Winter wheat yield evaluation method and system based on machine learning
Through one-dimensional time convolution and cross-modal Transformer fusion of multi-source data, combined with principal component analysis and improved CNN-BiGRU-Attention model, the problem of insufficient utilization of multi-source data correlation and time series information in traditional methods is solved, and a more accurate and generalized winter wheat yield evaluation is achieved.
Patent Information
- Application Number
- CN202510588170.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-08
- Publication Date
- 2025-08-19
AI Technical Summary
The traditional winter wheat yield evaluation method ignores the correlation and time series information between multi-source data, making it difficult to fully utilize the dynamic changes in winter wheat growth process and has poor generalization ability.
One-dimensional time convolution and cross-modal Transformer were used for multi-source data fusion, key features were extracted in combination with principal component analysis, and yield evaluation was performed using the improved CNN-BiGRU-Attention model.
It improves the accuracy of yield evaluation and the generalization ability of the model, can adapt to data changes at different time points and fields, reduces redundant calculations, and improves training efficiency.
Smart Images
Figure CN120508772A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of grain yield assessment, and in particular to a winter wheat yield assessment method and system based on machine learning. Background Art
[0002] Traditional winter wheat yield assessment methods often consider only a single data source (such as meteorological or historical yield data), ignoring the interdependencies between multiple data sources (e.g., imagery, meteorology, soil, and agronomic practices). The formats, dimensions, and temporal resolutions of these different data sources vary significantly, making direct fusion difficult. During data fusion, traditional methods often overlook the importance of time series information, failing to fully utilize the dynamic characteristics of winter wheat growth.
[0003] Traditional models often have poor generalization capabilities when dealing with complex winter wheat growing environments and multi-source data, and are difficult to adapt to different fields and years. Summary of the Invention
[0004] In order to solve the above problems, the present invention proposes a winter wheat yield assessment method and system based on machine learning, which specifically includes:
[0005] Step S1. Collect winter wheat images, meteorological data, soil conditions, agronomic measures, and historical yield data of the area to be measured to generate multi-source fusion data;
[0006] Step S2. extracting key features from the multi-source fusion data using principal component analysis to form a multi-dimensional feature matrix;
[0007] Step S3. extracting a time series multi-feature image based on the multi-dimensional feature matrix;
[0008] Step S4. Input the time series multi-feature image into the improved CNN-BiGRU-Attention model to obtain the winter wheat yield evaluation result.
[0009] Optionally, the process of generating multi-source fusion data specifically includes:
[0010] Performing one-dimensional temporal convolution on the winter wheat image, meteorological data, soil conditions, agronomic measures, and historical yield data to obtain wheat data features with time information and adding position codes to the wheat data features;
[0011] The cross-modal Transformer is used to perform multi-scale and multi-modal information fusion on the wheat data features with added position encoding to obtain multi-source fused data.
[0012] Optionally, the time convolution method is:
[0013]
[0014] Among them, X a represents the data of multimodal input sequence, k a Indicates the size of the convolution kernel under the corresponding mode a, Conv1D represents one-dimensional time convolution, and d represents the feature dimension.
[0015] Optionally, in step S4, the content of the improved CNN-BiGRU-Attention model specifically includes:
[0016] The temporal multi-feature image is input into the convolution layer for convolution and then pooled, and the pooled feature map is input into the fully connected layer to obtain a one-dimensional feature vector;
[0017] The one-dimensional feature vector is input into the BiGRU layer and the Attention layer to obtain output data.
[0018] The present invention also provides a winter wheat yield assessment system based on machine learning, the system comprising:
[0019] The data fusion module is used to collect winter wheat images, meteorological data, soil conditions, agronomic measures and historical yield data of the test area to generate multi-source fused data;
[0020] A feature extraction module is used to extract key features from the multi-source fusion data using principal component analysis to form a multi-dimensional feature matrix;
[0021] A time series feature extraction module, configured to extract a time series multi-feature image based on the multi-dimensional feature matrix;
[0022] The yield assessment module is used to input the time series multi-feature image into the improved CNN-BiGRU-Attent ion model to obtain the winter wheat yield assessment result.
[0023] Optionally, the process of generating multi-source fusion data specifically includes:
[0024] Performing one-dimensional temporal convolution on the winter wheat image, meteorological data, soil conditions, agronomic measures, and historical yield data to obtain wheat data features with time information and adding position codes to the wheat data features;
[0025] The cross-modal Transformer is used to perform multi-scale and multi-modal information fusion on the wheat data features with added position encoding to obtain multi-source fused data.
[0026] Optionally, the time convolution method is:
[0027]
[0028] Among them, Xa represents the data of multimodal input sequence, k a Indicates the size of the convolution kernel under the corresponding mode a, Conv1D represents one-dimensional time convolution, and d represents the feature dimension.
[0029] Optionally, the improved CNN-BiGRU-Attention model specifically includes:
[0030] The temporal multi-feature image is input into the convolution layer for convolution and then pooled, and the pooled feature map is input into the fully connected layer to obtain a one-dimensional feature vector;
[0031] The one-dimensional feature vector is input into the BiGRU layer and the Attention layer to obtain output data.
[0032] Compared with the prior art, the present invention has the following beneficial effects:
[0033] The present invention fuses multi-source data through one-dimensional temporal convolution and cross-modal Transformer, which can make full use of various information during the growth process of winter wheat and improve the accuracy of yield assessment.
[0034] The one-dimensional temporal convolution and BiGRU layer of the present invention can effectively capture the time series information during the growth process of winter wheat, enable the model to adapt to data changes at different time points, and enhance the generalization ability of the model.
[0035] The present invention has a reasonable structure through the improved CNN-BiGRU-Attention model, which can fully utilize spatial and temporal information, reduce redundant calculations, and further improve training efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0037] Figure 1 This is a flow chart of a method for winter wheat yield assessment based on machine learning according to an embodiment of the present invention;
[0038] Figure 2 is a schematic structural diagram of an electronic device according to an embodiment of the present invention;
[0039] Figure 3 Schematic diagram of the structure of the improved CNN-BiGRU-Attention model. DETAILED DESCRIPTION
[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0041] Example 1
[0042] A winter wheat yield assessment method based on machine learning, such as Figure 1 As shown, the method includes:
[0043] Step S1: Collect winter wheat images, meteorological data, soil conditions, agronomic measures and historical yield data of the area to be measured to generate multi-source fusion data.
[0044] During the winter wheat growing cycle, high-resolution satellite remote sensing technology, low-altitude drone remote sensing platforms, and field monitoring equipment were used to collect multispectral, hyperspectral, and visible light images of the winter wheat in the surveyed areas. These images provide information on the wheat's growth status, leaf area index, and vegetation cover. Meteorological monitoring stations were also used to obtain meteorological data for the region, including temperature, precipitation, daylight duration, wind speed, and relative humidity. These meteorological factors have a significant impact on the growth and development of winter wheat. Soil sampling and analysis were performed to determine soil conditions such as texture, organic matter content, pH, and nutrient content, such as nitrogen, phosphorus, and potassium, as these conditions directly affect root growth and nutrient absorption. Furthermore, data on agronomic practices in the region were collected, such as sowing density, fertilization rate, irrigation frequency, and pest and disease control measures, as these practices can alter the growing environment and growth patterns of winter wheat. Finally, historical yield data for the region were compiled to provide a reference for yield assessment. The collected winter wheat image data, meteorological data, soil condition data, agronomic practice data, and historical yield data are standardized to eliminate dimensional differences and inconsistent data formats between different data sources, and then fused together. The process of generating multi-source fused data specifically includes:
[0045] Performing one-dimensional temporal convolution on the winter wheat image, meteorological data, soil conditions, agronomic measures, and historical yield data to obtain wheat data features with time information and adding position codes to the wheat data features;
[0046] The cross-modal Transformer is used to perform multi-scale and multi-modal information fusion on the wheat data features with added position encoding to obtain multi-source fused data.
[0047] The temporal convolution method is:
[0048]
[0049] Among them, X a represents the data of multimodal input sequence, ka Indicates the size of the convolution kernel under the corresponding mode a, Conv1D represents one-dimensional time convolution, and d represents the feature dimension.
[0050] Multimodal fusion uses multi-head aggregation to aggregate directed pairwise cross-modal interactions between the target modality and the source modality:
[0051]
[0052] in, and is the weight parameter. When modality α constructs its i-th layer cross-module interaction from modality β, H [i] represents the set of multi-scale cross-modal interactions between mode α and mode β, specifically:
[0053]
[0054] in, It is the low-level feature of modality β, which is finally fused and outputted through the feedforward layer.
[0055] Step S2. Principal component analysis is used to extract key features from the multi-source fusion data and form a multidimensional feature matrix. Principal component analysis eliminates minor variables through linear combinations, focusing on core influencing factors and significantly improving subsequent modeling efficiency. Matrix construction techniques are used to transform the key feature set into a multidimensional feature matrix, resulting in a multidimensional feature matrix.
[0056] In this example, a matrix with 100 rows (number of plots) x 3 columns (precipitation, fertilizer application, and organic matter) was constructed, with each row representing the eigenvalue of a plot. This structured representation facilitates computation and visualization. If redundant dimensions exist in the matrix, such as a high correlation between fertilizer application and rainfall, dimensionality reduction techniques can be used to remove them.
[0057] For example, if correlation analysis reveals a correlation coefficient of 0.9, rainfall is retained and fertilizer application is removed, resulting in an optimized feature matrix. This dimensionality reduction reduces computational complexity and avoids overfitting. Cluster analysis is used to group the data based on the distribution characteristics of each dimension in the optimized feature matrix to obtain the grouping results.
[0058] Step S3: extracting a time series multi-feature image based on the multi-dimensional feature matrix.
[0059] Principal component analysis (PCA) was used to extract key features from the multi-source fused data, resulting in a multidimensional feature matrix F. This matrix, containing feature information after dimensionality reduction, effectively reduces data redundancy while retaining the most valuable information for winter wheat yield assessment. Next, we need to extract a time-series multi-feature image from this multidimensional feature matrix to better utilize this time series information for winter wheat yield assessment.
[0060] Assume that the dimensions of the multidimensional feature matrix F are m×k, where m represents the number of samples (usually corresponding to different time points or different plots of land), and k represents the number of features after PCA dimensionality reduction. To extract temporal information, we need to reorganize these features into chronological order.
[0061] Assume that the growth cycle of winter wheat is divided into T time points, and each time point has N field observation data. Therefore, the multidimensional feature matrix F can be reorganized into a three-dimensional matrix F 3D , whose dimension is T×N×k. Specifically, F 3D [t,n,:] represents the k eigenvalues at time point t and field n.
[0062] In order to transform the time series feature matrix F 3D Convert to image form, using the following method: Time series expansion: expand the feature value of each time point into a row of the image, and each feature corresponds to a column of the image. In this way, the feature matrix F at each time point is t (dimension N×k) can be regarded as a row of the image. Therefore, the entire temporal multi-feature image I T The dimension is T×(N×k).
[0063] Specifically, the temporal multi-feature image I T The generation formula can be expressed as:
[0064] I T =reshape(F 3D ,T,N×k)
[0065] Among them, reshape means to reshape the three-dimensional matrix F 3D Reorganize into a two-dimensional matrix I T .
[0066] The specific operation is as follows: For each time point t (t = 1, 2, ..., T): extract F 3D All the eigenvalues of the t-th time point in F t (Dimension is N×k).
[0067] F tExpand to a row, that is, I[t,:]=flatten(Ft), where flatten means expanding the matrix into a one-dimensional vector.
[0068] The final temporal multi-feature image I T The dimension is T×(N×k), where each row represents the feature information of a time point, and each column represents the value of a feature on different plots.
[0069] 3. Visualization and interpretation of time series multi-feature images T In the graph, the horizontal axis represents the time series (from t = 1 to t = T), and the vertical axis represents the expanded form of the features (the value of each feature in different plots). This graphical form can intuitively show the changes in winter wheat characteristics at different time points, while preserving the continuity of the time series and the correlation between features.
[0070] For example, assuming N = 5 (5 plots), k = 3 (3 features), the feature matrix F at each time point is t The dimension is 5×3. After expanding it into a row, the temporal multi-feature image I T Each row of will contain 5×3=15 pixel values, corresponding to the 3 feature values on 5 fields. T From the images, we can find the changing trends of the characteristic values at different time points and the differences between different plots.
[0071] Fusion of spatial and temporal information: Temporal multi-feature images not only retain the time series information, but also fuse the feature information of different fields in the form of images, enabling the model to simultaneously learn the temporal and spatial feature associations.
[0072] Facilitates the input of deep learning models: Image data can be directly input into deep learning models such as convolutional neural networks (CNNs), using the convolutional layers of CNNs to extract local features, thereby better capturing the complex relationships between temporal features.
[0073] Intuitive visualization: Time-series multi-feature images can intuitively display the characteristic changes during the growth process of winter wheat, making it easier for researchers to analyze and interpret.
[0074] Step S4. Input the time series multi-feature image into the improved CNN-BiGRU-Attention model to obtain the winter wheat yield evaluation result.
[0075] like Figure 3 As shown in Figure 2, the contents of the improved CNN-BiGRU-Attention model specifically include:
[0076] The temporal multi-feature image is input into the convolution layer for convolution and then pooled, and the pooled feature map is input into the fully connected layer to obtain a one-dimensional feature vector;
[0077] The one-dimensional feature vector is input into the BiGRU layer and the Attention layer to obtain output data.
[0078] The CNN-BiGRU-Attention yield estimation model is a complex neural network structure consisting of an input layer, a CNN layer, a BiGRU layer, an Attention layer, and an output layer. In the CNN layer, the model uses multiple convolution operations to deeply mine feature information from historical input data, increasing the depth of the features. A pooling layer then reduces the dimensionality of the features to extract key information. Finally, a fully connected layer converts the extracted feature vectors into a one-dimensional form, resulting in a low-dimensional, deep feature vector. The BiGRU and Attention layers work together to learn the data variation patterns in the low-dimensional, deep feature vectors, capturing dependencies in the time series and assigning different attention weights to different features. Finally, the model outputs a predicted value through the output layer, completing the entire yield estimation process.
[0079] Example 2
[0080] A winter wheat yield assessment system based on machine learning, the system comprising:
[0081] The data fusion module is used to collect winter wheat images, meteorological data, soil conditions, agronomic measures and historical yield data of the test area to generate multi-source fusion data.
[0082] During the winter wheat growing cycle, high-resolution satellite remote sensing technology, low-altitude drone remote sensing platforms, and field monitoring equipment were used to collect multispectral, hyperspectral, and visible light images of the winter wheat in the surveyed areas. These images provide information on the wheat's growth status, leaf area index, and vegetation cover. Meteorological monitoring stations were also used to obtain meteorological data for the region, including temperature, precipitation, daylight duration, wind speed, and relative humidity. These meteorological factors have a significant impact on the growth and development of winter wheat. Soil sampling and analysis were performed to determine soil conditions such as texture, organic matter content, pH, and nutrient content, such as nitrogen, phosphorus, and potassium, as these conditions directly affect root growth and nutrient absorption. Furthermore, data on agronomic practices in the region were collected, such as sowing density, fertilization rate, irrigation frequency, and pest and disease control measures, as these practices can alter the growing environment and growth patterns of winter wheat. Finally, historical yield data for the region were compiled to provide a reference for yield assessment. The collected winter wheat image data, meteorological data, soil condition data, agronomic practice data, and historical yield data are standardized to eliminate dimensional differences and inconsistent data formats between different data sources, and then fused together. The process of generating multi-source fused data specifically includes:
[0083] Performing one-dimensional temporal convolution on the winter wheat image, meteorological data, soil conditions, agronomic measures, and historical yield data to obtain wheat data features with time information and adding position codes to the wheat data features;
[0084] The cross-modal Transformer is used to perform multi-scale and multi-modal information fusion on the wheat data features with added position encoding to obtain multi-source fused data.
[0085] The temporal convolution method is:
[0086]
[0087] Among them, X a represents the data of multimodal input sequence, k a Indicates the size of the convolution kernel under the corresponding mode a, Conv1D represents one-dimensional time convolution, and d represents the feature dimension.
[0088] Multimodal fusion uses multi-head aggregation to aggregate directed pairwise cross-modal interactions between the target modality and the source modality:
[0089]
[0090] in, and is the weight parameter. When modality α constructs its i-th layer cross-module interaction from modality β, H [i]represents the set of multi-scale cross-modal interactions between mode α and mode β, specifically:
[0091]
[0092] in, It is the low-level feature of modality β, which is finally fused and outputted through the feedforward layer.
[0093] The feature extraction module is used to extract key features from the multi-source fusion data using principal component analysis to form a multi-dimensional feature matrix.
[0094] Principal component analysis eliminates minor variables through linear combination, focusing on core influencing factors, which can significantly improve the efficiency of subsequent modeling. Through matrix construction technology, the key feature set is converted into a multidimensional feature matrix to obtain a multidimensional feature matrix.
[0095] In this example, a matrix with 100 rows (number of plots) x 3 columns (precipitation, fertilizer application, and organic matter) was constructed, with each row representing the eigenvalue of a plot. This structured representation facilitates computation and visualization. If redundant dimensions exist in the matrix, such as a high correlation between fertilizer application and rainfall, dimensionality reduction techniques can be used to remove them.
[0096] For example, if correlation analysis reveals a correlation coefficient of 0.9, rainfall is retained and fertilizer application is removed, resulting in an optimized feature matrix. This dimensionality reduction reduces computational complexity and avoids overfitting. Cluster analysis is used to group the data based on the distribution characteristics of each dimension in the optimized feature matrix to obtain the grouping results.
[0097] The time series feature extraction module is used to extract a time series multi-feature image based on the multi-dimensional feature matrix.
[0098] Principal component analysis (PCA) was used to extract key features from the multi-source fused data, resulting in a multidimensional feature matrix F. This matrix, containing feature information after dimensionality reduction, effectively reduces data redundancy while retaining the most valuable information for winter wheat yield assessment. Next, we need to extract a time-series multi-feature image from this multidimensional feature matrix to better utilize this time series information for winter wheat yield assessment.
[0099] Assume that the dimensions of the multidimensional feature matrix F are m×k, where m represents the number of samples (usually corresponding to different time points or different plots of land), and k represents the number of features after PCA dimensionality reduction. To extract temporal information, we need to reorganize these features into chronological order.
[0100] Assume that the growth cycle of winter wheat is divided into T time points, and each time point has N field observation data. Therefore, the multidimensional feature matrix F can be reorganized into a three-dimensional matrix F 3D , whose dimension is T×N×k. Specifically, F 3D [t,n,:] represents the k eigenvalues at time point t and field n.
[0101] In order to transform the time series feature matrix F 3D Convert to image form, using the following method: Time series expansion: expand the feature value of each time point into a row of the image, and each feature corresponds to a column of the image. In this way, the feature matrix F at each time point is t (dimension N×k) can be regarded as a row of the image. Therefore, the entire temporal multi-feature image I T The dimension is T×(N×k).
[0102] Specifically, the temporal multi-feature image I T The generation formula can be expressed as:
[0103] I T =reshape(F 3D ,T,N×k)
[0104] Among them, reshape means to reshape the three-dimensional matrix F 3D Reorganize into a two-dimensional matrix I T .
[0105] The specific operation is as follows: For each time point t (t = 1, 2, ..., T): extract F 3D All the eigenvalues of the t-th time point in F t (Dimension is N×k).
[0106] F t Expand to a row, that is, I[t,:]=flatten(Ft), where flatten means expanding the matrix into a one-dimensional vector.
[0107] The final temporal multi-feature image I T The dimension is T×(N×k), where each row represents the feature information of a time point, and each column represents the value of a feature on different plots.
[0108] 3. Visualization and interpretation of time series multi-feature images T In the graph, the horizontal axis represents the time series (from t = 1 to t = T), and the vertical axis represents the expanded form of the features (the value of each feature in different plots). This graphical form can intuitively show the changes in winter wheat characteristics at different time points, while preserving the continuity of the time series and the correlation between features.
[0109] For example, assuming N = 5 (5 plots), k = 3 (3 features), the feature matrix F at each time point is t The dimension is 5×3. After expanding it into a row, the temporal multi-feature image I T Each row of will contain 5×3=15 pixel values, corresponding to the 3 feature values on 5 fields. T From the images, we can find the changing trends of the characteristic values at different time points and the differences between different plots.
[0110] Fusion of spatial and temporal information: Temporal multi-feature images not only retain the time series information, but also fuse the feature information of different fields in the form of images, enabling the model to simultaneously learn the temporal and spatial feature associations.
[0111] Facilitates the input of deep learning models: Image data can be directly input into deep learning models such as convolutional neural networks (CNNs), using the convolutional layers of CNNs to extract local features, thereby better capturing the complex relationships between temporal features.
[0112] Intuitive visualization: Time-series multi-feature images can intuitively display the characteristic changes during the growth process of winter wheat, making it easier for researchers to analyze and interpret.
[0113] The yield assessment module is used to input the time series multi-feature image into the improved CNN-BiGRU-Attention model to obtain the winter wheat yield assessment result.
[0114] like Figure 3 As shown in Figure 2, the contents of the improved CNN-BiGRU-Attention model specifically include:
[0115] The temporal multi-feature image is input into the convolution layer for convolution and then pooled, and the pooled feature map is input into the fully connected layer to obtain a one-dimensional feature vector;
[0116] The one-dimensional feature vector is input into the BiGRU layer and the Attention layer to obtain output data.
[0117] The CNN-BiGRU-Attention yield estimation model is a complex neural network structure consisting of an input layer, a CNN layer, a BiGRU layer, an Attention layer, and an output layer. In the CNN layer, the model uses multiple convolution operations to deeply mine feature information from historical input data, increasing the depth of the features. A pooling layer then reduces the dimensionality of the features to extract key information. Finally, a fully connected layer converts the extracted feature vectors into a one-dimensional form, resulting in a low-dimensional, deep feature vector. The BiGRU and Attention layers work together to learn the data variation patterns in the low-dimensional, deep feature vectors, capturing dependencies in the time series and assigning different attention weights to different features. Finally, the model outputs a predicted value through the output layer, completing the entire yield estimation process.
[0118] Example 3
[0119] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the winter wheat yield assessment method based on machine learning described in any of the above embodiments is implemented.
[0120] Figure 2 10 is a schematic diagram showing a more specific hardware structure of an electronic device provided in this embodiment. The device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are communicatively connected to each other within the device via the bus 1050.
[0121] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.
[0122] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, etc. The memory 1020 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.
[0123] The input / output interface 1030 is used to connect input / output modules to implement information input and output. The input / output modules can be configured as components within the device (not shown in the figure) or can be externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and output devices may include a display, speaker, vibrator, indicator light, etc.
[0124] The communication interface 1040 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via a wired method (such as USB (Universal Serial Bus), network cable, etc.) or a wireless method (such as mobile network, WIFI (Wireless Fidelity), Bluetooth, etc.).
[0125] The bus 1050 comprises a path for transmitting information between the various components of the device (eg, the processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 ).
[0126] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may only include the components necessary to implement the embodiments of this specification, and does not necessarily include all the components shown in the figure.
[0127] The system of the above embodiment is used to implement the corresponding winter wheat yield assessment method based on machine learning in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.
[0128] Example 4
[0129] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the winter wheat yield assessment method based on machine learning as described in any of the above embodiments.
[0130] The computer-readable media of this embodiment include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.
[0131] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the winter wheat yield assessment method based on machine learning as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0132] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples. Within the scope of the present disclosure, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present disclosure as described above, which are not provided in detail for the sake of simplicity.
[0133] In addition, to simplify the description and discussion, and so as not to obscure the embodiments of the present disclosure, known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided figures. In addition, devices may be shown in the form of block diagrams to avoid obscuring the embodiments of the present disclosure, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present disclosure are to be implemented (i.e., these details should be fully within the purview of those skilled in the art). Where specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present disclosure, it will be apparent to those skilled in the art that the embodiments of the present disclosure may be implemented without these specific details or with variations in these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.
[0134] Although the present disclosure has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.
[0135] Therefore, the units of each example described in the embodiments of this application can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0136] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.
Claims
1. A winter wheat yield assessment method based on machine learning, characterized in that: The method comprises: Step S1. Collect winter wheat images, meteorological data, soil conditions, agronomic measures, and historical yield data of the area to be measured to generate multi-source fusion data; Step S2. extracting key features from the multi-source fusion data using principal component analysis to form a multi-dimensional feature matrix; Step S3. extracting a time series multi-feature image based on the multi-dimensional feature matrix; Step S4. Input the time series multi-feature image into the improved CNN-BiGRU-Attention model to obtain the winter wheat yield evaluation result.
2. The winter wheat yield assessment method based on machine learning according to claim 1, characterized in that: The process of generating multi-source fusion data specifically includes: Performing one-dimensional temporal convolution on the winter wheat image, meteorological data, soil conditions, agronomic measures, and historical yield data to obtain wheat data features with time information and adding position codes to the wheat data features; The cross-modal Transformer is used to perform multi-scale and multi-modal information fusion on the wheat data features with added position encoding to obtain multi-source fused data.
3. The winter wheat yield assessment method based on machine learning according to claim 2, characterized in that: The temporal convolution method is: Among them, X a represents the data of multimodal input sequence, k a Indicates the size of the convolution kernel under the corresponding mode a, Conv1D represents one-dimensional time convolution, and d represents the feature dimension.
4. The winter wheat yield assessment method based on machine learning according to claim 1, characterized in that: In step S4, the content of the improved CNN-BiGRU-Attention model specifically includes: The temporal multi-feature image is input into the convolution layer for convolution and then pooled, and the pooled feature map is input into the fully connected layer to obtain a one-dimensional feature vector; The one-dimensional feature vector is input into the BiGRU layer and the Attention layer to obtain output data.
5. A winter wheat yield assessment system based on machine learning, the system being used to implement the assessment method according to any one of claims 1 to 4, characterized in that: The system includes: Step S1. Collect winter wheat images, meteorological data, soil conditions, agronomic measures, and historical yield data of the area to be measured to generate multi-source fusion data; Step S2. extracting key features from the multi-source fusion data using principal component analysis to form a multi-dimensional feature matrix; Step S3. extracting a time series multi-feature image based on the multi-dimensional feature matrix; Step S4. Input the time series multi-feature image into the improved CNN-BiGRU-Attention model to obtain the winter wheat yield evaluation result.
6. The winter wheat yield assessment system based on machine learning according to claim 5, characterized in that: The process of generating multi-source fusion data specifically includes: Performing one-dimensional temporal convolution on the winter wheat image, meteorological data, soil conditions, agronomic measures, and historical yield data to obtain wheat data features with time information and adding position codes to the wheat data features; The cross-modal Transformer is used to perform multi-scale and multi-modal information fusion on the wheat data features with added position encoding to obtain multi-source fused data.
7. The winter wheat yield assessment system based on machine learning according to claim 6, characterized in that: The temporal convolution method is: Among them, X a represents the data of multimodal input sequence, k a Indicates the size of the convolution kernel under the corresponding mode a, Conv1D represents one-dimensional time convolution, and d represents the feature dimension.
8. The winter wheat yield assessment system based on machine learning according to claim 6, characterized in that: The contents of the improved CNN-BiGRU-Attention model specifically include: The temporal multi-feature image is input into the convolution layer for convolution and then pooled, and the pooled feature map is input into the fully connected layer to obtain a one-dimensional feature vector; The one-dimensional feature vector is input into the BiGRU layer and the Attention layer to obtain output data.