Method for monitoring the linting rate of cotton and monitoring system

Abstract

This application discloses a method and system for monitoring cotton boll opening rate. Relating to the field of deep learning, the monitoring method includes: identifying the spatial distribution map of cotton in the target area based on multi-temporal multispectral remote sensing data and a first training model, wherein the first training model is constructed based on the sample point locations of cotton and standard growth characteristic curves; determining the cotton boll opening rate based on the spatial distribution map obtained from the first training model, single-scene multispectral remote sensing data, and a second training model, wherein the second training model is constructed based on multiple spectral data of the sample point locations and ground-level cotton boll opening rate sampling data. This invention first extracts the cotton growth area as basic data, and then uses machine learning methods to invert the cotton boll opening rate, which can improve the accuracy of cotton extraction. Furthermore, the monitoring model of this invention has greater generalization ability and higher inversion accuracy.

Description

Technical Field

[0001] This application relates to the field of deep learning, specifically to a method and system for monitoring cotton boll opening rate. Background Technology

[0002] The boll opening rate of cotton refers to the proportion of mature cotton fibers (commonly known as "fluff") that naturally break open from the cotton boll and become exposed. It is an important indicator of cotton maturity and harvest quality. A good boll opening rate means that the cotton is fully mature and suitable for mechanical or manual harvesting. Therefore, quickly obtaining a spatial distribution map of cotton planting areas and boll opening rates is of great significance.

[0003] Existing research has explored various methods for monitoring cotton boll opening rate using remote sensing technology, but problems remain, such as inaccurate extraction of cotton planting area and overly linear cotton boll opening rate inversion models with weak generalization ability. For example, existing technologies have proposed a method for predicting cotton boll opening rate based on UAV multispectral data, which inverts the cotton boll opening rate using remote sensing data and can be applied to cotton boll opening rate monitoring to a certain extent.

[0004] However, existing technologies still suffer from inaccurate cotton planting area extraction and overly linear cotton boll opening rate inversion models with weak generalization ability. Therefore, it is necessary to improve the accuracy of cotton boll opening rate inversion in order to provide relevant scientific decision-making for cotton planting management. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for monitoring cotton boll opening rate. First, the cotton growing area can be extracted as basic data. Then, machine learning methods are used to invert the cotton boll opening rate. The monitoring model of this invention has greater generalizability and higher inversion accuracy.

[0006] To achieve the above objectives, a first aspect of this application provides a method for monitoring cotton boll opening rate. The monitoring method includes: identifying a spatial distribution map of cotton in a target area based on multi-temporal multispectral remote sensing data of the target area and a first training model, wherein the first training model is constructed based on the sample point locations of cotton and standard growth characteristic curves; and determining the cotton boll opening rate based on the spatial distribution map obtained from the first training model, single-scene multispectral remote sensing data, and a second training model, wherein the second training model is constructed based on multiple spectral data of the sample point locations and ground boll opening rate sampling data of cotton.

[0007] In this embodiment of the application, the first training model is constructed through the following steps: determining the normalized vegetation index of the cotton at the sample point location based on the multi-temporal multispectral remote sensing data; constructing the standard growth characteristic curve of the cotton based on the historical accumulated temperature data of the cotton and the normalized vegetation index; and training the first training model using a deep neural network model based on the sample point location of the cotton and the standard growth characteristic curve.

[0008] In this embodiment, the deep learning framework of the deep neural network model is TensorFlow, and the deep neural network model includes an input layer, a hidden layer and an output layer. The activation function of the hidden layer is ReLU, the activation function of the output layer is Softmax, the optimizer is the Adam algorithm, and the loss function is the sparse multi-class cross-entropy function.

[0009] In this embodiment of the application, determining the normalized vegetation index (NDVI) of the cotton at the sample point location based on the multi-temporal multispectral remote sensing data includes: calculating the NDVI of the cotton at the sample point location based on the reflectance of multiple bands in the multi-temporal multispectral remote sensing data using the following formula:

[0010] Wherein, R represents the red band among the multiple band reflectances, and NIR represents the near-infrared band among the multiple band reflectances.

[0011] In the embodiments of this application, the first training model is evaluated and optimized using at least one of the following methods: accuracy, precision, recall, and intersection-union ratio.

[0012] In this embodiment of the application, the accuracy A is expressed by the following formula:

[0013] The accuracy P is expressed by the following formula:

[0014] The recall rate R is expressed by the following formula:

[0015] Wherein, TP is the number of positive samples correctly identified as valid samples; TN is the number of negative samples correctly identified as invalid samples; FP is the number of invalid samples incorrectly confused as valid samples; and FN is the number of valid samples incorrectly identified as invalid samples.

[0016] In this embodiment of the application, the second training model is constructed through the following steps: at the sample point location of the cotton, the ground boll opening rate sampling data of the cotton is obtained; based on the multi-temporal multispectral remote sensing data of the target area, multiple spectral data at the sample point location are obtained, wherein the multiple spectral values ​​include red, green, blue, and near-infrared spectral values; and based on the multiple spectral data and the ground boll opening rate sampling data of the cotton, an inversion dataset is constructed, and a random forest model is used to train the inversion dataset to obtain the second training model.

[0017] In this embodiment of the application, the inversion dataset includes a training set and a test set with a set ratio. The training set is input into a random forest regressor to model the random forest model. The test set is used to fine-tune the random forest model using grid search cross-validation. The root mean square error and R² value are used to evaluate the random forest model.

[0018] In this embodiment of the application, the monitoring method further includes: preprocessing the acquired multi-temporal multispectral remote sensing data of the target area in the following manner: correction, stitching, and cropping.

[0019] A second aspect of this application provides a monitoring system for cotton boll opening rate. The monitoring system includes: a spatial distribution identification device for identifying a spatial distribution map of cotton in a target area based on multi-temporal multispectral remote sensing data of the target area and a first training model, wherein the first training model is constructed based on the sample point locations of the cotton and a standard growth characteristic curve; and a boll opening rate determination device for determining the boll opening rate of the cotton based on the spatial distribution map obtained from the first training model, single-scene multispectral remote sensing data, and a second training model, wherein the second training model is constructed based on multiple spectral data of the sample point locations and ground boll opening rate sampling data of the cotton.

[0020] In summary, this invention proposes a method based on multispectral remote sensing data, employing multiple trained models to extract cotton planting areas and obtain a spatial distribution map of cotton, and then retrieving the cotton boll opening rate to obtain a spatial distribution map of the cotton boll opening rate. This invention first extracts the cotton growing areas as basic data, and then uses machine learning methods to retrieve the cotton boll opening rate, which improves the accuracy of cotton extraction. Furthermore, the monitoring model of this invention has greater generalization ability and higher retrieval accuracy.

[0021] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0022] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings: Figure 1 The illustration shows a flowchart of a method for monitoring cotton boll opening rate according to an embodiment of this application; Figure 2 This illustration schematically shows a technical roadmap of an embodiment of this application; Figure 3 A schematic diagram of a standard growth characteristic curve according to an embodiment of this application is shown; Figure 4 This illustration schematically shows a spatial distribution diagram of cotton according to an embodiment of this application; Figure 5 The schematic diagram illustrates the structure of a cotton boll opening rate monitoring system according to an embodiment of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0024] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0025] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0026] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0027] The following is an explanation of some of the terms used in this invention: Remote sensing refers to non-contact, long-distance detection technology. It generally refers to the detection of the electromagnetic radiation and reflection characteristics of objects using sensors / remote sensors. Remote sensing uses instruments sensitive to electromagnetic waves, such as remote sensors, to detect target features from a distance and without contact with the target object.

[0028] Deep neural networks (DNNs) are an emerging field of deep learning in academia. This algorithm consists of an input layer, hidden layers, and an output layer, with the layer between the input and output layers being a fully connected layer (or hidden layer). Training data is input into the neural network structure, passes through the hidden layers, and features are calculated by neurons, finally outputting a spatial distribution map of cotton. This study uses the TensorFlow deep learning framework to build, train, and verify the accuracy of the deep neural network.

[0029] Random forest is an ensemble learning method that improves the accuracy of classification or regression by constructing multiple decision trees and combining their predictions. Each tree is trained using randomly selected samples and features, and this randomness makes the model highly resistant to overfitting. The final prediction is obtained through voting (classification) or averaging (regression). Random forest is widely used in various machine learning tasks due to its efficiency and good performance. This study uses the random forest algorithm to invert the cotton boll opening rate.

[0030] Growing Degree Days (GDD) refers to the sum of the daily average temperatures exceeding the plant's basal growth temperature during the growing season. It measures the total amount of temperature suitable for plant growth.

[0031] The cotton boll opening rate refers to the ratio of cotton fibers (fluff) released from mature bolls to the total number of bolls during cotton growth. It is usually expressed as a percentage and reflects both cotton yield and quality. The boll opening rate is influenced by various factors, including climate conditions, cultivation management, soil conditions, and pests and diseases. Improving the boll opening rate is an important goal in cotton production because it directly affects economic benefits and crop yield.

[0032] The Normalized Difference Vegetation Index (NDVI) is an index used to assess the degree of vegetation cover and the health of vegetation. It is calculated using multispectral image data acquired through remote sensing technology. NDVI is based on the reflectance characteristics of plants in the infrared and visible light bands. This index is widely used in agriculture, forestry, ecology, and earth sciences to monitor vegetation growth, land cover change, drought monitoring, and other fields.

[0033] Cotton, as one of the world's most important natural fibers, is not only an indispensable raw material for the textile industry, supporting the production of clothing and home textiles, but also makes a significant contribution to the agricultural economies of many countries, especially developing countries. Furthermore, as a renewable resource, cotton is more environmentally friendly than synthetic fibers, and its byproducts, such as cottonseed oil and cottonseed cake, have wide applications in food processing and animal feed, demonstrating its multifaceted economic and social value. Remote sensing technology, as an emerging technological means, can provide high-resolution, large-scale surface information. Through multispectral remote sensing technology, information about cotton fields can be acquired non-contactly and over large areas. This helps farmers or managers make more scientific and rational management decisions based on the actual growth status of cotton (such as the progress of boll opening), and understanding the changing trends of cotton boll opening rate helps determine the optimal harvest time, avoiding losses caused by harvesting too early or too late, and ensuring the maximization of cotton quality and economic value.

[0034] The applicant has found that while existing remote sensing technologies have made some progress in extracting cotton planting area and obtaining boll opening rate, problems remain, such as inaccurate cotton planting area extraction and overly linear, weak generalization ability of cotton boll opening rate inversion models. Therefore, this invention will utilize various methods, including deep learning and machine learning, based on remote sensing data and ground sampling data to construct a cotton identification and boll opening rate inversion model, thereby improving the accuracy of cotton identification and boll opening rate inversion.

[0035] First, this application provides a method 100 for monitoring the cotton boll opening rate, such as... Figure 1 As shown, this may include steps S110-S120. The specific technical approach is described below. Figure 2 As shown.

[0036] Step S110: Based on the multi-temporal multispectral remote sensing data of the target area and the first training model, identify the spatial distribution map of cotton in the target area.

[0037] In this step, cotton planting areas can be extracted based on multi-temporal remote sensing data, using a first training model such as a deep neural network model, thereby providing basic data for monitoring cotton boll opening rate.

[0038] Specifically, the first training model can be constructed based on the sample point locations of cotton and the standard growth characteristic curve. For example, the first training model can be constructed through the following steps: 1) Based on multi-temporal and multispectral remote sensing data, determine the normalized vegetation index (NDI) of cotton at the sample point locations. The specific steps are as follows: First, multi-temporal satellite remote sensing data needs to be acquired. This invention can use Sentinel-2 satellite remote sensing data as an example, accessing and downloading multispectral time-series data of the target area and monitoring time range from the European Space Agency website.

[0039] Secondly, after acquiring multi-temporal multispectral remote sensing data of the target area, the acquired multi-temporal multispectral remote sensing data of the target area can be preprocessed in the following ways: correction, stitching, and cropping. That is, the acquired multispectral time-series data can be preprocessed by correction, stitching, and cropping.

[0040] Finally, the reflectance of multiple bands in the multi-temporal multispectral remote sensing data is obtained, such as the reflectance of the red (R), blue (B), green (G), and near-infrared (NIR) bands in the multispectral data. Based on the obtained reflectance of multiple bands in the multi-temporal multispectral remote sensing data, the Normalized Difference Vegetation Index (NDVI) of cotton at the sample point location is calculated using the following formula:

[0041] The Normalized Difference Vegetation Index (NDVI) can be used to monitor vegetation growth status and vegetation cover. NIR and R represent the near-infrared and red bands of multispectral remote sensing data, respectively. In other words, R represents the red band of reflectance across multiple bands, and NIR represents the near-infrared band of reflectance across multiple bands.

[0042] 2) Based on historical accumulated temperature data and normalized vegetation index, construct the standard growth characteristic curve of cotton.

[0043] First, historical accumulated temperature data can be used to calculate the effective accumulated temperature (GDD) for cotton. Effective accumulated temperature is crucial for cotton growth, promoting plant growth, flowering, and boll formation, increasing yield and fiber quality, and shortening the maturity period. Simultaneously, suitable accumulated temperature also enhances cotton's resistance to adverse conditions, helping it cope with pests, diseases, and inclement weather. Proper management of effective accumulated temperature can significantly improve the economic value of cotton. For example, if the commonly used benchmark temperature for cotton is 10 degrees Celsius, the calculation process is as follows: GDD =∑ i =1n( 2Tmax,i+Tmin,i−10 ) in, n It is the number of days the cotton grows. Tmax,i For the first i The highest temperature of the day (degrees Celsius). Tmin,i For the first i The lowest temperature of the day (degrees Celsius).

[0044] Then, a standard growth characteristic curve (GDD-NDVI) can be constructed. Specifically, the effective accumulated temperature (GDD) of cotton is used as the x-axis, and the mean NDVI of normally growing fields is used as the y-axis to establish the cotton standard growth characteristic curve. For example, this invention uses cotton extraction from a county in Northwest China as an example, and the constructed cotton standard growth characteristic curve is as follows. Figure 3 As shown.

[0045] Finally, a cotton extraction dataset can be constructed based on cotton sample point data and GDD-NDVI. That is, a cotton identification dataset can be constructed based on the location of cotton sample points and the GDD-NDVI obtained above, thereby constructing the training data for the first training model, including scattered points on the accumulated temperature curve of the sample points and whether they are cotton (cotton is 1, non-cotton is 0). The above data can be combined into data pairs and stored in a table for easy use later.

[0046] 3) Based on the sample point locations of cotton and the standard growth characteristic curve, a deep neural network model is used for training to obtain the first training model.

[0047] In this invention, the deep learning framework of the deep neural network model can be TensorFlow, and the deep neural network model can include an input layer, a hidden layer and an output layer. The activation function of the hidden layer can be ReLU, the activation function of the output layer can be Softmax, the optimizer can be the Adam algorithm, and the loss function can be the Sparse Categorical Crossentropy (SCCE).

[0048] Specifically, deep neural networks (DNNs) are an emerging algorithm in the field of deep learning. Their structure typically includes an input layer, hidden layers, and an output layer. The input layer passes training data to the hidden layers, where neurons compute features and ultimately output the corresponding result, i.e., identifying the cotton planting area. For example, this invention uses the TensorFlow deep learning framework for building, training, and verifying the accuracy of the DNN. The specific operation steps are as follows: a) Confirm the number of layers and neurons in the deep neural network.

[0049] The network structure consists of a flattening layer, a fully connected layer, and an output layer. The flattening layer compresses the input data into one-dimensional data. A fully connected layer (e.g., with 128 neurons) then receives the one-dimensional data, performs vector processing, and finally, the output layer outputs the result. Since there are two possible outputs (cotton and non-cotton), the number of neurons in the output layer is set to 2.

[0050] b) Identify the activation function of the deep neural network.

[0051] In the hidden layer, the ReLU activation function (Equation 1 below) can be used. This function can speed up the training process and avoid the influence of the Sigmoid function on network convergence in the saturation region.

[0052] (1) Additionally, the Softmax activation function (Equation 2 below) can be used in the output layer to convert the output of the hidden layer into a probability distribution.

[0053] (2) c) Confirm the deep neural network training method.

[0054] In this study, the Adam algorithm can be used as the optimizer, and training can be performed using its default parameters. Adam designs an independent adaptive learning rate for each parameter by calculating the first and second-order estimates of the gradient. Specifically, the exponential decay rate of the first-order matrix can be set to 0.9, the decay rate of the second-order matrix can be set to 0.99, and the stability constant can be set to 10. -8 Furthermore, the loss function can be a sparse multi-class cross-entropy function, which is suitable for handling multi-class tasks with integer labels.

[0055] d) Confirm the training parameters for the deep neural network.

[0056] To improve efficiency and explore the optimal iterations for the model, the number of iterations can be set to 140. Additionally, the training and test sets can be allocated according to a set ratio, for example, the training set comprising 70% of the total samples, with the remaining 30% used for testing. Simultaneously, the parameters can be set to epochs=140, validation_split=0.25, and validation performed every two epochs.

[0057] 4) Evaluate the first trained model. For example, one or more of the following methods can be used to evaluate the model: accuracy, precision, recall, intersection-over-union ratio, etc.

[0058] First, the following parameters can be defined: True Positives (TP): The number of samples that correctly identify the positive class as valid samples.

[0059] True Negatives (TN): The number of negative samples that are correctly identified as invalid samples.

[0060] False Positives (FP): The number of invalid samples that are mistakenly identified as valid samples.

[0061] False Negatives (FN): This refers to the number of samples that were not identified as valid samples but were instead identified as invalid samples.

[0062] That is, TP is the number of positive samples correctly identified as valid samples, TN is the number of negative samples correctly identified as invalid samples, FP is the number of invalid samples incorrectly confused as valid samples, and FN is the number of valid samples incorrectly identified as invalid samples.

[0063] Then the accuracy rate ( (A) can be represented as:

[0064] Accuracy is the proportion of samples correctly classified by the model out of the total number of samples. The higher the accuracy, the higher the overall classification accuracy of the model.

[0065] Precision rate ( (abbreviated as P) can be represented as:

[0066] Precision refers to the proportion of samples that a model predicts as positive, but are actually positive. A higher precision means a lower probability that the model will misclassify a negative example as a positive one, and thus a higher level of model accuracy.

[0067] Recall rate ( (abbreviated as R) can be represented as:

[0068] Recall refers to the proportion of samples that are actually positive that are correctly predicted as positive by the model. A higher recall indicates that the model is better able to identify positive examples and has higher robustness.

[0069] Furthermore, to comprehensively evaluate the model by combining precision and recall, this invention can also use F-Score as an evaluation metric. F-Score can be expressed as:

[0070] in F is a parameter, usually set to 1. P represents precision, and R represents recall. A higher F-Score indicates that the model performs better in both accuracy and robustness.

[0071] Through the above steps, a deep neural network model can be constructed, namely the first training model. Based on the model with optimal parameters, multi-temporal remote sensing data of the target area can be input to output a spatial distribution map of cotton in the target area, such as... Figure 4 The image shows a spatial distribution map of cotton in a county in northwestern China. It can be seen that this first training model can be used to extract cotton planting areas, thus providing basic data for monitoring cotton boll opening rates.

[0072] Step S120: Determine the cotton boll opening rate based on the spatial distribution map obtained from the first training model, single-scene multispectral remote sensing data, and the second training model.

[0073] In this step, multispectral remote sensing data and ground boll opening rate sample data can be used as the input. The input is single-scene multispectral remote sensing data (i.e., remote sensing data of the time when the user needs to identify the boll opening rate). Then, machine learning methods are used to invert the boll opening rate of cotton in the target area (the cotton area obtained in step S110).

[0074] Specifically, the second training model can be constructed based on multiple spectral data of the sample point locations and ground-level cotton boll opening rate sampling data. For example, the second training model can be constructed through the following steps: 1) At the cotton sampling point location, obtain the ground boll opening rate sampling data of the cotton.

[0075] This step requires obtaining sample data on the boll opening rate of cotton on the ground. The boll opening rate (BOR) is obtained by surveying each smallest unit plot to study the relationship between vegetation indices and the boll opening rate. For example, five sampling points can be determined in a "Z" pattern within each plot, with six cotton plants selected at each point. The total number of bolls and the number of opened bolls per plant are then counted. The formula for calculating BOR is:

[0076] Among them, NOB (Number of Open Bolls) is the number of open bolls, and NTB (Number of Total Bolls) is the total number of bolls on the marked cotton plant.

[0077] 2) Based on the multi-temporal multispectral remote sensing data of the target area, obtain multiple spectral data of the sample point locations.

[0078] Specifically, firstly, it is also necessary to acquire multi-temporal satellite remote sensing data. As mentioned above, Sentinel-2 satellite remote sensing data can be used as an example to download multispectral time-series data of the target area and monitoring time range from the European Space Agency website.

[0079] Secondly, the acquired multispectral time-series data is preprocessed. For example, preprocessing methods such as correction, stitching, and cropping can be used to preprocess the acquired multitemporal multispectral remote sensing data of the target area.

[0080] Finally, the reflectance of multiple bands in the multi-temporal multispectral remote sensing data is obtained, such as the reflectance of the red (R), blue (B), green (G), and near-infrared (NIR) bands of the multispectral data. That is, multiple spectral values ​​can include red, green, blue, and near-infrared spectral values.

[0081] 3) Construct an inversion dataset based on multiple spectral data and ground boll opening rate sampling data of cotton, and train the inversion dataset using a random forest model to obtain a second training model.

[0082] In this invention, the cotton boll opening rate inversion dataset can first be constructed based on ground-collected boll opening rate data and corresponding spectral data. That is, the training data for the second training model consists of the red, green, blue, and near-infrared spectral values ​​of the sample points and the corresponding measured boll opening rate values. The above data can also be constructed together into data pairs and stored in a table.

[0083] Secondly, a random forest model can be constructed, and the aforementioned inversion dataset can be imported into the model for training and model tuning. The inversion dataset can include a set ratio of training and test sets, for example, 70% training and 30% test.

[0084] Specifically, to optimize model performance, this invention can model a random forest by inputting the training set into a random forest regressor. Then, using a test set, the random forest model is tuned using grid search cross-validation (GC), evaluating its performance under different parameter combinations. The parameters used can include the number of trees (n_estimators), the maximum depth (max_depth), and the minimum number of sample splits per node (min_samples_split). Finally, the best-performing model parameters are selected, and the cross-validation results are output.

[0085] Finally, model evaluation can be performed, for example, by evaluating the random forest model using root mean square error (RMSE) and R² values, thereby measuring the fit of the second model on the training data.

[0086] Through the above steps, a random forest model, i.e., the second training model, can be constructed. Based on the cotton spatial distribution map of the target area, it can further predict the cotton boll opening rate under different single-scene multispectral remote sensing data, and produce a cotton boll opening rate spatial distribution map.

[0087] While existing remote sensing technologies have made some progress in extracting cotton planting area and obtaining boll opening rate, problems remain, such as inaccurate cotton planting area extraction and overly linear and weak generalization ability in cotton boll opening rate inversion models. Therefore, this invention utilizes a combination of methods, including deep learning and machine learning, based on remote sensing data and ground sampling data to construct a cotton identification and boll opening rate inversion model. Specifically, this invention first extracts cotton planting areas based on cotton growth standard curves and deep neural networks, and then monitors cotton boll opening rate using machine learning methods based on multispectral data and ground sampling data. Compared to existing technologies, this patent can first extract the cotton growth area as basic data, and then use machine learning methods to invert the cotton boll opening rate, resulting in greater generalization and higher inversion accuracy compared to the linear regression modeling of other comparative patents.

[0088] In summary, this invention proposes a method based on multispectral remote sensing data, employing a deep neural network (DNN) model to extract cotton planting areas and obtain a spatial distribution map of cotton, and then using a random forest model to invert the cotton boll opening rate and obtain a spatial distribution map of the cotton boll opening rate. The beneficial effects of this invention include: the cotton growth standard curve plotted based on multi-temporal remote sensing data and the identification of cotton planting areas using a deep neural network can improve the accuracy of cotton extraction. Furthermore, this invention can also invert the cotton boll opening rate using a random forest model based on remote sensing data, thereby improving the accuracy of the cotton boll opening rate inversion.

[0089] on the other hand, Figure 5 A schematic diagram of a cotton boll opening rate monitoring system according to an embodiment of this application is shown. In this embodiment, the optimization system 200 may include: The spatial distribution identification device 210 is used to identify the spatial distribution map of cotton in the target area based on multi-temporal multispectral remote sensing data of the target area and a first training model, wherein the first training model is constructed based on the sample point locations of cotton and standard growth characteristic curves.

[0090] The boll opening rate determination device 220 is used to determine the boll opening rate of cotton based on single-scene multispectral remote sensing data on a spatial distribution map and a second training model. The second training model is constructed based on multiple spectral data of sample point locations and ground boll opening rate sampling data of cotton.

[0091] In this embodiment of the application, the first training model is constructed through the following steps: determining the normalized vegetation index of the cotton at the sample point location based on the multi-temporal multispectral remote sensing data; constructing the standard growth characteristic curve of the cotton based on the historical accumulated temperature data of the cotton and the normalized vegetation index; and training the first training model using a deep neural network model based on the sample point location of the cotton and the standard growth characteristic curve.

[0092] In this embodiment, the deep learning framework of the deep neural network model is TensorFlow, and the deep neural network model includes an input layer, a hidden layer and an output layer. The activation function of the hidden layer is ReLU, the activation function of the output layer is Softmax, the optimizer is the Adam algorithm, and the loss function is the sparse multi-class cross-entropy function.

[0093] In this embodiment of the application, determining the normalized vegetation index (NDVI) of the cotton at the sample point location based on the multi-temporal multispectral remote sensing data includes: calculating the NDVI of the cotton at the sample point location based on the reflectance of multiple bands in the multi-temporal multispectral remote sensing data using the following formula:

[0094] Wherein, R represents the red band among the multiple band reflectances, and NIR represents the near-infrared band among the multiple band reflectances.

[0095] In the embodiments of this application, the first training model is evaluated and optimized using at least one of the following methods: accuracy, precision, recall, and intersection-union ratio.

[0096] In this embodiment of the application, the accuracy A is expressed by the following formula:

[0097] The accuracy P is expressed by the following formula:

[0098] The recall rate R is expressed by the following formula:

[0099] Wherein, TP is the number of positive samples correctly identified as valid samples; TN is the number of negative samples correctly identified as invalid samples; FP is the number of invalid samples incorrectly confused as valid samples; and FN is the number of valid samples incorrectly identified as invalid samples.

[0100] In this embodiment of the application, the second training model is constructed through the following steps: at the sample point location of the cotton, the ground boll opening rate sampling data of the cotton is obtained; based on the multi-temporal multispectral remote sensing data of the target area, multiple spectral data at the sample point location are obtained, wherein the multiple spectral values ​​include red, green, blue, and near-infrared spectral values; and based on the multiple spectral data and the ground boll opening rate sampling data of the cotton, an inversion dataset is constructed, and a random forest model is used to train the inversion dataset to obtain the second training model.

[0101] In this embodiment of the application, the inversion dataset includes a training set and a test set with a set ratio. The training set is input into a random forest regressor to model the random forest model. The test set is used to fine-tune the random forest model using grid search cross-validation. The root mean square error and R² value are used to evaluate the random forest model.

[0102] In this embodiment of the application, the monitoring system further includes a preprocessing device for preprocessing the acquired multi-temporal multispectral remote sensing data of the target area in the following manner: correction, stitching, and cropping.

[0103] The present invention proposes a system based on multispectral remote sensing data, employing a deep neural network (DNN) model to extract cotton planting areas and obtain a spatial distribution map of cotton, and using a random forest model to invert cotton boll opening rate and obtain a spatial distribution map of cotton boll opening rate. The beneficial effects of this invention include: the cotton growth standard curve plotted based on multi-temporal remote sensing data and the identification of cotton planting areas using a deep neural network can improve the accuracy of cotton extraction. Furthermore, this invention can also use a random forest model based on remote sensing data to invert cotton boll opening rate, thereby improving the accuracy of cotton boll opening rate inversion.

[0104] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0105] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for monitoring cotton boll opening rate, characterized in that, The monitoring method includes: Based on multi-temporal, multispectral remote sensing data of the target area and a first training model, the spatial distribution map of cotton in the target area is identified, wherein the first training model is constructed based on the sample point locations of cotton and standard growth characteristic curves; and The cotton boll opening rate is determined based on the spatial distribution map obtained from the first training model, single-scene multispectral remote sensing data, and the second training model. The second training model is constructed based on multiple spectral data of the sample point location and ground boll opening rate sampling data of the cotton.

2. The monitoring method according to claim 1, characterized in that, The first training model is constructed using the following steps: Based on the multi-temporal multispectral remote sensing data, the normalized vegetation index of the cotton at the sample point location is determined; Based on the historical accumulated temperature data of cotton and the normalized vegetation index, a standard growth characteristic curve of cotton is constructed. as well as Based on the sample point locations of the cotton and the standard growth characteristic curve, a deep neural network model is used for training to obtain the first training model.

3. The monitoring method according to claim 2, characterized in that, The deep neural network model uses TensorFlow as its deep learning framework and includes an input layer, a hidden layer, and an output layer. The activation function of the hidden layer is ReLU, the activation function of the output layer is Softmax, the optimizer is the Adam algorithm, and the loss function is the sparse multi-class cross-entropy function.

4. The monitoring method according to claim 2, characterized in that, The step of determining the normalized vegetation index of the cotton at the sample point location based on the multi-temporal multispectral remote sensing data includes: Based on the reflectance of multiple bands in the multi-temporal multispectral remote sensing data, the Normalized Difference Vegetation Index (NDVI) of the cotton at the sample point location is calculated using the following formula: Wherein, R represents the red band among the multiple band reflectances, and NIR represents the near-infrared band among the multiple band reflectances.

5. The monitoring method according to claim 2, characterized in that, The first trained model is evaluated and optimized using at least one of the following methods: accuracy, precision, recall, and intersection-union ratio.

6. The monitoring method according to claim 5, characterized in that, The accuracy A is expressed by the following formula: The accuracy P is expressed by the following formula: The recall rate R is expressed by the following formula: Wherein, TP is the number of positive samples correctly identified as valid samples; TN is the number of negative samples correctly identified as invalid samples; FP is the number of invalid samples incorrectly confused as valid samples; and FN is the number of valid samples incorrectly identified as invalid samples.

7. The monitoring method according to claim 1, characterized in that, The second training model is constructed using the following steps: At the sample point location of the cotton, obtain the ground boll opening rate sampling data of the cotton; Based on multi-temporal multispectral remote sensing data of the target area, multiple spectral data of the sample point location are obtained, wherein the multiple spectral values ​​include red, green, blue, and near-infrared spectral values; and An inversion dataset is constructed based on the multiple spectral data and the ground boll opening rate sampling data of the cotton, and a random forest model is used to train the inversion dataset to obtain the second training model.

8. The monitoring method according to claim 7, characterized in that, The inversion dataset includes a training set and a test set in a predetermined ratio. The random forest model is built by inputting the training set into a random forest regressor. The random forest model was tuned using grid search cross-validation on the test set, and evaluated using root mean square error and R² value.

9. The monitoring method according to any one of claims 1-8, characterized in that, The monitoring method further includes: preprocessing the acquired multi-temporal multispectral remote sensing data of the target area in the following manner: correction, stitching, and cropping.

10. A monitoring system for cotton boll opening rate, characterized in that, The monitoring system includes: A spatial distribution identification device is used to identify the spatial distribution map of cotton in a target area based on multi-temporal multispectral remote sensing data of the target area and a first training model, wherein the first training model is constructed based on the sample point locations of cotton and standard growth characteristic curves; and The cotton boll opening rate determination device is used to determine the cotton boll opening rate based on the spatial distribution map obtained from the first training model, single-scene multispectral remote sensing data, and the second training model, wherein the second training model is constructed based on multiple spectral data of the sample point location and ground cotton boll opening rate sampling data.

Images