Soil ecological restoration detection method and system based on artificial intelligence

Abstract

The artificial intelligence-based soil ecological restoration detection method and system provided in this application optimizes the network parameters of a first soil restoration status detection network and a second soil restoration status detection network based on comparison results, thereby obtaining trained first and second soil restoration status detection networks. The system receives soil ecological restoration data to be processed, extracts characteristic information of the soil ecological restoration data to be processed through the trained first soil restoration status detection network, and performs soil ecological restoration data detection to obtain detection results. Multi-label training is performed on the two networks using inclusion labels between soil object labels and soil ecological restoration data with similar soil ecological restoration data characteristics, combined with training soil ecological restoration data. The network parameters of the two networks are optimized based on the comparison results, thereby improving the working efficiency of the trained networks, thereby improving the accuracy of information detection and ensuring the reliability of ecological restoration.

Description

Technical Field

[0001] The present application relates to the field of soil data detection technology, and more specifically, to a soil ecological restoration detection method and system based on artificial intelligence. Background Art

[0002] Soil refers to the loose layer of material on the Earth's surface, composed of various granular minerals, organic matter, water, air, and microorganisms, capable of supporting plant growth. Soil is composed of minerals formed from weathered rocks, organic matter from the decomposition of plant and animal remains, microbial remains, soil organisms (solids), water (liquids), air (vapors), and oxidized humus.

[0003] Solid matter includes soil minerals, organic matter, and nutrients derived from microorganisms through light-induced sterilization. Liquid matter primarily refers to soil moisture. Gas is the air trapped in soil pores. These three types of soil matter form a contradictory unity. They are interconnected and mutually restrictive, providing the necessary living conditions for crops and forming the material basis for soil fertility.

[0004] In the actual process, after soil remediation, the soil needs to be brought back to the laboratory for testing. Various information about the soil needs to be tested, which requires a lot of time and labor costs. Therefore, the soil testing cycle is very long and may also be inaccurate (the reason for the inaccuracy is that the water content and microorganisms of the soil are difficult to accurately detect during the process of bringing the soil back to the laboratory). Therefore, a technical solution is urgently needed to improve the above technical problems. Summary of the Invention

[0005] In order to improve the technical problems existing in related technologies, this application provides a soil ecological restoration detection method and system based on artificial intelligence.

[0006] In a first aspect, a soil ecological restoration detection method based on artificial intelligence is provided, and a soil ecological restoration detection method based on artificial intelligence is provided, the method comprising: loading training soil ecological restoration data, soil object labels and inclusion labels into a first soil restoration state detection network and a second soil restoration state detection network for training, the inclusion labels being used to calibrate training soil ecological restoration data whose feature similarity of soil ecological restoration data is greater than a preset specified value; obtaining a comparison result between the predictions of the first soil restoration state detection network and the second soil restoration state detection network; optimizing the network parameters of the first soil restoration state detection network and the second soil restoration state detection network based on the comparison result, and obtaining the trained first soil restoration state detection network and the second soil restoration state detection network. Second soil remediation status detection network; wherein, the second soil remediation status detection network is used to load the important soil remediation indicators extracted by the second soil remediation status detection network into the first soil remediation status detection network, so as to realize mutual feature extraction and learning between the first soil remediation status detection network and the second soil remediation status detection network, thereby obtaining the trained first soil remediation status detection network and the second soil remediation status detection network, so that the trained first soil remediation status detection network has the important soil remediation indicators extracted by the features of the second soil remediation status detection network; receiving soil ecological remediation data to be processed, extracting feature information of the soil ecological remediation data to be processed through the trained first soil remediation status detection network, performing soil ecological remediation data detection, and obtaining detection results.

[0007] Furthermore, the artificial intelligence-based soil ecological restoration detection method also includes: obtaining training soil ecological restoration data, the training soil ecological restoration data having soil object labels; determining preset soil ecological restoration data corresponding to each soil object label type; classifying the preset soil ecological restoration data corresponding to different soil object label types, and classifying the preset soil ecological restoration data whose soil ecological restoration data feature similarity is greater than a preset specified value; and generating inclusion labels for the classified preset soil ecological restoration data.

[0008] Furthermore, the step of classifying and processing the preset soil ecological restoration data corresponding to different soil object label types, and classifying the preset soil ecological restoration data whose soil ecological restoration data feature similarity is greater than a preset specified value, includes: extracting soil ecological restoration data features of the preset soil ecological restoration data corresponding to each soil object label type; classifying and processing the soil ecological restoration data features of the preset soil ecological restoration data corresponding to each soil object label type by a classification method; and classifying the preset soil ecological restoration data whose soil ecological restoration data feature similarity is greater than a preset specified value.

[0009] Furthermore, the step of loading the training soil ecological restoration data, the soil object labels and the included labels into the first soil remediation status detection network and the second soil remediation status detection network for training includes: optimizing the training soil ecological restoration data; loading the optimized training soil ecological restoration data, the soil object labels and the included labels into the first soil remediation status detection network and the second soil remediation status detection network for training.

[0010] Furthermore, the step of obtaining a comparison result between the predictions of the first soil remediation state detection network and the second soil remediation state detection network includes: calculating the comparison result between the predictions of the first soil remediation state detection network and the second soil remediation state detection network through a comparison result calculation strategy, and constructing a quantitative evaluation thread corresponding to the comparison result.

[0011] Furthermore, the step of optimizing the network parameters of the first soil remediation state detection network and the second soil remediation state detection network in combination with the comparison result to obtain the trained first soil remediation state detection network and the second soil remediation state detection network includes: performing splicing training on the quantitative evaluation thread in combination with the initial loss value thread of the first soil remediation state detection network to obtain the trained first soil remediation state detection network; and performing splicing training on the quantitative evaluation thread in combination with the initial loss value thread of the second soil remediation state detection network to obtain the trained second soil remediation state detection network.

[0012] Furthermore, the step of receiving the soil ecological restoration data to be processed, extracting characteristic information of the soil ecological restoration data to be processed through the trained first soil restoration status detection network to perform soil ecological restoration data detection, and obtaining the detection result includes: receiving predicted soil ecological restoration data, identifying the data to be processed in the predicted soil ecological restoration data through the target parsing network; reading the data to be processed to generate the soil ecological restoration data to be processed; loading the soil ecological restoration data to be processed into the trained first soil restoration status detection network, extracting characteristic information after global convolution to perform soil ecological restoration data detection, and obtaining the detection result.

[0013] Furthermore, the step of extracting the feature information after global convolution to perform soil ecological restoration data detection and obtain a detection result includes: extracting feature information of a preset layer after global convolution; determining multiple directories to which the feature information belongs; obtaining a preset number of target soil ecological restoration data under each directory sorted from small to large in similarity with the feature information; loading the preset number of target soil ecological restoration data under each directory into the trained first soil restoration status detection network, respectively, and extracting the feature information of the preset layer after global convolution of the preset number of target soil ecological restoration data under each directory; calculating the splicing value of the distinguishing data between the feature information of each target soil ecological restoration data under the same directory and the feature information of the soil ecological restoration data to be processed; taking the directory with the smallest splicing value as the target directory, and obtaining the target soil ecological restoration data with the smallest distinguishing data under the target directory as the soil ecological restoration data detection result.

[0014] Furthermore, the step of calculating the concatenation value of the distinguishing data between the characteristic information of each target soil ecological restoration data in the same directory and the characteristic information of the soil ecological restoration data to be processed includes: calculating the distinguishing data between the characteristic information of each target soil ecological restoration data in the same directory and the characteristic information of the soil ecological restoration data to be processed; sorting the target soil ecological restoration data in the same directory in descending order of the distinguishing data; performing weighted processing on the distinguishing data of each target soil ecological restoration data in the same directory according to the sorting situation; and calculating the concatenation value of the distinguishing data of each target soil ecological restoration data in the same directory after the weighted processing.

[0015] In a second aspect, an artificial intelligence-based soil ecological restoration detection system is provided, comprising a processor and a memory that communicate with each other, wherein the processor is used to read a computer program from the memory and execute it to implement the above method.

[0016] The artificial intelligence-based soil ecological restoration detection method and system provided in the embodiment of the present application are as follows: the training soil ecological restoration data, the soil object labels and the inclusion labels are loaded into the first soil restoration state detection network and the second soil restoration state detection network for training; the comparison result between the predictions of the first soil restoration state detection network and the second soil restoration state detection network is obtained; the network parameters of the first soil restoration state detection network and the second soil restoration state detection network are optimized according to the comparison result to obtain the trained first soil restoration state detection network and the second soil restoration state detection network; the soil ecological restoration data to be processed is received, and the feature information of the soil ecological restoration data to be processed is extracted through the trained first soil restoration state detection network to perform soil ecological restoration data detection and obtain the detection result. Therefore, multi-label training is performed on the two networks by combining the inclusion labels between the soil ecological restoration data with similar soil object labels and soil ecological restoration data features with the training soil ecological restoration data, and the network parameters of the two networks are optimized according to the comparison result to improve the working efficiency of the trained network, thereby improving the accuracy of information detection and ensuring the reliability of ecological restoration. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A flowchart of an artificial intelligence-based soil ecological restoration detection method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to better understand the above technical solution, the technical solution of the present application is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0020] See also Figure 1 , shows a soil ecological restoration detection method based on artificial intelligence, which may include the technical solutions described in the following steps 101-104.

[0021] In step 101, the training soil ecological restoration data, the soil object labels and the inclusion labels are loaded into the first soil restoration status detection network and the second soil restoration status detection network for training respectively. The inclusion labels are used to calibrate the training soil ecological restoration data whose feature similarity of the soil ecological restoration data is greater than a preset specified value.

[0022] Furthermore, the soil object label can be understood as the soil identification of a certain area, including information such as soil mineral content, soil fertility, and soil water retention rate.

[0023] For example, soil ecological restoration data can be obtained from testing equipment.

[0024] The soil remediation status monitoring network can include artificial intelligence (AI), which is the theory, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0025] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, sortable storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0026] The first soil remediation status detection network is used to calculate soil material information, and the second soil remediation status detection network is used to calculate information about microorganisms in the soil.

[0027] The embodiment of the present application applies multi-label classification technology, that is, the training soil ecological restoration data can have one or two labels, and the first soil restoration status detection network can be a soil restoration status detection network with a deeper network layer and a narrower network channel.

[0028] Furthermore, the training soil ecological restoration data, the soil object labels and the inclusion labels are loaded into the first soil restoration status detection network and the second soil restoration status detection network for training respectively. Since the inclusion labels are introduced on the premise of the soil object labels, the learned first soil restoration status detection network and the second soil restoration status detection network can not only learn how to identify the soil object types of the training soil ecological restoration data, but also strengthen the feature correlation between similar soil ecological restoration data in the same directory.

[0029] In terms of a possible implementation embodiment, the step of loading the training soil ecological restoration data, the soil object labels, and the labels into the first soil remediation status detection network and the second soil remediation status detection network for training may include the following content.

[0030] (1) Optimize the training soil ecological restoration data.

[0031] (2) The optimized training soil ecological restoration data, the soil object labels and the included labels are loaded into the first soil restoration status detection network and the second soil restoration status detection network for training respectively.

[0032] Among them, the training soil ecological restoration data can be subjected to random degree of correlation, similarity, interaction degree, and transformation optimization processing to enhance the richness of the training set. The optimized training soil ecological restoration data, the soil object labels, and the included labels are loaded into the first soil remediation status detection network and the second soil remediation status detection network for training, which can enhance the robustness of the soil remediation status detection network.

[0033] In step 102, a comparison result between the predictions of the first soil remediation status detection network and the second soil remediation status detection network is obtained.

[0034] Among them, since the network structures of the first soil remediation status detection network and the second soil remediation status detection network are inconsistent, the important soil remediation indicators for feature extraction are different. The embodiment of the present application can draw on the advantages of the first soil remediation status detection network and the second soil remediation status detection network, that is, obtain a comparison result between the predictions of the first soil remediation status detection network and the second soil remediation status detection network for the same training soil ecological remediation data. The lower the comparison result, the closer the predictions of the first soil remediation status detection network and the second soil remediation status detection network are, and the higher the comparison result, the network parameters of the first soil remediation status detection network and the second soil remediation status detection network can be subsequently tuned according to the comparison result.

[0035] In one possible implementation embodiment, the step of obtaining a comparison result between the predictions of the first soil remediation status detection network and the second soil remediation status detection network may include: calculating the comparison result between the predictions of the first soil remediation status detection network and the second soil remediation status detection network using a comparison result calculation strategy, and constructing a quantitative evaluation thread corresponding to the comparison result.

[0036] For example, the quantitative evaluation thread can be understood as a divergence loss method.

[0037] In step 103, the network parameters of the first soil remediation state detection network and the second soil remediation state detection network are optimized according to the comparison result to obtain the trained first soil remediation state detection network and the second soil remediation state detection network.

[0038] Among them, in order to enable the first soil remediation status detection network and the second soil remediation status detection network to learn each other's threads for the important soil remediation indicators extracted by features, the network parameters of the first soil remediation status detection network and the second soil remediation status detection network can be optimized by reverse conduction according to the comparison results, so as to realize that the important soil remediation indicators extracted by features of the first soil remediation status detection network are loaded into the second soil remediation status detection network, and similarly, the important soil remediation indicators extracted by features of the second soil remediation status detection network are loaded into the first soil remediation status detection network.

[0039] Furthermore, the comparison results between the predictions of the first soil remediation status detection network and the second soil remediation status detection network for the same training soil ecological restoration data can be further obtained. Since the network parameter optimization was performed before, the first soil remediation status detection network and the second soil remediation status detection network perform feature extraction and learning on each other, the comparison result will become smaller. Therefore, the comparison result will become smaller and smaller until convergence, that is, the first soil remediation status detection network and the second soil remediation status detection network learn from each other, and the trained first soil remediation status detection network and the second soil remediation status detection network are obtained, so that the trained first soil remediation status detection network not only processes the details of the soil ecological restoration data more accurately, but also has better feature extraction capabilities due to the learning of the important soil remediation indicators of the second soil remediation status detection network. Moreover, because of multi-label training, the first soil remediation status detection network can better determine the feature correlation between similar soil ecological restoration data in the same directory on the premise that it can identify the soil objects of the soil ecological restoration data.

[0040] In one possible implementation embodiment, the step of optimizing the network parameters of the first soil remediation status detection network and the second soil remediation status detection network based on the comparison results to obtain the trained first soil remediation status detection network and the second soil remediation status detection network may include the following content.

[0041] (1) The quantitative evaluation thread is combined with the initial loss value thread of the first soil remediation status detection network for splicing training to obtain the trained first soil remediation status detection network.

[0042] (2) The quantitative evaluation thread is combined with the initial loss value thread of the second soil remediation status detection network for splicing training to obtain the trained second soil remediation status detection network.

[0043] Among them, the quantitative evaluation thread is combined with the initial loss value thread of the first soil remediation status detection network for splicing training, so that the initial loss value thread of the first soil remediation status detection network can be combined with the quantitative evaluation thread for splicing training, that is, the first soil remediation status detection network can continue to learn the important soil remediation indicators of the second soil remediation status detection network under normal training conditions until the initial loss value thread and the quantitative evaluation thread converge at the same time, thereby obtaining the trained first soil remediation status detection network.

[0044] Furthermore, the quantitative evaluation thread is combined with the initial loss value thread of the second soil remediation state detection network for splicing training, so that the initial loss value thread of the second soil remediation state detection network can be combined with the quantitative evaluation thread for splicing training, that is, the second soil remediation state detection network can continue to learn the important soil remediation indicators of the first soil remediation state detection network under normal training conditions until the initial loss value thread and the quantitative evaluation thread converge at the same time, thereby obtaining the trained second soil remediation state detection network.

[0045] In step 104, the soil ecological restoration data to be processed is received, and the feature information of the soil ecological restoration data to be processed is extracted through the trained first soil restoration status detection network to perform soil ecological restoration data detection and obtain a detection result.

[0046] Among them, after the first soil remediation status detection network and the second soil remediation status detection network are trained, the predicted soil ecological restoration data uploaded by the user can be received. The predicted soil ecological restoration data can include data to be processed, and the data to be processed can be read to generate soil ecological restoration data to be processed.

[0047] In order to solve the above technical problems, the embodiment of the present application can extract the characteristic information of the soil ecological restoration data to be processed through the first soil remediation status detection network after multi-label training. The difference between this characteristic information and the existing technology is that due to the introduction of training containing labels, the target soil ecological restoration data under the same directory can be better distinguished, avoiding the situation where abnormal soil ecological restoration data detection is caused by incorrect label selection.

[0048] In terms of a possible implementation embodiment, the steps of receiving soil ecological restoration data to be processed, extracting characteristic information of the soil ecological restoration data to be processed through a trained first soil restoration status detection network, performing soil ecological restoration data detection, and obtaining detection results may include the following contents.

[0049] (1) Receive predicted soil ecological restoration data and identify the data to be processed in the predicted soil ecological restoration data through the target parsing network.

[0050] (2) Read the data to be processed and generate soil ecological restoration data to be processed.

[0051] (3) The soil ecological restoration data to be processed is loaded into the trained first soil restoration status detection network, and the feature information after global convolution is extracted to perform soil ecological restoration data detection to obtain the detection results.

[0052] Among them, predicted soil ecological restoration data is received, which can be soil ecological restoration data uploaded by users and containing data to be processed. The target parsing network is used to identify the data to be processed in the predicted soil ecological restoration data, operate on the data to be processed, read part of the data to be processed, and generate the soil ecological restoration data to be processed.

[0053] Furthermore, the soil ecological restoration data to be processed is loaded into the trained first soil restoration status detection network, and the feature information after global convolution is extracted. According to the feature information, the most similar first multiple soil ecological restoration data in each directory in the background library are selected, and the feature information of the first multiple soil ecological restoration data in each directory is calculated through the trained first soil restoration status detection network. The difference splicing value between the feature information of the first multiple soil ecological restoration data in each directory and the feature information of the soil ecological restoration data to be processed is calculated, and the label with the lowest difference splicing value is used as the detection label, and the soil ecological restoration data with the smallest difference in the detection label is displayed as the detection soil ecological restoration data.

[0054] In terms of a possible implementation embodiment, the step of extracting feature information after global convolution to perform soil ecological restoration data detection and obtain detection results may include the following content.

[0055] (1.1) Extract feature information of the preset layer after global convolution.

[0056] (1.2) Determine the multiple directories to which the characteristic information belongs.

[0057] (1.3) Obtain a preset number of target soil ecological restoration data in each directory sorted from small to large in similarity with the feature information.

[0058] (1.4) Loading a preset number of target soil ecological restoration data in each directory into the trained first soil restoration status detection network, respectively, extracting feature information of a preset layer after global convolution of the preset number of target soil ecological restoration data in each directory.

[0059] (1.5) Calculate the splicing value of the distinguishing data between the characteristic information of each target soil ecological restoration data and the characteristic information of the soil ecological restoration data to be processed in the same directory.

[0060] (1.6) The directory with the smallest splicing value is used as the target directory, and the target soil ecological restoration data with the smallest distinguishing data under the target directory is obtained as the soil ecological restoration data detection result.

[0061] Among them, the feature information of the preset layer after the global convolution is extracted through the trained first soil remediation status detection network, and multiple directories related to the feature information are determined, that is, multiple directories to which the feature information most likely belongs are determined.

[0062] Furthermore, a preset number of target soil ecological restoration data are detected in each directory based on feature information, sorted from small to large in similarity. The preset number of target soil ecological restoration data in each directory are loaded into the trained first soil restoration state detection network, and feature information of a preset layer of target soil ecological restoration data in each directory after global convolution is extracted. Therefore, the splicing value of the distinguishing data between the feature information of each target soil ecological restoration data in the same directory and the feature information of the soil ecological restoration data to be processed is calculated. This splicing value can reflect the correlation between the soil ecological restoration data to be processed and each directory. The directory with the smallest splicing value is used as the target directory with the greatest correlation, and the target soil ecological restoration data with the smallest distinguishing data under this target directory is obtained as the soil ecological restoration data detection result. Therefore, when performing soil ecological restoration data detection, sorting and comparing the proximity of multiple directories can more accurately detect the target directory and thus find the best soil ecological restoration data for detection.

[0063] As can be seen from the above, the embodiment of the present application loads the training soil ecological restoration data, the soil object labels and the inclusion labels into the first soil restoration state detection network and the second soil restoration state detection network for training respectively; obtains the comparison result between the predictions of the first soil restoration state detection network and the second soil restoration state detection network; optimizes the network parameters of the first soil restoration state detection network and the second soil restoration state detection network according to the comparison result to obtain the trained first soil restoration state detection network and the second soil restoration state detection network; receives the soil ecological restoration data to be processed, extracts the feature information of the soil ecological restoration data to be processed through the trained first soil restoration state detection network to perform soil ecological restoration data detection and obtain the detection result. Therefore, the multi-label training of the two networks is performed by combining the inclusion labels between the soil ecological restoration data with similar soil object labels and soil ecological restoration data features with the training soil ecological restoration data, and the network parameters of the two networks are optimized according to the comparison result to improve the working efficiency of the trained network, thereby improving the accuracy of information detection and ensuring the reliability of ecological restoration.

[0064] The artificial intelligence-based soil ecological restoration detection method provided in the embodiments of the present application may include the following content.

[0065] In step 201, training soil ecological restoration data is obtained, and preset soil ecological restoration data corresponding to each soil object label type is determined.

[0066] Among them, a large amount of training soil ecological restoration data can be obtained, and the training soil ecological restoration data has soil object labels.

[0067] In step 202, soil ecological restoration data features of the preset soil ecological restoration data corresponding to each soil object label type are extracted, and the soil ecological restoration data features of the preset soil ecological restoration data corresponding to each soil object label type are classified and processed by a classification method. The preset soil ecological restoration data whose soil ecological restoration data feature similarity is greater than a preset specified value is classified, and an inclusion label for the classified preset soil ecological restoration data is generated.

[0068] In step 203, the training soil ecological restoration data is optimized, and the optimized training soil ecological restoration data, the soil object labels, and the included labels are loaded into the first soil restoration status detection network and the second soil restoration status detection network for training.

[0069] In step 204, a comparison result between the predictions of the first soil remediation state detection network and the second soil remediation state detection network is calculated using a comparison result calculation strategy, and a quantitative evaluation thread corresponding to the comparison result is constructed.

[0070] In step 205, the quantitative evaluation thread is combined with the initial loss value thread of the first soil remediation state detection network for splicing training to obtain the trained first soil remediation state detection network, and the quantitative evaluation thread is combined with the initial loss value thread of the second soil remediation state detection network for splicing training to obtain the trained second soil remediation state detection network.

[0071] In step 206, the predicted soil ecological restoration data is received, the to-be-processed data in the predicted soil ecological restoration data is identified through the target parsing network, the to-be-processed data is read, and the to-be-processed soil ecological restoration data is generated.

[0072] In step 207, the soil ecological restoration data to be processed is loaded into the trained first soil restoration status detection network, feature information of the preset layer after global convolution is extracted, and multiple directories to which the feature information belongs are determined.

[0073] In step 208, a preset number of target soil ecological restoration data in each directory is obtained and sorted from small to large in similarity with the feature information, and the preset number of target soil ecological restoration data in each directory is loaded into the trained first soil remediation status detection network respectively, and the feature information of the preset layer after global convolution of the preset number of target soil ecological restoration data in each directory is extracted.

[0074] In step 209, the distinguishing data between the characteristic information of each target soil ecological restoration data in the same directory and the characteristic information of the soil ecological restoration data to be processed are calculated, and the target soil ecological restoration data in the same directory are sorted in descending order of the distinguishing data. The distinguishing data of each target soil ecological restoration data in the same directory are weighted according to the sorting situation, and the splicing value of the distinguishing data of each target soil ecological restoration data in the same directory after the weighted processing is calculated.

[0075] Among them, the distinguishing data between the characteristic information of each target soil ecological restoration data in the same directory and the characteristic information of the soil ecological restoration data to be processed is calculated. The distinguishing data can be Euclidean difference or cosine difference. The smaller the distinguishing data, the closer the two are, and the larger the distinguishing data, the greater the difference between the two. The target soil ecological restoration data in the same directory can be sorted in descending order according to the distinguishing data.

[0076] Furthermore, the distinguishing data of each target soil ecological restoration data under each directory can be weighted according to the sorting situation. Since the trained first soil restoration status detection network has the ability to distinguish similar soil ecological restoration data under the same directory, the difference value between the soil ecological restoration data to be processed and similar soil ecological restoration data under the most detected label will be significantly smaller than the difference value with the soil ecological restoration data under other labels. In order to expand the discrimination degree of the difference values ​​of different labels, the distinguishing data of each target soil ecological restoration data under the same directory can be weighted according to the ranking order.

[0077] In step 210, the splicing value of the distinguishing data of each target soil ecological restoration data in the same directory after weighted processing is calculated, the directory with the smallest splicing value is used as the target directory, and the target soil ecological restoration data with the smallest distinguishing data in the target directory is obtained as the soil ecological restoration data detection result.

[0078] The weighted splicing value of the distinguishing data of each target soil ecological restoration data in the same directory is calculated. This splicing value reflects the degree of detection between the soil ecological restoration data to be processed and each directory. The directory with the smallest splicing value is used as the target directory for the most detection. The shoe label with the smallest splicing value can be used as the target directory, and the target soil ecological restoration data with the smallest distinguishing data under this target directory is obtained as the soil ecological restoration data detection result. Therefore, through the above multi-label training method and label detection method, the target directory to which the soil ecological restoration data to be processed belongs can be accurately determined, thereby generating soil ecological restoration data detection results with high accuracy.

[0079] As can be seen from the above, the embodiment of the present application loads the training soil ecological restoration data, the soil object labels and the inclusion labels into the first soil restoration state detection network and the second soil restoration state detection network for training respectively; obtains the comparison result between the predictions of the first soil restoration state detection network and the second soil restoration state detection network; optimizes the network parameters of the first soil restoration state detection network and the second soil restoration state detection network according to the comparison result to obtain the trained first soil restoration state detection network and the second soil restoration state detection network; receives the soil ecological restoration data to be processed, extracts the feature information of the soil ecological restoration data to be processed through the trained first soil restoration state detection network to perform soil ecological restoration data detection and obtain the detection result. Therefore, the multi-label training of the two networks is performed by combining the inclusion labels between the soil ecological restoration data with similar soil object labels and soil ecological restoration data features with the training soil ecological restoration data, and the network parameters of the two networks are optimized according to the comparison result to improve the working efficiency of the trained network, thereby improving the accuracy of information detection and ensuring the reliability of ecological restoration.

[0080] Based on the above, a soil ecological restoration detection device based on artificial intelligence is provided, which includes:

[0081] A network training module is used to load the training soil ecological restoration data, the soil object labels and the inclusion labels into the first soil restoration status detection network and the second soil restoration status detection network for training, respectively. The inclusion labels are used to calibrate the training soil ecological restoration data whose feature similarity of the soil ecological restoration data is greater than a preset specified value;

[0082] a result comparison module, configured to obtain a comparison result between the predictions of the first soil remediation status detection network and the second soil remediation status detection network;

[0083] a parameter optimization module, configured to optimize the network parameters of the first and second soil remediation state detection networks based on the comparison result, thereby obtaining trained first and second soil remediation state detection networks; wherein the second soil remediation state detection network is configured to load the important soil remediation indicators extracted from the second soil remediation state detection network into the first soil remediation state detection network, thereby achieving mutual feature extraction and learning between the first and second soil remediation state detection networks, thereby obtaining trained first and second soil remediation state detection networks, so that the trained first soil remediation state detection network has the important soil remediation indicators extracted from the second soil remediation state detection network;

[0084] The information detection module is used to receive the soil ecological restoration data to be processed, extract the characteristic information of the soil ecological restoration data to be processed through the trained first soil restoration status detection network, and perform soil ecological restoration data detection to obtain a detection result.

[0085] Based on the above, an artificial intelligence-based soil ecological restoration detection system is shown, which includes a processor and a memory that communicate with each other. The processor is used to read a computer program from the memory and execute it to implement the above method.

[0086] Based on the above, a computer-readable storage medium is also provided, on which a computer program stored implements the above method when running.

[0087] In summary, based on the above scheme, the training soil ecological restoration data, the soil object labels, and the inclusion labels are loaded into the first soil restoration state detection network and the second soil restoration state detection network for training; a comparison result between the predictions of the first soil restoration state detection network and the second soil restoration state detection network is obtained; the network parameters of the first soil restoration state detection network and the second soil restoration state detection network are optimized based on the comparison result to obtain the trained first soil restoration state detection network and the second soil restoration state detection network; the soil ecological restoration data to be processed is received, and the feature information of the soil ecological restoration data to be processed is extracted through the trained first soil restoration state detection network to perform soil ecological restoration data detection and obtain the detection result. Therefore, multi-label training is performed on the two networks by combining the inclusion labels between soil ecological restoration data with similar soil object labels and soil ecological restoration data features with the training soil ecological restoration data. The network parameters of the two networks are optimized based on the comparison result, thereby improving the working efficiency of the trained networks, thereby improving the accuracy of information detection and ensuring the reliability of ecological restoration.

[0088] It should be understood that the system and its modules shown above can be implemented in various ways. For example, in some embodiments, the system and its modules can be implemented by hardware, software, or a combination of software and hardware. Among them, the hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated hardware. Those skilled in the art will understand that the above-mentioned methods and systems can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the system and its modules of the present application. Not only can hardware circuits such as ultra-large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc. be implemented, they can also be implemented using software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software (for example, firmware).

[0089] It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects that may be produced may be any one or a combination of the above, or any other possible beneficial effects.

Claims

1. A soil ecological restoration detection method based on artificial intelligence, characterized in that: The method comprises: The training soil ecological restoration data, the soil object labels and the inclusion labels are loaded into the first soil restoration state detection network and the second soil restoration state detection network for training respectively, wherein the inclusion labels are used to calibrate the training soil ecological restoration data whose feature similarity of the soil ecological restoration data is greater than a preset specified value; Obtaining a comparison result between predictions of the first soil remediation status detection network and the second soil remediation status detection network; Based on the comparison result, the network parameters of the first soil remediation state detection network and the second soil remediation state detection network are optimized to obtain the trained first soil remediation state detection network and the second soil remediation state detection network; wherein the second soil remediation state detection network is used to load the important soil remediation indicators extracted by the second soil remediation state detection network into the first soil remediation state detection network, so as to achieve mutual feature extraction learning between the first soil remediation state detection network and the second soil remediation state detection network, thereby obtaining the trained first soil remediation state detection network and the second soil remediation state detection network, so that the trained first soil remediation state detection network has the important soil remediation indicators extracted by the second soil remediation state detection network; Soil ecological restoration data to be processed is received, and characteristic information of the soil ecological restoration data to be processed is extracted through a trained first soil restoration status detection network to perform soil ecological restoration data detection and obtain a detection result.

2. The soil ecological restoration detection method based on artificial intelligence according to claim 1, characterized in that: The artificial intelligence-based soil ecological restoration detection method further includes: Acquire training soil ecological restoration data, wherein the training soil ecological restoration data has soil object labels; determine preset soil ecological restoration data corresponding to each soil object label type; Classify the preset soil ecological restoration data corresponding to different soil object label types, and classify the preset soil ecological restoration data whose soil ecological restoration data feature similarity is greater than a preset specified value; Generate inclusion labels for classified preset soil ecological restoration data.

3. The soil ecological restoration detection method based on artificial intelligence according to claim 2, characterized in that: The step of classifying the preset soil ecological restoration data corresponding to different soil object label types and classifying the preset soil ecological restoration data whose feature similarity of the soil ecological restoration data is greater than a preset specified value includes: Extracting soil ecological restoration data features of preset soil ecological restoration data corresponding to each soil object label type; Classify and process the soil ecological restoration data features of the preset soil ecological restoration data corresponding to each soil object label type by classification; The preset soil ecological restoration data whose soil ecological restoration data feature similarity is greater than the preset specified value are classified.

4. The soil ecological restoration detection method based on artificial intelligence according to claim 1, characterized in that: The step of loading the training soil ecological restoration data, the soil object labels, and the labels into the first soil restoration status detection network and the second soil restoration status detection network for training includes: Optimizing the training soil ecological restoration data; The optimized training soil ecological restoration data, the soil object labels and the included labels are loaded into the first soil restoration status detection network and the second soil restoration status detection network for training respectively.

5. The soil ecological restoration detection method based on artificial intelligence according to claim 4, characterized in that: The step of obtaining a comparison result between the predictions of the first soil remediation state detection network and the second soil remediation state detection network includes: calculating the comparison result between the predictions of the first soil remediation state detection network and the second soil remediation state detection network using a comparison result calculation strategy, and constructing a quantitative evaluation thread corresponding to the comparison result.

6. The soil ecological restoration detection method based on artificial intelligence according to claim 5, characterized in that: The step of optimizing the network parameters of the first soil remediation state detection network and the second soil remediation state detection network based on the comparison result to obtain the trained first soil remediation state detection network and the second soil remediation state detection network comprises: The quantitative evaluation thread is combined with the initial loss value thread of the first soil remediation status detection network for splicing training to obtain a trained first soil remediation status detection network; The quantitative evaluation thread is combined with the initial loss value thread of the second soil remediation status detection network for splicing training to obtain a trained second soil remediation status detection network.

7. The soil ecological restoration detection method based on artificial intelligence according to claim 4, characterized in that: The step of receiving the soil ecological restoration data to be processed, extracting characteristic information of the soil ecological restoration data to be processed through the trained first soil restoration status detection network, performing soil ecological restoration data detection, and obtaining a detection result includes: Receiving predicted soil ecological restoration data, and identifying data to be processed in the predicted soil ecological restoration data through a target parsing network; Reading the data to be processed to generate soil ecological restoration data to be processed; The soil ecological restoration data to be processed is loaded into the trained first soil restoration status detection network, and the feature information after global convolution is extracted to perform soil ecological restoration data detection to obtain a detection result.

8. The soil ecological restoration detection method based on artificial intelligence according to claim 7, characterized in that: The step of extracting the feature information after global convolution to perform soil ecological restoration data detection and obtain the detection result includes: Extract feature information of the preset layer after global convolution; Determine multiple directories to which the characteristic information belongs; obtain a preset number of target soil ecological restoration data from each directory sorted in ascending order of similarity with the characteristic information; load the preset number of target soil ecological restoration data from each directory into a trained first soil restoration status detection network, and extract characteristic information of a preset layer after global convolution of the preset number of target soil ecological restoration data from each directory; Calculate the splicing value of the distinguishing data between the characteristic information of each target soil ecological restoration data and the characteristic information of the soil ecological restoration data to be processed in the same directory; The directory with the smallest splicing value is used as the target directory, and the target soil ecological restoration data with the smallest distinguishing data under the target directory is obtained as the soil ecological restoration data detection result.

9. The soil ecological restoration detection method based on artificial intelligence according to claim 8, characterized in that: The step of calculating the splicing value of distinguishing data between the characteristic information of each target soil ecological restoration data and the characteristic information of the soil ecological restoration data to be processed in the same directory includes: Calculate the distinguishing data between the characteristic information of each target soil ecological restoration data and the characteristic information of the soil ecological restoration data to be processed in the same directory; Sort the target soil ecological restoration data in the same directory in descending order of distinguishing data; The differentiated data of each target soil ecological restoration data in the same directory are weighted according to the ranking; Calculate the splicing value of the distinguishing data of each target soil ecological restoration data in the same directory after weighted processing.

10. A soil ecological restoration detection system based on artificial intelligence, characterized in that: The invention comprises a processor and a memory communicating with each other, wherein the processor is used to read a computer program from the memory and execute the computer program to implement the method according to any one of claims 1 to 9.

Images