Comprehensive early warning method, platform and storage medium for soil damage

Abstract

The present invention belongs to the field of soil damage early warning technology, and specifically relates to a comprehensive early warning method, platform and storage medium for soil damage. The method includes: S1, obtaining several remote sensing images of soil in advance, using several training data consisting of remote sensing images and real results corresponding to the remote sensing images to train and generate an early warning model; S2, continuously obtaining remote sensing images of soil, inputting different remote sensing images into the early warning model in sequence, outputting first early warning results respectively, judging whether the situation corresponding to the new type of damage of the remote sensing image occurs, and in the case of yes, proceeding to the next step; S3, continuously obtaining remote sensing images of soil, inputting different remote sensing images into the backup early warning model in sequence, generating second early warning results respectively according to the output results, and retraining the early warning model based on all the original training data and several first training data. The present invention can ensure the efficient execution of the early warning method.

Description

Technical Field

[0001] The present invention belongs to the technical field of soil damage early warning, and in particular relates to a comprehensive early warning method, platform and storage medium for soil damage. Background Art

[0002] After human beings dump or landfill pollutants into the soil environment, it is necessary to continuously monitor the soil damage to the soil environment to provide basic data for subsequent soil environment remediation work.

[0003] Similar prior art includes a Chinese patent application with publication number CN117952321A, which discloses a soil erosion intelligent monitoring and early warning method and system based on land engineering, and relates to the field of land monitoring technology. The method includes: obtaining engineering equipment components of land engineering, collecting resource monitoring index samples of each engineering equipment, obtaining the preset land area of ​​the land engineering project and conducting soil data monitoring, obtaining an erosion feature sample set according to a soil erosion model, and then training a soil erosion intensity assessment model, and finally conducting an erosion early warning. In addition, similar prior art includes a Chinese patent application with publication number CN117391548A, which provides a land resource carrying capacity monitoring and early warning method. By using the land resource and environmental carrying capacity evaluation results, the monitoring object is divided into three different carrying states: loadable, critical overload, and overload. From the three aspects of resource utilization, ecological protection, and spatial structure, adaptive weights of influencing factors based on different carrying states are formulated, and a land resource carrying capacity remote sensing monitoring and early warning indicator system that takes into account both economic development and ecological protection is constructed, so as to realize the monitoring and early warning work of serving land resource supervision and resource and environmental carrying capacity. However, when a new warning result appears, the warning methods in the above two patent applications will be stopped and will not be resumed until the warning model is retrained, so the execution efficiency is relatively low. Summary of the invention

[0004] The present invention trains the early warning model in advance, uses the early warning model to warn of soil damage under normal circumstances, and uses the backup early warning model to continue to warn of soil damage when a new type of damage occurs corresponding to the remote sensing image. After the retraining of the early warning model is completed, the early warning model is used again to warn of soil damage. The present invention aims to improve the execution efficiency of the early warning method without interrupting the execution of the early warning method.

[0005] In order to achieve the above-mentioned purpose of the invention, the present invention provides the following comprehensive early warning method for soil damage, which mainly includes the following steps:

[0006] S1. Acquire several remote sensing images of soil in advance, and for each remote sensing image, set a real result corresponding to the remote sensing image to generate training data, where the real result is the real damage type corresponding to the remote sensing image, and use several training data consisting of remote sensing images and the real results corresponding to the remote sensing images to train and generate an early warning model;

[0007] S2, continuously acquiring remote sensing images of soil, inputting different remote sensing images into the early warning model in sequence, and the early warning model outputting first early warning results respectively, wherein the first early warning results are different damage types corresponding to the remote sensing images, and reliability values ​​corresponding to the different damage types respectively, and judging whether the situation corresponding to the new damage type of the remote sensing image occurs, if yes, proceeding to the next step, if no, judging whether the end condition is satisfied, if yes, ending all steps, if not, repeating this step;

[0008] S3. Continuously acquire remote sensing images of soil, input different remote sensing images into the backup warning model in sequence, generate second warning results according to the output results of the backup warning model, the second warning result is a damage type corresponding to the remote sensing image, and generate a number of first training data at the same time. Retrain the warning model based on all the original training data and the first training data. After the retraining is completed, jump to S2 to continue execution.

[0009] As a preferred technical solution of the present invention, the early warning model is composed of a first-layer early warning model, a second-layer early warning model, and a third-layer early warning model, wherein the image element values ​​of different image elements in the remote sensing image are input into the first-layer early warning model for calculation, the output result of the first-layer early warning model is input into the second-layer early warning model for recalculation, the output result of the second-layer early warning model is further input into the third-layer early warning model for final calculation, and the third-layer early warning model outputs the first warning result.

[0010] As a preferred technical solution of the present invention, judging whether a remote sensing image corresponds to a new type of damage occurs includes the following steps:

[0011] S21, generating different data records in sequence according to the order in which the warning model outputs the first warning results, the data records including the first warning results and representative values ​​corresponding to the first warning results;

[0012] S22, sorting different data records according to the order in which different data records are generated, calculating the average of the representative values ​​in the data records of the preset first number to obtain an average representative value for the data records of the preset first number that are ranked first, and determining whether the average representative value is less than a preset average representative value threshold value, if so, determining that the remote sensing image corresponds to a new type of damage has occurred, and ending all steps, if not, proceeding to the next step;

[0013] S23. For the data records that are sorted later, determine the data record whose number before the data record is the preset first number minus one, calculate the average of the representative value in the data record and the representative values ​​in the determined different data records to obtain the average representative value, and determine whether the average representative value is less than the preset average representative value threshold. If so, determine that a new type of damage corresponding to the remote sensing image has occurred, and terminate all steps. If not, determine whether all data records have been processed. If so, terminate all steps. If not, repeat this step.

[0014] As a preferred technical solution of the present invention, the backup warning model includes a first-layer backup warning model and a second-layer backup warning model, wherein the first-layer backup warning model further includes a first sub-model and a second sub-model in parallel, and the structures of the first sub-model and the second sub-model are the same as the structure composed of the first-layer warning model and the second-layer warning model, and the output results of the first sub-model and the second sub-model are simultaneously input into the second-layer backup warning model, and the second-layer backup warning model calculates the similarity value between the output result of the first sub-model and the output result of the second sub-model.

[0015] As a preferred technical solution of the present invention, before inputting different remote sensing images into the backup warning model in sequence, it also includes: generating a plurality of second training data, the true results in the second training data are different from the true results in all the original training data, and making the plurality of second training data and all the original training data together form a training data set.

[0016] As a preferred technical solution of the present invention, the remote sensing image is input into the backup warning model, and a second warning result is generated according to the output result of the backup warning model, including the following steps:

[0017] S311, based on each training data in the training data set, inputting the remote sensing image into the first sub-model, and simultaneously inputting the remote sensing image in the training data into the second sub-model, and the second-layer backup warning model outputs a similarity value;

[0018] S312. Sort the remote sensing images in different training data in the training data set in descending order of the corresponding similarity values, determine the remote sensing images with a preset second number of remote sensing images ranked first, and use the true result with the largest total number of corresponding remote sensing images as the second warning result.

[0019] As a preferred technical solution of the present invention, the remote sensing images in the first training data include: remote sensing images whose corresponding second warning results are different from the actual results in all the original training data, and remote sensing images whose similarity values ​​with the second training data exceed a preset similarity value threshold.

[0020] As a preferred technical solution of the present invention, after the retraining is completed, the following steps are further included:

[0021] S321, collecting a plurality of first warning results output by the warning model during the retraining process, recording the final warning result of the remote sensing image input to the warning model as the damage type corresponding to the maximum reliable value in the corresponding first warning result, and recording the maximum reliable value in the corresponding first warning result;

[0022] S322, dividing different remote sensing images corresponding to the same final warning result into a group, sorting the different remote sensing images in each group in ascending order of the corresponding maximum reliability values, and selecting the remote sensing image with a preset third number at the top of the sorting;

[0023] S323, checking whether the final warning results of the preset third number of remote sensing images in each group are correct, correcting the incorrect final warning results, and generating a number of third training data based on a number of remote sensing images whose corresponding final warning results have been corrected, and using the number of third training data to train the warning model again.

[0024] The present invention also provides a comprehensive early warning platform for soil damage, including the following modules:

[0025] A training module is used to obtain several remote sensing images of soil in advance, and for each remote sensing image, set a real result corresponding to the remote sensing image to generate training data, where the real result is the real damage type corresponding to the remote sensing image, and use several remote sensing images and the training data consisting of the real results corresponding to the remote sensing images to train and generate an early warning model;

[0026] The first early warning module is used to continuously obtain remote sensing images of soil, input different remote sensing images into the early warning model in sequence, and the early warning model outputs first early warning results respectively. The first early warning results are different damage types corresponding to the remote sensing images and the reliability values ​​corresponding to the different damage types, and judge whether the situation corresponding to the new damage type of the remote sensing image occurs. If yes, the second early warning module processes it. If not, it judges whether the end condition is met. If so, all processing is terminated. If not, the current module continues to process it.

[0027] The second early warning module is used to continuously obtain remote sensing images of soil, input different remote sensing images into the backup early warning model in sequence, and generate second early warning results according to the output results of the backup early warning model. The second early warning result is a damage type corresponding to the remote sensing image. At the same time, a number of first training data are generated. Based on all the original training data and a number of first training data, the early warning model is retrained. After the retraining is completed, it is processed by the first early warning module.

[0028] The present invention also provides a storage medium, wherein the storage medium stores program instructions, wherein when the program instructions are executed, the device where the storage medium is located is controlled to execute any one of the above methods.

[0029] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0030] In the present invention, firstly, several remote sensing images of soil are acquired in advance, and for each remote sensing image, a real result corresponding to the remote sensing image is set to generate training data, and a warning model is generated by training using several training data consisting of remote sensing images and real results corresponding to the remote sensing images. Secondly, remote sensing images of soil are continuously acquired, and different remote sensing images are sequentially input into the warning model, and the warning model outputs first warning results respectively, and it is judged whether the situation corresponding to the new damage type of the remote sensing image occurs. If yes, the subsequent processing is continued. Finally, remote sensing images of soil are continuously acquired, and different remote sensing images are sequentially input into the backup warning model, and second warning results are respectively generated according to the output results of the backup warning model, and several first training data are generated at the same time. Based on all the original training data and several first training data, the warning model is retrained, and after the retraining is completed, the warning model is continued to be used for warning processing. Through the present invention, not only the execution of the warning method will not be interrupted, the execution efficiency of the warning method can be ensured, but also a warning method with high warning accuracy can be provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A flowchart of the steps of the comprehensive early warning method for soil damage of the present invention;

[0032] Figure 2 This is a structural diagram of the integrated early warning platform for soil damage of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0034] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0035] The present invention provides Figure 1 The comprehensive early warning method for soil damage shown is mainly implemented by executing the following steps:

[0036] S1. Acquire several remote sensing images of soil in advance, and for each remote sensing image, set a real result corresponding to the remote sensing image to generate training data, where the real result is the real damage type corresponding to the remote sensing image, and use several training data consisting of remote sensing images and the real results corresponding to the remote sensing images to train and generate an early warning model;

[0037] S2, continuously acquiring remote sensing images of soil, inputting different remote sensing images into the early warning model in sequence, and the early warning model outputting first early warning results respectively, the first early warning results being different damage types corresponding to the remote sensing images, and reliability values ​​corresponding to the different damage types, and judging whether the situation corresponding to the new damage type of the remote sensing image occurs, if yes, proceeding to the next step, if no, judging whether the end condition is met, if yes, ending all steps, if not, repeating this step;

[0038] S3. Continuously acquire remote sensing images of soil, input different remote sensing images into the backup warning model in sequence, and generate second warning results according to the output results of the backup warning model. The second warning result is a damage type corresponding to the remote sensing image. Meanwhile, generate a number of first training data, and retrain the warning model based on all the original training data and a number of first training data. After the retraining is completed, jump to S2 to continue execution.

[0039] Specifically, in the prior art, a comprehensive early warning of soil damage can be achieved based on the analysis of remote sensing images of soil. The early warning method includes: first, obtaining a high-resolution remote sensing image of soil, which is obtained by remote sensing equipment carried by an aircraft or an unmanned aerial vehicle, and the remote sensing image of soil includes the spectral characteristics of soil; second, performing radiation correction, atmospheric correction, geometric correction, etc. on the remote sensing image of soil; finally, inputting the remote sensing image of soil into a trained early warning model, which can be a neural network model, to identify the type of damage corresponding to the remote sensing image of soil, that is, the early warning result, such as heavy metal pollution, organic pollutants, salinization, etc. However, when a new type of damage appears, the original early warning model needs to be retrained, and the early warning process must be suspended during the training period, so the execution efficiency of the early warning method is relatively low.

[0040] In order to solve the above technical problems, the present invention firstly obtains multiple remote sensing images of soil in advance, and the real results corresponding to the multiple remote sensing images are known. The remote sensing images and the corresponding real results are used to form training data respectively. The early warning model is generated according to different training data, so that the remote sensing images are input into the early warning model, and the early warning model outputs different damage types corresponding to the remote sensing images, and the reliability values ​​corresponding to the different damage types, wherein the real results refer to the real damage types corresponding to the remote sensing images. Secondly, the remote sensing images of soil are continuously obtained, for example, the remote sensing images of soil are obtained once every preset fixed time interval, and different remote sensing images are sequentially input into the early warning model, and the early warning model outputs the first early warning results respectively, wherein the first early warning results refer to the different damage types corresponding to the remote sensing images, and the reliability values ​​corresponding to the different damage types, and the damage type corresponding to the maximum reliability value is usually used as the final early warning result. While the early warning model continuously processes different remote sensing images, it also determines whether the remote sensing image corresponds to a new damage type. If yes, proceed to the next step. If not, determine whether the end condition is met. If so, end all steps. If not, repeat this step. The end condition is, for example, that there is no need to warn of soil damage. Finally, remote sensing images of the soil are continuously acquired. The remote sensing images acquired at this time may correspond to new types of damage. Different remote sensing images are sequentially input into the backup warning model, and second warning results are generated according to the output results of the backup warning model. The second warning result refers to a type of damage corresponding to the remote sensing image. While the backup warning model continuously processes different remote sensing images, it also generates several first training data. The warning model is retrained based on all the original training data and several first training data. This is done because when the remote sensing image corresponds to a new type of damage, the initially trained warning model cannot adapt to this situation. After the retraining is completed, the previous step is jumped to continue execution. In other words, the warning model is still used for processing because the warning accuracy of the warning model is higher.

[0041] Through the above method, when the remote sensing image corresponds to a new type of damage, although the early warning model cannot be used because it cannot adapt to the situation, the backup early warning model can continue to be used to process the remote sensing image without interrupting the execution of the early warning method, thereby improving the execution efficiency of the early warning method.

[0042] Furthermore, the early warning model consists of a first-layer early warning model, a second-layer early warning model, and a third-layer early warning model, wherein the image element values ​​of different image elements in the remote sensing image are input into the first-layer early warning model for calculation, the output result of the first-layer early warning model is input into the second-layer early warning model for recalculation, the output result of the second-layer early warning model is further input into the third-layer early warning model for final calculation, and the third-layer early warning model outputs the first warning result.

[0043] Specifically, the structure of the early warning model is introduced here. The early warning model includes a first-layer early warning model, a second-layer early warning model, and a third-layer early warning model. The processing of a remote sensing image is taken as an example to illustrate the role of early warning models at different layers. The image element values ​​of different image elements in the remote sensing image are input into the first-layer early warning model for calculation, and the output result of the first-layer early warning model is input into the second-layer early warning model for recalculation. The output result of the second-layer early warning model is further input into the third-layer early warning model for final calculation. The third-layer early warning model outputs the above-mentioned first warning result. It should be noted that before the third-layer early warning model, a layer of early warning model can be added to continue calculating the output result of the second-layer early warning model, and then its output result is input into the third-layer early warning model for final calculation. The early warning model can be Convolutional Neural Networks.

[0044] Further, judging whether the remote sensing image corresponds to a new type of damage occurs includes the following steps:

[0045] S21, generating different data records in sequence according to the order in which the warning model outputs the first warning result, the data records including the first warning result and a representative value corresponding to the first warning result;

[0046] S22, sorting the different data records according to the order in which the different data records are generated, calculating the average of the representative values ​​in the data records of the preset first number that are ranked first to obtain an average representative value, and determining whether the average representative value is less than a preset average representative value threshold. If so, determining that the remote sensing image corresponds to a new type of damage that has occurred, and ending all steps. If not, proceeding to the next step.

[0047] S23, for the data records that are ranked later, determine the data record that is a preset first number minus one number before the data record, calculate the average of the representative value in the data record and the representative values ​​in the determined different data records to obtain an average representative value, and determine whether the average representative value is less than a preset average representative value threshold. If so, determine that the remote sensing image corresponds to a new type of damage that has occurred, and end all steps. If not, determine whether all data records have been processed. If so, end all steps. If not, repeat this step.

[0048] Furthermore, the representative numerical value corresponding to the above-mentioned first warning result is the maximum reliable value in the above-mentioned first warning result.

[0049] Specifically, the process of judging whether the remote sensing image corresponds to a new type of damage is introduced here. First, different data records are generated in sequence according to the order in which the warning model outputs the first warning result. Each data record includes the first warning result and the representative value corresponding to the first warning result. The representative value corresponding to the first warning result is the maximum reliable value in the first warning result. Secondly, different data records are sorted according to the order in which different data records are generated, and for the data records with the preset first number that are ranked first, the average value of the representative values ​​in the data records with the preset first number is calculated to obtain the average representative value. The first number is set in the actual application scenario. Then, it is checked whether the average representative value is less than the preset average representative value threshold. The average representative value threshold is also set in the actual application scenario. If so, it is determined that the remote sensing image corresponds to a new type of damage. The situation has occurred and all steps are terminated. If not, proceed to the next step. Finally, for the data records that are sorted later, for ease of understanding, for example, the sequence of data records is: data record 1, data record 2, data record 3, until data record N, assuming that the first number is 5, then the data record that is sorted later for the first time is data record 6, determine the data record that is the preset first number minus one number before the data record, that is, determine data records 2 to data records 5, calculate the average of the representative value in the data record and the representative values ​​in the determined different data records to obtain the average representative value, and determine whether the average representative value is less than the preset average representative value threshold. If so, it is determined that the remote sensing image corresponds to a new type of damage. End all steps. If not, determine whether all data records have been processed above. If so, end all steps. If not, repeat this step, that is, continue to perform the processing in this step for data record 7. Through the above method, it is possible to automatically and accurately determine whether the remote sensing image corresponds to a new type of damage according to the first warning result continuously output by the warning model.

[0050] Furthermore, while determining that a new type of damage corresponding to the remote sensing image has occurred, a data record of a preset first number used to calculate the average representative value is recorded, different remote sensing images corresponding to the data record of the preset first number are determined, and the different remote sensing images are marked as problematic remote sensing images.

[0051] Specifically, when it is determined that a new type of damage has occurred corresponding to a remote sensing image, a data record of a preset first number used to calculate an average representative value can also be recorded, where the average representative value refers to an average representative value that is less than the above-mentioned average representative value threshold, and different remote sensing images corresponding to the data record of the preset first number are determined, and the different remote sensing images are marked as problematic remote sensing images. Because a new type of damage has occurred corresponding to the remote sensing image, the early warning model cannot adapt to the situation, so the first early warning result corresponding to the problematic remote sensing image may be wrong, and the first early warning result corresponding to the problematic remote sensing image can then be manually corrected.

[0052] Furthermore, the backup warning model includes a first-layer backup warning model and a second-layer backup warning model, wherein the first-layer backup warning model further includes a first sub-model and a second sub-model in parallel, and the structures of the first sub-model and the second sub-model are the same as the structure composed of the first-layer warning model and the second-layer warning model, and the output results of the first sub-model and the second sub-model are simultaneously input into the second-layer backup warning model, and the second-layer backup warning model calculates the similarity value between the output results of the first sub-model and the output results of the second sub-model.

[0053] Specifically, the structure of the backup warning model is introduced here. The backup warning model consists of a first-layer backup warning model and a second-layer backup warning model. The first-layer backup warning model consists of a first sub-model and a second sub-model in parallel. The first sub-model and the second sub-model are independent of each other. The structure of the first sub-model is actually a structure composed of the above-mentioned first-layer warning model and the above-mentioned second-layer warning model. The structure of the second sub-model is the same as that of the first sub-model. The output results of the first sub-model and the second sub-model are simultaneously input into the second-layer backup warning model. The second-layer backup warning model calculates the similarity value between the output result of the first sub-model and the output result of the second sub-model, and then outputs the similarity value.

[0054] Furthermore, before inputting different remote sensing images into the backup warning model in sequence, it also includes: generating a plurality of second training data, the real results in the second training data are different from the real results in all the original training data, and making the plurality of second training data and all the original training data together form a training data set.

[0055] Specifically, before different remote sensing images are input into the backup warning model in sequence, a process of generating a training data set is also included. However, since there is no need to train the backup warning model, the training data set is not used to train the backup warning model. The role of the training data set will be described below. The process of generating a training data set includes: first generating multiple second training data, the second training data also includes remote sensing images, and real results corresponding to the remote sensing images, but the real results in the second training data are different from the real results in all the original training data, that is, the remote sensing images in the second training data correspond to new types of damage, and then the multiple second training data and all the original training data are combined to form a training data set.

[0056] Furthermore, the remote sensing image is input into the backup warning model, and a second warning result is generated according to the output result of the backup warning model, including the following steps:

[0057] S311, based on each training data in the training data set, the remote sensing image is input into the first sub-model, and the remote sensing image in the training data is input into the second sub-model, and the second-layer backup warning model outputs a similarity value;

[0058] S312. Sort the remote sensing images in different training data in the training data set in descending order of the corresponding similarity values, determine the remote sensing images with a preset second number of remote sensing images ranked first, and use the true result with the largest total number of corresponding remote sensing images as the second warning result.

[0059] Specifically, here, taking the processing of a remote sensing image as an example, the process of generating the second warning result of this remote sensing image is explained. First, the remote sensing image to be processed is input into the first sub-model of the first-layer backup warning model, and at the same time, a remote sensing image in a training data set is input into the second sub-model of the first-layer backup warning model, so that the second-layer backup warning model outputs the similarity value between the output result of the first sub-model and the output result of the second sub-model. For other training data in the training data set, the same method is repeated. Secondly, according to the corresponding similarity value from large to small, the remote sensing images in different training data in the training data set are sorted, and the remote sensing images with a preset second number that are ranked first are determined. These remote sensing images all correspond to real results. The second number is set in the actual application scenario. Based on the remote sensing images of the second number, the real results with the largest total number of corresponding remote sensing images are regarded as the second warning results corresponding to the remote sensing images to be processed. Through the above method, while retraining the warning model, the remote sensing image can still be processed.

[0060] Furthermore, the remote sensing images in the first training data include: remote sensing images whose corresponding second warning results are different from the real results in all the original training data, and remote sensing images whose similarity values ​​with the second training data exceed a preset similarity value threshold.

[0061] Specifically, while the backup warning model processes remote sensing images, it is necessary to generate multiple first training data. The multiple first training data and all the original training data are used together to retrain the warning model. Here, how the first training data is generated is introduced. The remote sensing images in the first training data include remote sensing images whose corresponding second warning results are different from the actual results in all the original training data, that is, remote sensing images whose corresponding second warning results are new types of damage. The remote sensing images in the first training data also include remote sensing images whose similarity values ​​with the second training data exceed a preset similarity value threshold. The similarity value can be calculated using existing technology, which will not be repeated here. The similarity value threshold is set in the actual application scenario. In addition, the actual result in the first training data refers to the actual type of damage corresponding to the remote sensing image in the first training data.

[0062] Furthermore, after the retraining is completed, the following steps are also included:

[0063] S321, collecting a number of first warning results output by the warning model during the retraining process, recording the final warning result of the remote sensing image input to the warning model as the damage type corresponding to the maximum reliable value in the corresponding first warning result, and recording the maximum reliable value in the corresponding first warning result;

[0064] S322, dividing different remote sensing images corresponding to the same final warning result into a group, sorting the different remote sensing images in each group in ascending order of the corresponding maximum reliability values, and selecting the remote sensing image with a preset third number at the top of the sorting;

[0065] S323. Check whether the final warning results of the preset third number of remote sensing images in each group are correct, correct the incorrect final warning results, and generate a number of third training data based on the corresponding remote sensing images whose final warning results have been corrected, and use the number of third training data to train the warning model again.

[0066] Specifically, in order to further improve the warning accuracy of the early warning model, first, multiple first warning results output by the early warning model during the retraining process are collected, and the final warning result of the remote sensing image input into the early warning model is determined to be the damage type corresponding to the maximum reliable value in the corresponding first warning result, and the maximum reliable value in the corresponding first warning result is recorded at the same time. Secondly, different remote sensing images corresponding to the same final warning result are divided into a group. In each group, different remote sensing images are sorted in ascending order according to the corresponding maximum reliable value, and the remote sensing images with the preset third number that is ranked first are selected. The third number is set in the actual application scenario. Finally, check whether the final warning results of the preset third number of remote sensing images in each group are correct, and correct the incorrect final warning results. On this basis, generate multiple third training data according to the multiple remote sensing images whose corresponding final warning results in all groups have been corrected. The real results in the third training data refer to the corresponding corrected final warning results. Use the multiple third training data to train the warning model again. It should be noted that it is also possible to generate multiple fourth training data according to the multiple remote sensing images whose corresponding final warning results in all groups have not been corrected, and use the multiple third training data and the multiple fourth training data to train the warning model again.

[0067] According to another aspect of the embodiment of the present invention, Figure 2 As shown, a comprehensive early warning platform for soil damage is also provided, including a training module, a first early warning module, and a second early warning module, which are used to implement the comprehensive early warning method for soil damage as described above.

[0068] The specific functions of each module are as follows:

[0069] A training module is used to obtain several remote sensing images of soil in advance, and for each remote sensing image, set a real result corresponding to the remote sensing image to generate training data, where the real result is the real damage type corresponding to the remote sensing image, and use several remote sensing images and the training data consisting of the real results corresponding to the remote sensing images to train and generate an early warning model;

[0070] The first early warning module is used to continuously obtain remote sensing images of soil, input different remote sensing images into the early warning model in sequence, and the early warning model outputs first early warning results respectively. The first early warning results are different damage types corresponding to the remote sensing images and the reliability values ​​corresponding to the different damage types, and judge whether the situation corresponding to the new damage type of the remote sensing image occurs. If yes, the second early warning module processes it. If not, it judges whether the end condition is met. If so, all processing is terminated. If not, the current module continues to process it.

[0071] The second early warning module is used to continuously obtain remote sensing images of soil, input different remote sensing images into the backup early warning model in sequence, and generate second early warning results according to the output results of the backup early warning model. The second early warning result is a damage type corresponding to the remote sensing image. At the same time, a number of first training data are generated. Based on all the original training data and a number of first training data, the early warning model is retrained. After the retraining is completed, it is processed by the first early warning module.

[0072] According to another aspect of an embodiment of the present invention, a storage medium is provided, wherein the storage medium stores program instructions, wherein when the program instructions are executed, a device where the storage medium is located is controlled to execute any one of the above methods.

[0073] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have strict order restrictions, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0074] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0075] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0077] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A comprehensive early warning method for soil damage, characterized in that: The method comprises the following steps: S1. Acquire several remote sensing images of soil in advance, and for each remote sensing image, set a real result corresponding to the remote sensing image to generate training data, where the real result is the real damage type corresponding to the remote sensing image, and use several training data consisting of remote sensing images and the real results corresponding to the remote sensing images to train and generate an early warning model; S2, continuously acquiring remote sensing images of soil, inputting different remote sensing images into the early warning model in sequence, and the early warning model outputting first early warning results respectively, wherein the first early warning results are different damage types corresponding to the remote sensing images, and reliability values ​​corresponding to the different damage types respectively, and judging whether the situation corresponding to the new damage type of the remote sensing image occurs, if yes, proceeding to the next step, if no, judging whether the end condition is satisfied, if yes, ending all steps, if not, repeating this step; S3. Continuously acquire remote sensing images of soil, input different remote sensing images into the backup warning model in sequence, generate second warning results according to the output results of the backup warning model, the second warning result is a damage type corresponding to the remote sensing image, and generate a number of first training data at the same time. Retrain the warning model based on all the original training data and the first training data. After the retraining is completed, jump to S2 to continue execution.

2. The method according to claim 1, characterized in that The early warning model consists of a first-layer early warning model, a second-layer early warning model, and a third-layer early warning model, wherein the image element values ​​of different image elements in the remote sensing image are input into the first-layer early warning model for calculation, the output result of the first-layer early warning model is input into the second-layer early warning model for recalculation, the output result of the second-layer early warning model is further input into the third-layer early warning model for final calculation, and the third-layer early warning model outputs the first warning result.

3. The method according to claim 1, characterized in that Determining whether a new type of damage corresponding to the remote sensing image occurs includes the following steps: S21, generating different data records in sequence according to the order in which the warning model outputs the first warning results, the data records including the first warning results and representative values ​​corresponding to the first warning results; S22, sorting different data records according to the order in which different data records are generated, calculating the average of the representative values ​​in the data records of the preset first number to obtain an average representative value for the data records of the preset first number that are ranked first, and determining whether the average representative value is less than a preset average representative value threshold value, if so, determining that the remote sensing image corresponds to a new type of damage has occurred, and ending all steps, if not, proceeding to the next step; S23. For the data records that are sorted later, determine the data record whose number before the data record is the preset first number minus one, calculate the average of the representative value in the data record and the representative values ​​in the determined different data records to obtain the average representative value, and determine whether the average representative value is less than the preset average representative value threshold. If so, determine that a new type of damage corresponding to the remote sensing image has occurred, and terminate all steps. If not, determine whether all data records have been processed. If so, terminate all steps. If not, repeat this step.

4. The method according to claim 2, characterized in that: The backup warning model includes a first-layer backup warning model and a second-layer backup warning model, wherein the first-layer backup warning model further includes a first sub-model and a second sub-model in parallel, the structures of the first sub-model and the second sub-model are the same as the structure composed of the first-layer warning model and the second-layer warning model, the output results of the first sub-model and the second sub-model are simultaneously input into the second-layer backup warning model, and the second-layer backup warning model calculates the similarity value between the output result of the first sub-model and the output result of the second sub-model.

5. The method according to claim 4, characterized in that Before inputting different remote sensing images into the backup warning model in sequence, the method also includes: generating a plurality of second training data, wherein the true results in the second training data are different from the true results in all the original training data, and making the plurality of second training data and all the original training data together form a training data set.

6. The method according to claim 5, characterized in that Inputting the remote sensing image into the backup warning model, and generating a second warning result according to the output result of the backup warning model, comprises the following steps: S311, based on each training data in the training data set, inputting the remote sensing image into the first sub-model, and simultaneously inputting the remote sensing image in the training data into the second sub-model, and the second-layer backup warning model outputs a similarity value; S312. Sort the remote sensing images in different training data in the training data set in descending order of the corresponding similarity values, determine the remote sensing images with a preset second number of remote sensing images ranked first, and use the true result with the largest total number of corresponding remote sensing images as the second warning result.

7. The method according to claim 5, characterized in that The remote sensing images in the first training data include: remote sensing images whose corresponding second warning results are different from the real results in all the original training data, and remote sensing images whose similarity values ​​with the second training data exceed a preset similarity value threshold.

8. The method according to claim 1, characterized in that After the retraining is completed, the following steps are also included: S321, collecting a plurality of first warning results output by the warning model during the retraining process, recording the final warning result of the remote sensing image input to the warning model as the damage type corresponding to the maximum reliable value in the corresponding first warning result, and recording the maximum reliable value in the corresponding first warning result; S322, dividing different remote sensing images corresponding to the same final warning result into a group, sorting the different remote sensing images in each group in ascending order of the corresponding maximum reliability values, and selecting the remote sensing image with a preset third number at the top of the sorting; S323, checking whether the final warning results of the preset third number of remote sensing images in each group are correct, correcting the incorrect final warning results, and generating a number of third training data based on a number of remote sensing images whose corresponding final warning results have been corrected, and using the number of third training data to train the warning model again.

9. A comprehensive early warning platform for soil damage, used to implement the method according to any one of claims 1 to 8, characterized in that: Includes the following modules: A training module is used to obtain several remote sensing images of soil in advance, and for each remote sensing image, set a real result corresponding to the remote sensing image to generate training data, where the real result is the real damage type corresponding to the remote sensing image, and use several remote sensing images and the training data consisting of the real results corresponding to the remote sensing images to train and generate an early warning model; The first early warning module is used to continuously obtain remote sensing images of soil, input different remote sensing images into the early warning model in sequence, and the early warning model outputs first early warning results respectively. The first early warning results are different damage types corresponding to the remote sensing images and the reliability values ​​corresponding to the different damage types, and judge whether the situation corresponding to the new damage type of the remote sensing image occurs. If yes, the second early warning module processes it. If not, it judges whether the end condition is met. If so, all processing is terminated. If not, the current module continues to process it. The second early warning module is used to continuously obtain remote sensing images of soil, input different remote sensing images into the backup early warning model in sequence, and generate second early warning results according to the output results of the backup early warning model. The second early warning result is a damage type corresponding to the remote sensing image. At the same time, a number of first training data are generated. Based on all the original training data and a number of first training data, the early warning model is retrained. After the retraining is completed, it is processed by the first early warning module.

10. A storage medium, characterized in that: The storage medium stores program instructions, wherein when the program instructions are executed, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 8.

Images