Green loan rating method and related device

Abstract

The invention discloses a green loan rating method and a related device, and the method comprises the steps: obtaining N region sample images, obtaining a first sample feature and a second sample feature based on the N region sample images through an initial feature extraction model, generating a first loss function according to the difference between the first sample feature and the second sample feature, the method comprises the steps of obtaining an initial encoder, performing parameter adjustment on the initial encoder through a first loss function to obtain an encoder, determining a classification model according to the encoder and a classifier, and determining an identification result through the classification model for a region image of an enterprise production region of a to-be-identified enterprise, the identification result being used for identifying whether the to-be-identified enterprise meets a green loan requirement or not. Therefore, accurate identification and evaluation of enterprise green loans are realized, and a more accurate green loan identification and evaluation system is constructed.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a green loan rating method and related devices. Background Art

[0003] At present, relevant technologies have designed a multi-level green credit classification system based on loan descriptions, and constructed a multi-level green credit classification model based on this classification system to predict new samples.

[0004] However, the multi-level green credit classification model constructed by related technologies relies on historical images provided by the applicant. When the image availability is poor, it is impossible to accurately identify and evaluate the green credit of enterprises. Summary of the invention

[0005] In order to solve the above technical problems, the present application provides a green loan rating method and related devices, which reduce the dependence on historical images provided by the applicant and realize the accurate identification and evaluation of corporate green loans.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] In a first aspect, the present application provides a green loan rating method, the method comprising:

[0008] Obtaining N regional sample images, where the N regional sample images are collected from the production area of ​​the enterprise;

[0009] Based on the N regional sample images, a first sample feature and a second sample feature are obtained through an initial feature extraction model, wherein the initial feature model includes a first feature extraction branch and a second feature extraction branch, wherein the second feature extraction branch is provided with a masker and an initial encoder based on an input order, wherein the masker is used to perform image masking processing on the regional sample image to obtain a masked image, and the initial encoder is used to perform feature extraction on the masked image to obtain the second sample feature;

[0010] Generate a first loss function according to the difference between the first sample feature and the second sample feature, and adjust the parameters of the initial encoder by using the first loss function to obtain an encoder;

[0011] Determine a classification model according to the encoder and the classifier;

[0012] For the regional image of the enterprise production area of ​​the enterprise to be identified, the identification result is determined by the classification model, and the identification result is used to indicate whether the enterprise to be identified meets the green loan requirements.

[0013] Optionally, determining a classification model according to the encoder and the classifier includes:

[0014] Classify and annotate C regional sample images among the N regional sample images to obtain an annotated image including a classification label, wherein the classification label is used to identify whether the enterprise production area in the corresponding annotated image meets the green loan requirements;

[0015] Setting the encoder and the classifier based on the input sequence to achieve an initial classification model;

[0016] Based on the C labeled images, obtaining corresponding classification prediction results through the initial classification model;

[0017] A second loss function is generated according to the difference between the classification prediction result and the classification label of the corresponding annotated image, and the parameters of the initial classification model are adjusted by the second loss function to obtain the classification model.

[0018] Optionally, both the regional sample image and the regional image are hyperspectral images.

[0019] Optionally, in the process of obtaining the mask image, the masker is further used to perform masking processing of spatial information on the regional sample image.

[0020] Optionally, the method further comprises:

[0021] During the time period from when the enterprise to be identified applies for a green loan project to when the enterprise completes the green loan project, the regional image used for identifying green loan requirements is periodically acquired.

[0022] Optionally, the method further comprises:

[0023] When the regional image is obtained during the period when the enterprise to be identified applies for a green loan project, in response to determining through the identification result that the enterprise to be identified does not meet the green loan requirements, the application of the enterprise to be identified for the green loan project is rejected;

[0024] When the regional image is obtained after the application for the green loan project by the enterprise to be identified is approved, in response to determining through the identification result that the enterprise to be identified does not meet the green loan requirements, an alarm is issued to the enterprise to be identified regarding the green loan project.

[0025] In a second aspect, the present application provides a green loan rating device, the device comprising:

[0026] An acquisition unit, used to acquire N regional sample images, where the N regional sample images are acquired from the production area of ​​the enterprise;

[0027] An extraction unit, configured to obtain a first sample feature and a second sample feature through an initial feature extraction model based on the N regional sample images, wherein the initial feature model includes a first feature extraction branch and a second feature extraction branch, wherein the second feature extraction branch is provided with a masker and an initial encoder based on an input sequence, wherein the masker is configured to perform image masking processing on the regional sample image to obtain a masked image, and the initial encoder is configured to perform feature extraction on the masked image to obtain the second sample feature;

[0028] a generating unit, configured to generate a first loss function according to a difference between the first sample feature and the second sample feature, and adjust parameters of the initial encoder by using the first loss function to obtain an encoder;

[0029] A first determining unit, configured to determine a classification model according to the encoder and the classifier;

[0030] The second determination unit is used to determine the recognition result through the classification model for the regional image of the enterprise production area of ​​the enterprise to be identified, and the recognition result is used to identify whether the enterprise to be identified meets the green loan requirements.

[0031] In a third aspect, the present application provides an electronic device, the electronic device comprising:

[0032] Memory for storing computer programs or computer instructions;

[0033] The processor is used to execute the computer program or computer instructions stored in the memory, so that the electronic device executes the method as described in the first aspect.

[0034] In a fourth aspect, an embodiment of the present application further provides a computer program product comprising instructions, which, when executed on a computing device, enables the computing device to execute the method described in the first aspect above.

[0035] In a fifth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method described in the first aspect above.

[0036] It can be seen from the above technical solutions that the present application has the following beneficial effects:

[0037] The technical solution of the present application obtains N regional sample images, which are collected from the production area of ​​the enterprise. Based on the N regional sample images, the first sample feature and the second sample feature are obtained through the initial feature extraction model. The initial feature model includes a first feature extraction branch and a second feature extraction branch. The second feature extraction branch is provided with a masker and an initial encoder based on the input order. The masker is used to perform image masking on the regional sample image to obtain a mask image. The initial encoder is used to extract features from the mask image to obtain the second sample feature. Since the initial encoder inputs a mask image after masking by the masker, the second sample feature extracted by the initial encoder from the mask image is different from the first sample feature extracted by the first feature branch. A first loss function is generated according to the difference between the first sample feature and the second sample feature, and the parameters of the initial encoder are adjusted by the first loss function to obtain an encoder. The encoder is obtained by adjusting the parameters of the initial encoder according to the first loss function, so that the encoder learns how to restore the defective content from sample images with different degrees of defect under the guidance of the first loss function and extract high-quality image features. The ability of the encoder to extract more accurate features using the mask image is improved, and the generalization and robustness of the encoder to extract features is enhanced. According to the encoder and classifier, the classification model is determined. For the regional image of the enterprise production area of ​​the enterprise to be identified, the recognition result is determined by the classification model. The recognition result is used to identify whether the enterprise to be identified meets the green loan requirements. From the above, it can be seen that since the encoder adopts a self-supervised training method and does not rely on the labeled data provided by the applicant, a large number of regional sample images can be used for training during the training process, so as to obtain a high-quality encoder, which provides strong support for the subsequent identification of the green loan requirements of the enterprise to be identified. . BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0039] Figure 1 A schematic diagram of a green loan rating method in an embodiment of the present application;

[0040] Figure 2 A schematic diagram of a green loan rating method in an embodiment of the present application;

[0041] Figure 3 This is a schematic diagram of the structure of a green loan rating model in an embodiment of the present application;

[0042] Figure 4This is a schematic diagram of the structure of a green loan rating device in an embodiment of the present application. DETAILED DESCRIPTION

[0043] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.

[0044] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0045] First, some technical terms involved in the embodiments of the present application are introduced.

[0046] Green loans: Green loans are a type of financial policy implemented by commercial banks and other financial institutions in accordance with the country's environmental economic policies and industrial policies. Its main goal is to guide the flow of funds through conscious credit behavior, encourage social entities to reduce pollution to the environment, protect ecological balance, and conserve natural resources.

[0047] Hyperspectral image: Hyperspectral image is an image with hundreds of spectral channels, which contains rich spatial and spectral information. Its data is in the form of cubes and can generally be obtained through data collection by airborne or satellite-borne hyperspectral imagers. Since substances with different components have different effects on spectral bands, the composition of substances can be inferred by analyzing the spectrum. The above characteristics make hyperspectral images have important applications in air pollution monitoring, water eutrophication monitoring, mineral distribution investigation, etc.

[0048] MSTV algorithm: The MSTV algorithm is an image processing algorithm that combines the Minimum Spanning Tree (MST) and relative total variation (RTV). The MSTV algorithm can construct multi-scale structural features that are insensitive to image noise. It mainly extracts multi-scale structural features based on relative total variation (RTV), and then performs kernel principal component analysis on the superimposed multi-scale structural features to enhance the separability of pixels of different categories. Relative total variation can reduce the influence of texture and noise as well as reduce intra-class differences, thereby achieving the purpose of increasing inter-class differences.

[0049] Kernel Principal Component Analysis (KPCA): is a principal component analysis method based on kernel function, which can be used to process nonlinearly separable data. Its main principle is to map the original data into a high-dimensional feature space, so that the linear separability of the data in this space is stronger.

[0050] Encoder: The encoder is used to convert the input sequence into high-dimensional representation vectors, which contain rich contextual information. It is mainly composed of position encoding, self-attention layer, feedforward neural network, residual connection and layer normalization, and can capture more complex sequence models by stacking multiple encoders.

[0052] After studying the relevant rating methods for green loans, it was found that the traditional methods only predict loan risks after summarizing and analyzing the application information of green loan projects, and then concluded that there is a lack of monitoring means and methods for green loans after they are issued. At the same time, most of the existing methods need to build a green financial corpus, use a large number of historical images, and rely on historical images provided by the applicant. When the images provided by the applicant are of poor availability, it is impossible to accurately identify and evaluate corporate green credit.

[0053] In order to conduct comprehensive and effective risk management and monitoring of the entire cycle of green loan projects, ensure that green financial policies truly play a guiding role in resources, and reduce dependence on historical data and data provided by applicants, and achieve accurate identification and evaluation of corporate green loans, this application proposes a green loan rating method and related devices. The encoder with adjusted parameters extracts more accurate image features of the mask image, and determines whether the enterprise meets the green loan requirements based on the classification model. When the image availability is poor, it can also achieve accurate identification and evaluation of the company's green credit.

[0054] Exemplary Methods

[0055] In order to make the technical solution of the present application clearer and easier to understand, the method provided by the present application is introduced below in conjunction with specific embodiments.

[0056] Figure 1 A schematic diagram of a green loan rating method provided in an embodiment of the present application, the method specifically comprises the following steps:

[0057] S101: Obtain N regional sample images.

[0058] S102: Based on N regional sample images, obtain a first sample feature and a second sample feature through an initial feature extraction model.

[0059] S103: Generate a first loss function according to the difference between the first sample feature and the second sample feature, and adjust the parameters of the initial encoder by using the first loss function to obtain an encoder.

[0060] S104: Determine a classification model according to the encoder and the classifier.

[0061] S105: For the regional image of the production area of ​​the enterprise to be identified, determine the identification result through the classification model.

[0062] The N regional sample images are sample images collected from the production area of ​​the enterprise. The initial feature extraction model includes a first feature extraction branch and a second feature extraction branch. The first feature extraction branch uses a conventional feature extraction method to extract features from the regional sample images to obtain first sample features. There is no limitation on which algorithm is used for the first feature extraction branch. For example, the first feature extraction branch uses the MSTV algorithm to extract features from the N regional sample images. For the regional feature image I, the structural feature of the regional feature image I is set to S, and then formula (1) is as follows:

[0063]

[0064] Where N is the total number of pixels in the image. S is the resulting image. λ is a weight parameter used to control the smoothness of the resulting image. ε is a small positive number to avoid being divided by 0. On the one hand, (S i -I i ) 2 Limit the difference between the output image and the input image. On the other hand, Ensure that the output image reduces texture features while retaining the main structural information.

[0065] In detail, D x (i) and D y (i) are the total changes in the window in the x and y directions, respectively, which can capture the absolute spatial difference in the window and are used to represent the local window texture information.x (i) and L y (i) represents the inherent variation in the window in the x and y directions respectively, and can be used to count the overall spatial variation to represent the structural information in the local window. Multi-scale structural features can be extracted by specifying different window sizes. For the stacked multi-scale structural features, KPCA is used to enhance the separability of pixels of different categories, providing more features required for green credit rating.

[0066] The second feature extraction branch is provided with a masker and an initial encoder based on the input sequence. The masker is used to perform image masking on the regional sample image to obtain a masked image. The initial encoder is used to perform feature extraction on the masked image to obtain a second sample feature.

[0067] Since the initial encoder inputs a mask image after mask processing by the masker, the second sample feature extracted by the initial encoder from the mask image is different from the first sample feature extracted by the first feature branch. A first loss function is generated according to the difference between the first sample feature and the second sample feature, and the parameters of the initial encoder are adjusted by the first loss function to obtain an encoder.

[0068] The mask image after mask processing by the masker is input into the initial encoder of the second feature extraction branch to obtain the second sample feature f, and the regional sample image is input into the first feature extraction branch to obtain the first sample feature f'. The parameters of the initial encoder are adjusted according to the first loss function to obtain the encoder. The first loss function is shown in formula (2):

[0069] L pre-train =||f-f'|| 2 Formula (2)

[0070] Among them, the parameters of the initial encoder are adjusted according to the first loss function. The purpose of the parameter adjustment is to make the mask image after masking by the masker input into the initial encoder of the second feature extraction branch to obtain the second sample feature f approximate the first sample feature f' obtained by inputting the regional sample image into the first feature extraction branch, thereby obtaining the encoder.

[0071] According to the encoder and the classifier, a classification model is determined. The encoder in the classification model is used to extract features from the regional image of the production area of ​​the enterprise to be identified. The classifier in the classification model is used to determine the recognition result according to the features extracted by the encoder. The recognition result is used to indicate whether the enterprise to be identified meets the green loan requirements.

[0072] As can be seen from the above, the technical solution of the present application obtains N regional sample images, and the N regional sample images are collected from the production area of ​​the enterprise. Based on the N regional sample images, the first sample feature and the second sample feature are obtained through the initial feature extraction model. The initial feature model includes a first feature extraction branch and a second feature extraction branch. The second feature extraction branch is provided with a masker and an initial encoder based on the input order. The masker is used to perform image masking on the regional sample image to obtain a mask image. The initial encoder is used to perform feature extraction on the mask image to obtain the second sample feature. Since the initial encoder inputs a mask image after masking by the masker, the second sample feature extracted by the initial encoder from the mask image is different from the first sample feature extracted by the first feature branch. A first loss function is generated according to the difference between the first sample feature and the second sample feature, and the parameters of the initial encoder are adjusted by the first loss function to obtain an encoder. The encoder is obtained by adjusting the parameters of the initial encoder according to the first loss function, so that the encoder learns how to restore the defective content from sample images with different degrees of defect under the guidance of the first loss function and extract high-quality image features. The second sample feature extracted by the encoder based on the mask image is close to the first sample feature, which improves the ability of the encoder to extract more accurate features using the mask image, and enhances the generalization and robustness of the feature extraction of the encoder. According to the encoder and the classifier, a classification model is determined, and the recognition result is determined by the classification model for the regional image of the enterprise production area of ​​the enterprise to be identified. The recognition result is used to identify whether the enterprise to be identified meets the green loan requirements. From the above, it can be seen that since the encoder adopts a self-supervised training method and does not rely on the labeled data provided by the applicant, a large number of regional sample images can be used for training during the training process to obtain a high-quality encoder, which provides strong support for the subsequent identification of the green loan requirements of the enterprise to be identified.

[0073] In a possible implementation, the present application provides a specific implementation of a green loan rating method, see A101-A104:

[0074] A101: Classify and label C regional sample images among the N regional sample images to obtain labeled images including classification labels.

[0075] A102: Set the encoder and classifier based on the input order to obtain the initial classification model.

[0076] A103: Based on the C labeled images, obtain the corresponding classification prediction results through the initial classification model.

[0077] A104: Generate a second loss function according to the difference between the classification prediction result and the classification label of the corresponding annotated image, and adjust the parameters of the initial classification model through the second loss function to obtain the classification model.

[0078] The classification label is used to identify whether the enterprise production area in the corresponding annotated image meets the green loan requirements.

[0079] As mentioned earlier, due to Figure 1 The corresponding embodiment trains a high-quality encoder, so when training the initial classification model, it is only necessary to annotate a small number of regional sample images (such as the aforementioned C regional sample images), and then simply fine-tune the initial classification model based on the annotated images to meet the classification requirements.

[0080] For the labeled image x i and classification label c, the second loss function is shown in formula (3):

[0081]

[0082] Where N is the number of samples in N regions. C is the number of classification labels. i c represents the labeled image x i The difference between the classification prediction result and the classification label c. ic is a symbolic function, if the labeled image x i If the category of is equal to c, it is 1, otherwise it is 0.

[0083] By classifying and labeling C regional sample images among N regional sample images, labeled images are obtained that identify whether the enterprise's production area meets the green loan requirements. Based on the C labeled images, the corresponding classification prediction results are obtained through the initial classification model. If the classification prediction result is too different from the classification label of the corresponding labeled image, the parameters of the initial classification model are adjusted according to the second loss function. The goal of the parameter adjustment is to narrow the difference between the classification prediction result and the classification label of the corresponding labeled image, so as to obtain a classification model that can better classify and predict the regional sample images, and realize the accurate identification and classification of green loans for enterprises.

[0084] Figure 3 This is a structural diagram of a green loan rating model in an embodiment of the present application.

[0085] like Figure 3 As shown, there are three branches from the regional sample image to the right. The first two branches from top to bottom are the pre-training stage of the initial feature extraction model, and the last branch is the fine-tuning stage of the initial classification model.

[0086] The first branch may be the first feature extraction branch of the aforementioned initial feature extraction model, and the second branch may be the second feature extraction branch of the aforementioned initial feature extraction model.

[0087] Through the pre-training stage, the initial encoder is trained as an encoder, which will be applied to the last branch and combined with the initial classifier to complete the fine-tuning of the initial classification model. That is, the model parameters of the encoder in the third branch are consistent with the model parameters of the encoder obtained by pre-training.

[0088] In a possible implementation manner, both the regional sample image and the regional image are hyperspectral images.

[0089] Hyperspectral images are images with hundreds of spectral channels. Hyperspectral images have high spatial resolution and high spectral resolution. They can capture subtle spectral features and spatial information of ground targets, and accurately identify high-energy consumption and high-pollution behaviors in the production areas of enterprises. Hyperspectral images can be collected in a variety of ways, such as remote sensing satellites, aircraft, and drones. By selecting or extracting spectral information of specific bands, hyperspectral images can highlight target features and effectively distinguish the spectral characteristics of different substances, thereby accurately locating high-energy consumption and high-pollution substances. Using hyperspectral images as the information source of regional sample images and regional images has broadened the source of green loan monitoring information and improved the accuracy of green loan monitoring information.

[0090] In a possible implementation, in the process of obtaining the mask image, the masker is also used to perform masking processing of spatial information on the regional sample image.

[0091] The regional sample image is a hyperspectral image, and the masker masks the spectral information and spatial information of the hyperspectral image. The initial encoder extracts features from the regional sample image after the spectral information and spatial information are masked to obtain a second sample feature. A first loss function is generated according to the difference between the first sample feature and the second sample feature, and the initial encoder is parameter-adjusted through the first loss function to obtain an encoder. Since the encoder obtains the second sample feature by extracting features from the regional sample image after the spectral information and spatial information are masked, the second sample feature extracted from the regional sample image with more missing information is made to approach the first sample feature extracted from the complete regional sample image through parameter adjustment, which can effectively improve the feature extraction generalization ability of the encoder.

[0092] In a possible implementation, the regional image used for green loan requirement identification is periodically acquired during the time period from when the enterprise to be identified applies for a green loan project to when the enterprise completes the green loan project.

[0093] During the time period from when the enterprise to be identified applies for a green loan project to when it completes the green loan project, regional images used for identifying green loan requirements are periodically acquired. Based on the identified regional images, effective monitoring can be carried out before, during and after the green loan period to ensure that during the entire cycle of the enterprise to be identified applying for a green loan project, the regional images used for identifying green loan requirements of the enterprise to be identified are effectively monitored, thereby improving the transparency and regulatory efficiency of green loan projects.

[0094] In a possible implementation, when the regional images are obtained during the application of the green loan project by the enterprise to be identified, these images will be used as an important basis for evaluating whether the enterprise to be identified meets the various standards of green loans. If the identification results determine that the enterprise to be identified does not meet the green loan requirements, the green loan project application of the enterprise to be identified will be rejected, that is, the green loan application of the enterprise to be identified will be rejected; when the regional images are obtained after the application of the green loan project by the enterprise to be identified is passed, the regional images become an important means to monitor the implementation of the project and ensure the compliance of the loan use. When it is found that the enterprise fails to implement the project in accordance with the requirements of the green loan, that is, it is determined by the identification results that the enterprise to be identified does not meet the green loan requirements, such as there are environmental pollution, waste of resources or ecological damage, the enterprise to be identified will be warned against the green loan project. This ensures that the implementation of the green loan project always remains under strict green loan standards, and improves the transparency and regulatory efficiency of the green loan project.

[0095] Figure 2 This is a flow chart of a green loan rating method in an embodiment of the present application.

[0096] 1. Model initialization

[0097] At the beginning, the green loan classification model (i.e. the initial feature extraction model and initial classification model mentioned above) is initialized.

[0098] 2. First data collection

[0099] The first hyperspectral image is collected for the production area of ​​the enterprise to be identified.

[0100] 3. Model pre-training

[0101] Obtain N regional sample images, obtain the first sample feature and the second sample feature through the initial feature extraction model based on the N regional sample images, generate the first loss function according to the difference between the first sample feature and the second sample feature, and adjust the parameters of the initial encoder through the first loss function to obtain the encoder, and determine the classification model according to the encoder and the classifier. For the regional image of the enterprise production area of ​​the enterprise to be identified, determine the recognition result through the classification model.

[0102] 4. Fine-tuning the model

[0103] C of the N regional sample images are classified and labeled to obtain labeled images including classification labels. The encoder and the classifier are set based on the input order to obtain an initial classification model. Based on the C labeled images, the corresponding classification prediction results are obtained through the initial classification model. According to the difference between the classification prediction results and the classification labels of the corresponding labeled images, a second loss function is generated, and the parameters of the initial classification model are adjusted through the second loss function to obtain a classification model.

[0104] 5. Periodic data collection and green credit rating

[0105] Periodically acquire regional image data for identifying green loan requirements and conduct green loan rating, that is, acquire regional image data for identifying green loan requirements before, during and after the green loan, and conduct green loan rating.

[0106] 6. Determine whether the loan cycle has ended

[0107] Determine whether the loan cycle has ended. If not, continue to periodically obtain regional image data for green loan requirement identification and perform green loan rating. Otherwise, end.

[0108] Exemplary Devices

[0109] Figure 4 A green loan rating device provided for this application comprises:

[0110] An acquisition unit 401 is used to acquire N regional sample images, where the N regional sample images are collected from the production area of ​​the enterprise;

[0111] An extraction unit 402 is used to obtain a first sample feature and a second sample feature through an initial feature extraction model based on the N regional sample images, wherein the initial feature model includes a first feature extraction branch and a second feature extraction branch, wherein the second feature extraction branch is provided with a masker and an initial encoder based on an input sequence, wherein the masker is used to perform image masking processing on the regional sample image to obtain a mask image, and the initial encoder is used to perform feature extraction on the mask image to obtain the second sample feature;

[0112] A generating unit 403 is configured to generate a first loss function according to a difference between the first sample feature and the second sample feature, and adjust parameters of the initial encoder by using the first loss function to obtain an encoder;

[0113] A first determining unit 404 is used to determine a classification model according to the encoder and the classifier;

[0114] The second determination unit 405 is used to determine a recognition result through the classification model for the regional image of the enterprise production area of ​​the enterprise to be identified, and the recognition result is used to identify whether the enterprise to be identified meets the green loan requirements.

[0115] As can be seen from the above, the acquisition unit 401 of the technical solution of the present application acquires N regional sample images, and the N regional sample images are collected from the production area of ​​the enterprise. The extraction unit 402 obtains the first sample feature and the second sample feature through the initial feature extraction model based on the N regional sample images. The initial feature model includes a first feature extraction branch and a second feature extraction branch. The second feature extraction branch is provided with a masker and an initial encoder based on the input order. The masker is used to perform image masking on the regional sample image to obtain a mask image. The initial encoder is used to perform feature extraction on the mask image to obtain the second sample feature. Since the initial encoder inputs the mask image after the masking process of the masker, the second sample feature extracted by the initial encoder from the mask image is different from the first sample feature extracted by the first feature branch. The generation unit 403 generates a first loss function according to the difference between the first sample feature and the second sample feature, and adjusts the parameters of the initial encoder through the first loss function to obtain an encoder. The encoder is obtained by adjusting the parameters of the initial encoder according to the first loss function, so that the encoder learns how to restore the defective content from sample images with different degrees of defect under the guidance of the first loss function and extracts high-quality image features. The second sample feature extracted by the encoder according to the mask image is close to the first sample feature, the ability of the encoder to extract more accurate features using the mask image is improved, and the generalization and robustness of the feature extraction of the encoder are enhanced. The first determination unit 404 determines the classification model according to the encoder and the classifier, and the second determination unit 405 determines the recognition result through the classification model for the regional image of the enterprise production area of ​​the enterprise to be identified, and the recognition result is used to identify whether the enterprise to be identified meets the green loan requirements. From the above, it can be seen that since the encoder adopts a self-supervised training method and does not rely on the labeled data provided by the applicant, a large number of regional sample images can be used for training during the training process, so as to obtain a high-quality encoder, which provides strong support for the subsequent identification of the green loan requirements of the enterprise to be identified.

[0116] As a possible implementation manner, the first determining unit 404 is specifically configured to:

[0117] Classify and annotate C regional sample images among the N regional sample images to obtain an annotated image including a classification label, wherein the classification label is used to identify whether the enterprise production area in the corresponding annotated image meets the green loan requirements;

[0118] Setting the encoder and the classifier based on the input order to obtain an initial classification model;

[0119] Based on the C labeled images, obtaining corresponding classification prediction results through the initial classification model;

[0120] A second loss function is generated according to the difference between the classification prediction result and the classification label of the corresponding annotated image, and the parameters of the initial classification model are adjusted by the second loss function to obtain the classification model.

[0121] As a possible implementation manner, both the regional sample image and the regional image are hyperspectral images.

[0122] As a possible implementation manner, in the process of obtaining the mask image, the masker is further used to perform masking processing of spatial information on the regional sample image.

[0123] As a possible implementation manner, the device further includes a first acquiring unit, configured to:

[0124] During the time period from when the enterprise to be identified applies for a green loan project to when the enterprise completes the green loan project, the regional image used for identifying green loan requirements is periodically acquired.

[0125] As a possible implementation manner, the device further includes a passing unit and an alarm unit, which are used to:

[0126] A passing unit, configured to, when the regional image is obtained during the period when the enterprise to be identified applies for a green loan project, in response to determining through the identification result that the enterprise to be identified does not meet the green loan requirements, perform a non-passing process on the application of the enterprise to be identified for the green loan project;

[0127] An alarm unit is used to, when the area image is obtained after the application of the enterprise to be identified for the green loan project is approved, in response to determining through the identification result that the enterprise to be identified does not meet the green loan requirements, issue an alarm to the enterprise to be identified for the green loan project.

[0128] The present application provides an electronic device, the electronic device comprising:

[0129] Memory for storing computer programs or computer instructions;

[0130] The processor is used to execute the computer program or computer instruction stored in the memory, so that the electronic device executes the method provided by the above embodiment.

[0131] The embodiment of the present application also provides a computer program product including instructions, which, when executed on a computing device, enables the computing device to execute the method provided in the above embodiment.

[0132] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program so that the computer program executes the method provided in the above embodiment.

[0133] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0134] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, in which the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0135] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A green loan rating method, characterized in that: The method comprises: Obtaining N regional sample images, where the N regional sample images are collected from the production area of ​​the enterprise; Based on the N regional sample images, a first sample feature and a second sample feature are obtained through an initial feature extraction model, wherein the initial feature model includes a first feature extraction branch and a second feature extraction branch, wherein the second feature extraction branch is provided with a masker and an initial encoder based on an input order, wherein the masker is used to perform image masking processing on the regional sample image to obtain a masked image, and the initial encoder is used to perform feature extraction on the masked image to obtain the second sample feature; Generate a first loss function according to the difference between the first sample feature and the second sample feature, and adjust the parameters of the initial encoder by using the first loss function to obtain an encoder; Determine a classification model according to the encoder and the classifier; For the regional image of the enterprise production area of ​​the enterprise to be identified, the identification result is determined by the classification model, and the identification result is used to indicate whether the enterprise to be identified meets the green loan requirements.

2. The method according to claim 1, characterized in that Determining a classification model according to the encoder and the classifier includes: Classify and annotate C regional sample images among the N regional sample images to obtain an annotated image including a classification label, wherein the classification label is used to identify whether the enterprise production area in the corresponding annotated image meets the green loan requirements; Setting the encoder and the classifier based on the input order to obtain an initial classification model; Based on the C labeled images, obtaining corresponding classification prediction results through the initial classification model; A second loss function is generated according to the difference between the classification prediction result and the classification label of the corresponding annotated image, and the parameters of the initial classification model are adjusted by the second loss function to obtain the classification model.

3. The method according to claim 1, characterized in that: The regional sample image and the regional image are both hyperspectral images.

4. The method according to claim 3, characterized in that In the process of obtaining the mask image, the masker is further used to perform mask processing of spatial information on the regional sample image.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: During the time period from when the enterprise to be identified applies for a green loan project to when the enterprise completes the green loan project, the regional image used for identifying green loan requirements is periodically acquired.

6. The method according to claim 5, characterized in that The method further comprises: When the regional image is obtained during the period when the enterprise to be identified applies for a green loan project, in response to determining through the identification result that the enterprise to be identified does not meet the green loan requirements, the application of the enterprise to be identified for the green loan project is rejected; When the regional image is obtained after the application for the green loan project by the enterprise to be identified is approved, in response to determining through the identification result that the enterprise to be identified does not meet the green loan requirements, an alarm is issued to the enterprise to be identified regarding the green loan project.

7. A green loan rating device, characterized in that: The device comprises: An acquisition unit, used to acquire N regional sample images, where the N regional sample images are acquired from the production area of ​​the enterprise; An extraction unit, configured to obtain a first sample feature and a second sample feature through an initial feature extraction model based on the N regional sample images, wherein the initial feature model includes a first feature extraction branch and a second feature extraction branch, wherein the second feature extraction branch is provided with a masker and an initial encoder based on an input sequence, wherein the masker is configured to perform image masking processing on the regional sample image to obtain a masked image, and the initial encoder is configured to perform feature extraction on the masked image to obtain the second sample feature; a generating unit, configured to generate a first loss function according to a difference between the first sample feature and the second sample feature, and adjust parameters of the initial encoder by using the first loss function to obtain an encoder; A first determining unit, configured to determine a classification model according to the encoder and the classifier; The second determination unit is used to determine the recognition result through the classification model for the regional image of the enterprise production area of ​​the enterprise to be identified, and the recognition result is used to identify whether the enterprise to be identified meets the green loan requirements.

8. An electronic device, characterized in that: The electronic device comprises: Memory for storing computer programs or computer instructions; A processor, configured to execute a computer program or computer instruction stored in the memory, so that the electronic device executes the method as claimed in any one of claims 1 to 6.

9. A computer program product, characterized in that When the computer program product is run on a computer, the computer is caused to execute the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The method comprises computer instructions, wherein the computer instructions instruct a computer to execute the method according to any one of claims 1 to 6.

Images