Winter wheat leaf area index estimation method and system based on unmanned aerial vehicle remote sensing

Abstract

The invention provides a winter wheat leaf area index estimation method and system based on unmanned aerial vehicle remote sensing, and the method comprises the following steps: S1, data acquisition: obtaining unmanned aerial vehicle multispectral remote sensing image data of a research region, and collecting corresponding field actual measurement data in different growth periods of winter wheat; according to the winter wheat leaf area index estimation method and system based on unmanned aerial vehicle remote sensing provided by the invention, through the step design of S1 to S7, especially the collaborative cooperation of data preprocessing, data fusion, model construction and optimization, during use, on one hand, the leaf area index of the winter wheat can be estimated; in S2, splicing and geographic coordinate system embedding processing are carried out on unmanned aerial vehicle multispectral remote sensing image data, spatial positions of different wavebands can be automatically aligned, the problem that the number of rows and columns of multiband data is inconsistent is solved fundamentally, and a spatially unified data source is provided for subsequent vegetation index calculation and feature extraction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural remote sensing monitoring, and in particular to a winter wheat leaf area index estimation method and system based on unmanned aerial vehicle remote sensing. BACKGROUND

[0002] Leaf area index (LAI) is a key parameter for representing crop growth conditions, directly reflecting the photosynthetic capacity and biomass accumulation of vegetation, and is of great significance for agricultural production management and yield estimation. Traditional LAI measurement relies on field measurement (such as LAI-2200 instrument), which is time-consuming and labor-intensive and difficult to achieve dynamic monitoring on a large scale. Remote sensing technology, with its advantages of large-scale and multi-temporal, has become the main means of LAI inversion. However, single data source or band combination is easily disturbed by vegetation coverage, soil background, etc., resulting in limited inversion accuracy.

[0003] In the prior art, there are problems such as inconsistent row and column numbers of multi-band data, low classification accuracy of ground objects, etc. in the processing of unmanned aerial vehicle remote sensing data. In the LAI inversion process, saturation phenomenon is easy to occur in high LAI area, and multi-scale data fusion deviation also affects the inversion accuracy. In addition, the traditional model is difficult to effectively extract and fuse the spectral features and spatial features in multi-source data, resulting in insufficient LAI estimation accuracy and failing to meet the needs of precision agriculture.

[0004] Therefore, it is necessary to provide a winter wheat leaf area index estimation method and system based on unmanned aerial vehicle remote sensing to solve the above technical problems. SUMMARY

[0005] The present application provides a winter wheat leaf area index estimation method and system based on unmanned aerial vehicle remote sensing, which solves the problems of inconsistent row and column numbers in unmanned aerial vehicle remote sensing data processing, inversion saturation in high LAI area, multi-scale data fusion deviation, difficulty in effectively extracting and fusing spectral features and spatial features in multi-source data, resulting in insufficient LAI estimation accuracy and failing to meet the needs of precision agriculture. To solve the above technical problems, the present application provides a winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing, comprising the following steps: S1, data acquisition: acquiring unmanned aerial vehicle multispectral remote sensing image data of the research area, and collecting corresponding field measurement data at different growth stages of winter wheat, the field measurement data including leaf area index (LAI), soil moisture, conductivity, chlorophyll content, and sample site geographical location, soil type, fertilization and irrigation information; S2, data preprocessing: performing stitching, geographic coordinate system embedding and reflectivity conversion processing on the unmanned aerial vehicle multispectral remote sensing image data; screening, entering, missing value processing, and normalization and standardization processing on the field measurement data; S3, vegetation index calculation: according to the pre-processed unmanned aerial vehicle multi-spectral remote sensing image data, calculate the normalized difference vegetation index (NDVI) and the enhanced vegetation index (EVI); S4, data fusion: using convolutional neural network (CNN) to extract multi-scale space and spectral features of the unmanned aerial vehicle multi-spectral remote sensing image data, using multi-layer perceptron (MLP) or feature embedding to encode high-dimensional features of the field measured data, combining wavelet transform, convolution pooling and spatial resampling technology to realize scale conversion and feature alignment of the unmanned aerial vehicle multi-spectral remote sensing image data and the field measured data, obtaining the fusion data through feature-level fusion or decision-level fusion; S5, feature extraction: extracting spectral features and spatial features from the pre-processed unmanned aerial vehicle multi-spectral remote sensing image data, the spectral features including the NDVI and EVI, and the spatial features including the shape and texture of the crops; extracting crop condition features from the pre-processed field measured data; S6, model construction and optimization: constructing a ResNet deep learning model, adjusting hyperparameters using grid search, introducing a channel attention module into the ResNet deep learning model, processing the fusion data to obtain a winter wheat leaf area index estimation model; S7, model verification and estimation: verifying the winter wheat leaf area index estimation model using independent verification set, spatio-temporal cross-validation and uncertainty quantification method, and estimating the winter wheat leaf area index of the research area using the verified winter wheat leaf area index estimation model.

[0006] Preferably, in the data acquisition, the unmanned aerial vehicle multi-spectral remote sensing image data and the field measured data are acquired synchronously, and the different growth periods of the winter wheat include the green-up period, the jointing period, the heading period and the milk stage.

[0007] Preferably, the calculation formula of the normalized difference vegetation index (NDVI) is: , Wherein, ρNIR is the near-infrared band reflectivity, and ρRed is the red band reflectivity.

[0008] Preferably, the calculation formula of the enhanced vegetation index (EVI) is: Wherein, ρBlue is the blue band reflectivity.

[0009] Preferably, the ResNet deep learning model is realized by stacking residual blocks, and the residual blocks are used to solve the problems of gradient vanishing and gradient explosion in deep network training.

[0010] Preferably, the channel attention module is used for weight distribution of spectral channels of the fused data, highlighting the importance of the spectral sensitive band.

[0011] Preferably, the hyperparameters of the grid search include the number of neuron layers, the number of neurons and the batch size.

[0012] Preferably, the precision evaluation index of the model verification includes root mean square error (RMSE) and determination coefficient (R²).

[0013] To solve the above problems, the application further provides a winter wheat leaf area index estimation system based on unmanned aerial vehicle remote sensing, comprising: A data acquisition module is configured to acquire unmanned aerial vehicle multispectral remote sensing image data and corresponding field measured data of a research area. A data preprocessing module is configured to preprocess the unmanned aerial vehicle multispectral remote sensing image data and the field measured data. A vegetation index calculation module is configured to calculate normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI). A data fusion module is configured to fuse the unmanned aerial vehicle multispectral remote sensing image data and the field measured data. A feature extraction module is configured to extract spectral features, spatial features and crop condition features. A model construction and optimization module is configured to construct and optimize a winter wheat leaf area index estimation model. A model verification and estimation module is configured to verify the winter wheat leaf area index estimation model and perform estimation.

[0014] Preferably, the data acquisition module, the data preprocessing module, the vegetation index calculation module, the feature extraction module, the data fusion module, the model construction and optimization module and the model verification and estimation module are sequentially connected.

[0015] Compared with the related art, the winter wheat leaf area index estimation method and system based on unmanned aerial vehicle remote sensing provided by the application have the following beneficial effects: This invention provides a method and system for estimating the leaf area index (LAI) of winter wheat based on UAV remote sensing. Through the design of steps S1 to S7, especially the coordinated cooperation of data preprocessing, data fusion, model construction, and optimization, the system achieves the following: First, in S2, the UAV multispectral remote sensing image data is stitched together and embedded with a geographic coordinate system, automatically aligning the spatial locations of different bands and fundamentally solving the problem of inconsistent row and column numbers in multi-band data. This provides a spatially unified data source for subsequent vegetation index calculation and feature extraction. Second, S3 simultaneously calculates both NDVI and EVI vegetation indices. EVI, by introducing blue band reflectance to suppress soil background interference, is more sensitive to areas with high vegetation cover. Combined with the residual blocks and channel attention module of the ResNet model in S6, the channel attention module can assign weights to the spectral channels of the fused data, highlighting near-infrared and red L... The contribution of AI-sensitive bands effectively alleviates the inversion saturation phenomenon that easily occurs in high LAI regions with traditional single NDVI. Meanwhile, S4 extracts multi-scale spatial and spectral features of UAV images through CNN, uses MLP to encode field measurement data in high dimensions, and combines wavelet transform and convolutional pooling to achieve scale transformation and feature alignment, completely solving the problem of multi-scale data fusion bias. Then, S5 specifically extracts spectral features, spatial features, and agricultural features to achieve deep fusion of multi-source data features, avoiding the information loss caused by traditional models relying on only a single feature. Finally, S6 uses grid search to optimize the hyperparameters of the ResNet model, and S7 ensures the reliability of the model through independent validation sets and spatiotemporal cross-validation, greatly improving the accuracy of LAI estimation. This fundamentally solves the pain points of UAV remote sensing data processing and the problem of multi-source feature fusion, meeting the needs of precision agriculture for high-precision, wide-coverage, and dynamic monitoring of LAI. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a preferred embodiment of a method for estimating the leaf area index of winter wheat based on UAV remote sensing provided by the present invention. Figure 2 The convolutional neural network feature extraction diagram provided by this invention; Figure 3 A schematic diagram of the attention module provided by the present invention; Figure 4 The spatial distribution map of LAI provided by this invention; Figure 5 This is a scatter plot of LAI provided by the present invention; Figure 6 This is a flowchart of a preferred embodiment of a winter wheat leaf area index estimation system based on UAV remote sensing provided by the present invention. Detailed Implementation

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0018] Please refer to Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 And Figure 5 Wherein, Figure 1 is a flow chart of a preferred embodiment of a winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing provided by the application; Figure 2 is a convolutional neural network feature extraction diagram provided by the application; Figure 3 is a schematic diagram of an attention module provided by the application; Figure 4 is a LAI spatial distribution diagram provided by the application; Figure 5 is a LAI scatter diagram provided by the application.

[0019] A winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing, comprising the following steps: S1, data acquisition: acquiring unmanned aerial vehicle multispectral remote sensing image data of a research area, and collecting corresponding field measured data at different growth stages of winter wheat, the field measured data including leaf area index (LAI), soil moisture, conductivity, chlorophyll content, and sample site geographical location, soil type, fertilization and irrigation information; S2, data preprocessing: splicing, geographic coordinate system embedding and reflectivity conversion processing are performed on the unmanned aerial vehicle multispectral remote sensing image data; screening, inputting, missing value processing, and normalization and standardization processing are performed on the field measured data; S3, vegetation index calculation: according to the preprocessed unmanned aerial vehicle multispectral remote sensing image data, normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) are calculated; S4, data fusion: multi-scale spatial and spectral features of the unmanned aerial vehicle multispectral remote sensing image data are extracted using convolutional neural network (CNN), high-dimensional feature coding is performed on the field measured data using multilayer perceptron (MLP) or feature embedding, scale conversion and feature alignment of the unmanned aerial vehicle multispectral remote sensing image data and the field measured data are realized by combining wavelet transform, convolution pooling and spatial resampling technology, and fusion data is obtained through feature-level fusion or decision-level fusion; S5, feature extraction: spectral features and spatial features are extracted from the preprocessed unmanned aerial vehicle multispectral remote sensing image data, the spectral features including the NDVI and EVI, and the spatial features including the shape and texture of the crops; crop condition features are extracted from the preprocessed field measured data; S6, model construction and optimization: a ResNet deep learning model is constructed, hyperparameters are adjusted by grid search, a channel attention module is introduced into the ResNet deep learning model, the fusion data is processed to obtain a winter wheat leaf area index estimation model; S7, model verification and estimation: the winter wheat leaf area index estimation model is verified using an independent verification set, a spatio-temporal cross-validation method and an uncertainty quantification method, and the winter wheat leaf area index of the study area is estimated using the verified winter wheat leaf area index estimation model.

[0020] In the data acquisition, the unmanned aerial vehicle multispectral remote sensing image data and the field measured data are acquired synchronously, and the different growth stages of winter wheat include the green return period, the jointing period, the heading period and the milk ripening period.

[0021] The calculation formula of the normalized vegetation index (NDVI) is: Wherein, ρNIR is the near-infrared band reflectivity, and ρRed is the red band reflectivity.

[0022] The calculation formula of the enhanced vegetation index (EVI) is: Wherein, ρBlue is the blue band reflectivity.

[0023] The ResNet deep learning model is realized by stacking residual blocks, and the residual blocks are used to solve the problems of gradient disappearance and gradient explosion in deep network training.

[0024] The channel attention module is used to allocate weights to the spectral channels of the fusion data, highlighting the importance of spectral sensitive bands.

[0025] The hyperparameters of the grid search include the number of neuron layers, the number of neurons and the batch size.

[0026] The precision evaluation index of the model verification includes the root mean square error (RMSE) and the determination coefficient (R²).

[0027] Research area profile: the research area is Guangrao County, Dongying City, Shandong Province, located in the Yellow River Delta plain, with geographical coordinates of east longitude 118°53' and north latitude 37°21'. The region belongs to warm and semi-humid monsoon climate, with annual precipitation of 580-650mm, humidity of 41.6%, soil mainly being moist brown soil and salinized moist soil, flat terrain, suitable for winter wheat growth. The place is an important winter wheat production area, suitable for being a typical research area for winter wheat LAI inversion, with LAI ranging from 0.5 to 6.0m² / m², covering different growth stages of winter wheat; Data acquisition: ground measured data:​ In the study area, select representative winter wheat sample plots, at different growth stages of winter wheat (anti-clear period, jointing stage, heading stage, milk stage), using LAI-2200 leaf area meter, SPAD instrument and other professional instruments, for different growth conditions of plants, leaf area index (LAI), soil moisture, electrical conductivity and chlorophyll content and other key agricultural parameters were measured and averaged; Simultaneously record the geographical location of the sample plot, soil type, fertilization and irrigation information and other environmental factors, as the ground truth value of model calibration and verification.

[0028] In order to clearly present the data processing flow, the raw data obtained is as follows: After preliminary screening, Excel input and normalization and standardization pretreatment, high quality support is provided for remote sensing inversion of LAI and winter wheat growth parameters, and subsequent model construction and verification will be based on these processed data, to ensure the reliability of the inversion results.

[0029] Unmanned aerial vehicle remote sensing data: Using unmanned aerial vehicle equipped with multispectral sensor, according to the growth cycle of winter wheat and weather conditions, flexible arrangement of flight task, and synchronous acquisition of high resolution multispectral image with ground sampling.

[0030] Field measured data preprocessing: when processing field measured data, first, the data is screened, and the data with large measurement error is removed, to ensure the data quality, then, the data is input into Excel or database according to unified format, to ensure clear field and unified unit, for the samples with more missing values, can be considered to be removed, or use mean interpolation method to complete, finally, all data is normalized and standardized, to facilitate subsequent registration with remote sensing data and model analysis, the whole process emphasizes the standardization of measurement and the spatiotemporal consistency of data, to ensure that field measured data can provide high quality support for remote sensing inversion of LAI and winter wheat growth parameters; Unmanned aerial vehicle data preprocessing: this study establishes a complete unmanned aerial vehicle image processing technology flow, through multi-link collaborative processing to realize data optimization. First, the data image is spliced, the original data is converted to GeoTIFF format and embedded in WGS84 geographic coordinate system, and the DN value is converted to reflectivity; Vegetation index calculation: the following key vegetation indices are used: Normalized difference vegetation index (NDVI): , Where ρNIR is the near-infrared band reflectivity, and ρRed is the red band reflectivity; Enhanced vegetation index (EVI): , where pBlue is the reflectance of the blue band; Data fusion: For remote sensing image data, multi-scale spatial and spectral features are automatically extracted using convolutional neural networks (CNN). Meanwhile, the representation of crop canopy structure and physiological characteristics is enriched through band combination, vegetation index, and texture features. For tabular ground-truth data, multi-layer perceptron (MLP) or feature embedding is used for high-dimensional feature encoding. To address the scale difference between UAV and field measurement data, techniques such as wavelet transform, convolution pooling, and spatial resampling are used to achieve scale conversion and feature alignment, ensuring spatial consistency for subsequent fusion. In the design of multi-modal deep fusion models, a "feature-level fusion + dynamic attention mechanism" framework is adopted. Each modality data is first input into an independent encoding network for feature extraction. Then, in the fusion layer, the feature vectors are spliced or weighted integrated, and an attention mechanism is introduced to adaptively focus on key growth periods and spatial heterogeneity areas, effectively improving the model's sensitivity to LAI dynamic changes. In some scenarios, decision-level fusion can also be used, where each modality sub-network is trained separately and the output is integrated through weighted averaging, stacking, or adaptive fusion layers to further improve model robustness. Feature extraction: From the fused multi-modal data, rich features are extracted. On the one hand, traditional spectral indices such as NDVI and EVI are extracted to reflect the vegetation growth status of winter wheat canopy. On the other hand, inspired by deep learning methods, convolutional neural networks are used to extract features from multispectral images. Each band data of the multispectral image is taken as input Figure 2 , and through the sliding of convolution kernels on the image, spatial features (crop shape, texture) within different bands and spectral features (reflectance differences between different bands) across bands are extracted. Through downsampling, the spatial dimension of the feature map is reduced, reducing computational complexity while preserving the most significant features. By stacking multiple convolution and pooling layers, deep-level features are gradually extracted. A multi-dimensional feature space is constructed to provide comprehensive data support for subsequent model construction.

[0031] Model construction and optimization: Hyperparameter adjustment using network search: To improve model accuracy, the best hyperparameters of the model are found using grid search in the hyperparameter space. The value range of the hyperparameters is divided into discrete grid points, where each grid corresponds to a hyperparameter, such as the number of neurons, the number of neurons, and the batch size. All possible parameter combinations are exhaustively searched, and the model is trained and evaluated one by one to select the optimal combination: ResNet model is used to extract image features: ResNet solves the problems of gradient disappearance and gradient explosion in deep network training by using residual blocks (Residual Blocks), so it can train very deep neural networks. Deep ResNet network is adopted, which is realized by stacking residual blocks. Multi-level feature extraction: shallow residual blocks capture basic features such as leaf edge and light spot, and deep residual blocks extract abstract features such as crown structure and growth period dynamics. Cross-scale information fusion: Through residual connection, the features of different levels are integrated, and the adaptability to multi-scale data is enhanced. Design attention module to allocate weights to channel features to improve the expression ability of key features. Traditional convolutional neural networks treat all channel features equally, which makes it difficult to highlight the importance of spectral sensitive bands (such as near-infrared bands). This project introduces a channel attention module (Channel Attention Module) to learn the importance weights of each spectral channel, suppress invalid noise, and strengthen key features, making important details in the image more prominent. Deep fusion model based on multi-head attention mechanism: Figure 3 The attention module structure is shown in the figure: first, for the input feature map, three groups of 1x1 convolution are used for channel compression, and the channel dimension is kept. The width and height are expanded to one dimension, second, the feature map of branch f(x) is transposed and multiplied with the feature map of branch g(x), and the result is normalized by softmax, finally, the attention matrix after normalization output and the feature map obtained by branch are multiplied, according to the similarity, the weight of each channel is redistributed, then the softmax function is used for processing, and the channel number is expanded to the same channel number as the input feature map by 1x1 convolution. The output feature map processed in this way can provide more sufficient expression on key detail features, so as to realize attention re-allocation.

[0032] Model verification and estimation: Verification method and process: independent verification set, using "field measurement + unmanned aerial vehicle" data not involved in training, to evaluate the generalization ability of the model in different environments, spatio-temporal cross-validation: dividing data by growth period (such as seedling stage, heading stage, mature stage) and spatial region (such as different farmland plots), testing the spatio-temporal stability of the model, uncertainty quantification: using deep learning method, uncertainty analysis is carried out on the prediction results, and confidence evaluation (such as confidence interval) is output to assist agricultural decision-making; Precision evaluation index: adopt RMSE (root mean squared error), R2 (determination coefficient) and other indexes, so that R2> 0.8, RMSE is about 1, compare the predicted value of the model with the measured LAI value to ensure that the inversion accuracy is greater than 80%; the root mean squared error (RMSE) is an index for measuring the prediction accuracy of the prediction model on continuous data. It measures the root mean squared difference between the predicted value and the true value, indicating the average deviation between the predicted value and the true value, which is one of the commonly used performance evaluation indexes in regression tasks , R2 (R square) score, also known as determination coefficient, is an index for evaluating the performance of a regression model, which represents the proportion of the dependent variable variation explained by the independent variable. The value of R2 score is usually between 0 and 1, and the closer the value is to 1, the better the model fitting effect .

[0033] Experimental results and analysis: Band combination effectiveness analysis: the inversion result of the optimal combination of "near-infrared + red band" shows that: For the area with LAI = 0.5-6 m2 / m2, the synergistic effect of NDVI and EVI makes R2 reach 0.85; Inversion result: the ResNet model is used to generate the LAI spatial distribution map of the research area ( Figure 4 ) and the LAI inversion scatter diagram ( Figure 5 ), Figure 4 showing the change of LAI of winter wheat from the green period to the jointing period to the heading period to the milk ripening period, Figure 4 It can be seen that the LAI of winter wheat presents a trend of first increasing and then decreasing, Figure 5 The scatter diagram of LAI inversion shows that Figure 5 It can be seen that the inversion result has good consistency compared with the measured value; By screening "near-infrared + red band" as the optimal band combination (simplifying the dependence on red edge band), cooperating with normalized difference vegetation index and enhanced vegetation index, using the sensitivity of NDVI to vegetation coverage and the anti-saturation of EVI to high biomass area, the saturation problem of high LAI area inversion is effectively alleviated, the model performs best, the verification set index reaches, and the model is suitable for large-scale LAI dynamic monitoring demand. At present, the inversion result of the model is underestimated, and the model parameters need to be further optimized in the future.

[0034] Compared with the related art, the winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing provided by the present application has the following beneficial effects: Through the step design of S1 to S7, especially the cooperative matching of data preprocessing, data fusion, model construction and optimization, when in use, on the one hand, the unmanned aerial vehicle multi-spectral remote sensing image data in S2 is spliced and embedded in the geographic coordinate system, which can automatically align the spatial positions of different bands, thus solving the problem of inconsistent row and column numbers of multi-band data from the root, and providing a spatially unified data source for subsequent vegetation index calculation and feature extraction; on the other hand, S3 simultaneously calculates NDVI and EVI two vegetation indexes, EVI introduces blue band reflectivity to suppress soil background interference, and has higher sensitivity to high vegetation coverage area, and cooperates with the residual block and channel attention module of the ResNet model in S6, the channel attention module can allocate weights to the spectral channels of the fused data, highlight the contribution of near-infrared, red and other LAI sensitive bands, and effectively alleviate the saturation phenomenon of traditional single NDVI in high LAI area; at the same time, S4 extracts multi-scale spatial and spectral features of unmanned aerial vehicle image through CNN, encodes field measured data with high dimension through MLP, realizes scale conversion and feature alignment through wavelet transform and convolution pooling, and completely solves the fusion deviation problem of multi-scale data, then S5 extracts spectral features, spatial features and crop condition features, realizes deep fusion of multi-source data features, and avoids information loss caused by traditional model relying on single feature; finally, S6 optimizes the ResNet model hyperparameters through grid search, and S7 ensures the reliability of the model through independent validation set and space-time cross-validation, greatly improves the LAI estimation accuracy, fundamentally solves the problems of unmanned aerial vehicle remote sensing data processing and multi-source feature fusion, and meets the demand of precision agriculture for high-precision, wide-coverage and dynamic monitoring of LAI.

[0035] Please refer to Figure 6 To solve the above problems, the application also provides a winter wheat leaf area index estimation system based on unmanned aerial vehicle remote sensing, comprising: A data acquisition module is used to acquire unmanned aerial vehicle multi-spectral remote sensing image data and corresponding field measured data of a research area. A data preprocessing module is used to preprocess the unmanned aerial vehicle multi-spectral remote sensing image data and field measured data. A vegetation index calculation module is used to calculate normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI). A data fusion module is used to fuse the unmanned aerial vehicle multi-spectral remote sensing image data and field measured data. A feature extraction module is used to extract spectral features, spatial features and crop condition features. A model construction and optimization module is used to construct and optimize a winter wheat leaf area index estimation model. A model verification and estimation module is used to verify the winter wheat leaf area index estimation model and perform estimation.

[0036] The data acquisition module, the data preprocessing module, the vegetation index calculation module, the feature extraction module, the data fusion module, the model construction and optimization module, and the model verification and estimation module are sequentially connected.

[0037] The above is only an embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation using the content of the specification and drawings of the present application, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for estimating winter wheat leaf area index based on unmanned aerial vehicle remote sensing, characterized in that, The method comprises the following steps: S1, data acquisition: acquiring unmanned aerial vehicle multi-spectral remote sensing image data of a research area, and collecting corresponding field measured data at different growth stages of winter wheat, wherein the field measured data includes leaf area index (LAI), soil moisture, conductivity, chlorophyll content, and sample plot geographical location, soil type, fertilization and irrigation information; S2, data preprocessing: performing splicing, geographical coordinate system embedding and reflectivity conversion processing on the unmanned aerial vehicle multi-spectral remote sensing image data; performing screening, inputting, missing value processing, and normalization and standardization processing on the field measured data; S3, vegetation index calculation: calculating normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) according to the preprocessed unmanned aerial vehicle multi-spectral remote sensing image data; S4, data fusion: extracting multi-scale spatial and spectral features of the unmanned aerial vehicle multi-spectral remote sensing image data by using a convolutional neural network (CNN), performing high-dimensional feature coding on the field measured data by using a multi-layer perceptron (MLP) or feature embedding, combining wavelet transform, convolution pooling and spatial resampling technology to realize scale conversion and feature alignment of the unmanned aerial vehicle multi-spectral remote sensing image data and the field measured data, and obtaining fused data through feature-level fusion or decision-level fusion; S5, feature extraction: extracting spectral features and spatial features from the preprocessed unmanned aerial vehicle multi-spectral remote sensing image data, wherein the spectral features include the NDVI and EVI, and the spatial features include the shape and texture of the crops; extracting crop condition features from the preprocessed field measured data; S6, model construction and optimization: constructing a ResNet deep learning model, adjusting hyperparameters by using a grid search, introducing a channel attention module into the ResNet deep learning model, processing the fused data, and obtaining a winter wheat leaf area index estimation model; S7, model verification and estimation: verifying the winter wheat leaf area index estimation model by using an independent verification set, spatio-temporal cross-validation and uncertainty quantification method, and estimating the winter wheat leaf area index of the research area by using the verified winter wheat leaf area index estimation model. 2.The winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing of claim 1, wherein, In the data acquisition, the unmanned aerial vehicle multi-spectral remote sensing image data and the field measured data are acquired synchronously, and the different growth stages of the winter wheat include the green-up stage, the jointing stage, the heading stage and the milk stage. 3.The winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing of claim 1, wherein, The calculation formula of the normalized difference vegetation index (NDVI) is: ， wherein, ρNIR is the near-infrared band reflectivity, and ρRed is the red band reflectivity. 4.The winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing of claim 1, wherein, The calculation formula of the enhanced vegetation index (EVI) is: where pBlue is the blue waveband reflectance. 5.The winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing of claim 1, wherein, The ResNet deep learning model is realized by stacking residual blocks, and the residual blocks are used to solve the gradient vanishing and gradient explosion problems in deep network training. 6.The winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing of claim 1, wherein, The channel attention module is used to allocate weights to the spectral channels of the fused data, and highlight the importance of the spectral sensitive bands.

7. The method of claim 1, wherein the method comprises the steps of: a) obtaining the UAV images of the winter wheat; b) calculating the NDVI of the winter wheat; c) calculating the LAI of the winter wheat; and d) calculating the LAI of the winter wheat using the UAV images. The hyperparameters of the grid search include the number of neuron layers, the number of neurons and the batch size. 8.The winter wheat leaf area index estimation method based on unmanned aerial vehicle remote sensing of claim 1, wherein, The precision evaluation indexes of the model verification include the root mean square error (RMSE) and the determination coefficient (R²). 9.A system for estimating winter wheat leaf area index based on unmanned aerial vehicle remote sensing, characterized in that, The method comprises the following steps: The data acquisition module is configured to acquire unmanned aerial vehicle multi-spectral remote sensing image data and corresponding field measured data of a research area. The data preprocessing module is configured to preprocess the unmanned aerial vehicle multi-spectral remote sensing image data and the field measured data. The vegetation index calculation module is configured to calculate normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI). The data fusion module is configured to fuse the unmanned aerial vehicle multi-spectral remote sensing image data and the field measured data. The feature extraction module is configured to extract spectral features, spatial features and crop condition features. The model construction and optimization module is configured to construct and optimize a winter wheat leaf area index estimation model. The model verification and estimation module is configured to verify the winter wheat leaf area index estimation model and perform estimation. 10.The winter wheat leaf area index estimation system based on UAV remote sensing of claim 9, wherein, The data acquisition module, the data preprocessing module, the vegetation index calculation module, the feature extraction module, the data fusion module, the model construction and optimization module and the model verification and estimation module are sequentially connected.

Images