A sugarcane sharpness prediction method, device, medium and product

Abstract

The application discloses a sugarcane brix prediction method, device, medium and product, relates to the field of agricultural remote sensing and crop phenotype detection, and comprises the following steps: preprocessing unmanned aerial vehicle multispectral remote sensing images to obtain a ground reflectance image; based on the ground reflectance image, spectral features and texture features are extracted respectively to construct an initial feature set; a feature selection method is used to perform feature screening on the initial feature set to obtain a sensitive feature subset; the sensitive feature subset is taken as an input variable, a sugarcane brix measured data set is taken as a target variable, a machine learning regression model is used for training, and a sugarcane brix prediction model is obtained; unmanned aerial vehicle multispectral remote sensing images of a to-be-predicted sugarcane field and corresponding sensitive features are acquired; and according to the sensitive features, the sugarcane brix prediction model is used to predict the sugarcane brix. The application can realize lossless, high-throughput and cross-condition stable prediction of the sugarcane brix.

Description

Technical Field

[0001] This application relates to the fields of agricultural remote sensing and crop phenotyping, and in particular to a method, equipment, medium, and product for predicting sugarcane hammer weight. Background Technology

[0002] Sugarcane is the world's most important sugar crop, accounting for over 70% of global sugar production. As the world's third-largest sugarcane producer, China's sugar production accounts for over 90% of the country's total sugar output, playing a vital role in ensuring national sugar supply security, promoting agricultural development in sugarcane-growing areas, and increasing farmers' income.

[0003] Brix refers to the soluble solids content (Bx) in sugarcane juice at 20°C, of ​​which sucrose accounts for 85% to 95%. Brix is ​​highly significantly positively correlated with sucrose content and is a recognized rapid proxy indicator for sucrose content. It is widely used as a key economic phenotype to characterize the level of sugar accumulation, maturity, and raw material quality in sugarcane.

[0004] Currently, the determination of sugarcane saturation mainly relies on manual field sampling and juice extraction followed by testing with a handheld refractometer or a laboratory polarized light saccharimeter. Although this method has high accuracy, it is highly destructive and inefficient, cannot continuously monitor the same plant, and is time-consuming and labor-intensive, making it difficult to meet the needs of high-throughput sugarcane phenotyping. At the same time, the number of manual sampling points is limited, resulting in insufficient spatial representativeness and difficulty in fully reflecting the spatial heterogeneity of saturation in a large sugarcane field, leading to poor representativeness of monitoring results. Current technologies have not yet achieved rapid, non-destructive, and continuous monitoring of sugarcane saturation, limiting its application in precision agricultural management.

[0005] In recent years, remote sensing technology has gradually become an important means of crop phenotypic monitoring due to its advantages of non-destructive, large-scale and high-throughput acquisition of crop canopy information. Among them, UAV remote sensing platforms have provided a new technical approach for dynamic monitoring of crops under complex terrain conditions with advantages such as centimeter-level spatial resolution, flexible acquisition, relatively low cost and less impact from cloud and rain weather. In existing remote sensing-based studies on sugarcane slack weight monitoring, regression models are mainly constructed using spectral features, which have achieved certain results under specific conditions. However, the following shortcomings still exist: First, sugarcane is a typical tall, densely planted C4 crop with distinct vertical canopy stratification, severe leaf overlap, and tall plant size, which easily leads to canopy spectral signal saturation. Relying solely on spectral information is insufficient to effectively characterize subtle changes in slack weight. Second, sugarcane slack weight is not only related to chlorophyll content and plant growth, but also influenced by factors such as source-sink coordination, sugar translocation efficiency, stalk fullness, and canopy structural heterogeneity. Existing models based on single spectral features are insufficient to fully capture this structural information. Third, existing models exhibit poor generalization ability under different varieties, fertilization conditions, and growth stages, making it difficult to achieve stable cross-condition application.

[0006] Subsequently, texture features are introduced. Texture features can quantify the spatial distribution pattern of image pixel grayscale, reflecting information such as canopy density, leaf arrangement, canopy evenness, and structural heterogeneity. They can provide a supplementary dimension to crop growth status monitoring, distinct from spectral information. Existing studies have shown that texture features have high potential in monitoring leaf area index, biomass, chlorophyll content, and nitrogen, and to some extent outperform simple band reflectance or vegetation indices. Furthermore, the fusion of vegetation indices and texture features can integrate the advantages of both types of features in spectral response and structural characterization, thereby improving the accuracy of crop parameter inversion. However, current technologies lack research on applying texture features to sugarcane hammer weight prediction, especially a systematic evaluation of the applicability of spectral, texture, and their fused features in hammer weight monitoring.

[0007] Furthermore, the introduction of texture features leads to increasingly prominent problems of multicollinearity and information redundancy among high-dimensional variables, which can easily cause model overfitting, decreased computational efficiency, and weakened cross-temporal and spatial generalization ability. Traditional regression methods often struggle to balance feature correlation and redundancy, thus limiting model accuracy and extrapolation ability. Although machine learning methods have significant advantages in high-dimensional feature processing, characterization of complex nonlinear relationships, and improvement of model robustness, there are currently no applications of machine learning models that integrate texture features in remote sensing monitoring of sugarcane hammer weight. Systematic research on sensitive feature screening, model optimization, explanation of mechanisms of action, and robustness across varieties, fertilization treatments, and years still needs to be deepened.

[0008] In view of the shortcomings of the existing technology, there is an urgent need to provide a new method for predicting sugarcane hammer weight. Summary of the Invention

[0009] The purpose of this application is to provide a method, equipment, medium, and product for predicting sugarcane hammer weight, which can achieve non-destructive, high-throughput, and cross-conditional stable prediction of sugarcane hammer weight.

[0010] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for predicting sugarcane hammer weight, including: Acquire a measured dataset of sugarcane hammer weight in sugarcane fields and simultaneously acquire UAV multispectral remote sensing images covering the sugarcane fields; The UAV multispectral remote sensing image is preprocessed to obtain a surface reflectance image; Based on the surface reflectance image, spectral features and texture features are extracted respectively to construct an initial feature set; The initial feature set is filtered using a feature selection method to obtain a sensitive feature subset. Using the sensitive feature subset as input variables and the measured sugarcane hammer weight dataset as target variables, a machine learning regression model is used for training to obtain a sugarcane hammer weight prediction model. Acquire UAV multispectral remote sensing images of sugarcane fields to be predicted and their corresponding sensitive features; Based on the sensitive characteristics, the sugarcane hammer weight prediction model is used to predict the sugarcane hammer weight of the sugarcane field to be predicted.

[0011] Secondly, this application provides a sugarcane hammer weight prediction device, comprising: The data acquisition module is used to acquire the measured dataset of sugarcane hammer weight in sugarcane fields and simultaneously acquire UAV multispectral remote sensing images covering the sugarcane fields. The data preprocessing module is used to preprocess the UAV multispectral remote sensing images to obtain surface reflectance images; The feature extraction module is used to extract spectral features and texture features based on the surface reflectance image to construct an initial feature set; The feature filtering module is used to filter the initial feature set using a feature selection method to obtain a sensitive feature subset; The prediction model building module is used to take the sensitive feature subset as input variables, the measured sugarcane hammer weight dataset as target variables, and train it using a machine learning regression model to obtain a sugarcane hammer weight prediction model. The sugarcane hammer weight prediction module is used to acquire UAV multispectral remote sensing images of the sugarcane field to be predicted and the corresponding sensitive features; based on the sensitive features, the sugarcane hammer weight prediction model is used to predict the sugarcane hammer weight of the sugarcane field to be predicted.

[0012] Thirdly, this application 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 computer program to implement the sugarcane weight prediction method.

[0013] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the sugarcane weight prediction method.

[0014] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the sugarcane weight prediction method.

[0015] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, device, medium, and product for predicting sugarcane weight. By combining texture features (TFs) and spectral features (SFs) extracted from UAV multispectral remote sensing images, and employing a feature selection algorithm to screen sensitive spectral and texture features suitable for sugarcane weight monitoring, and then combining them with a machine learning regression model, this method solves the problems of spectral saturation, single information dimension, and weak model generalization ability caused by the complex canopy structure in existing single spectral models for sugarcane weight monitoring. This method achieves non-destructive, high-throughput, and cross-condition stable prediction of sugarcane weight. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of a sugarcane hammer weight prediction method in one embodiment of this application; Figure 2 This is a 10-fold cross-validation line chart; Figure 3 Schematic diagram of model performance analysis under different conditions ( Figure 3 Part (a) compares the distribution characteristics of the measured and predicted values. Figure 3 Part (b) is a scatter plot of regression analysis for different varieties; Figure 3 Part (c) represents the model performance under different nitrogen fertilizer treatments; Figure 3 (Part (d) compares the predictive effects of different planting types). Figure 4 The screening results based on the mRMR algorithm ( Figure 4 Part (a) consists of the six selected sensitive spectral features; Figure 4 (b) consists of the eight selected sensitive texture features. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] In one exemplary embodiment, such as Figure 1 As shown, a method for predicting sugarcane hammer weight is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the sugarcane hammer weight prediction method includes the following steps S101 to S107. Wherein: S101, acquire the measured dataset of sugarcane hammer weight in the sugarcane field, and simultaneously acquire UAV multispectral remote sensing images covering the sugarcane field; As another specific embodiment, UAV multispectral remote sensing images can also be replaced by UAV RGB or hyperspectral images, or if the spatial resolution of satellite remote sensing images (such as Gaofen satellite, Jilin satellite, PlanetScope series, etc.) is high enough (e.g., better than 5 meters) to extract effective canopy texture features, they can also be used as alternative data sources.

[0021] As a specific example, multiple sampling points (e.g., 1 meter) are obtained within the sugarcane field. The measured weight of sugarcane was obtained from a 1-meter-long, uniformly grown, weed-free sugarcane plot. The measured weight was obtained by measuring the sugarcane stalk sap using a refractometer. To enhance the model's generalization ability, the training sample data should cover weight values ​​under different sugarcane varieties (e.g., Guiliu 05136, Guitang 42), different nitrogen fertilizer application levels, and different planting years (newly planted sugarcane, ratooned sugarcane). On the same day as the hammer sampling, multispectral images covering the target sugarcane field were acquired using a drone platform equipped with a multispectral sensor. The multispectral images included at least blue light (450±16 nm), green light (560±16 nm), red light (650±16 nm), red edge (730±16 nm), and near-infrared (840±26 nm), with a pixel resolution of 2.08 million pixels. The aerial photography mission was conducted between 10:00 and 14:00 local time under clear, cloudless conditions. Flight path planning and image acquisition were completed using the DJI Pilot program on the drone, with a flight altitude of 130 m, a flight speed of 4.6 m / s, a forward overlap of 90%, and a lateral overlap of 80%. The raw images acquired by the aerial photography were preprocessed using DJI Terra software. The main processes included image alignment, ground control point spiculation, image geometric registration, and execution of bundled photogrammetry to produce dense point clouds and orthophotos. The stitched single-band multispectral images are uniformly projected onto the WGS_84 UTM Zone 49N coordinate system, with a spatial resolution accuracy better than 10cm.

[0022] Four diffuse reflection calibration cloths were placed flat within the study area, within the field of view of the UAV. The reflectivity of each cloth was measured using the corresponding instrument. Subsequently, radiometric correction was performed using an empirical model. Based on the known reflectivity of a standard diffuse reflection reference target, the DN values ​​of the images for each band were calibrated to minimize reflectivity errors caused by factors such as lens position, atmospheric conditions, and differences in radiation intensity. These four reflectivities were used to convert the DN values ​​of the UAV images to reflectivity values ​​in an empirically linear manner, eliminating radiometric distortion caused by varying illumination levels in images from different periods. The empirical formula is: ; In the formula, and These represent the corresponding bands of the multispectral image. Reflectivity and original DN value; and They represent the corresponding bands. The transformation coefficients are calculated using the least squares method; S102, preprocess the UAV multispectral remote sensing image to obtain a surface reflectance image; S102 specifically includes: S21 performs image alignment, geometric correction, and radiometric correction on UAV multispectral remote sensing images; S22 converts the original grayscale values ​​of the processed UAV multispectral remote sensing image into surface reflectance, and unifies the projection coordinate system and spatial resolution to generate multi-band surface reflectance images.

[0023] S103, Based on the surface reflectance image, extract spectral features and texture features respectively to construct an initial feature set; S103 specifically includes: S31, Based on the original band reflectance of the surface reflectance image, determine multiple vegetation indices; Specifically, as shown in Table 1, the vegetation indices (a total of 41 were calculated) include, but are not limited to: Normalized Difference Vegetation Index (NDVI), Normalized Red Edge Index (NDRE), Normalized Difference Green Light Vegetation Index (GNDVI), Soil-Adjusted Vegetation Index (SAVI), Enhanced Vegetation Index 2 (EVI2), and MERIS Terrestrial Chlorophyll Index (MTCI). In Table 1, Blue, Green, Red, Rededge, and Nir represent the reflectance values ​​of the corresponding bands in the multispectral image. Table 1 only lists some representative indices; the actual calculation can be extended to 41 or more.

[0024] Table 1

[0025] S32, combine multiple vegetation indices with the original band reflectance to form the spectral features; S33, The gray-level co-occurrence matrix (GLCM) method is used to perform texture analysis on each band of the surface reflectance image; the sliding window, step distance and calculation direction are set, and multiple texture statistics are extracted based on the gray-level co-occurrence matrix of each band; Specifically, the window size was set to 3×3, the step size was 1, and the direction was 45°. The following eight statistics were calculated for each band: contrast (Con), correlation (Cor), entropy (Ent), second moment (SM), mean (MEAN), variance (Variance, Var), homogeneity (Hom), and dissimilarity (Dis), as shown in Table 2.

[0026] Table 2

[0027] S34, combine the texture statistics of each band to form the texture feature.

[0028] S104, The initial feature set is filtered using a feature selection method to obtain a sensitive feature subset; The feature selection methods include, but are not limited to: maximum relevance minimum redundancy (mRMR), recursive feature elimination, feature importance ranking based on random forest, and LASSO regression (L1 regularization), to achieve the purpose of screening sensitive variables and reducing redundancy.

[0029] As a specific implementation, when the feature selection method is the maximum correlation minimum redundancy algorithm, S104 specifically includes: S41, Construct the objective optimization function; the objective optimization function is the difference between the correlation measure between the features and the objective variable in the initial feature set and the redundancy measure between the features; Specifically, using formulas Determine the objective optimization function; in, D represents the correlation between the feature and the hammer value, and R represents the redundancy between features. They are defined as follows: ; ; In the formula ，S For the selected feature subset, Bx The target variable is the hammer value. I ( x i , Bx ) as a feature x i With the goal Bx Inter-information I(x i , j ) Features x i and x j Mutual information between them.

[0030] S42, calculate the mutual information between each feature and the measured sugarcane weight dataset, and calculate the mutual information matrix between any two features; mutual information is used to measure the degree of interdependence between two random variables; For any two random variables x i and x j Their mutual information is defined as: ; In the formula, p(x) i ,x j ) is the variable x i The joint probability distribution of x.i ) and p(x j ) are x i With x j The marginal probability distribution. F is the set of all features, F={x1,x2,...,x...} i ,..x j ,..x m}, where i≠j, and m is the number of features.

[0031] S43, employing an incremental forward search strategy, features that maximize the objective optimization function value are sequentially added to the candidate feature subset, and a basic evaluation model (e.g., random forest, decision tree, etc.) is used to record the model evaluation metric after each feature addition; the model evaluation metric includes the coefficient of determination (R²). 2 ), Root Mean Square Error (RMSE) and Akaike Information Criterion (AIC); S44. Based on the preset evaluation index threshold optimization principle, determine the optimal number of features and the corresponding feature names, and output the sensitive feature subset; the evaluation index threshold optimization principle is that the information criterion is maximized, the determination coefficient is maximized, and the root mean square error is minimized.

[0032] S105, the sensitive feature subset is used as the input variable, the measured sugarcane weight dataset is used as the target variable, and a machine learning regression model is used for training to obtain a sugarcane weight prediction model. The machine learning regression models include, but are not limited to: Support Vector Machine Regression (SVR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Lightweight Gradient Boosting (LightGBM).

[0033] As a specific example, SVR is selected as the core prediction model. SVR maps the input features to a high-dimensional space through a kernel function (such as the radial basis function RBF) to construct the optimal hyperplane to minimize the prediction error. Using fused spectral and texture features (SFs+TFs) as input features, combined with SVR, yields the best prediction results. S106, Acquire UAV multispectral remote sensing images of the sugarcane field to be predicted and the corresponding sensitive features; S107. Based on the sensitive characteristics, the sugarcane hammer weight prediction model is used to predict the sugarcane hammer weight of the sugarcane field to be predicted.

[0034] The effectiveness of the proposed technical solution is verified by comparing the performance of different feature types (SFs, TFs, SFs+TFs) and different modeling methods (linear models and nonlinear machine learning models) in sugarcane hammer weight prediction. Experimental data were obtained from field trials conducted over two complete crushing seasons from 2022 to 2024, covering different sugarcane varieties (Guiliu 05136, Guitang 42), different nitrogen fertilizer treatment levels (N0, N1, N2, N3), and different planting types (newly planted sugarcane, ratooned sugarcane), totaling 228 samples. A 10-fold cross-validation method was used to evaluate model performance, with the coefficient of determination (R²) as the evaluation index. 2 ) and root mean square error (RMSE).

[0035] like Figure 4 As shown, the mRMR algorithm was used to select 6 sensitive spectral features (rededge, REDVI, DCabcac, DATT, NGRVI, GM) and 8 sensitive texture features (RE_MEAN, RE_Cor, RE_Var, R_Cor, G_Cor, B_MEAN, NIR_MEAN, NIR_SM). The selected feature subset effectively eliminated multicollinearity and information redundancy among features, providing high-quality input for subsequent modeling, reducing the risk of model overfitting, and improving computational efficiency.

[0036] Six spectral features and eight texture features were selected and fused to obtain fused features (SFs+TFs, a total of 14). Ten machine learning algorithms were used to construct prediction models: Multiple linear regression (MLR), Stepwise multiple regression (SMR), Random forest regression (RFR), Support vector regression (SVR), Ridge regression (RR), Adaboost (ADB), Lessasolute shrinkage and selection operator (LAASSO), Lightweight gradient boosting machine (LGB), Extreme gradient boosting (XGB), and Elastic net (EN). Nonlinear algorithms included SVR, LGB, ADB, XGB, and RFR, while linear algorithms included MLR, RR, LASSO, EN, and SMR. The technical results are shown in Table 4.

[0037] Table 3

[0038] As shown in Table 3, the following conclusions can be drawn: (1) Texture features are superior to spectral features: When using TFs for modeling, the R of the SVR model is better. 2 The performance of all models improved significantly, from 0.39 to 0.78 and from 4.36 to 2.56, demonstrating that texture features play a crucial role in predicting sugarcane hammer weight.

[0039] (2) Feature fusion further improves accuracy: When using SFs+TFs for modeling, the SVR model achieves the best results, R 2 =0.82, RMSE=2.35, which is better than the single feature type, indicating that spectral information and texture information are complementary.

[0040] (3) Nonlinear algorithms are superior to linear algorithms: The prediction performance of nonlinear machine learning models such as SVR is significantly better than that of linear regression models such as MLR and RR, proving that there is a complex nonlinear relationship between sugarcane weight and remote sensing features.

[0041] All samples were randomly divided into 10 mutually exclusive and complete subsets. Each time, one subset was used as the validation set, and the remaining nine subsets were used as the training set. This process was repeated 10 times to ensure that each subset served as the validation set in the evaluation. All models used the exact same hyperparameter settings to ensure fairness in the comparison. The achieved technical results, such as... Figure 2 Showing R values ​​of 10 models under three feature combinations 2 From the changes in RMSE, we can see that: (1) When using TFs, SVR, XGB, ADB and LGB are in most trade-offs R 2 It reaches 0.8 or higher.

[0042] (2) When using SFs+TFs, the overall performance of the model reaches its best, and SVR achieves optimal performance in multiple trade-offs. 2 Above 0.90, RMSE is as low as 1.69.

[0043] (3) SVR showed high R under different feature combinations. 2 With a low RMSE and minimal fluctuations between different folds, it is the most stable algorithm among the 10 models.

[0044] like Figure 3 As shown, the optimal prediction model (SFs+TFs+SVR) was applied to validation data of different sugarcane varieties (Guiliu 05136, Guitang 42), different nitrogen fertilizer treatments (N0, N1, N2, N3), and different planting types (newly planted sugarcane PC, ratooned sugarcane RC) to evaluate the model's predictive performance under different conditions, and the following conclusions were drawn: (1) Cross-variety: The model is used in Guiliu 05136 (R 2 =0.94) and Guitang No. 42 (R 2 The model performs well on all values ​​(=0.89), demonstrating its good adaptability to different varieties.

[0045] (2) Cross-fertilization treatments: The model R under N2 and N3 treatments 2 The average value reached 0.95, indicating that the model can adapt to different nitrogen fertilizer management levels.

[0046] (3) Across planting years: The model R on newly planted sugarcane and ratoon sugarcane 2 Both were 0.92, and although the RMSE for ratooned sugarcane was slightly higher (3.36 vs 1.42), it still maintained high prediction accuracy.

[0047] As a specific embodiment, as shown in Table 4, the sugarcane hammer weight prediction method based on UAV multispectral remote sensing proposed in this application achieves rapid, non-destructive, accurate, and high-throughput monitoring of sugarcane hammer weight by fusing spectral and texture features, using the mRMR algorithm to screen sensitive features, and combining the SVR nonlinear machine learning model. This method provides important technical support for the precise management and intelligent production of sugarcane.

[0048] Table 4

[0049] Based on the same inventive concept, this application also provides a sugarcane weight prediction device for implementing the sugarcane weight prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more sugarcane weight prediction device embodiments provided below can be found in the limitations of the sugarcane weight prediction method described above, and will not be repeated here.

[0050] In one exemplary embodiment, a sugarcane hammer weight prediction device is provided, comprising: The data acquisition module is used to acquire the measured dataset of sugarcane hammer weight in sugarcane fields and simultaneously acquire UAV multispectral remote sensing images covering the sugarcane fields. The data preprocessing module is used to preprocess the UAV multispectral remote sensing images to obtain surface reflectance images; The feature extraction module is used to extract spectral features and texture features based on the surface reflectance image to construct an initial feature set; The feature filtering module is used to filter the initial feature set using a feature selection method to obtain a sensitive feature subset; The prediction model building module is used to take the sensitive feature subset as input variables, the measured sugarcane hammer weight dataset as target variables, and train it using a machine learning regression model to obtain a sugarcane hammer weight prediction model. The sugarcane hammer weight prediction module is used to acquire UAV multispectral remote sensing images of the sugarcane field to be predicted and the corresponding sensitive features; based on the sensitive features, the sugarcane hammer weight prediction model is used to predict the sugarcane hammer weight of the sugarcane field to be predicted.

[0051] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a sugarcane weight prediction method.

[0052] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0053] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0054] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0055] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0056] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0057] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0058] In this application, all actions to acquire signals, information, or data are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with the authorization granted by the owner of the relevant device.

[0059] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0060] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the weight of sugarcane, characterized in that, include: Acquire a measured dataset of sugarcane hammer weight in sugarcane fields and simultaneously acquire UAV multispectral remote sensing images covering the sugarcane fields; The UAV multispectral remote sensing image is preprocessed to obtain a surface reflectance image; Based on the surface reflectance image, spectral features and texture features are extracted respectively to construct an initial feature set; The initial feature set is filtered using a feature selection method to obtain a sensitive feature subset. Using the sensitive feature subset as input variables and the measured sugarcane hammer weight dataset as target variables, a machine learning regression model is used for training to obtain a sugarcane hammer weight prediction model. Acquire UAV multispectral remote sensing images of sugarcane fields to be predicted and their corresponding sensitive features; Based on the sensitive characteristics, the sugarcane hammer weight prediction model is used to predict the sugarcane hammer weight of the sugarcane field to be predicted.

2. The sugarcane hammer weight prediction method according to claim 1, characterized in that, The preprocessing of the UAV multispectral remote sensing image to obtain a surface reflectance image specifically includes: Image alignment, geometric correction, and radiometric correction are performed on UAV multispectral remote sensing images; The original grayscale values ​​of the processed UAV multispectral remote sensing image are converted into surface reflectance, and the projection coordinate system and spatial resolution are unified to generate multi-band surface reflectance image.

3. The sugarcane hammer weight prediction method according to claim 1, characterized in that, The step of extracting spectral and textural features from the surface reflectance image to construct an initial feature set specifically includes: Based on the original band reflectance of the surface reflectance image, multiple vegetation indices are determined. The spectral features are formed by merging multiple vegetation indices with the original band reflectance. Texture analysis of each band of the surface reflectance image was performed using the gray-level co-occurrence matrix method; a sliding window, step size, and calculation direction were set, and multiple texture statistics were extracted based on the gray-level co-occurrence matrix of each band. The texture features are formed by combining the texture statistics of each band.

4. The sugarcane hammer weight prediction method according to claim 1, characterized in that, The feature selection methods include: maximum relevance minimum redundancy algorithm, recursive feature elimination, feature importance ranking based on random forest, and LASSO regression.

5. The sugarcane hammer weight prediction method according to claim 4, characterized in that, When the feature selection method is the maximum relevance minimum redundancy algorithm, the step of using the feature selection method to filter the initial feature set to obtain a sensitive feature subset specifically includes: Construct a target optimization function; the target optimization function is the difference between the correlation measure between features and the target variable and the redundancy measure between features in the initial feature set; Calculate the mutual information between each feature and the measured sugarcane weight dataset, and calculate the mutual information matrix between any two features; An incremental forward search strategy is adopted to sequentially add the features that maximize the value of the objective optimization function to the candidate feature subset, and the model evaluation index after each feature addition is recorded using a basic evaluation model; the model evaluation index includes the coefficient of determination, root mean square error, and Akaike information criterion. Based on the preset evaluation index threshold optimization principle, the optimal number of features and the corresponding feature names are determined, and the sensitive feature subset is output; the evaluation index threshold optimization principle is that the information criterion is maximized, the determination coefficient is maximized, and the root mean square error is minimized.

6. The sugarcane hammer weight prediction method according to claim 1, characterized in that, The machine learning regression models include support vector machine regression models, random forest models, extreme gradient boosting models, and lightweight gradient boosting machines.

7. A sugarcane weight prediction device, characterized in that, include: The data acquisition module is used to acquire the measured dataset of sugarcane hammer weight in sugarcane fields and simultaneously acquire UAV multispectral remote sensing images covering the sugarcane fields. The data preprocessing module is used to preprocess the UAV multispectral remote sensing images to obtain surface reflectance images; The feature extraction module is used to extract spectral features and texture features based on the surface reflectance image to construct an initial feature set; The feature filtering module is used to filter the initial feature set using a feature selection method to obtain a sensitive feature subset; The prediction model building module is used to take the sensitive feature subset as input variables, the measured sugarcane hammer weight dataset as target variables, and train it using a machine learning regression model to obtain a sugarcane hammer weight prediction model. The sugarcane hammer weight prediction module is used to acquire UAV multispectral remote sensing images of the sugarcane field to be predicted and the corresponding sensitive features; Based on the sensitive characteristics, the sugarcane hammer weight prediction model is used to predict the sugarcane hammer weight of the sugarcane field to be predicted.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the sugarcane hammer weight prediction method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the sugarcane hammer weight prediction method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the sugarcane hammer weight prediction method as described in any one of claims 1-6.

Images