A post-loan risk monitoring method and system for agricultural credit

Abstract

A post-loan risk monitoring method and system for agricultural credit, relating to the field of intelligent risk monitoring technology, obtains loan application information of target customers and remote sensing images of target plots, and introduces semantic coding models and image processing algorithms at the back end to analyze the loan application information and the remote sensing images, thereby generating a risk monitoring report for the credit project based on the growth status of crops and loan information. In this way, the monitoring efficiency and accuracy can be improved, more comprehensive decision support can be provided for financial institutions, loan risks can be reduced, and sustainable agricultural development can be promoted.

Description

Technical Field

[0001] The present application relates to the field of intelligent risk monitoring technology, and more specifically, to a post-loan risk monitoring method and system for agricultural credit. Background Art

[0002] With the development of agricultural modernization, the demand for agricultural loans is increasing. Agricultural production units need to continuously increase financial support, so the number of agricultural loan issuance institutions, projects and quantities has also increased significantly. The agricultural loan industry is facing the demand for agricultural production that requires a lot of money, a long time and high technical requirements, such as emerging agricultural cash crops and agricultural and sideline product processing. To meet these needs, agricultural loans should provide a large amount, long term, low interest rate and stable loan strategy. However, agricultural credit has certain post-loan risks, including poor crop growth, natural disasters, market fluctuations and other factors that may lead to the failure to repay the loan on time or fail to achieve the expected returns.

[0003] At present, post-loan risk monitoring and analysis of agricultural loans mainly rely on on-site inspections and manual analysis by account managers. However, this method has some disadvantages. First, if the plot is far away from the credit account manager, real-time monitoring cannot be carried out, especially in remote areas, where monitoring is more difficult, resulting in increased loan risks. Second, a single account manager needs to manage many customers and cannot regularly visit each customer for inspection. Third, customers applying for agricultural loans usually have larger plots with different terrain conditions, making it impossible to conduct in-depth and detailed inspections of each plot. Finally, the credit account manager himself lacks the professional knowledge of various crop assessments and cannot make professional assessments.

[0004] Therefore, an optimized post-loan risk monitoring system for agricultural credit is desired. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed.

[0006] In a first aspect, a post-loan risk monitoring system for agricultural credit is provided, which includes:

[0007] A loan application information collection module is used to obtain loan application information of target customers, wherein the loan application information includes project information and plot information;

[0008] Remote sensing image acquisition module, used to obtain remote sensing images of target plots;

[0009] A loan application semantic coding module, used for semantically coding the project information and the plot information in the loan application information to obtain a project information semantic coding feature vector and a plot information semantic coding feature vector;

[0010] A loan application semantic information fusion module, used for concatenating the project information semantic coding feature vector and the plot information semantic coding feature vector to obtain a loan application information semantic coding feature vector;

[0011] A remote sensing image feature extraction module, used for extracting features from the remote sensing image by a crop growth state feature extractor based on a deep neural network model to obtain a multi-scale feature map of crop growth state;

[0012] A loan semantics-guided crop growth state feature characterization module is used to perform cross-modal information fusion on the multi-scale feature map of the crop growth state and the semantic encoding feature vector of the loan application information to obtain loan semantics-guided crop growth state features;

[0013] The loan project risk detection module is used to generate a risk monitoring report for the credit project based on the loan semantics to guide the crop growth status characteristics.

[0014] In the second aspect, a post-loan risk monitoring method for agricultural credit is provided, which includes:

[0015] Obtaining loan application information of target customers, wherein the loan application information includes project information and plot information;

[0016] Acquire remote sensing images of target plots;

[0017] Performing semantic coding on the project information and the plot information in the loan application information to obtain a semantic coding feature vector of the project information and a semantic coding feature vector of the plot information;

[0018] Concatenating the project information semantic coding feature vector and the plot information semantic coding feature vector to obtain a loan application information semantic coding feature vector;

[0019] Extracting features from the remote sensing image using a crop growth state feature extractor based on a deep neural network model to obtain a multi-scale feature map of crop growth state;

[0020] Performing cross-modal information fusion on the multi-scale feature map of crop growth status and the semantic encoding feature vector of loan application information to obtain loan semantics-guided crop growth status features;

[0021] Based on the loan semantics and guiding the crop growth status characteristics, a risk monitoring report for the credit project is generated.

[0022] The beneficial effects of the present application are as follows: the embodiments of the present application provide a post-loan risk monitoring method and system for agricultural credit, which obtains the loan application information of the target customer and the remote sensing image of the target plot, and introduces a semantic coding model and an image processing algorithm at the back end to analyze the loan application information and the remote sensing image, thereby generating a risk monitoring report for the credit project based on the crop growth status and loan information. In this way, the monitoring efficiency and accuracy can be improved, providing financial institutions with more comprehensive decision-making support, reducing loan risks, and promoting the sustainable development of agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 It is a block diagram of a post-loan risk monitoring system for agricultural credit according to an embodiment of the present application.

[0025] Figure 2 It is a flow chart of a post-loan risk monitoring method for agricultural credit according to an embodiment of the present application.

[0026] Figure 3 It is a schematic diagram of the architecture of the post-loan risk monitoring method for agricultural credit according to an embodiment of the present application.

[0027] Figure 4 This is an application scenario diagram of the post-loan risk monitoring system for agricultural credit according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] The following will describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0029] Unless otherwise specified, all technical and scientific terms used in the embodiments of the present application have the same meaning as those commonly understood by those skilled in the art of the present application. The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit the scope of this application.

[0030] In the embodiments of the present application, it should be noted that, unless otherwise specified and limited, the term "connection" should be understood in a broad sense. For example, it can be an electrical connection or internal connectivity between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0031] It should be noted that the terms "first\second\third" involved in the embodiments of the present application are only used to distinguish similar objects, and do not represent a specific order for the objects. It is understandable that the specific order or sequence of "first\second\third" can be interchanged where permitted. It should be understood that the objects distinguished by "first\second\third" can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0032] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0033] In one embodiment of the present application, Figure 1 FIG. 1 is a block diagram of a post-loan risk monitoring system for agricultural credit according to an embodiment of the present application. Figure 1 As shown, according to the agricultural credit post-loan risk monitoring system 100 of the embodiment of the present application, it includes: a loan application information acquisition module 110, which is used to obtain the loan application information of the target customer, and the loan application information includes project information and plot information; a remote sensing image acquisition module 120, which is used to obtain the remote sensing image of the target plot; a loan application semantic encoding module 130, which is used to semantically encode the project information and plot information in the loan application information to obtain the project information semantic encoding feature vector and the plot information semantic encoding feature vector; a loan application semantic information fusion module 140, which is used to semantically encode the project information semantic encoding feature vector and the plot information semantic encoding feature vector The remote sensing image feature extraction module 150 is used to extract features from the remote sensing image through a crop growth state feature extractor based on a deep neural network model to obtain a multi-scale feature map of crop growth state; the loan semantics-guided crop growth state feature characterization module 160 is used to perform cross-modal information fusion on the multi-scale feature map of crop growth state and the loan application information semantic encoding feature vector to obtain loan semantics-guided crop growth state features; the loan project risk detection module 170 is used to generate a risk monitoring report for a credit project based on the loan semantics-guided crop growth state features.

[0034] In the loan application information collection module 110, the target customer's loan application information, including project information and plot information, is collected. During implementation, it is ensured that the collected information is accurate, complete, and meets the requirements and specifications of the loan application. Through the effective implementation of this module, the target customer's loan application information can be collected, providing the necessary data basis for subsequent analysis and processing.

[0035] In the remote sensing image acquisition module 120, a remote sensing image of the target plot is acquired. During implementation, a suitable remote sensing data source, acquisition method and technology are selected to ensure that high-quality remote sensing image data is acquired. Through the implementation of this module, a remote sensing image of the target plot can be acquired, providing image data support for subsequent feature extraction and analysis.

[0036] In the loan application semantic encoding module 130, the project information and the plot information in the loan application information are semantically encoded to obtain the semantic encoding feature vectors of the project information and the plot information. During implementation, a suitable semantic encoding method and model are selected to ensure that the semantic meaning of the information is accurately captured. Through the implementation of this module, the project information and the plot information in the loan application information can be converted into a semantic encoding feature vector, providing semantic representation for subsequent information fusion and feature characterization.

[0037] In the loan application semantic information fusion module 140, the semantic coding feature vectors of the project information and the plot information are concatenated to obtain the semantic coding feature vector of the loan application information. During implementation, the dimension and order of the feature vectors are ensured to be correct to maintain the integrity and consistency of the information. Through the implementation of this module, the semantic coding feature vectors of the project information and the plot information can be fused to obtain the semantic coding feature vector of the loan application information, providing a comprehensive information representation for subsequent feature extraction and risk detection.

[0038] In the remote sensing image feature extraction module 150, a feature extractor based on a deep neural network model is used to extract features from the remote sensing image to obtain a multi-scale feature map of the crop growth state. During implementation, a suitable deep neural network model and feature extraction method are selected to ensure that representative crop growth state features are extracted. Through the implementation of this module, a multi-scale feature map of the crop growth state can be extracted from the remote sensing image, providing information representation of the crop growth state for subsequent feature fusion and risk detection.

[0039] In the loan semantics-guided crop growth state feature characterization module 160, cross-modal information fusion is performed on the multi-scale feature map of the crop growth state and the semantic coding feature vector of the loan application information to obtain the loan semantics-guided crop growth state features. During implementation, a suitable cross-modal information fusion method and model are selected to ensure that the fused features can accurately characterize the crop growth state. Through the implementation of this module, the multi-scale feature map of the crop growth state can be fused with the semantic coding feature vector of the loan application information to obtain the loan semantics-guided crop growth state features, providing a more comprehensive and accurate feature representation for subsequent risk detection.

[0040] In the loan project risk detection module 170, a risk monitoring report of a credit project is generated based on the crop growth status characteristics guided by the loan semantics. During implementation, an effective risk detection model and algorithm are established, and risk assessment and report generation are performed in combination with characteristics and rules. Through the implementation of this module, risk detection of credit projects can be performed based on the crop growth status characteristics guided by the loan semantics, and a risk monitoring report can be provided to help decision makers evaluate the feasibility and risk level of loan projects.

[0041] In response to the above technical problems, the technical concept of the present application is to obtain the loan application information of the target customer and the remote sensing image of the target plot, and introduce a semantic coding model and an image processing algorithm at the back end to analyze the loan application information and the remote sensing image, thereby generating a risk monitoring report for the credit project based on the crop growth status and loan information. In this way, the monitoring efficiency and accuracy can be improved, providing financial institutions with more comprehensive decision-making support, reducing loan risks, and promoting the sustainable development of agriculture.

[0042] Specifically, in the technical solution of the present application, first, the target customer's loan application information is obtained, and the loan application information includes project information and plot information. Next, considering that in the target customer's loan application information, the project information and the plot information both contain their own semantic features, in order to be able to semantically understand the project information and the plot information respectively, so as to subsequently integrate the semantic features of the two to perform semantic analysis of the target customer's loan application information, in the technical solution of the present application, the project information and the plot information in the loan application information are semantically encoded to obtain a project information semantic encoding feature vector and a plot information semantic encoding feature vector.

[0043] It should be understood that by semantically encoding the project information and the plot information, they can be converted into vector representations and their semantic features can be captured. The project information may include the purpose of the loan, crop planting plans, expected returns, etc., while the plot information may include soil type, topography, climate conditions, etc. After encoding this information, the resulting feature vector can reflect the key semantic features of the loan application. Furthermore, the semantically encoded feature vector of the project information and the semantically encoded feature vector of the plot information are spliced ​​to fuse the semantic feature information of these two parts together to form a more comprehensive semantically encoded feature vector of the loan application information. In this way, the impact of the project information and the plot information on loan risks can be comprehensively considered to improve the accuracy and comprehensiveness of post-loan risk monitoring.

[0044] In a specific embodiment of the present application, the loan application semantic encoding module includes: a word segmentation unit, used to perform word segmentation processing on the project information and the plot information in the loan application information to convert the project information and the plot information in the loan application information into a first word sequence and a second word sequence composed of multiple words, respectively; an embedding encoding unit, used to use the word embedding layer of the semantic encoder including the word embedding layer to map each word in the first word sequence and the second word sequence to a word vector to obtain a sequence of first word vectors and a sequence of second word vectors; and, a context encoding unit, used to use the semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of the first word vectors and the sequence of the second word vectors, respectively, to obtain the project information semantic encoding feature vector and the plot information semantic encoding feature vector.

[0045] Then, a remote sensing image of the target plot is obtained, and a convolutional neural network model with excellent performance in implicit feature extraction of the image is used to perform feature mining of the remote sensing image of the target plot. In particular, considering that when performing feature analysis of remote sensing images to detect the growth status of crops, not only deep features of crops are needed, but also shallow features on the surface should be focused. Therefore, the remote sensing image is passed through a crop growth status feature extractor based on a pyramid network to obtain a multi-scale feature map of crop growth status. In the process of crop growth status detection, the pyramid network can extract different levels of features related to crops through branches of different depths, for example, low-level branches can capture the texture and edge information of crops, while high-level branches can capture more abstract semantic features of crops. In this way, the pyramid network can provide rich hierarchical feature information, such as vegetation index, leaf area index, vegetation coverage, etc., for describing the growth status of crops.

[0046] In a specific embodiment of the present application, the deep neural network model is a pyramid network.

[0047] Furthermore, in order to fuse the loan information with the crop growth state information to improve the efficiency and accuracy of post-loan risk monitoring, the multi-scale feature map of the crop growth state and the semantic encoding feature vector of the loan application information are further fused through a cross-modal information element fusion module based on MetaNet to obtain a loan semantics-guided crop growth state feature map. It should be understood that the MetaNet is a method of cross-modal information fusion, which can effectively fuse and integrate features of different modalities. In this case, the multi-scale feature map of the crop growth state represents the image modality, and the semantic encoding feature vector of the loan application information represents the text modality. Through the cross-modal information element fusion module based on MetaNet, the features of these two different modalities can be fused to obtain the loan semantics-guided crop growth state feature map. In particular, here, the fused loan semantics-guided crop growth state feature representation can comprehensively consider the correlation between the growth state features of crops and the semantic features of loan application information, thereby facilitating the provision of more comprehensive and accurate post-loan risk monitoring results.

[0048] In a specific embodiment of the present application, the loan semantics-guided crop growth status feature characterization module is used to: pass the multi-scale feature map of the crop growth status and the semantic encoding feature vector of the loan application information through a cross-modal information element fusion module based on MetaNet to obtain a loan semantics-guided crop growth status feature map.

[0049] Then, based on the loan semantics-guided crop growth state feature map, a risk monitoring report for the credit project is generated. In other words, the loan semantic features are used to guide the cross-modal fusion feature information of the crop growth state features for classification processing, so as to generate a risk monitoring report for the credit project based on the crop growth state and loan information. In this way, the monitoring efficiency and accuracy can be improved, and more comprehensive decision support can be provided for financial institutions.

[0050] In a specific embodiment of the present application, the loan project risk detection module is used to generate a risk monitoring report for a credit project based on the loan semantics-guided crop growth status characteristic graph.

[0051] The loan project risk detection module is used to: pass the loan semantics-guided crop growth state characteristic map through a classifier to obtain a classification result, and the classification result is used to represent the risk level of the credit project.

[0052] In one embodiment of the present application, the post-loan risk monitoring system for agricultural credit also includes a training module for training the semantic encoder including the word embedding layer, the crop growth status feature extractor based on the pyramid network, the cross-modal information element fusion module based on MetaNet, and the classifier. The training module includes: a training loan application information acquisition unit, which is used to obtain training loan application information of a target customer, wherein the training loan application information includes training project information and training plot information, as well as the true value of the risk level of the credit project; a training remote sensing image acquisition unit, which is used to obtain a training remote sensing image of a target plot; a training loan application semantic encoding unit, which is used to semantically encode the training project information and the training plot information in the training loan application information to obtain a training project information semantic encoding feature vector and a training plot information semantic encoding feature vector; a training loan application semantic information fusion unit, which is used to splice the training project information semantic encoding feature vector and the training plot information semantic encoding feature vector to obtain a training loan application information semantic encoding feature vector; a training remote sensing image feature extraction unit, which is used to extract features from the training remote sensing image through a crop growth state feature extractor based on a deep neural network model to obtain Training a multi-scale feature map of crop growth status; training the loan semantics-guided crop growth status feature characterization unit, used to: pass the training crop growth status multi-scale feature map and the training loan application information semantic encoding feature vector through a cross-modal information element fusion module based on MetaNet to obtain a training loan semantics-guided crop growth status feature map; a clustering optimization unit, used to cluster and optimize the training loan semantics-guided crop growth status feature map to obtain an optimized training loan semantics-guided crop growth status feature map; training a loan project risk detection unit, used to pass the optimized training loan semantics-guided crop growth status feature map through a classifier to obtain a classification loss function value; and, based on the classification loss function value, training the semantic encoder including a word embedding layer, the crop growth status feature extractor based on a pyramid network, the cross-modal information element fusion module based on MetaNet and the classifier.

[0053] In particular, in the technical solution of the present application, here, each feature matrix of the training crop growth state multi-scale feature map expresses the cross-scale and cross-depth image semantic features of the remote sensing image of the training target plot based on the pyramid network, and the training loan application information semantic encoding feature vector expresses the text semantic features of the training project information and the training plot information in the training loan application information. After passing through the MetaNet-based cross-modal information element fusion module, the cross-scale and cross-depth expressions of the image semantic features of the training crop growth state multi-scale feature map can be constrained along the channel based on the text semantic feature distribution of the training loan application information semantic encoding feature vector, so that the training loan semantics-guided crop growth state feature map contains the text semantic feature information of the training project information and the training plot information in the training loan application information in the channel dimension, thereby improving the expression effect of the training loan semantics-guided crop growth state feature map.

[0054] On the other hand, considering the essential difference between the text semantic feature distribution of the project information and plot information in the training loan application information and the cross-scale and cross-depth distribution of the image semantic features of the pyramid network, the training loan semantics-guided crop growth status feature map will also have multi-dimensional sparsification of the cross-channel expression of image semantic features, so that when the training loan semantics-guided crop growth status feature map as a whole is subjected to class probability regression mapping through a classifier, the probability density distribution of the regression probability of each feature value of the training loan semantics-guided crop growth status feature map has poor convergence, resulting in difficulty in convergence to the predetermined class probability when class regression is performed through the classifier, thereby affecting the training speed of the classifier and the accuracy of the final classification result.

[0055] Based on this, the present application performs clustering optimization on the training loan semantics-guided crop growth state feature map to obtain an optimized training loan semantics-guided crop growth state feature map, that is, firstly clustering the various feature values ​​of the training loan semantics-guided crop growth state feature map, for example, clustering based on the distance between feature values, and then optimizing based on the in-class and out-of-class representations of the clustered features to obtain the optimized training loan semantics-guided crop growth state feature map.

[0056] In a specific embodiment of the present application, optimizing based on the in-class and out-of-class representations of the features after clustering to obtain the optimized training loan semantics guided crop growth state feature map includes: optimizing based on the in-class and out-of-class representations of the features after clustering to obtain the optimized training loan semantics guided crop growth state feature map according to the following clustering optimization formula; wherein the clustering optimization formula is:

[0057]

[0058] in, are the eigenvalues ​​of the training loan semantics guided crop growth state feature map, is the number of feature sets corresponding to the training loan semantics-guided crop growth state feature map, is the number of clustering features, represents the clustering feature set, It is the various feature values ​​of the crop growth state feature map guided by the optimization training loan semantics.

[0059] Specifically, by taking the in-class features and out-of-class features of the training loan semantics guided crop growth state feature map as different instance roles to perform class instance description based on clustering ratio distribution, and introducing clustering response history based on in-class and out-of-class dynamic context, a coordinated global perspective is maintained for the in-class distribution and out-of-class distribution of the overall features of the training loan semantics guided crop growth state feature map, so that the optimized feature clustering operation for the training loan semantics guided crop growth state feature map can maintain a consistent response of the in-class and out-of-class features, so that the regression convergence path based on feature clustering in the class regression process remains consistent, and the convergence effect of the training loan semantics guided crop growth state feature map towards the predetermined class probability is improved, so as to improve the training speed of the classifier and the accuracy of the classification result. In this way, a risk monitoring report of a credit project can be generated based on crop growth status and loan information, and in this way, the monitoring efficiency and accuracy can be improved, and more comprehensive decision support can be provided for financial institutions, loan risks can be reduced, and sustainable agricultural development can be promoted.

[0060] In summary, the post-loan risk monitoring system 100 for agricultural credit based on the embodiment of the present application is explained, which can improve monitoring efficiency and accuracy, provide more comprehensive decision support for financial institutions, reduce loan risks, and promote sustainable agricultural development.

[0061] As described above, the post-loan risk monitoring system 100 for agricultural credit according to the embodiment of the present application can be implemented in various terminal devices, such as a server for post-loan risk monitoring of agricultural credit. In one example, the post-loan risk monitoring system 100 for agricultural credit according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the post-loan risk monitoring system 100 for agricultural credit can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the post-loan risk monitoring system 100 for agricultural credit can also be one of the many hardware modules of the terminal device.

[0062] Alternatively, in another example, the agricultural credit post-loan risk monitoring system 100 and the terminal device may also be separate devices, and the agricultural credit post-loan risk monitoring system 100 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0063] In one embodiment of the present application, Figure 2 It is a flow chart of a post-loan risk monitoring method for agricultural credit according to an embodiment of the present application. Figure 3 Schematic diagram of the post-loan risk monitoring method architecture of agricultural credit according to the embodiment of the present application. Figure 2 and Figure 3 As shown, the post-loan risk monitoring method of agricultural credit includes: 210, obtaining loan application information of a target customer, wherein the loan application information includes project information and plot information; 220, obtaining a remote sensing image of a target plot; 230, semantically encoding the project information and plot information in the loan application information to obtain a semantic encoding feature vector of the project information and a semantic encoding feature vector of the plot information; 240, concatenating the semantic encoding feature vector of the project information and the semantic encoding feature vector of the plot information to obtain a semantic encoding feature vector of the loan application information; 250, extracting features from the remote sensing image through a crop growth state feature extractor based on a deep neural network model to obtain a multi-scale feature map of crop growth state; 260, cross-modal information fusion of the multi-scale feature map of crop growth state and the semantic encoding feature vector of the loan application information to obtain a loan semantically guided crop growth state feature; 270, generating a risk monitoring report for a credit project based on the loan semantically guided crop growth state feature.

[0064] In the post-loan risk monitoring method for agricultural credit, semantic encoding is performed on the project information and the plot information in the loan application information to obtain a semantic encoding feature vector for the project information and a semantic encoding feature vector for the plot information, including: performing word segmentation processing on the project information and the plot information in the loan application information to convert the project information and the plot information in the loan application information into a first word sequence and a second word sequence composed of multiple words, respectively; using the word embedding layer of the semantic encoder including the word embedding layer to map each word in the first word sequence and the second word sequence to a word vector to obtain a sequence of first word vectors and a sequence of second word vectors; and using the semantic encoder including the word embedding layer to perform global context-based semantic encoding on the sequence of the first word vectors and the sequence of the second word vectors to obtain the semantic encoding feature vector for the project information and the semantic encoding feature vector for the plot information.

[0065] Those skilled in the art will appreciate that the specific operations of each step in the above agricultural credit post-loan risk monitoring method have been described in the above reference. Figure 1 The description of the post-loan risk monitoring system of agricultural credit has been introduced in detail, and therefore, its repeated description will be omitted.

[0066] Figure 4 FIG. 1 is an application scenario diagram of the post-loan risk monitoring system for agricultural credit according to an embodiment of the present application. Figure 4 As shown, in this application scenario, first, the loan application information of the target customer is obtained, and the loan application information includes project information (for example, Figure 4 C1) and plot information (e.g. Figure 4 C2 as shown in the figure), and obtaining remote sensing images of the target plot (e.g. Figure 4 Then, the acquired project information, plot information and remote sensing images are input into a server where an agricultural credit post-loan risk monitoring algorithm is deployed (for example, Figure 4 In S) shown in , the server is capable of processing the project information, the plot information and the remote sensing image based on the post-loan risk monitoring algorithm of agricultural credit to generate a risk monitoring report for the credit project.

[0067] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0068] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0069] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0070] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A post-loan risk monitoring system for agricultural credit, characterized in that: include: A loan application information collection module is used to obtain loan application information of target customers, wherein the loan application information includes project information and plot information; Remote sensing image acquisition module, used to obtain remote sensing images of target plots; A loan application semantic coding module, used for semantically coding the project information and the plot information in the loan application information to obtain a project information semantic coding feature vector and a plot information semantic coding feature vector; A loan application semantic information fusion module, used for concatenating the project information semantic coding feature vector and the plot information semantic coding feature vector to obtain a loan application information semantic coding feature vector; A remote sensing image feature extraction module, used for extracting features from the remote sensing image by a crop growth state feature extractor based on a deep neural network model to obtain a multi-scale feature map of crop growth state; A loan semantics-guided crop growth state feature characterization module is used to perform cross-modal information fusion on the multi-scale feature map of the crop growth state and the semantic encoding feature vector of the loan application information to obtain loan semantics-guided crop growth state features; A loan project risk detection module, used to generate a risk monitoring report for a credit project based on the loan semantics to guide the crop growth status characteristics; The system also includes a training module for training a semantic encoder including a word embedding layer, a crop growth state feature extractor based on a pyramid network, a cross-modal information element fusion module based on MetaNet, and a classifier; The training module comprises: A training loan application information acquisition unit, used to obtain training loan application information of a target customer, wherein the training loan application information includes training project information and training plot information, as well as a true value of a risk level of a credit project; A training remote sensing image acquisition unit, used to acquire a training remote sensing image of a target plot; A training loan application semantic encoding unit, used for semantically encoding the training project information and the training plot information in the training loan application information to obtain a semantic encoding feature vector of the training project information and a semantic encoding feature vector of the training plot information; A training loan application semantic information fusion unit, used for concatenating the training project information semantic encoding feature vector and the training plot information semantic encoding feature vector to obtain a training loan application information semantic encoding feature vector; A training remote sensing image feature extraction unit, used for extracting features from the training remote sensing image through a crop growth state feature extractor based on a deep neural network model to obtain a multi-scale feature map of the growth state of the training crops; Training the loan semantics-guided crop growth state feature characterization unit, used to obtain the training loan semantics-guided crop growth state feature map by fusing the training crop growth state multi-scale feature map and the training loan application information semantic encoding feature vector through a cross-modal information element fusion module based on MetaNet; A clustering optimization unit, used for clustering and optimizing the training loan semantics-guided crop growth state feature map to obtain an optimized training loan semantics-guided crop growth state feature map; A training loan project risk detection unit is used to: pass the optimized training loan semantics-guided crop growth state feature map through a classifier to obtain a classification loss function value; and Based on the classification loss function value, the semantic encoder including the word embedding layer, the crop growth state feature extractor based on the pyramid network, the cross-modal information element fusion module based on MetaNet, and the classifier are trained; A clustering optimization unit, used for clustering each feature value of the training loan semantics-guided crop growth state feature map, and then optimizing based on the clustered feature in-class and out-of-class representations to obtain the optimized training loan semantics-guided crop growth state feature map; The following clustering optimization formula is used to optimize the in-class and out-of-class representations of the clustered features to obtain the optimized training loan semantics-guided crop growth state feature map; wherein the clustering optimization formula is: ； in, are the eigenvalues ​​of the training loan semantics guided crop growth state feature map, is the number of feature sets corresponding to the training loan semantics-guided crop growth state feature map, is the number of clustering features, represents the clustering feature set, are the characteristic values ​​of the crop growth state characteristic map guided by the optimization training loan semantics; The in-class features and out-of-class features of the training loan semantics-guided crop growth state feature map are used as different instance roles to perform class instance description based on clustering proportion distribution, and clustering response history based on in-class and out-of-class dynamic context is introduced. A coordinated global perspective is maintained for the in-class distribution and out-of-class distribution of the overall features of the training loan semantics-guided crop growth state feature map, so that the optimized feature clustering operation of the training loan semantics-guided crop growth state feature map can maintain a consistent response of the in-class and out-of-class features, thereby maintaining a consistent regression convergence path based on feature clustering in the class regression process, thereby improving the convergence effect of the training loan semantics-guided crop growth state feature map towards the predetermined class probability.

2. The post-loan risk monitoring system for agricultural credit according to claim 1 is characterized in that: The loan application semantic encoding module includes: A word segmentation unit is used to perform word segmentation processing on the project information and the plot information in the loan application information so as to convert the project information and the plot information in the loan application information into a first word sequence and a second word sequence composed of a plurality of words, respectively; an embedding encoding unit, configured to map each word in the first word sequence and the second word sequence to a word vector using a word embedding layer of a semantic encoder including a word embedding layer to obtain a sequence of first word vectors and a sequence of second word vectors; and A context encoding unit is used to use the semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of the first word vectors and the sequence of the second word vectors to obtain the project information semantic encoding feature vector and the plot information semantic encoding feature vector.

3. The post-loan risk monitoring system for agricultural credit according to claim 2 is characterized in that: The deep neural network model is a pyramid network.

4. The post-loan risk monitoring system for agricultural credit according to claim 3 is characterized in that: The loan semantics-guided crop growth state feature characterization module is used to: obtain the loan semantics-guided crop growth state feature map by fusing the multi-scale feature map of the crop growth state and the semantic encoding feature vector of the loan application information through a cross-modal information element fusion module based on MetaNet.

5. The post-loan risk monitoring system for agricultural credit according to claim 4 is characterized in that: The loan project risk detection module is used to generate a risk monitoring report for a credit project based on the loan semantics-guided crop growth status characteristic graph.

6. The post-loan risk monitoring system for agricultural credit according to claim 5 is characterized in that: The loan project risk detection module is used to: pass the loan semantics-guided crop growth state characteristic map through a classifier to obtain a classification result, and the classification result is used to represent the risk level of the credit project.

7. A post-loan risk monitoring method for agricultural credit, characterized in that: include: Obtaining loan application information of target customers, wherein the loan application information includes project information and plot information; Acquire remote sensing images of target plots; Performing semantic coding on the project information and the plot information in the loan application information to obtain a semantic coding feature vector of the project information and a semantic coding feature vector of the plot information; Concatenating the project information semantic coding feature vector and the plot information semantic coding feature vector to obtain a loan application information semantic coding feature vector; Extracting features from the remote sensing image using a crop growth state feature extractor based on a deep neural network model to obtain a multi-scale feature map of crop growth state; Performing cross-modal information fusion on the multi-scale feature map of crop growth status and the semantic encoding feature vector of loan application information to obtain loan semantics-guided crop growth status features; Generate a risk monitoring report for a credit project based on the loan semantics to guide the crop growth status characteristics; The method also includes a training step for training a semantic encoder including a word embedding layer, a crop growth state feature extractor based on a pyramid network, a cross-modal information element fusion module based on MetaNet, and a classifier; The training step comprises: Acquire training loan application information of the target customer, wherein the training loan application information includes training project information and training plot information, as well as the actual value of the risk level of the credit project; Obtain training remote sensing images of the target plot; Performing semantic coding on the training project information and the training plot information in the training loan application information to obtain a semantic coding feature vector of the training project information and a semantic coding feature vector of the training plot information; Concatenating the semantic coding feature vector of the training project information and the semantic coding feature vector of the training plot information to obtain a semantic coding feature vector of the training loan application information; Extracting features from the training remote sensing image using a crop growth state feature extractor based on a deep neural network model to obtain a multi-scale feature map of the training crop growth state; The training crop growth state multi-scale feature map and the training loan application information semantic encoding feature vector are fused through a cross-modal information element fusion module based on MetaNet to obtain a training loan semantics-guided crop growth state feature map; Performing cluster optimization on the training loan semantics-guided crop growth state feature map to obtain an optimized training loan semantics-guided crop growth state feature map; Passing the optimized training loan semantics-guided crop growth state feature map through a classifier to obtain a classification loss function value; and Based on the classification loss function value, the semantic encoder including the word embedding layer, the crop growth state feature extractor based on the pyramid network, the cross-modal information element fusion module based on MetaNet, and the classifier are trained; Clustering and optimizing the training loan semantics-guided crop growth state feature map to obtain an optimized training loan semantics-guided crop growth state feature map, including: clustering each feature value of the training loan semantics-guided crop growth state feature map, and then optimizing based on the clustered feature in-class and out-of-class representations to obtain the optimized training loan semantics-guided crop growth state feature map; The following clustering optimization formula is used to optimize the in-class and out-of-class representations of the clustered features to obtain the optimized training loan semantics-guided crop growth state feature map; wherein the clustering optimization formula is: ； in, are the eigenvalues ​​of the training loan semantics guided crop growth state feature map, is the number of feature sets corresponding to the training loan semantics-guided crop growth state feature map, is the number of clustering features, represents the clustering feature set, are the characteristic values ​​of the crop growth state characteristic map guided by the optimization training loan semantics; The in-class features and out-of-class features of the training loan semantics-guided crop growth state feature map are used as different instance roles to perform class instance description based on clustering proportion distribution, and clustering response history based on in-class and out-of-class dynamic context is introduced. A coordinated global perspective is maintained for the in-class distribution and out-of-class distribution of the overall features of the training loan semantics-guided crop growth state feature map, so that the optimized feature clustering operation of the training loan semantics-guided crop growth state feature map can maintain a consistent response of the in-class and out-of-class features, thereby maintaining a consistent regression convergence path based on feature clustering in the class regression process, thereby improving the convergence effect of the training loan semantics-guided crop growth state feature map towards the predetermined class probability.

8. The method for monitoring the post-loan risk of agricultural credit according to claim 7, characterized in that: The project information and the plot information in the loan application information are semantically encoded to obtain a semantically encoded feature vector of the project information and a semantically encoded feature vector of the plot information, including: Performing word segmentation processing on the project information and the plot information in the loan application information to convert the project information and the plot information in the loan application information into a first word sequence and a second word sequence consisting of a plurality of words, respectively; Mapping each word in the first word sequence and the second word sequence to a word vector using a word embedding layer of a semantic encoder including a word embedding layer to obtain a sequence of first word vectors and a sequence of second word vectors; and A semantic encoder including a word embedding layer is used to perform global context semantic encoding on the sequence of the first word vectors and the sequence of the second word vectors to obtain the project information semantic encoding feature vector and the plot information semantic encoding feature vector.

Images