A loan risk assessment method and device based on agricultural remote sensing images

Abstract

The application provides a loan risk assessment method and device based on agricultural remote sensing images, and relates to the technical field of artificial intelligence.The method comprises the following steps: obtaining a hyperspectral image data set of a target region; obtaining the crop type corresponding to each pixel in the hyperspectral image data set according to the hyperspectral image data set of the target region and a crop recognition model; obtaining the estimated planting area of each crop according to the crop type corresponding to each pixel and the land area corresponding to each pixel, and predicting the total agricultural output value of the target region according to the estimated planting area of each crop, the expected unit yield and the expected unit price; and obtaining a loan risk assessment result according to the total agricultural output value of the target region and the loan amount of a farmer in the target region.The device is used to execute the above method.The loan risk assessment method and device based on agricultural remote sensing images provided in the application embodiment improve the accuracy of post-loan risk assessment of agricultural loans.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method and apparatus for loan risk assessment based on agricultural remote sensing images. Background Technology

[0002] Agricultural loans are loans issued by banks and other financial institutions to farmers, typically used for agricultural production. Farmers repay the loans after selling their produce.

[0003] Agricultural loans undergo regular post-loan risk assessments after issuance. Current technology allows for such assessments through regular manual inspections; however, the difficulty in accurately assessing farmers' actual agricultural production conditions leads to inaccurate risk assessments. Summary of the Invention

[0004] To address the problems in the prior art, embodiments of the present invention provide a method and apparatus for loan risk assessment based on agricultural remote sensing images, which can at least partially solve the problems existing in the prior art.

[0005] In a first aspect, this invention proposes a loan risk assessment method based on agricultural remote sensing images, comprising:

[0006] Obtain the hyperspectral image dataset of the target region;

[0007] Based on the hyperspectral image dataset of the target region and the crop recognition model, the crop type corresponding to each pixel in the hyperspectral image dataset of the target region is obtained; wherein, the crop recognition model is trained based on the hyperspectral image sample dataset;

[0008] Based on the crop type corresponding to each pixel and the land area corresponding to each pixel, the estimated planting area of ​​each crop is obtained, and the total agricultural output value of the target area is predicted based on the estimated planting area, expected unit yield and expected unit price of each crop.

[0009] Loan risk assessment results are obtained based on the total agricultural output of the target region and the loan amount of farmers in the target region.

[0010] Secondly, the present invention provides a loan risk assessment device based on agricultural remote sensing images, comprising:

[0011] The acquisition module is used to acquire the hyperspectral image dataset of the target region;

[0012] The identification module is used to obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target region based on the hyperspectral image dataset of the target region and the crop identification model; wherein, the crop identification model is trained based on the hyperspectral image sample dataset;

[0013] The prediction module is used to obtain the estimated planting area of ​​each crop based on the crop type corresponding to each pixel and the land area corresponding to each pixel, and to predict the total agricultural output value of the target area based on the estimated planting area, expected unit yield and expected unit price of each crop.

[0014] The assessment module is used to obtain loan risk assessment results based on the total agricultural output value of the target area and the loan amount of farmers in the target area.

[0015] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the loan risk assessment method based on agricultural remote sensing images as described in any of the above embodiments.

[0016] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the loan risk assessment method based on agricultural remote sensing images as described in any of the above embodiments.

[0017] Fifthly, the present invention provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the loan risk assessment method based on agricultural remote sensing images as described in any of the above embodiments.

[0018] The loan risk assessment method and apparatus based on agricultural remote sensing images provided in this invention can acquire a hyperspectral image dataset of a target area. Based on the hyperspectral image dataset of the target area and a crop identification model, it can obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target area. Based on the crop type corresponding to each pixel and the land area corresponding to each pixel, it can obtain the estimated planting area of ​​each crop. Based on the estimated planting area, estimated unit yield, and estimated unit price of each crop, it can predict the total agricultural output value of the target area. Based on the total agricultural output value of the target area and the loan amount of farmers in the target area, it can obtain the loan risk assessment result, thereby improving the accuracy of post-loan risk assessment of agricultural loans. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0020] Figure 1 This is a flowchart illustrating the loan risk assessment method based on agricultural remote sensing images provided in the first embodiment of the present invention.

[0021] Figure 2 This is a flowchart illustrating the loan risk assessment method based on agricultural remote sensing images provided in the second embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of the convolutional neural network model provided in the third embodiment of the present invention.

[0023] Figure 4 This is a flowchart illustrating the loan risk assessment method based on agricultural remote sensing images provided in the fourth embodiment of the present invention.

[0024] Figure 5 This is a flowchart illustrating the loan risk assessment method based on agricultural remote sensing images provided in the fifth embodiment of the present invention.

[0025] Figure 6 This is a schematic diagram of the structure of the loan risk assessment device based on agricultural remote sensing images provided in the sixth embodiment of the present invention.

[0026] Figure 7 This is a schematic diagram of the structure of the loan risk assessment device based on agricultural remote sensing images provided in the seventh embodiment of the present invention.

[0027] Figure 8 This is a schematic diagram of the structure of the loan risk assessment device based on agricultural remote sensing images provided in the eighth embodiment of the present invention.

[0028] Figure 9 This is a schematic diagram of the structure of a loan risk assessment device based on agricultural remote sensing images provided in the ninth embodiment of the present invention.

[0029] Figure 10 This is a schematic diagram of the physical structure of the electronic device provided in the tenth embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain the present invention, but are not intended to limit the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. The acquisition, storage, use, and processing of data in the technical solutions of this application all comply with relevant laws and regulations. The user information in the embodiments of this application is obtained through legal and compliant means, and the acquisition, storage, use, and processing of user information have been authorized and agreed upon by the customer.

[0031] For a hyperspectral image dataset, it is defined as ξ∈R H×W×D H is the height of the image, W is the width of the image, and D is the number of spectral bands. The number of pixels in the hyperspectral image dataset is N = H × W. When identifying crops in the hyperspectral image dataset using a crop recognition model, the crop corresponding to each pixel is identified.

[0032] The following describes the specific implementation process of the loan risk assessment method based on agricultural remote sensing images provided in this embodiment of the invention, using a server as the execution subject as an example.

[0033] Figure 1 This is a flowchart illustrating the loan risk assessment method based on agricultural remote sensing images provided in the first embodiment of the present invention, as shown below. Figure 1 As shown in the embodiment of the present invention, the loan risk assessment method based on agricultural remote sensing images includes:

[0034] S101. Obtain the hyperspectral image dataset of the target region;

[0035] Specifically, a hyperspectral image dataset of the target region can be collected; the target region is the area where the agricultural loan borrower grows crops. The server can obtain the hyperspectral image dataset of the target region.

[0036] S102. Based on the hyperspectral image dataset of the target region and the crop recognition model, obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target region; wherein, the crop recognition model is trained based on the hyperspectral image sample dataset;

[0037] Specifically, the server inputs the hyperspectral image dataset of the target region into the crop recognition model. The crop recognition model identifies the crop for each pixel in the hyperspectral image dataset to obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target region. The crop types include, but are not limited to, wheat, corn, and soybeans. The crop recognition model is trained based on a hyperspectral image sample dataset, which includes hyperspectral image data of different crops.

[0038] S103. Based on the crop type corresponding to each pixel and the land area corresponding to each pixel, obtain the estimated planting area of ​​each crop, and predict the total agricultural output value of the target area based on the estimated planting area, expected unit yield and expected unit price of each crop.

[0039] Specifically, after obtaining the crop type corresponding to each pixel in the hyperspectral image dataset of the target region, the server counts the number of pixels of the same crop type and, combined with the land area corresponding to each pixel, calculates the estimated planting area for each crop. Based on the estimated planting area, expected yield per unit area, and expected price per unit area, the server obtains the output value of each crop. Summing the output values ​​of all crops yields the total agricultural output value of the target region. The land area corresponding to each pixel is obtained based on the land area of ​​the target region and the number of pixels included in the hyperspectral image dataset of the target region. The expected yield per unit area and expected price per unit area for each crop are obtained in advance.

[0040] For example, if the planting area of ​​crop A is S, the expected yield per unit is q, and the expected price per unit is p, then the output value of crop A is V = pqS.

[0041] S104. Obtain loan risk assessment results based on the total agricultural output value of the target area and the loan amount of the lenders in the target area.

[0042] Specifically, the server can obtain the loan amounts of borrowers in the target region and compare these loan amounts with the total agricultural output value of the target region to conduct a loan risk assessment. For example, if the total agricultural output value of the target region is greater than the loan amount of the borrowers in the target region, the loan risk assessment result can be described as low delinquency risk. If the total agricultural output value of the target region is less than or equal to the loan amount of the borrowers in the target region, the loan risk assessment result can be described as high delinquency risk.

[0043] The loan risk assessment method based on agricultural remote sensing images provided in this invention can acquire a hyperspectral image dataset of a target area. Based on the hyperspectral image dataset of the target area and a crop identification model, it can obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target area. Based on the crop type corresponding to each pixel and the land area corresponding to each pixel, it can obtain the estimated planting area of ​​each crop. Based on the estimated planting area, estimated unit yield, and estimated unit price of each crop, it can predict the total agricultural output value of the target area. Based on the total agricultural output value of the target area and the loan amount of farmers in the target area, it can obtain the loan risk assessment result, thereby improving the accuracy of post-loan risk assessment of agricultural loans.

[0044] Figure 2 This is a flowchart illustrating the loan risk assessment method based on agricultural remote sensing images provided in the second embodiment of the present invention, as shown below. Figure 2 As shown, based on the above embodiments, the further steps of training a crop recognition model based on a hyperspectral image sample dataset include:

[0045] S201. Obtain an initial training sample set from the hyperspectral image sample dataset and perform sample annotation;

[0046] Specifically, hyperspectral image data of different types of crops are collected, and then individual pixels are extracted from these hyperspectral image data as samples. These samples constitute the hyperspectral image sample dataset. Each sample in the hyperspectral image sample dataset is not labeled with a corresponding crop type. The server can obtain the hyperspectral image sample dataset and extract a portion of the samples from it as an initial training sample set. Each sample in the initial training sample set is then labeled to obtain the corresponding crop type as the label for each sample.

[0047] For example, 3% of the samples are drawn from the hyperspectral image sample dataset to form the initial training sample set.

[0048] Furthermore, to increase the number of samples in the initial training sample set, data augmentation (DA) can be used to expand the sample size and enhance the initial training sample set. For example, this involves performing transformation operations such as horizontal flipping, vertical flipping, 90° clockwise rotation, 180° clockwise rotation, and 270° clockwise rotation on the sample images. After the samples extracted from the hyperspectral image sample dataset constitute the initial training sample set, DA operations can be used to expand the initial training sample set, increasing the number of samples. For instance, five different transformation operations can be performed on each sample. The expanded initial training sample set contains five times more samples than those directly extracted from the hyperspectral image sample dataset.

[0049] S202. Based on the labeled initial training sample set and the original model, train to obtain a pre-trained model;

[0050] Specifically, the server trains the original model based on the labeled initial training sample set to obtain a pre-trained model. The original model is selected according to actual needs, such as a convolutional neural network model; this embodiment of the invention does not impose a limitation. The pre-trained model is used to label unlabeled samples in the hyperspectral image sample dataset, obtaining pseudo-labels for the samples.

[0051] For example, the original model uses Figure 3 The convolutional neural network (CNN) model shown includes one input layer, two convolutional layers, two max-pooling layers, two fully connected layers, and one output layer. The input layer takes a hyperspectral image with a pixel size of 8×8×D. The first convolutional layer contains 20 convolutional kernels of size 3×3, and the second convolutional layer contains 20 convolutional kernels of size 2×2. After each convolutional layer, a max-pooling layer with a kernel size of 2×2 and a stride of 2 is connected. The first fully connected layer contains 500 units, and the second fully connected layer contains units equal to the number of categories K, i.e., the number of crop types. The batch size is set to 50, and the learning rate is 0.001. Stochastic gradient descent can be used to improve training efficiency during the training of the CNN model.

[0052] S203. The initial training sample set is supplemented according to the remaining training sample set and the pre-trained model to obtain an updated initial training sample set; wherein, the remaining training sample set is obtained by removing the initial training sample set from the hyperspectral image sample dataset;

[0053] Specifically, after the server removes the initial training sample set from the hyperspectral image sample dataset, the remaining samples constitute the remaining training sample set. Based on the remaining training sample set and the pre-trained model, the server extracts minority class samples from the remaining training sample set to supplement the initial training sample set, obtaining an updated initial sample set. Here, minority class samples refer to samples in the remaining training sample set that have a smaller number of identical pseudo-labels. The pseudo-label corresponding to each sample in the remaining training sample set can be obtained through the pre-trained model.

[0054] The precision of minority class samples is much higher than that of majority class samples, indicating that the model's predictions of minority class behavior are more conservative. Therefore, including minority class pseudo-labels in the initial training sample set carries less risk of error. Due to the scarcity of minority class samples, supplementing the initial training sample set with minority class samples makes the categories of the labels corresponding to the samples in the initial training sample set more balanced. This results in a lighter bias in the model in the next round of pseudo-label distribution, which is beneficial to improving the accuracy of model training.

[0055] S204. Retrain the intermediate model based on the annotated and updated initial training sample set;

[0056] Specifically, unlabeled samples in the updated initial training sample set are labeled to obtain a labeled updated initial training sample set. The server trains the original model based on the labeled updated initial training sample set to obtain an intermediate training model. The intermediate training model is used to label the unlabeled samples in the hyperspectral image sample dataset to obtain pseudo-labels for the samples.

[0057] S205. If the training termination condition is not met, the intermediate model is used as a pre-trained model to supplement the initial training sample set and the intermediate model is retrained; if the training termination condition is met, the intermediate model is used as a crop recognition model.

[0058] Specifically, after obtaining the intermediate training model, the server determines whether the training termination condition is met. If the training termination condition is not met, the intermediate model is used as a pre-trained model, and step S203 is repeated to supplement the initial training sample set to obtain an updated initial training sample set. Step S204 is repeated to retrain and obtain the intermediate model, and then the determination of whether the training termination condition is met is made again. If the training termination condition is met, the model training ends, and the intermediate model is used as the crop recognition model. The training termination condition is set according to actual needs, and this embodiment of the invention does not impose any limitations.

[0059] For example, the training termination condition is that the number of training rounds reaches a preset number. Alternatively, the training termination condition is that the number of updates to the initial training sample set reaches a preset number.

[0060] Figure 4 This is a flowchart illustrating the loan risk assessment method based on agricultural remote sensing images provided in the fourth embodiment of the present invention, as shown below. Figure 4 As shown, the step of supplementing the initial training sample set with the remaining training sample set and the pre-trained model to obtain the updated initial training sample set includes:

[0061] S401. Based on the remaining training sample set and the pre-trained model, obtain the pseudo-label corresponding to each sample in the remaining training sample set;

[0062] Specifically, the server inputs each sample from the remaining training sample set into the pre-trained model, and can output the probability of each crop corresponding to each sample. The crop with the highest probability for each sample is taken as the type and used as the pseudo-label for each sample.

[0063] S402. Obtain a first supplementary training set from the remaining training sample set according to the pseudo-labels corresponding to each sample in the remaining training sample set; wherein, the samples in the first supplementary training set correspond to the same pseudo-label and the number of samples in the first supplementary training set is the smallest among all the sample numbers corresponding to all pseudo-labels.

[0064] Specifically, the server classifies each sample in the remaining training sample set according to the pseudo-labels corresponding to each sample, groups samples with the same pseudo-label into one class, and then counts the number of samples under each pseudo-label. The samples corresponding to the pseudo-label with the fewest samples constitute the first supplementary training set.

[0065] For example, there are a total of 16 categories of pseudo-labels corresponding to each sample in the remaining training sample set. These pseudo-labels are sorted from largest to smallest according to the number of samples corresponding to each category. The pseudo-label ranked first has the largest number of samples, and the pseudo-label ranked sixteenth has the smallest number of samples. Therefore, the samples corresponding to the pseudo-label ranked sixteenth are used to form the first supplementary training set.

[0066] S403. If the number of samples corresponding to the first supplementary training set is greater than or equal to the supplementary sample threshold, then update the initial training sample set according to the first supplementary training set.

[0067] Specifically, the server counts the number of samples included in the first supplementary training set, which is taken as the sample count corresponding to the first supplementary training set. The sample count corresponding to the first supplementary training set is compared with a supplementary sample threshold. If the sample count corresponding to the first supplementary training set is greater than or equal to the supplementary sample threshold, then the samples in the first supplementary training set are added to the initial training sample set, updating the initial training sample set. The supplementary sample threshold is preset and can be set according to actual needs; this embodiment of the invention does not impose limitations.

[0068] Figure 5 This is a flowchart illustrating the loan risk assessment method based on agricultural remote sensing images provided in the fifth embodiment of the present invention, as shown below. Figure 5 As shown, based on the above embodiments, the step of supplementing the initial training sample set with the remaining training sample set and the pre-trained model to obtain an updated initial training sample set includes:

[0069] S501. Based on the remaining training sample set and the pre-trained model, obtain the pseudo-label corresponding to each sample in the remaining training sample set;

[0070] Specifically, the server inputs each sample from the remaining training sample set into the pre-trained model, and can output the crop type corresponding to each sample, with the crop type corresponding to each sample serving as the pseudo-label for each sample.

[0071] S502. Obtain a first supplementary training set from the remaining training sample set according to the pseudo-labels corresponding to each sample in the remaining training sample set; wherein, the samples in the first supplementary training set correspond to the same pseudo-label and the number of samples in the first supplementary training set is the smallest among all the sample numbers corresponding to all pseudo-labels.

[0072] Specifically, the server classifies each sample in the remaining training sample set according to the pseudo-labels corresponding to each sample, groups samples with the same pseudo-label into one class, and then counts the number of samples under each pseudo-label. The samples corresponding to the pseudo-label with the fewest samples constitute the first supplementary training set.

[0073] S503. If the number of samples corresponding to the first supplementary training set is less than the supplementary sample threshold, then based on the sample supplementation rule, a second supplementary training set is obtained from the secondary remaining training sample set so that the sum of the number of samples corresponding to the first supplementary training set and the number of samples corresponding to the second supplementary training set is greater than or equal to the supplementary sample threshold; wherein, the sample supplementation rule is preset; the secondary remaining training sample set is obtained by removing the first supplementary training set from the remaining training sample set;

[0074] Specifically, the server compares the number of samples corresponding to the first supplementary training set with a supplementary sample threshold. If the number of samples corresponding to the first supplementary training set is less than the supplementary sample threshold, the server selects a certain number of samples from the secondary remaining training sample set to form a second supplementary training set according to the sample supplementation rule. The sum of the number of samples corresponding to the second supplementary training set and the number of samples corresponding to the first supplementary training set is greater than or equal to the supplementary sample threshold. The secondary remaining training sample set is obtained by removing the first supplementary training set from the remaining training sample set; the sample supplementation rule is preset.

[0075] S504. Update the initial training sample set based on the first supplementary training set and the second supplementary training set.

[0076] Specifically, the server adds the samples from the first supplementary training set and the samples from the second supplementary training set to the initial training sample set, updates the initial training sample set, and obtains the updated initial training sample set.

[0077] Based on the above embodiments, the sample supplementation rules further include:

[0078] Obtain samples with smaller best-to-second-best class values ​​from the secondary residual training sample set.

[0079] Specifically, each sample in the secondary residual training sample set is input into the pre-trained model, which outputs the probability of each sample for various crops. From the probabilities corresponding to each sample, the maximum probability and the second-highest probability for each sample are obtained. The difference between the maximum probability and the second-highest probability for each sample is calculated as the Best vs. Second Best (BvSB) value for each sample. The samples in the secondary residual training sample set are arranged in descending order of BvSB value, with smaller BvSB values ​​being prioritized for selection into the second supplementary training set.

[0080] If a sample has a small BvSB value, the difference between the best and second-best classes is smaller, and the sample contains more information. By selecting samples with smaller BvSB values ​​and prioritizing those samples that have high uncertainty for the current classification model and significantly impact the classification boundary, the accuracy of subsequent model training can be improved.

[0081] Based on the above embodiments, the step of obtaining the initial training sample set from the hyperspectral image sample dataset further includes:

[0082] Samples of a predetermined proportion are obtained from the hyperspectral image sample dataset to form the initial training sample set.

[0083] Specifically, the server obtains a preset proportion of samples from the hyperspectral image sample dataset to form the initial training sample set. The preset proportion is set according to actual needs, and this embodiment of the invention does not impose any limitation.

[0084] Based on the above embodiments, the loan risk assessment method based on agricultural remote sensing images provided by the embodiments of the present invention further includes:

[0085] If the loan risk assessment result indicates that there is a risk of delinquency, a warning message predicting loan delinquency will be output.

[0086] Specifically, if the loan risk assessment result indicates that there is a risk of delinquency, the server will output a warning message predicting loan delinquency, which can be provided to loan risk assessment personnel for reference.

[0087] Figure 6 This is a schematic diagram of the loan risk assessment device based on agricultural remote sensing images provided in the sixth embodiment of the present invention, as shown below. Figure 6 As shown, the loan risk assessment device based on agricultural remote sensing images provided in this embodiment of the invention includes an acquisition module 601, an identification module 602, a prediction module 603, and an assessment module 604, wherein:

[0088] The acquisition module 601 is used to acquire a hyperspectral image dataset of the target area; the identification module 602 is used to obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target area based on the hyperspectral image dataset of the target area and the crop identification model; wherein, the crop identification model is trained based on the hyperspectral image sample dataset; the prediction module 603 is used to obtain the estimated planting area of ​​each crop based on the crop type corresponding to each pixel and the land area corresponding to each pixel, and predict the total agricultural output value of the target area based on the estimated planting area, expected unit yield and expected unit price of each crop; the evaluation module 604 is used to obtain the loan risk assessment result based on the total agricultural output value of the target area and the loan amount of the farmers in the target area.

[0089] Specifically, a hyperspectral image dataset of the target region can be collected; the target region is the area where the agricultural loan borrower grows crops. The acquisition module 601 can acquire the hyperspectral image dataset of the target region.

[0090] The recognition module 602 inputs the hyperspectral image dataset of the target region into the crop recognition model. The crop recognition model identifies the crop for each pixel in the hyperspectral image dataset to obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target region. The crop types include, but are not limited to, wheat, corn, and soybeans. The crop recognition model is trained based on a hyperspectral image sample dataset, which includes hyperspectral image data of different crops.

[0091] After obtaining the crop type corresponding to each pixel in the hyperspectral image dataset of the target region, the prediction module 603 counts the number of pixels of the same crop type and, combined with the land area corresponding to each pixel, calculates the estimated planting area for each crop. Based on the estimated planting area, expected yield per unit area, and expected price per unit area, the prediction module 603 obtains the output value of each crop. Summing the output values ​​of all crops yields the total agricultural output value of the target region. The land area corresponding to each pixel is obtained based on the land area of ​​the target region and the number of pixels included in the hyperspectral image dataset of the target region. The expected yield per unit area and expected price per unit area for each crop are obtained in advance.

[0092] The assessment module 604 can obtain the loan amount of borrowers in the target area and compare it with the total agricultural output value of the target area to conduct a loan risk assessment. For example, if the total agricultural output value of the target area is greater than the loan amount of the borrowers in the target area, the loan risk assessment result can be low delinquency risk. If the total agricultural output value of the target area is less than or equal to the loan amount of the borrowers in the target area, the loan risk assessment result can be high delinquency risk.

[0093] The loan risk assessment device based on agricultural remote sensing images provided in this invention can acquire a hyperspectral image dataset of a target area. Based on the hyperspectral image dataset of the target area and a crop identification model, it can obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target area. Based on the crop type corresponding to each pixel and the land area corresponding to each pixel, it can obtain the estimated planting area of ​​each crop. Based on the estimated planting area, estimated unit yield, and estimated unit price of each crop, it can predict the total agricultural output value of the target area. Based on the total agricultural output value of the target area and the loan amount of farmers in the target area, it can obtain the loan risk assessment result, thereby improving the accuracy of post-loan risk assessment of agricultural loans.

[0094] Figure 7This is a schematic diagram of the loan risk assessment device based on agricultural remote sensing images provided in the seventh embodiment of the present invention, as shown below. Figure 7 As shown, based on the above embodiments, the loan risk assessment device based on agricultural remote sensing images provided in this embodiment of the invention further includes a sample acquisition module 605, a training module 606, a supplementation module 607, a retraining module 608, and a judgment module 609, wherein:

[0095] The sample acquisition module 605 is used to acquire and label an initial training sample set from the hyperspectral image sample dataset; the training module 606 is used to train a pre-trained model based on the labeled initial training sample set and the original model; the supplementation module 607 is used to supplement the initial training sample set based on the remaining training sample set and the pre-trained model to obtain an updated initial training sample set; wherein, the remaining training sample set is obtained by removing the initial training sample set from the hyperspectral image sample dataset; the retraining module 608 is used to retrain an intermediate model based on the updated initial training sample set; the judgment module 609 is used to, if the training termination condition is not met, use the intermediate model as a pre-trained model to supplement the initial training sample set again and retrain the intermediate model; if the training termination condition is met, use the intermediate model as a crop recognition model.

[0096] Figure 8 This is a schematic diagram of the loan risk assessment device based on agricultural remote sensing images provided in the eighth embodiment of the present invention, as shown below. Figure 8 As shown, based on the above embodiments, the supplementary module 607 further includes a first obtaining unit 6071, a first supplementing unit 6072, and a first updating unit 6073, wherein:

[0097] The first obtaining unit 6071 is used to obtain pseudo-labels corresponding to each sample in the remaining training sample set based on the remaining training sample set and the pre-trained model; the first supplementing unit 6072 obtains a first supplementary training set from the remaining training sample set based on the pseudo-labels corresponding to each sample in the remaining training sample set; wherein, the samples in the first supplementary training set correspond to the same pseudo-label and the number of samples in the first supplementary training set is the smallest among the number of samples corresponding to all pseudo-labels; the first updating unit 6073 is used to update the initial training sample set based on the first supplementary training set if the number of samples corresponding to the first supplementary training set is greater than or equal to the supplementary sample threshold.

[0098] Figure 9 This is a schematic diagram of the loan risk assessment device based on agricultural remote sensing images provided in the ninth embodiment of the present invention, as shown below. Figure 9As shown, based on the above embodiments, the supplementary module 607 further includes a second obtaining unit 6074, a second supplementary unit 6075, a third supplementary unit 6076, and a second updating unit 6077, wherein:

[0099] The second obtaining unit 6074 is used to obtain pseudo-labels corresponding to each sample in the remaining training sample set based on the remaining training sample set and the pre-trained model; the second supplementing unit 6075 is used to obtain a first supplementary training set from the remaining training sample set based on the pseudo-labels corresponding to each sample in the remaining training sample set; wherein, the samples in the first supplementary training set correspond to the same pseudo-label and the number of samples in the first supplementary training set is the smallest among the number of samples corresponding to all pseudo-labels; the third supplementing unit 6076 is used to obtain a second supplementary training set from the secondary remaining training sample set based on the sample supplementing rule if the number of samples corresponding to the first supplementary training set is less than the supplementary sample threshold, so that the sum of the number of samples corresponding to the first supplementary training set and the number of samples corresponding to the second supplementary training set is greater than or equal to the supplementary sample threshold; wherein, the sample supplementing rule is preset; the secondary remaining training sample set is obtained by removing the first supplementary training set from the remaining training sample set; the second updating unit 6077 is used to update the initial training sample set based on the first supplementary training set and the second supplementary training set.

[0100] Based on the above embodiments, the sample supplementation rules further include:

[0101] Obtain samples with smaller best-to-second-best class values ​​from the secondary residual training sample set.

[0102] Based on the above embodiments, the sample acquisition module 605 is further specifically used for:

[0103] Samples of a predetermined proportion are obtained from the hyperspectral image sample dataset to form the initial training sample set.

[0104] Based on the above embodiments, the loan risk assessment device based on agricultural remote sensing images provided in this embodiment of the invention further includes an output module, wherein:

[0105] The output module is used to output a warning message predicting loan delinquency if the loan risk assessment result indicates a high risk of delinquency.

[0106] The server embodiments provided in this invention can be used to execute the processing flow of the above method embodiments. Its functions will not be repeated here, but can be referred to the detailed description of the above method embodiments.

[0107] It should be noted that the loan risk assessment method and apparatus based on agricultural remote sensing images provided in this embodiment of the invention can be used in the financial field, or in any technical field other than the financial field. This embodiment of the invention does not limit the application field of the loan risk assessment method and apparatus based on agricultural remote sensing images.

[0108] Figure 10 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, as shown below. Figure 10 As shown, the electronic device may include: a processor 1001, a communication interface 1002, a memory 1003, and a communication bus 1004, wherein the processor 1001, the communication interface 1002, and the memory 1003 communicate with each other through the communication bus 1004. The processor 1001 can call logical instructions in the memory 1003 to execute the following methods: acquiring a hyperspectral image dataset of a target region; obtaining the crop type corresponding to each pixel in the hyperspectral image dataset of the target region based on the hyperspectral image dataset of the target region and a crop recognition model; wherein the crop recognition model is trained based on a hyperspectral image sample dataset; obtaining the estimated planting area of ​​each crop based on the crop type corresponding to each pixel and the land area corresponding to each pixel, and predicting the total agricultural output value of the target region based on the estimated planting area, estimated unit yield, and estimated unit price of each crop; and obtaining a loan risk assessment result based on the total agricultural output value of the target region and the loan amount of farmers in the target region.

[0109] Furthermore, the logical instructions in the aforementioned memory 1003 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] This embodiment discloses a computer program product, which includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer can perform the methods provided in the above-described method embodiments, such as: acquiring a hyperspectral image dataset of a target region; obtaining the crop type corresponding to each pixel in the hyperspectral image dataset of the target region based on the hyperspectral image dataset of the target region and a crop recognition model; wherein the crop recognition model is trained based on a hyperspectral image sample dataset; obtaining the estimated planting area of ​​each crop based on the crop type corresponding to each pixel and the land area corresponding to each pixel, and predicting the total agricultural output value of the target region based on the estimated planting area, expected unit yield, and expected unit price of each crop; and obtaining a loan risk assessment result based on the total agricultural output value of the target region and the loan amount of farmers in the target region.

[0111] This embodiment provides a computer-readable storage medium storing a computer program that causes a computer to execute the methods provided in the above-described method embodiments. For example, the methods include: acquiring a hyperspectral image dataset of a target region; obtaining the crop type corresponding to each pixel in the hyperspectral image dataset of the target region based on the hyperspectral image dataset of the target region and a crop identification model; wherein the crop identification model is trained based on a hyperspectral image sample dataset; obtaining the estimated planting area of ​​each crop based on the crop type corresponding to each pixel and the land area corresponding to each pixel, and predicting the total agricultural output value of the target region based on the estimated planting area, expected unit yield, and expected unit price of each crop; and obtaining a loan risk assessment result based on the total agricultural output value of the target region and the loan amount of farmers in the target region.

[0112] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0113] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0114] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0115] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0116] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0117] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A loan risk assessment method based on agricultural remote sensing images, characterized in that, include: Obtain the hyperspectral image dataset of the target region; Based on the hyperspectral image dataset of the target region and the crop recognition model, the crop type corresponding to each pixel in the hyperspectral image dataset of the target region is obtained; wherein, the crop recognition model is trained based on the hyperspectral image sample dataset; Based on the crop type corresponding to each pixel and the land area corresponding to each pixel, the estimated planting area of ​​each crop is obtained, and the total agricultural output value of the target area is predicted based on the estimated planting area, expected unit yield and expected unit price of each crop. Based on the total agricultural output value of the target area and the loan amount of farmers in the target area, obtain the loan risk assessment results; The steps for training a crop recognition model based on a hyperspectral image sample dataset include: Obtain an initial training sample set from the hyperspectral image sample dataset and label it; Based on the labeled initial training sample set and the original model, a pre-trained model is obtained through training. The initial training sample set is supplemented with the remaining training sample set and the pre-trained model to obtain an updated initial training sample set; wherein, the remaining training sample set is obtained by removing the initial training sample set from the hyperspectral image sample dataset; Based on the annotated and updated initial training sample set, retrain to obtain an intermediate model; If the training termination condition is not met, the intermediate model is used as a pre-trained model to supplement the initial training sample set and retrain the intermediate model; if the training termination condition is met, the intermediate model is used as a crop recognition model. The step of supplementing the initial training sample set with the remaining training sample set and the pre-trained model to obtain an updated initial training sample set includes: Based on the remaining training sample set and the pre-trained model, obtain the pseudo-label corresponding to each sample in the remaining training sample set; wherein, the pseudo-label corresponding to each sample is the crop type corresponding to each sample; A first supplementary training set is obtained from the remaining training sample set based on the pseudo-labels corresponding to each sample in the remaining training sample set; wherein, the samples in the first supplementary training set correspond to the same pseudo-label and the number of samples in the first supplementary training set is the smallest among all the sample numbers corresponding to all pseudo-labels. If the number of samples corresponding to the first supplementary training set is greater than or equal to the supplementary sample threshold, then the initial training sample set is updated based on the first supplementary training set. If the number of samples corresponding to the first supplementary training set is less than the supplementary sample threshold, then a second supplementary training set is obtained from the secondary remaining training sample set based on the sample supplementation rule, so that the sum of the number of samples corresponding to the first supplementary training set and the number of samples corresponding to the second supplementary training set is greater than or equal to the supplementary sample threshold; wherein, the sample supplementation rule is preset; the secondary remaining training sample set is obtained by removing the first supplementary training set from the remaining training sample set; Update the initial training sample set based on the first supplementary training set and the second supplementary training set.

2. The method according to claim 1, characterized in that, The sample replenishment rules include: Obtain samples with smaller best-to-second-best class values ​​from the secondary residual training sample set.

3. The method according to claim 1, characterized in that, The step of obtaining the initial training sample set from the hyperspectral image sample dataset includes: Samples of a predetermined proportion are obtained from the hyperspectral image sample dataset to form the initial training sample set.

4. The method according to any one of claims 1 to 3, characterized in that, Also includes: If the loan risk assessment result indicates a high risk of delinquency, a warning message predicting loan delinquency will be output.

5. A loan risk assessment device based on agricultural remote sensing images, characterized in that, include: The acquisition module is used to acquire the hyperspectral image dataset of the target region; The identification module is used to obtain the crop type corresponding to each pixel in the hyperspectral image dataset of the target region based on the hyperspectral image dataset of the target region and the crop identification model; wherein, the crop identification model is trained based on the hyperspectral image sample dataset; The prediction module is used to obtain the estimated planting area of ​​each crop based on the crop type corresponding to each pixel and the land area corresponding to each pixel, and to predict the total agricultural output value of the target area based on the estimated planting area, expected unit yield and expected unit price of each crop. The assessment module is used to obtain loan risk assessment results based on the total agricultural output value of the target area and the loan amount of farmers in the target area; The device further includes: The sample acquisition module is used to acquire an initial training sample set from the hyperspectral image sample dataset and label it; The training module is used to train a pre-trained model based on the labeled initial training sample set and the original model. The supplementation module is used to supplement the initial training sample set according to the remaining training sample set and the pre-trained model to obtain an updated initial training sample set; wherein, the remaining training sample set is obtained by removing the initial training sample set from the hyperspectral image sample dataset; The retraining module is used to retrain the intermediate model based on the annotated and updated initial training sample set. The judgment module is used to, if the training termination condition is not met, supplement the initial training sample set with the intermediate model as a pre-trained model and retrain the intermediate model; if the training termination condition is met, use the intermediate model as a crop recognition model. The supplementary module includes: The first obtaining unit is used to obtain the pseudo label corresponding to each sample in the remaining training sample set based on the remaining training sample set and the pre-trained model; wherein, the pseudo label corresponding to each sample is the crop type corresponding to each sample; The first supplementary unit is used to obtain a first supplementary training set from the remaining training sample set according to the pseudo-labels corresponding to each sample in the remaining training sample set; wherein, the samples in the first supplementary training set correspond to the same pseudo-label and the number of samples in the first supplementary training set is the smallest among all the number of samples corresponding to all pseudo-labels. The first update unit is used to update the initial training sample set according to the first supplementary training set if the number of samples corresponding to the first supplementary training set is greater than or equal to the supplementary sample threshold. The third supplementary unit is configured to, if the number of samples corresponding to the first supplementary training set is less than the supplementary sample threshold, obtain a second supplementary training set from the secondary remaining training sample set based on the sample supplementation rule, such that the sum of the number of samples corresponding to the first supplementary training set and the number of samples corresponding to the second supplementary training set is greater than or equal to the supplementary sample threshold; wherein, the sample supplementation rule is preset; the secondary remaining training sample set is obtained by removing the first supplementary training set from the remaining training sample set; The second update unit is used to update the initial training sample set based on the first supplementary training set and the second supplementary training set.

6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 4.

8. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 4.

Images