Method, device and equipment for evaluating repayment capability of loan farmer, medium and product

Abstract

The invention provides a loan farmer repayment capability assessment method, device and equipment, a medium and a product, and relates to the field of financial science and technology. Firstly, agricultural product data and a first image set of a planting field of a loan farmer are acquired; wherein the agricultural product data comprises at least one of planting area, planting variety, growth environment, climate condition and historical yield; inputting the first image set into a pre-trained diffusion model, and generating a prediction frame of agricultural product fruit detection and a category label of the agricultural product fruit; performing detection and counting according to the prediction frame and the category label to obtain the agricultural product yield of the borrowing farmer in the current time period; according to the agricultural product yield and the agricultural product data in the current time period, the agricultural product yield of the borrowing farmer in the target time period is predicted; and determining the repayment capability of the borrowing farmer according to the agricultural product yield in the target time period.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a method, device, equipment, medium and product for evaluating the repayment ability of borrowing farmers. Background Art

[0002] With the rise of digital finance and inclusive finance in recent years, financial institutions have actively responded to rural and nature conservation needs, balanced multiple financial resources, explored practical and convenient rural inclusive finance and efficient and sustainable green inclusive finance, thereby accelerating rural modernization and achieving integrated urban and rural development.

[0003] The traditional credit process is that after a borrower submits a credit application to a financial institution or lending institution, the financial institution will review and evaluate the borrower's credit status and repayment ability through pre-loan investigations, on-site inspections, and verification of various materials. The financial institution will decide whether to approve the loan application based on the results of the review and evaluation, and will pay close attention to the borrower's assets during and after the loan.

[0004] This strict and cautious approach requires bank credit department managers to conduct on-site inspections of lender departments and other industrial sites, and manually calculate and review borrower assets, which results in a lengthy review process. Summary of the invention

[0005] The present application provides a method, device, equipment, medium and product for evaluating the repayment ability of borrowing farmers, which is used to solve the technical problem of the long review process when farmers borrow money.

[0006] In a first aspect, the present application provides a method for evaluating the repayment ability of a borrowing farmer, comprising: obtaining agricultural product data and a first image set of the planting land of the borrowing farmer; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climatic conditions, and historical yield; inputting the first image set into a pre-trained diffusion model to generate a prediction box for agricultural product fruit detection and a category label for the agricultural product fruit; performing detection and counting based on the prediction box and the category label to obtain the agricultural product yield of the borrowing farmer in the current time period; predicting the agricultural product yield of the borrowing farmer in the target time period based on the agricultural product yield and agricultural product data in the current time period; and determining the repayment ability of the borrowing farmer based on the agricultural product yield in the target time period.

[0007] In a second aspect, the present application provides an evaluation device for the repayment ability of a borrowing farmer, comprising: an acquisition module, used to obtain agricultural product data and an image set of the borrowing farmer's planting site; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climatic conditions, and historical yield; a generation model; used to input the image set into a pre-trained diffusion model to generate a prediction box for agricultural product fruit detection and a category label for agricultural product fruit; detection and counting are performed based on the prediction box and the category label to obtain the agricultural product yield of the borrowing farmer in the current time period; a processing module, used to predict the agricultural product yield of the borrowing farmer in the target time period based on the agricultural product yield and agricultural product data in the current time period; and determine the repayment ability of the borrowing farmer based on the agricultural product yield in the target time period.

[0008] In a third aspect, the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the above method.

[0009] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the above method.

[0010] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0011] The present application provides a method, device, equipment, medium and product for evaluating the repayment ability of a borrowing farmer, which obtains agricultural product data and a first image set of the planting area of ​​the borrowing farmer; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climate conditions, and historical yield; the first image set is input into a pre-trained diffusion model to generate a prediction frame for agricultural product fruit detection and a category label for agricultural product fruit; detection and counting are performed according to the prediction frame and category label to obtain the agricultural product yield of the borrowing farmer in the current period; based on the agricultural product yield and agricultural product data in the current period, the agricultural product yield of the borrowing farmer in the target period is predicted; based on the agricultural product yield in the target period, the repayment ability of the borrowing farmer is determined. This solution detects and counts the fruit targets in the orchard by using the method of generating targets through a diffusion model, so that early yield prediction can be achieved when the fruit is not mature, and the farmer's future income and repayment ability can be calculated and evaluated based on this method. It is simple and efficient, and can greatly improve the calculation speed and accuracy, and shorten the bank's review time. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0013] Figure 1 A schematic diagram of the process of evaluating the repayment ability of farmers who borrow money provided for this application Figure 1 ;

[0014] Figure 2 A schematic diagram of a method for evaluating the repayment ability of borrowing farmers Figure 2 ;

[0015] Figure 3 This is the overall architecture diagram of the diffusion model;

[0016] Figure 4 A schematic diagram of the process of extracting features for the feature extraction network model;

[0017] Figure 5 This is a structural diagram of the weighted bidirectional feature pyramid network model;

[0018] Figure 6 A schematic diagram of the structure of a device for evaluating the repayment ability of a borrowing farmer provided in this application;

[0019] Figure 7 A schematic diagram of the structure of a device for evaluating the repayment ability of a farmer who has borrowed money provided in this application.

[0020] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0021] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0022] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0023] In addition, this application involves conducting big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and using artificial intelligence technology to make automated decisions, and making technical solutions that have a significant impact on personal rights and interests based on the results of automated decisions. The application provides users with corresponding operation entrances for them to choose to agree or reject the results of automated decisions; if the user chooses to reject, the expert decision-making process will be entered.

[0024] It should be noted that the method, device, equipment, medium and product for evaluating the repayment ability of borrowing farmers provided in this application can be used in the field of financial technology, and can also be used in any field other than financial technology. The application field of the method, device, equipment, storage medium and product for evaluating the repayment ability of borrowing farmers in this application is not limited.

[0025] With the rise of digital finance and inclusive finance in recent years, financial institutions have actively responded to rural and nature conservation needs, balanced multiple financial resources, explored practical and convenient rural inclusive finance and efficient and sustainable green inclusive finance, thereby accelerating rural modernization and achieving integrated urban and rural development.

[0026] The traditional credit process is that after a borrower submits a credit application to a financial institution or lending institution, the financial institution will review and evaluate the borrower's credit status and repayment ability through pre-loan investigations, on-site inspections, and verification of various materials. The financial institution will decide whether to approve the loan application based on the results of the review and evaluation, and will pay close attention to the borrower's assets during and after the loan.

[0027] This strict and cautious approach requires bank credit department managers to conduct on-site inspections of lender departments and other industrial sites, and manually calculate and review borrower assets, which results in a lengthy review process.

[0028] The technical solution provided by the present application is intended to solve the above technical problems of the prior art. The present application provides a method, device, equipment, medium and product for evaluating the repayment ability of a borrowing farmer, which obtains the agricultural product data and the first image set of the planting land of the borrowing farmer; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climatic conditions, and historical yield; the first image set is input into a pre-trained diffusion model to generate a prediction frame for agricultural product fruit detection and a category label for agricultural product fruit; detection and counting are performed according to the prediction frame and category label to obtain the agricultural product yield of the borrowing farmer in the current period; based on the agricultural product yield and agricultural product data in the current period, the agricultural product yield of the borrowing farmer in the target period is predicted; based on the agricultural product yield in the target period, the repayment ability of the borrowing farmer is determined. This solution detects and counts the agricultural product fruit target by the method of generating targets through a diffusion model, so that early yield prediction can be achieved when the fruit is not mature, and the farmer's future income and repayment ability can be calculated and evaluated based on this method. It is simple and efficient, and can greatly improve the calculation speed and accuracy, and shorten the bank's review time.

[0029] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0030] Embodiment 1

[0031] Figure 1 A schematic diagram of the process of evaluating the repayment ability of farmers who borrow money provided for this application Figure 1 ,like Figure 1 As shown, including:

[0032] S101, obtaining agricultural product data and a first image set of the planting land of the borrowing farmer; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climate conditions, and historical yield;

[0033] S102, inputting the first image set into the pre-trained diffusion model to generate a prediction frame for agricultural product fruit detection and a category label for agricultural product fruit; performing detection and counting based on the prediction frame and the category label to obtain the agricultural product output of the borrowing farmer in the current period;

[0034] S103, predicting the agricultural product output of the borrowing farmer in the target period based on the agricultural product output and agricultural product data in the current period; and determining the repayment capacity of the borrowing farmer based on the agricultural product output in the target period.

[0035] Among them, the agricultural product data and the first image set of the loan farmers' planting areas can be obtained through questionnaire surveys, field surveys, drone aerial photography and public data analysis.

[0036] Diffusion models are a type of probability-based generative model; its basic idea is to learn the distribution of data by simulating a diffusion process, and then generate new data. Diffusion models are used to generate prediction boxes and category labels for agricultural products, based on which agricultural products can be accurately detected and counted, thereby achieving early yield prediction and determining the repayment ability of borrowing farmers.

[0037] In practical applications, taking orchard farmers as an example, drones are first used to move in the orchard to obtain real-time fruit images with complex backgrounds as the first image set, and then the diffusion model is used to generate prediction boxes and fruit category labels for fruit detection. Then, the fruit targets in the orchard are detected and counted based on the prediction boxes and fruit category labels for fruit detection, so that early yield prediction can be achieved when the fruit is not yet mature, and this can be used to calculate and evaluate farmers' future income and repayment ability.

[0038] The present embodiment provides a method for evaluating the repayment ability of a borrowing farmer, which obtains agricultural product data and a first image set of the planting land of the borrowing farmer; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climate conditions, and historical yield; the first image set is input into a pre-trained diffusion model to generate a prediction frame for agricultural product fruit detection and a category label for agricultural product fruit; detection and counting are performed according to the prediction frame and the category label to obtain the agricultural product yield of the borrowing farmer in the current period; the agricultural product yield of the borrowing farmer in the target period is predicted according to the agricultural product yield and agricultural product data in the current period; the repayment ability of the borrowing farmer is determined according to the agricultural product yield in the target period. The agricultural product fruit target is detected and counted by the method of generating targets through the diffusion model, so that early yield prediction can be achieved when the fruit is not mature, and the farmer's future income and repayment ability can be calculated and evaluated based on this method, which is simple and efficient, can greatly improve the calculation speed and accuracy, and shorten the bank's review time.

[0039] Optionally, images in the second image set whose clarity does not meet a preset standard are eliminated; data enhancement is performed on the retained second image set, and the second image set is divided into a training set, a test set, and a validation set; the training set, the test set, and the validation set are used to train a diffusion model.

[0040] Among them, the preset standard of clarity is set according to the requirements based on resolution and edge sharpness. In practical applications, in addition to the standard of clarity set according to the requirements based on resolution and edge sharpness, a large number of methods for evaluating image clarity are also used through human eye observation.

[0041] The second image set includes images of different angles, different lighting conditions, and different types. After preliminary screening, images with insufficient clarity or too low pixels are eliminated. Data enhancement is performed on the retained second image set, including rotation, flipping, scaling, cropping, brightness adjustment, etc. The second image set can be divided into training set, test set, and validation set according to a certain ratio, among which the ratio is 70% - 15% - 15%: 70% as training set, 15% for test set, and 15% as validation set. This division ratio allows the model to fully learn and build a relatively stable feature extraction and decision-making mechanism, while not overly relying on all data, thereby retaining a certain generalization ability. In addition, it helps to control the risk of overfitting in the early training stage; there is also an 80% - 10% - 10% ratio division: 80% as training set, 10% for test set, and 10% as validation set. This division method provides the model with richer data resources to learn deep features and helps to improve the performance upper limit of the model.

[0042] In addition to these two proportion division methods, it can also be divided according to specific task requirements and data size, distribution, etc. For example, if the amount of data is small, in order to allow the model to have enough data for training, a 60% - 20% - 20% division method may be adopted.

[0043] By screening, enhancing, and dividing the image set, images that do not meet the clarity standards can be screened out to prevent these images from interfering with model training. In addition, enhancing the image set can increase the diversity of the image set, and dividing the image set can not only make rational use of data for model training and evaluation, but also ensure the fairness and reliability of the evaluation.

[0044] Optional, Figure 2 A schematic diagram of a method for evaluating the repayment ability of borrowing farmers Figure 2 ,like Figure 2 As shown, before the first image set is input into the pre-trained diffusion model, it also includes: the diffusion model includes an image encoder, a neural network model, and a detection decoder;

[0045] S201, annotating according to the training set to obtain image bounding box information; wherein the image bounding box information includes a bounding box, an image with a bounding box, and a category label;

[0046] S202, inputting the image with the bounding box into an image encoder to obtain a feature map of the image; inputting the bounding box into a neural network model to obtain a noise box; and cutting out features of a region of interest of the noise box according to the noise box and the feature map of the image;

[0047] S203, inputting the region of interest features and category labels of the noise frame into the detection decoder to predict and generate a true bounding box and a true category label;

[0048] S204, calculating the difference between the bounding box, the true bounding box, the category label and the true category label according to the loss function to obtain a loss value; and updating the model parameters by performing gradient descent optimization according to the loss value to train the diffusion model.

[0049] In the detection and counting of agricultural products and fruits, it is very important to accurately locate the position of the bounding box. Therefore, the task of agricultural product and fruit detection is handled based on the diffusion model, and the detection task is regarded as a task of generating the position (center coordinates) and size (width and height) of the target bounding box. In this process, the model can accurately identify the fruit target from a series of noisy bounding boxes and predict and generate the corresponding precise bounding box.

[0050] The region of interest of the noise box refers to the area containing the target object or important information related to the task. The loss function is , Represents a neural network; neural network After training, z0 can be predicted from the noise box zt for the corresponding image x, and the corresponding target category label can be generated at the same time.

[0051] In addition, the main function of the detection decoder is to convert the region of interest features of the noise box into the final output of target detection, including the location (bounding box coordinates) and category information of the target object. In order to gradually learn the conversion process from the noisy bounding box to the accurate target bounding box, the detection decoder uses deep features as conditions to gradually refine the prediction of the box, recover the real bounding box from the noise box, and predict the category of the target object at the same time. In practical applications, in addition to using the detection decoder, structures such as Region Proposal Network (RPN) or Fully Convolutional Networks (FCN) can also be used to generate the location and category information of the target by performing more in-depth processing and analysis of the feature.

[0052] Figure 3 This is the overall architecture diagram of the diffusion model, such as Figure 3 As shown in Figure 1, the diffusion model includes an image encoder, a neural network model, and a detection decoder. As a deep generative model based on the likelihood method, the diffusion model starts from randomly distributed samples, gradually removes the noise in the image by reversing the diffusion process of the noise box, and restores the original state of the data sample, thereby generating an accurate boundary prediction box.

[0053] The neural network model adds noise to the bounding box to obtain a noise box. The noise addition process is defined as:

[0054]

[0055] in, It is used to control the noise scale by using a monotonically decreasing cosine schedule in different time steps t. Represents the noise variance table.

[0056] In practical applications, during the training phase, the real bounding box is continuously diffused into the random noise distribution, which enables the model to learn the noise modeling process, thereby adding Gaussian noise to the real bounding box to generate a noise box, thereby constructing the diffusion process from the real border to the noise box; then, according to the generated noise box and the feature map of the image, the region of interest of the noise box is cropped, and the region of interest features and category labels of the noise box are input into the detection decoder to predict and generate the real bounding box and the real category label; according to the loss function L train The difference between the bounding box, the true bounding box, the category label and the true category label is calculated to obtain the loss value; the model parameters are updated by gradient descent optimization according to the loss value to train the diffusion model.

[0057] Specifically, in the detection task, by adding noise to the input noise box, we can simulate various interference factors that may appear in real scenes, such as lighting changes, target occlusion, blur, etc., and train the model to be a more stable and adaptable detector.

[0058] By training the diffusion model composed of a neural network model, an image encoder, and a detection decoder, the model can maintain high accuracy even in complex and changeable orchard environments, providing strong support for actual agricultural product fruit detection and early yield prediction.

[0059] Optionally, the diffusion model is tested and optimized based on the test set and the validation set to obtain a trained diffusion model.

[0060] By evaluating the performance of the model on the validation set, the model's hyperparameters (such as learning rate, parameters of the network architecture, regularization strength, etc.) can be adjusted. For example, when training a deep neural network model, if the loss value of the model on the validation set is found to continue to rise with the increase in training rounds, this may mean that the model is overfitting. At this time, the performance of the model can be improved by adjusting the hyperparameters (such as increasing the regularization strength). During the model training process, the validation set can be used to monitor overfitting. By observing the performance difference between the model on the validation set and the training set, overfitting can be discovered in time. Using the test set and validation set to test and optimize the diffusion model avoids the situation where the model overfits on the training data and causes poor performance in actual applications, and ensures that the model has good generalization ability.

[0061] Optionally, the image encoder includes a feature extraction network model and a weighted bidirectional feature pyramid network model;

[0062] A feature extraction network model is used to extract features based on the input image with a bounding box, and obtain the depth features corresponding to the image with the bounding box;

[0063] The weighted bidirectional feature pyramid network model is used to generate a feature map of an image based on the deep features output by the feature extraction network model.

[0064] The image encoder aims to extract deep feature representation from the original fruit image. The image encoder includes a feature extraction network model (Swin Transforme) and a weighted bidirectional feature pyramid network model (Bidirectional Feature Pyramid Network, BiFPN). Swin Transformer is a cutting-edge backbone network based on the Transformer architecture. Through multi-level Transformer encoding blocks, it deeply mines the multi-dimensional deep features of the image, such as color, texture, shape, and size, to achieve deep interaction of features and precise fusion of information.

[0065] Specifically, Figure 4 The schematic diagram of the process of extracting features for the feature extraction network model is as follows: Figure 4 As shown in the figure, the feature extraction network model runs in stages, gradually extracting the deep features of the image with the bounding box. At the same time, the downsampling operation of block merging is used to reduce the resolution of the image, adjust the number of channels, and improve the computational efficiency of the model. The deep features extracted by SwinTransformer are then sent to the bidirectional feature pyramid network BiFPN, which performs multiple bidirectional fusions of high-level semantic information and low-level position information to generate the feature map of the image. Figure 5The figure is a structural diagram of the weighted bidirectional feature pyramid network model. Among them, high-level semantic information is a high-level abstract understanding of the image, and low-level position information mainly refers to the specific position and arrangement order of each element in the data (such as pixels, words, etc.); in the image, this includes the coordinate position of the pixel, the spatial position of the object in the image (such as the position of the bounding box), the spatial distribution of the texture, etc.

[0066] In the feature fusion stage, through bidirectional connections and cross-layer connections, the feature maps of the image can efficiently exchange and fuse information, thereby obtaining richer semantic information and achieving accurate extraction and effective fusion of deep features.

[0067] By using an image encoder that combines a feature extraction network model with a weighted bidirectional feature pyramid network model to extract the feature map of the image, it is possible to effectively capture various levels of features in the image and feature maps that are more discriminative.

[0068] Optionally, the repayment capacity of the borrowing farmers can be determined based on the agricultural product output during the target period, including:

[0069] The agricultural product output and agricultural product data in the target period are input into the evaluation model for evaluation to obtain the repayment capacity of the borrowing farmers; wherein the evaluation model is trained based on historical agricultural product data, agricultural product output in historical time periods and corresponding repayment capacity.

[0070] The agricultural product yields during the entire target period are detected and counted, and the future orchard yields are predicted based on the agricultural product data. The yield prediction results are then linked to the farmers' repayment ability to more accurately assess the farmers' financial health. Specifically, a high yield of agricultural products often means that farmers have stronger repayment power, while a poor yield reflects the relatively weak repayment ability of farmers.

[0071] In addition, we can also comprehensively consider many dimensions such as the production dynamics of the planting area, the credit history of the farmers, and the income and expenditure structure to fully understand the credit status of the farmers and accurately measure their future repayment potential.

[0072] Assessing farmers' repayment ability based on the assessment model can more comprehensively and systematically analyze farmers' financial status and potential risks and ensure that farmers obtain loans that match their own economic strength.

[0073] Optionally, formulate a loan plan for the borrowing farmers based on their repayment capacity.

[0074] Banks can flexibly adjust their credit strategies and tailor loan plans for farmers accordingly. For example, for farmers with excellent repayment ability, it is recommended to provide more favorable loan conditions, such as higher loan amounts; while for farmers with slightly insufficient repayment ability, it may be possible to adopt strategies such as reducing loan amounts or moderately increasing interest rates.

[0075] In addition, farmers can also more clearly understand their repayment ability based on the assessed repayment ability, and thus formulate a more reasonable repayment plan to ensure the health and stability of cash flow.

[0076] By formulating loan plans based on the assessed repayment capacity, farmers can avoid debt risks caused by excessive borrowing and financial institutions can maintain financial health and sustainable business development, thereby enhancing their competitiveness in the rural credit market.

[0077] Embodiment 2

[0078] Figure 6 A schematic diagram of a device for evaluating the repayment ability of a farmer who borrows money provided in this application, such as Figure 6 As shown, including:

[0079] The acquisition module 11 is used to acquire agricultural product data and image sets of the planting areas of the borrowing farmers; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climate conditions, and historical yield;

[0080] Generate model 12; input the image set into the pre-trained diffusion model to generate prediction frames for agricultural product fruit detection and category labels for agricultural product fruit; perform detection and counting based on the prediction frames and category labels to obtain the agricultural product output of the borrowing farmers in the current period;

[0081] The processing module 13 is used to predict the agricultural product output of the borrowing farmer in the target period according to the agricultural product output and agricultural product data in the current period; and determine the repayment ability of the borrowing farmer according to the agricultural product output in the target period.

[0082] Among them, the agricultural product data and the first image set of the loan farmers' planting areas can be obtained through questionnaire surveys, field surveys, drone aerial photography and public data analysis.

[0083] Diffusion models are a type of probability-based generative model; its basic idea is to learn the distribution of data by simulating a diffusion process, and then generate new data. Diffusion models are used to generate prediction boxes and category labels for agricultural products, based on which agricultural products can be accurately detected and counted, thereby achieving early yield prediction and determining the repayment ability of borrowing farmers.

[0084] The present embodiment provides a method for evaluating the repayment ability of a borrowing farmer, which obtains agricultural product data and a first image set of the planting land of the borrowing farmer; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climate conditions, and historical yield; the first image set is input into a pre-trained diffusion model to generate a prediction frame for agricultural product fruit detection and a category label for agricultural product fruit; detection and counting are performed according to the prediction frame and the category label to obtain the agricultural product yield of the borrowing farmer in the current period; the agricultural product yield of the borrowing farmer in the target period is predicted according to the agricultural product yield and agricultural product data in the current period; the repayment ability of the borrowing farmer is determined according to the agricultural product yield in the target period. The agricultural product fruit target is detected and counted by the method of generating targets through the diffusion model, so that early yield prediction can be achieved when the fruit is not mature, and the farmer's future income and repayment ability can be calculated and evaluated based on this method, which is simple and efficient, can greatly improve the calculation speed and accuracy, and shorten the bank's review time.

[0085] Optionally, the acquisition module 11 is also used to: eliminate images in the second image set whose clarity does not meet a preset standard; perform data enhancement on the retained second image set, and divide the second image set into a training set, a test set, and a validation set; the training set, the test set, and the validation set are used to train a diffusion model.

[0086] Among them, the preset standard of clarity is set according to the requirements based on resolution and edge sharpness. In practical applications, in addition to the standard of clarity set according to the requirements based on resolution and edge sharpness, a large number of methods for evaluating image clarity are also used through human eye observation.

[0087] The second image set includes images of different angles, different lighting conditions, and different types. After preliminary screening, images with insufficient clarity or too low pixels are eliminated. Data enhancement is performed on the retained second image set, including rotation, flipping, scaling, cropping, brightness adjustment, etc. The second image set can be divided into training set, test set, and validation set according to a certain ratio, among which the ratio is 70% - 15% - 15%: 70% as training set, 15% for test set, and 15% as validation set. This division ratio allows the model to fully learn and build a relatively stable feature extraction and decision-making mechanism, while not overly relying on all data, thereby retaining a certain generalization ability. In addition, it helps to control the risk of overfitting in the early training stage; there is also an 80% - 10% - 10% ratio division: 80% as training set, 10% for test set, and 10% as validation set. This division method provides the model with richer data resources to learn deep features and helps to improve the performance upper limit of the model.

[0088] In addition to these two proportion division methods, you can also divide it according to specific project requirements and data size, distribution, etc. For example, if the amount of data is small, in order to allow the model to have enough data for training, a 60% - 20% - 20% division method may be adopted.

[0089] By screening, enhancing, and dividing the image set, images that do not meet the clarity standards can be screened out to prevent these images from interfering with model training. In addition, enhancing the image set can increase the diversity of the image set, and dividing the image set can not only make rational use of data for model training and evaluation, but also ensure the fairness and reliability of the evaluation.

[0090] Optionally, before inputting the first image set into the pre-trained diffusion model, the method further includes: the diffusion model includes an image encoder, a neural network model, and a detection decoder;

[0091] Annotate the training set to obtain image bounding box information; the image bounding box information includes a bounding box, an image with a bounding box, and a category label;

[0092] Input the image with bounding box into the image encoder to obtain the feature map of the image; input the bounding box into the neural network model to obtain the noise box; according to the noise box and the feature map of the image, crop the features of the region of interest of the noise box;

[0093] Input the region of interest features and category labels of the noise box into the detection decoder to predict and generate the true bounding box and true category label;

[0094] The difference between the bounding box, the true bounding box, the category label and the true category label is calculated according to the loss function to obtain the loss value; the model parameters are updated by performing gradient descent optimization according to the loss value to train the diffusion model.

[0095] In the detection and counting of agricultural products and fruits, it is very important to accurately locate the position of the bounding box. Therefore, the task of agricultural product and fruit detection is handled based on the diffusion model, and the detection task is regarded as a task of generating the position (center coordinates) and size (width and height) of the target bounding box. In this process, the model can accurately identify the fruit target from a series of noisy bounding boxes and predict and generate the corresponding precise bounding box.

[0096] The region of interest of the noise box refers to the area containing the target object or important information related to the task. The loss function is , Represents a neural network; neural network After training, z0 can be predicted from the noise box zt for the corresponding image x, and the corresponding target category label can be generated at the same time.

[0097] In addition, the main function of the detection decoder is to convert the region of interest features of the noise box into the final output of target detection, including the location (bounding box coordinates) and category information of the target object. In order to gradually learn the conversion process from the noisy bounding box to the accurate target bounding box, the detection decoder uses deep features as conditions to gradually refine the prediction of the box, recover the real bounding box from the noise box, and predict the category of the target object at the same time. In practical applications, in addition to using the detection decoder, structures such as Region Proposal Network (RPN) or Fully Convolutional Networks (FCN) can also be used to generate the location and category information of the target by performing more in-depth processing and analysis of the feature.

[0098] The diffusion model includes an image encoder, a neural network model, and a detection decoder. As a deep generative model based on the likelihood method, the diffusion model starts from randomly distributed samples, gradually removes the noise in the image by reversing the diffusion process of the noise box, and restores the original state of the data sample, thereby generating an accurate boundary prediction box.

[0099] The neural network model adds noise to the bounding box to obtain a noise box. The noise addition process is defined as:

[0100]

[0101] in, It is used to control the noise scale by using a monotonically decreasing cosine schedule in different time steps t. Represents the noise variance table.

[0102] In practical applications, during the training phase, the real bounding box is continuously diffused into the random noise distribution, which enables the model to learn the noise modeling process, thereby adding Gaussian noise to the real bounding box to generate a noise box, thereby constructing the diffusion process from the real border to the noise box; then, according to the generated noise box and the feature map of the image, the region of interest of the noise box is cropped, and the region of interest features and category labels of the noise box are input into the detection decoder to predict and generate the real bounding box and the real category label; according to the loss function L train The difference between the bounding box, the true bounding box, the category label and the true category label is calculated to obtain the loss value; the model parameters are updated by gradient descent optimization according to the loss value to train the diffusion model.

[0103] Specifically, in the detection task, by adding noise to the input noise box, we can simulate various interference factors that may appear in real scenes, such as lighting changes, target occlusion, blur, etc., and train the model to be a more stable and adaptable detector.

[0104] By training the diffusion model composed of a neural network model, an image encoder, and a detection decoder, the model can maintain high accuracy even in complex and changeable orchard environments, providing strong support for actual agricultural product fruit detection and early yield prediction.

[0105] Optionally, the diffusion model is tested and optimized based on the test set and the validation set to obtain a trained diffusion model.

[0106] By evaluating the performance of the model on the validation set, the model's hyperparameters (such as learning rate, parameters of the network architecture, regularization strength, etc.) can be adjusted. For example, when training a deep neural network model, if the loss value of the model on the validation set is found to continue to rise with the increase in training rounds, this may mean that the model is overfitting. At this time, the performance of the model can be improved by adjusting the hyperparameters (such as increasing the regularization strength). During the model training process, the validation set can be used to monitor overfitting. By observing the performance difference between the model on the validation set and the training set, overfitting can be discovered in time. Using the test set and validation set to test and optimize the diffusion model avoids the situation where the model overfits on the training data and causes poor performance in actual applications, and ensures that the model has good generalization ability.

[0107] Optionally, the image encoder includes a feature extraction network model and a weighted bidirectional feature pyramid network model;

[0108] A feature extraction network model is used to extract features based on the input image with a bounding box, and obtain the depth features corresponding to the image with the bounding box;

[0109] The weighted bidirectional feature pyramid network model is used to generate a feature map of an image based on the deep features output by the feature extraction network model.

[0110] The image encoder aims to extract deep feature representation from the original fruit image. The image encoder includes a feature extraction network model (Swin Transforme) and a weighted bidirectional feature pyramid network model (Bidirectional Feature Pyramid Network, BiFPN). Swin Transformer is a cutting-edge backbone network based on the Transformer architecture. Through multi-level Transformer encoding blocks, it deeply mines the multi-dimensional deep features of the image, such as color, texture, shape, and size, to achieve deep interaction of features and precise fusion of information.

[0111] Specifically, the feature extraction network model runs in stages, gradually extracting the deep features of the image with the bounding box. At the same time, the downsampling operation of block merging is used to reduce the resolution of the image, adjust the number of channels, and improve the computational efficiency of the model. The deep features extracted by Swin Transformer are then sent to the bidirectional feature pyramid network BiFPN, which performs multiple bidirectional fusions of high-level semantic information and low-level position information to generate the feature map of the image. Among them, high-level semantic information is a high-level abstract understanding of the image, and low-level position information mainly refers to the specific position and arrangement order of each element in the data (such as pixels, words, etc.); in the image, this includes the coordinate position of the pixel, the spatial position of the object in the image (such as the bounding box position), the spatial distribution of the texture, etc.

[0112] In the feature fusion stage, through bidirectional connections and cross-layer connections, the feature maps of the image can efficiently exchange and fuse information, thereby obtaining richer semantic information and achieving accurate extraction and effective fusion of deep features.

[0113] By using an image encoder that combines a feature extraction network model with a weighted bidirectional feature pyramid network model to extract the feature map of the image, it is possible to effectively capture various levels of features in the image and feature maps that are more discriminative.

[0114] Optionally, the repayment capacity of the borrowing farmers can be determined based on the agricultural product output during the target period, including:

[0115] The agricultural product output and agricultural product data in the target period are input into the evaluation model for evaluation to obtain the repayment capacity of the borrowing farmers; wherein the evaluation model is trained based on historical agricultural product data, agricultural product output in historical time periods and corresponding repayment capacity.

[0116] The agricultural product yields during the entire target period are detected and counted, and the future orchard yields are predicted based on the agricultural product data. The yield prediction results are then linked to the farmers' repayment ability to more accurately assess the farmers' financial health. Specifically, a high yield of agricultural products often means that farmers have stronger repayment power, while a poor yield reflects the relatively weak repayment ability of farmers.

[0117] In addition, we can also comprehensively consider many dimensions such as the production dynamics of the planting area, the credit history of the farmers, and the income and expenditure structure to fully understand the credit status of the farmers and accurately measure their future repayment potential.

[0118] Assessing farmers' repayment ability based on the assessment model can more comprehensively and systematically analyze farmers' financial status and potential risks and ensure that farmers obtain loans that match their own economic strength.

[0119] Optionally, formulate a loan plan for the borrowing farmers based on their repayment capacity.

[0120] Banks can flexibly adjust their credit strategies and tailor loan plans for farmers accordingly. For example, for farmers with excellent repayment ability, it is recommended to provide more favorable loan conditions, such as higher loan amounts; while for farmers with slightly insufficient repayment ability, it may be possible to adopt strategies such as reducing loan amounts or moderately increasing interest rates.

[0121] In addition, farmers can also more clearly understand their repayment ability based on the assessed repayment ability, and thus formulate a more reasonable repayment plan to ensure the health and stability of cash flow.

[0122] By formulating loan plans based on the assessed repayment capacity, farmers can avoid debt risks caused by excessive borrowing and financial institutions can maintain financial health and sustainable business development, thereby enhancing their competitiveness in the rural credit market.

[0123] Figure 7 This is a schematic diagram of a device for evaluating the repayment ability of a farmer who borrows money provided in this application. Figure 7 As shown, the electronic device 70 provided in this embodiment includes: at least one processor 701 and a memory 702. Optionally, the device 70 further includes a communication component 703. The processor 701, the memory 702 and the communication component 703 are connected via a bus 704.

[0124] In a specific implementation process, at least one processor 701 executes the computer-executable instructions stored in the memory 702, so that at least one processor 701 executes the above method.

[0125] The specific implementation process of the processor 701 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0126] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0127] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0128] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0129] It should be further noted that, although the various steps in the flowchart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0130] It should be understood that the above-mentioned device embodiments are only illustrative, and the device of the present application can also be implemented in other ways. For example, the division of units / modules in the above-mentioned embodiments is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units, modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed.

[0131] In the above embodiments, the description of each embodiment has its own emphasis. For the part not described in detail in a certain embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0132] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0133] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for evaluating the repayment ability of borrowing farmers, characterized in that: include: Acquire agricultural product data and a first image set of the planting land of the borrowing farmer; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growing environment, climate conditions, and historical yield; Inputting the first image set into a pre-trained diffusion model to generate a prediction frame for agricultural product fruit detection and a category label for agricultural product fruit; performing detection and counting based on the prediction frame and the category label to obtain the agricultural product output of the borrowing farmer in the current period; The agricultural product output of the borrowing farmer in the target period is predicted based on the agricultural product output in the current period and the agricultural product data; and the repayment ability of the borrowing farmer is determined based on the agricultural product output in the target period.

2. The method according to claim 1, characterized in that: The method further comprises: The images whose clarity does not meet the preset standard in the second image set are eliminated; data enhancement is performed on the retained second image set, and the second image set is divided into a training set, a test set and a validation set; the training set, the test set and the validation set are used to train a diffusion model.

3. The method according to claim 2, characterized in that Before inputting the first image set into the pre-trained diffusion model, the method further includes: the diffusion model includes an image encoder, a neural network model, and a detection decoder; Annotating the training set to obtain image bounding box information; wherein the image bounding box information includes a bounding box, an image with a bounding box, and a category label; Inputting the image with the bounding box into the image encoder to obtain a feature map of the image; inputting the bounding box into the neural network model to obtain a noise box; and cutting out the features of the region of interest of the noise box according to the noise box and the feature map of the image; Inputting the region of interest features of the noise frame and the category label into the detection decoder to predict and generate a true bounding box and a true category label; The difference between the bounding box, the true bounding box, the category label and the true category label is calculated according to the loss function to obtain a loss value; and the model parameters are updated by performing gradient descent optimization according to the loss value to train the diffusion model.

4. The method according to claim 3, characterized in that After the diffusion model is trained, the method further includes: The diffusion model is tested and optimized based on the test set and the validation set to obtain a trained diffusion model.

5. The method according to claim 3, characterized in that: The image encoder includes a feature extraction network model and a weighted bidirectional feature pyramid network model; The feature extraction network model is used to extract features according to the input image with a bounding box to obtain a depth feature corresponding to the image with a bounding box; The weighted bidirectional feature pyramid network model is used to generate a feature map of an image based on the deep features output by the feature extraction network model.

6. The method according to claim 1, characterized in that Determining the repayment capacity of the borrowing farmer according to the agricultural product output in the target period specifically includes: The agricultural product output in the target period and the agricultural product data are input into the evaluation model for evaluation to obtain the repayment capacity of the borrowing farmer; wherein the evaluation model is trained based on historical agricultural product data, agricultural product output in historical time periods and corresponding repayment capacity.

7. The method according to claims 1-6, characterized in that: include: A loan plan for the borrowing farmers is formulated according to the repayment ability of the borrowing farmers.

8. A device for evaluating the repayment ability of a farmer who borrows money, characterized in that: include: An acquisition module is used to acquire agricultural product data and image sets of the planting areas of the borrowing farmers; wherein the agricultural product data includes at least one of the following: planting area, planting variety, growth environment, climate conditions, and historical yield; Generate a model; input the image set into a pre-trained diffusion model to generate a prediction frame for agricultural product fruit detection and a category label for agricultural product fruit; perform detection and counting based on the prediction frame and the category label to obtain the agricultural product output of the borrowing farmer in the current period; The processing module is used to predict the agricultural product output of the borrowing farmer in the target period according to the agricultural product output in the current period and the agricultural product data; and determine the repayment ability of the borrowing farmer according to the agricultural product output in the target period.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

11. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 7 when being executed by a processor.

Images