and the project prediction data vector of each predicted project in the second number of measured projects is not limited and can be selected according to actual conditions. For example, in an alternative embodiment, in order to fully remove interference information in the associated soil and water loss vector through external association mining of semantic information, that is, to achieve full fusion of semantic information between the associated soil and water loss vector and the project prediction data vector, the following content may be further included:

[0124] First, for each of the project prediction data vectors corresponding to the second number of prediction projects, the project prediction data vector is mapped according to the second mapping parameter formed by learning the sample data to form a corresponding fourth mapping parameter distribution and a fifth mapping parameter distribution, and the associated soil and water loss vector is mapped to form a corresponding sixth mapping parameter distribution, as described above.

[0125] Secondly, a dot product operation is performed on the fourth mapping parameter distribution and the sixth mapping parameter distribution to form a second associated parameter distribution corresponding to the project prediction data vector, as described above;

[0126] Then, each first associated parameter in the second associated parameter distribution corresponding to the project prediction data vector is used as a weighting coefficient, and the fifth mapping parameter distribution is weighted summed to form a second local soil and water loss vector corresponding to the project prediction data vector, as described above.

[0127] It is understandable that, in the above-mentioned step S131b, the specific manner of aggregating the first local soil and water loss vector and the second local soil and water loss vector is not limited and can be selected according to actual conditions. For example, in an alternative embodiment, in order to achieve reliable aggregation of the first local soil and water loss vector and the second local soil and water loss vector, the first local soil and water loss vector and the second local soil and water loss vector can be associated with each other to ensure the accuracy of the semantic representation of the obtained multimodal soil and water loss vector. The above-mentioned step S131b can further include the following content:

[0128] First, for each of the second local soil and water loss vectors, the second local soil and water loss vector can be mapped according to the third mapping parameter formed by learning the sample data to form a corresponding seventh mapping parameter distribution, and the first local soil and water loss vector can be mapped to form a corresponding eighth mapping parameter distribution, wherein the project prediction data vectors of the second number of prediction projects correspond to a second number of second local soil and water loss vectors; that is, the third mapping parameter can include two mapping parameter sub-matrices, so that the first mapping parameter sub-matrix can be multiplied by the second local soil and water loss vector (multiplying the parameters at the corresponding positions) to obtain the corresponding seventh mapping parameter distribution; and the second mapping parameter sub-matrix can be multiplied by the first local soil and water loss vector to obtain the eighth mapping parameter distribution; thus, for the second number of second local soil and water loss vectors, there are a second number of seventh mapping parameter distributions;

[0129] Secondly, the distribution similarity between the seventh mapping parameter distribution and the eighth mapping parameter distribution can be calculated, and the distribution similarity can be used as the corresponding weighting coefficient to perform a weighted summation operation on each of the second local soil and water loss vectors to form a target local soil and water loss vector; that is, since the second local soil and water loss vector and the first local soil and water loss vector belong to different semantic spaces, they can be mapped first through the third mapping parameter so that they can be in a similar semantic space, so that the calculated distribution similarity can be more reliable, wherein the distribution similarity can refer to a parameter such as cosine similarity that can characterize the similarity relationship; in this way, since the similarity with the first local soil and water loss vector is different, the corresponding second local soil and water loss vector has different importance, so that a weighted summation operation can be performed based on the distribution similarity, thereby ensuring that the formed target local soil and water loss vector has a higher reliability;

[0130] Finally, the first local soil and water loss vector and the target local soil and water loss vector can be aggregated to form a multimodal soil and water loss vector of the bth interference removal unit. For example, the first local soil and water loss vector and the target local soil and water loss vector can be spliced, averaged, or weighted summed to obtain the corresponding multimodal soil and water loss vector.

[0131] On the basis of the above content, in order to ensure the reliable execution of step S110, step S120, step S130 and step S140, transfer learning can be performed first to obtain a reliable target prediction network, so that corresponding processing can be performed based on the target prediction network. Based on this, the automatic monitoring and early warning method for soil and water loss can also include a step of training to form a target prediction network, which includes:

[0132] First, the soil and water loss monitoring samples of the sample area can be determined, corresponding to the soil and water loss monitoring data mentioned above, and the specific content is as described above;

[0133] Secondly, the candidate prediction network can be used to mine the project monitoring data vectors of the first number of sample measured projects in the soil and water loss monitoring samples based on the soil and water loss monitoring samples, and to mine the project prediction data vectors of the second number of sample prediction projects, as described above.

[0134] Then, the candidate prediction network can be used to determine a sample interference vector, and based on the sample interference vector, a sample remote sensing image of the sample area is subjected to A stages of interference application operations to form a sample area remote sensing image vector, as described above.

[0135] Afterwards, the candidate prediction network can be used to perform A stages of interference removal operations on the sample area remote sensing image vector based on the project monitoring data vectors of the first number of sample measured projects and the project prediction data vectors of the second number of sample prediction projects, thereby forming a sample multimodal soil and water loss vector corresponding to the a-th stage of interference removal operations, as described above.

[0136] Finally, the candidate prediction network can be used to perform a soil and water loss prediction operation based on the sample multimodal soil and water loss vector corresponding to the interference removal operation in the a-th stage to form a soil and water loss prediction sample corresponding to the sample area, and based on the error between the soil and water loss prediction sample and the actual soil and water loss data of the sample area (such as the mean square error or cross entropy error can be selected according to needs), the network parameters of the candidate prediction network are updated to form a target prediction network (for example, the network parameters can be updated in the direction of reducing the error, such as updating the aforementioned first mapping parameters, second mapping parameters and third mapping parameters, etc., until the error converges, such as the error is less than a preset value or the error reduction amplitude is less than a preset amplitude, thereby determining that the update is completed, that is, obtaining the target prediction network), wherein the target prediction network is used to perform a soil and water loss prediction operation to form soil and water loss prediction data corresponding to the target area.

[0137] In summary, the present invention provides an automated soil and water loss monitoring and early warning method based on transfer learning. First, based on the soil and water loss monitoring data, a project monitoring data vector and a project prediction data vector are mined; secondly, based on the target interference vector, an interference application operation is performed on the target remote sensing image of the target area for A stages to form a regional remote sensing image vector; then, based on the project monitoring data vector and the project prediction data vector, an interference removal operation is performed on the regional remote sensing image vector for A stages to form a multimodal soil and water loss vector; finally, a soil and water loss prediction operation is performed based on the multimodal soil and water loss vector to form soil and water loss prediction data, and a soil and water loss early warning operation is performed based on the soil and water loss prediction data. Based on the above content, on the one hand, since the semantic information of the measured project and the predicted project is combined with the semantic information in the target remote sensing image, it is possible to make predictions and early warnings based on the obtained multimodal soil and water loss vector. In this way, when the multimodal soil and water loss vector has a relatively rich semantic information representation capability, the reliability of the prediction and early warning can be guaranteed. On the other hand, the application of interference information can break the integrity of the target remote sensing image itself, so that when fusing other information (i.e., performing interference removal operations based on other information), more reliance is placed on relevant information to restore the target remote sensing image. In this way, it is not just about directly operating on the target remote sensing image, but also about integrating various information to improve the understanding and processing capabilities of the target. In this way, the quality of information fusion can be improved, thereby obtaining more accurate and rich inference results. Therefore, the problem of relatively low reliability of automated monitoring and early warning of soil and water loss in existing technologies can be improved.

[0138] In the several embodiments provided in the embodiments of the present invention, 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 the devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, which contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0139] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0140] If the functions are implemented in the form of software function 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 the present invention is essentially or partly contributed to the prior art or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, electronic device, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk. It should be noted that, in this article, the terms "include", "comprising" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such a process, method, article or device. Without further constraints, an element defined by the phrase "comprises a..." does not preclude the existence of additional identical elements in the process, method, article or apparatus that includes the element.

[0141] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for automatic monitoring and early warning of soil and water loss based on transfer learning, characterized in that: The automatic monitoring and early warning method for soil and water loss includes: Determining soil and water loss monitoring data for a target area, and mining, based on the soil and water loss monitoring data, project monitoring data vectors for a first number of measured items in the soil and water loss monitoring data, and mining project prediction data vectors for a second number of predicted items, wherein the first number of measured items includes at least one of terrain, soil, and vegetation, and the second number of predicted items includes at least one of predicted precipitation, predicted wind speed, and predicted wind direction; Determining a target interference vector, and, based on the target interference vector, performing A stages of interference application operations on the target remote sensing image of the target area to form a regional remote sensing image vector; Based on the project monitoring data vectors of the first number of measured projects and the project prediction data vectors of the second number of predicted projects, performing A stages of interference removal operations on the regional remote sensing image vector to form a multimodal soil and water loss vector corresponding to the a-th stage of interference removal operations; A soil and water loss prediction operation is performed based on the multimodal soil and water loss vector corresponding to the interference removal operation in the ath stage to form soil and water loss prediction data corresponding to the target area; and, based on the soil and water loss prediction data, a soil and water loss early warning operation is performed on the target area.

2. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to claim 1 is characterized in that: The step of performing A stages of interference removal operations on the regional remote sensing image vector based on the project monitoring data vectors of the first number of measured projects and the project prediction data vectors of the second number of predicted projects to form a multimodal soil and water loss vector corresponding to the interference removal operation of the ath stage includes: Performing A stages of interference removal operations on the regional remote sensing image vector to form a multimodal soil and water loss vector corresponding to the interference removal operation of the a-th stage, wherein, for the interference removal operation of the a-th stage among the A-stage interference removal operations, if a is equal to 1, then the input of the interference removal operation of the a-th stage is the regional remote sensing image vector; if a is greater than 1 and less than or equal to A, then the input of the interference removal operation of the a-th stage is the multimodal soil and water loss vector of the interference removal operation of the a-1-th stage; In the interference removal operation of the a-th stage, B interference removal units are used to perform an interference removal sub-operation, wherein, for the b-th interference removal unit among the B interference removal units, if b is equal to 1, then the input of the b-th interference removal unit is the input of the interference removal operation of the a-th stage; if b is greater than 1 and less than or equal to B, then the input of the b-th interference removal unit includes the multimodal soil and water loss vector of the b-1-th interference removal unit; if b is equal to B, then the multimodal soil and water loss vector of the interference removal operation of the a-th stage is the multimodal soil and water loss vector of the b-th interference removal unit; For the b-th interference removal unit, an interference removal sub-operation is performed on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1-th interference removal unit and the project monitoring data vectors of the first number of measured projects to form a first local soil and water loss vector; and an interference removal sub-operation is performed on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1-th interference removal unit and the project prediction data vectors of the second number of prediction projects to form a second local soil and water loss vector; The first local soil and water loss vector and the second local soil and water loss vector are aggregated to form a multimodal soil and water loss vector of the bth interference removal unit.

3. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to claim 2 is characterized in that: The step of performing an interference removal sub-operation on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1th interference removal unit and the project monitoring data vectors of the first number of measured projects to form a first local soil and water loss vector, and performing an interference removal sub-operation on the regional remote sensing image vector based on the multimodal soil and water loss vector of the b-1th interference removal unit and the project prediction data vectors of the second number of prediction projects to form a second local soil and water loss vector includes: For the b-th interference removal unit, using the b-th interference removal unit, compressing the multimodal soil and water loss vector of the b-1-th interference removal unit to form a corresponding compressed soil and water loss vector; Performing internal association mining on the semantic information in the compressed soil and water loss vector to form a corresponding associated soil and water loss vector; performing an interference removal sub-operation based on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector; An interference removal sub-operation is performed respectively according to the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects to form second local soil and water loss vectors corresponding to the second number of prediction projects.

4. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to claim 3 is characterized in that: The step of performing an interference removal sub-operation based on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector includes: External association mining of semantic information is performed on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector.

5. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to claim 4 is characterized in that: The step of performing external association mining of semantic information on the associated soil and water loss vector and the project monitoring data vectors of the first number of measured projects to form a corresponding first local soil and water loss vector includes: Performing external correlation mining of semantic information on the associated soil and water loss vector and the project monitoring data vector of each of the first number of measured projects, to form a project external correlation vector corresponding to the project monitoring data vector of each measured project; The project external correlation vectors corresponding to the project monitoring data vectors of the first number of measured projects are aggregated to form corresponding first local soil and water loss vectors.

6. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to claim 5 is characterized in that: The step of performing external association mining of semantic information on the associated soil and water loss vector and the project monitoring data vector of each of the first number of measured projects to form a project external association vector corresponding to the project monitoring data vector of each measured project includes: For each of the project monitoring data vectors of the first number of measured projects, mapping the project monitoring data vector according to a first mapping parameter formed by learning the sample data to form a corresponding first mapping parameter distribution and a second mapping parameter distribution, and mapping the associated soil and water loss vector to form a corresponding third mapping parameter distribution; performing a dot product operation on the first mapping parameter distribution and the third mapping parameter distribution to form a first associated parameter distribution corresponding to the project monitoring data vector; Each first correlation parameter in the first correlation parameter distribution corresponding to the project monitoring data vector is used as a weighting coefficient, and a weighted sum operation is performed on the second mapping parameter distribution to form a project external correlation vector corresponding to the project monitoring data vector.

7. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to claim 3 is characterized in that: The step of performing interference removal sub-operations based on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects to form second local soil and water loss vectors corresponding to the second number of prediction projects includes: External association mining of semantic information is performed on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects respectively, to form a second local soil and water loss vector corresponding to each of the second number of prediction projects.

8. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to claim 7 is characterized in that: The step of performing external association mining of semantic information on the associated soil and water loss vector and the project prediction data vectors corresponding to the second number of prediction projects to form a second local soil and water loss vector corresponding to each of the second number of prediction projects includes: For each of the project prediction data vectors corresponding to the second number of prediction projects, mapping the project prediction data vector according to the second mapping parameter formed by learning the sample data to form a corresponding fourth mapping parameter distribution and a fifth mapping parameter distribution, and mapping the associated soil and water loss vector to form a corresponding sixth mapping parameter distribution; performing a dot product operation on the fourth mapping parameter distribution and the sixth mapping parameter distribution to form a second associated parameter distribution corresponding to the project prediction data vector; Each first associated parameter in the second associated parameter distribution corresponding to the project prediction data vector is used as a weighting coefficient, and a weighted sum operation is performed on the fifth mapping parameter distribution to form a second local soil and water loss vector corresponding to the project prediction data vector.

9. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to claim 2 is characterized in that: The step of aggregating the first local soil and water loss vector and the second local soil and water loss vector to form a multimodal soil and water loss vector of the bth interference removal unit includes: For each of the second local soil and water loss vectors, mapping the second local soil and water loss vector according to a third mapping parameter formed by learning the sample data to form a corresponding seventh mapping parameter distribution, and mapping the first local soil and water loss vector to form a corresponding eighth mapping parameter distribution, wherein the second number of project prediction data vectors for the prediction projects corresponds to a second number of second local soil and water loss vectors; calculating a distribution similarity between the seventh mapping parameter distribution and the eighth mapping parameter distribution, and using the distribution similarity as a corresponding weighting coefficient to perform a weighted sum operation on each of the second local soil and water loss vectors to form a target local soil and water loss vector; The first local soil and water loss vector and the target local soil and water loss vector are aggregated to form a multimodal soil and water loss vector of the bth interference removal unit.

10. The automatic monitoring and early warning method for soil and water loss based on transfer learning according to any one of claims 1 to 9, characterized in that: The automatic monitoring and early warning method for soil and water loss also includes: Determine the soil and water loss monitoring samples in the sample area; Using the candidate prediction network, mining project monitoring data vectors of a first number of sample measured projects in the soil and water loss monitoring samples based on the soil and water loss monitoring samples, and mining project prediction data vectors of a second number of sample prediction projects; Determining a sample interference vector using the candidate prediction network, and performing A stages of interference application operations on the sample remote sensing image of the sample area based on the sample interference vector to form a sample area remote sensing image vector; Using the candidate prediction network, based on the project monitoring data vectors of the first number of sample measured projects and the project prediction data vectors of the second number of sample prediction projects, perform A stages of interference removal operations on the remote sensing image vectors of the sample area to form a sample multimodal soil and water loss vector corresponding to the a-th stage of interference removal operations; Using the candidate prediction network, a soil and water loss prediction operation is performed based on the sample multimodal soil and water loss vector corresponding to the interference removal operation in the a-th stage to form a soil and water loss prediction sample corresponding to the sample area, and based on the error between the soil and water loss prediction sample and the actual soil and water loss data of the sample area, the network parameters of the candidate prediction network are updated to form a target prediction network, wherein the target prediction network is used to perform the soil and water loss prediction operation to form the soil and water loss prediction data corresponding to the target area.

Images