A rice yield prediction method, device, equipment and medium

Abstract

The present invention discloses a rice yield prediction method, device, equipment and medium, including: obtaining multispectral data of a rice yield to be predicted area, and performing preprocessing to obtain an RGB image of the rice yield to be predicted area and an infrared image of the rice yield to be predicted area; fusing the color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area to obtain fusion features; inputting the fusion features into a preset yield prediction model to obtain the predicted yield of the rice yield to be predicted area. The present invention belongs to the field of rice yield prediction. The present invention utilizes VAE to fuse the color features and infrared temperature features of multispectral images, and extracts deep-level features through latent space, so as to more comprehensively evaluate the growth status and health level of crops. The method provided by the present invention can effectively utilize the spatiotemporal spectral features of images to accurately predict rice yield, and has stronger applicability and higher recognition accuracy.

Description

Technical Field

[0001] The invention relates to the field of rice yield prediction, and in particular to a rice yield prediction method, device, equipment and medium. Background Art

[0002] Rice is an important food crop, belonging to the genus Oryza of the Poaceae family, with the scientific name Oryza sativa. It is one of the most important food crops in the world, providing the main food source for many people around the world. Rice is a perennial herb that grows in warm climates. Its main part is grain, which we call rice, and is one of the important staple foods for humans. The cultivation history of rice is very long. It originated in Southeast Asia and is now widely planted in many countries and regions on all continents of the world. The cultivation of rice requires sufficient water resources and a warm climate, so it is mainly distributed in subtropical and tropical regions. Rice cultivation can be divided into two types: wet rice and upland rice, mainly based on different irrigation methods. Wet rice requires a lot of paddy field irrigation, while upland rice relies on natural precipitation. In general, rice is not just a crop, but also an important pillar of global food security and agricultural economy, which has an important impact on global food security.

[0003] Traditional rice sampling surveys are time-consuming and may damage crops. At the same time, subjective management methods are not accurate, which can easily lead to resource waste and environmental damage. In addition, the current method of obtaining phenotypic information such as crop leaf area index and chlorophyll content is mainly based on field sampling. The data obtained by this method has poor robustness, resulting in low prediction accuracy of crop yield estimation in existing studies, which cannot meet the requirements of precision agriculture for accurate, real-time, fast and lossless information acquisition. In recent years, some researchers have used a piecewise linear regression method with breakpoints to predict corn and soybean yields using NDVI, surface temperature, precipitation and soil moisture. There is also a stepwise regression method to estimate winter wheat yield using MODIS NDVI data. However, the above methods rely heavily on ground survey data, with a small amount of data, which is not enough to effectively monitor crops, and the accuracy of rice yield prediction is low. In view of this, how to accurately and efficiently predict rice yield is an urgent problem to be solved. Summary of the invention

[0004] The embodiments of the present application solve the technical problem of inaccurate rice yield prediction in the prior art by providing a rice yield prediction method, device, equipment and medium, and achieve the technical effect of accurately predicting rice yield.

[0005] In a first aspect, the present application provides a rice yield prediction method, the method comprising:

[0006] Acquire multispectral data of the area to be predicted for rice yield, and perform data preprocessing to obtain an RGB image of the area to be predicted for rice yield and an infrared image of the area to be predicted for rice yield;

[0007] The color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area are fused to obtain fusion features;

[0008] The fusion features are input into the preset yield prediction model to obtain the predicted yield of rice in the area to be predicted.

[0009] Furthermore, multispectral data of the area to be predicted for rice yield is obtained, and data preprocessing is performed to obtain an RGB image of the area to be predicted for rice yield and an infrared image of the area to be predicted for rice yield, including:

[0010] The multispectral data of the rice yield prediction area is calibrated by radiation, atmospherically corrected and geometrically corrected;

[0011] Determine the spectral reflectance of the processed multispectral data of the area to be predicted for rice yield;

[0012] According to the spectral reflectance, the spectral feature statistical analysis is performed to obtain the color features of the RGB image of the area to be predicted for rice yield and the infrared temperature features of the infrared image of rice in the prediction area;

[0013] Output the RGB image of the area to be predicted for rice yield and the infrared image of the area to be predicted for rice yield.

[0014] Furthermore, the color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area are fused to obtain fused features, including:

[0015] Input the color features of the RGB image of the area to be predicted for rice yield and the infrared temperature features of the infrared image of the area to be predicted for rice yield, and perform pre-processing;

[0016] In the preset encoder, reparameterization and KL divergence minimization are performed, and the fusion features are obtained by decoding through the preset decoder.

[0017] Furthermore, the preset yield prediction model is trained, including:

[0018] Obtain historical fusion features and the historical actual output corresponding to the historical fusion features;

[0019] Input the historical fusion features into the 3D-CNN network architecture to obtain the historical predicted output corresponding to the historical fusion features;

[0020] According to the historical predicted output corresponding to the historical fusion features, the error loss of the 3D-CNN network architecture is determined by comparing it with the historical actual output corresponding to the historical fusion features;

[0021] According to the error loss of the 3D-CNN network architecture, the operating parameters of the 3D-CNN network architecture are adjusted.

[0022] Further, including:

[0023] The 3D-CNN network architecture uses three-dimensional convolutions, including width, height, and time dimensions.

[0024] Furthermore, after performing spectral feature statistical analysis according to the spectral reflectance to obtain color features and infrared temperature features, the method further includes:

[0025] Perform feature enhancement on color features and infrared temperature features.

[0026] Furthermore, KL divergence minimization includes:

[0027] Mean and log variance.

[0028] In a second aspect, the present application provides a rice yield prediction device, the device comprising:

[0029] An acquisition module is used to acquire multispectral data of the rice yield to be predicted area, and perform data preprocessing to obtain an RGB image of the rice yield to be predicted area and an infrared image of the rice yield to be predicted area;

[0030] A fusion module is used to fuse the color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area to obtain a fusion feature;

[0031] The prediction module is used to input fusion features into a preset yield prediction model to obtain the predicted yield of the rice yield in the area to be predicted.

[0032] In a third aspect, the present application provides an electronic device, including:

[0033] processor;

[0034] a memory for storing processor-executable instructions;

[0035] The processor is configured to execute to implement a rice yield prediction method as provided in the first aspect.

[0036] In a fourth aspect, the present application provides a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to implement a rice yield prediction method as provided in the first aspect.

[0037] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0038] The present application provides a rice yield prediction method, including: obtaining multispectral data of a rice yield prediction area, and performing data preprocessing to obtain an RGB image of the rice yield prediction area and an infrared image of the rice yield prediction area; fusing the color features of the RGB image of the rice yield prediction area and the infrared temperature features of the infrared image of the rice yield prediction area to obtain a fusion feature; inputting the fusion feature into a preset yield prediction model to obtain a predicted yield of the rice yield prediction area. Compared with the traditional method, the present application combines multispectral images with deep learning technology to collect data in a large range and with high precision. At the same time, VAE is used to fuse the RGB color features and infrared temperature features of the multispectral images, extract deep features through latent space, learn and use the relationship between different features, and more comprehensively evaluate the growth status and health level of crops. In the prediction model, 3D-CNN is used to analyze the patterns and trends in multi-time series remote sensing data, without the need for complex preprocessing of input variables, and has higher prediction accuracy and better modeling convenience. In addition, an additional time dimension is introduced to capture multi-time series features that are not based solely on spectral information, so that seasonal data can be better distinguished from spectral data, providing more accurate feature indicators for the prediction of annual rice yield. The method provided in this application can effectively use the spatiotemporal spectral features of images to accurately predict rice yield, and has stronger applicability and higher recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 A schematic diagram of a rice yield prediction method provided in this application;

[0041] Figure 2 A schematic diagram of a multispectral data processing process provided for this application;

[0042] Figure 3 A schematic diagram of a flow chart of output fusion features provided in this application;

[0043] Figure 4 A schematic diagram of a flow chart of a preset yield prediction model training provided for this application;

[0044] Figure 5 A schematic diagram of the structure of a rice yield prediction device provided in this application;

[0045] Figure 6 A schematic diagram of the structure of an electronic device provided in this application. DETAILED DESCRIPTION

[0046] The embodiment of the present application solves the technical problem of inaccurate rice yield prediction in the prior art by providing a rice yield prediction method.

[0047] The technical solution of the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0048] A rice yield prediction method comprises: acquiring multispectral data of a rice yield to be predicted area, and performing data preprocessing to obtain an RGB image of the rice yield to be predicted area and an infrared image of the rice yield to be predicted area; fusing color features of the RGB image of the rice yield to be predicted area and infrared temperature features of the infrared image of the rice yield to be predicted area to obtain fused features; and inputting the fused features into a preset yield prediction model to obtain a predicted yield of the rice yield to be predicted area.

[0049] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0050] First of all, the term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0051] This application provides Figure 1 A rice yield prediction method shown includes steps S11-S13.

[0052] Step S11, acquiring multispectral data of the area to be predicted for rice yield, and performing data preprocessing to obtain an RGB image of the area to be predicted for rice yield and an infrared image of the area to be predicted for rice yield.

[0053] Step S12, fusing the color features of the RGB image of the area to be predicted for rice yield and the infrared temperature features of the infrared image of the area to be predicted for rice yield to obtain a fused feature.

[0054] Step S13, inputting the fusion features into the preset yield prediction model to obtain the predicted yield of the rice yield in the area to be predicted.

[0055] Regarding step S11, multispectral data of the area to be predicted for rice yield is obtained, and data preprocessing is performed to obtain an RGB image of the area to be predicted for rice yield and an infrared image of the area to be predicted for rice yield.

[0056] For step S11, please refer to Figure 2 The rice yield prediction area refers to the area where rice yield prediction is required, which includes a number of rice plants. Multispectral data of the rice yield prediction area can be obtained in a variety of ways, such as remote sensing data obtained by sensors carried by drones.

[0057] The method obtains multispectral data of a rice yield prediction area and performs data preprocessing to obtain an RGB image of the rice yield prediction area and an infrared image of the rice yield prediction area, including: performing radiation calibration, atmospheric correction and geometric correction processing on the multispectral data of the rice yield prediction area; determining the spectral reflectance of the processed multispectral data of the rice yield prediction area; performing spectral feature statistical analysis according to the spectral reflectance to obtain color features of the RGB image of the rice yield prediction area and infrared temperature features of an infrared image of rice in the prediction area; and outputting the RGB image of the rice yield prediction area and the infrared image of the rice yield prediction area.

[0058] After obtaining the multispectral data of the area to be predicted for rice yield, data preprocessing can be performed, including: radiation calibration, atmospheric correction and geometric correction.

[0059] Radiometric calibration: The raw data in remote sensing images are usually the radiation energy values ​​received by the sensor. The process of radiometric calibration is to convert the raw data into reflectivity data, which is the ratio of the reflectivity of the surface material. Radiometric calibration can correct the sensitivity of the sensor, remove background noise, etc. to ensure the accuracy and comparability of the data.

[0060] Atmospheric correction: Atmospheric correction is to eliminate the influence of the atmosphere on the acquisition and interpretation of remote sensing data. The atmosphere absorbs and scatters the radiation from the surface, making the received sensor signal different from the actual reflectivity of the surface. Through atmospheric correction, the interference of the atmosphere on the image can be inferred and eliminated to obtain more accurate surface reflectivity information.

[0061] Geometric correction: Remote sensing images may be affected by factors such as platform attitude and terrain undulations when they are acquired, resulting in inaccurate positions and directions of image pixels. The purpose of geometric correction is to correct the image so that it can accurately correspond to the position in the geographic coordinate system to ensure that the image can accurately match the geographic information system data.

[0062] After data preprocessing of the multispectral data of the area to be predicted for rice yield, the spectral reflectance can be calculated based on the multispectral data, and spectral feature statistical analysis can be performed based on the spectral reflectance to determine the spectral feature with the highest correlation with yield.

[0063] After performing spectral feature statistical analysis based on spectral reflectance to obtain color features and infrared temperature features, the process also includes: enhancing the color features and infrared temperature features. For example, the features are enhanced by color adjustment and contrast enhancement. By enhancing spectral features, the contrast and clarity of different bands in the image can be enhanced, making the subtle features and patterns hidden in the original image more prominent. Spectral feature enhancement can highlight the differences between different bands.

[0064] Regarding step S12, the color features of the RGB image of the area to be predicted for rice yield and the infrared temperature features of the infrared image of the area to be predicted for rice yield are fused to obtain fused features.

[0065] Regarding step S12, you can also refer to Figure 3 , the color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area are fused to obtain fused features, including: inputting the color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area, and performing pre-processing; in a preset encoder, re-parameterization and KL divergence minimization (considering the mean and logarithmic variance) are performed, and they are decoded by a preset decoder to obtain fused features.

[0066] To improve the efficiency and accuracy of the fusion, the univariate Gaussian product of the features of the RGB image and the infrared image can optimally combine the information of the two modes, ensuring that the network remains resilient when one of the data streams is not ideal.

[0067] The encoder can use a variational autoencoder, which is a generative model used to learn the potential representation of data. VAE maps input data (such as images) to a latent space and generates new data samples. The variational autoencoder (VAE) consists of two main parts: an encoder and a decoder. The encoder converts the input data into the mean and variance parameters of the latent space, and the decoder reconstructs the input data from the samples of the latent space.

[0068] By reparameterizing the mean and variance of the encoder output, the network training can be optimized based on gradient descent. This makes the gradient propagation more efficient and avoids the problems caused by directly sampling random variables during training.

[0069] VAEs work by minimizing the reconstruction error (usually the difference between the reconstructed image and the original image) while maximizing the homogeneity of the latent space, which allows for a better incorporation by taking both the mean and variance into account.

[0070] The latent variables of VAE are assumed to be multivariate Gaussian distributions. By considering the mean and variance, VAE can better capture the complex structure of the data and generate representations that are similar to the original data but more compact in the latent space.

[0071] In this way, the color features of the RGB image of the area to be predicted for rice yield and the infrared temperature features of the infrared image of the area to be predicted for rice yield can be fused. VAE enhances the ability to capture complex multimodal data relationships by constructing a multivariate latent space, thereby optimizing the final prediction accuracy. In addition, VAE also enhances the robustness of the model in the case of incomplete or inconsistent data, which is particularly critical for yield prediction in the agricultural field under complex scenarios.

[0072] Regarding step S13, the fusion features are input into the preset yield prediction model to obtain the predicted yield of the rice yield in the area to be predicted.

[0073] After obtaining the fused features, the fused features of the rice yield area to be predicted can be directly input into the preset yield prediction model to obtain the predicted yield of the rice yield area to be predicted.

[0074] This application also provides a method for training a preset yield prediction model, which can be referred to in conjunction with Figure 4 , training the preset production prediction model, including: obtaining historical fusion features and the historical actual production corresponding to the historical fusion features; inputting the historical fusion features into the 3D-CNN network architecture to obtain the historical predicted production corresponding to the historical fusion features; comparing the historical predicted production corresponding to the historical fusion features with the historical actual production corresponding to the historical fusion features to determine the error loss of the 3D-CNN network architecture; adjusting the operating parameters of the 3D-CNN network architecture according to the error loss of the 3D-CNN network architecture.

[0075] The historical fusion features may be obtained in the same manner as steps S11-S12. It may be understood that the historical fusion features refer to the fusion features formed in the historically acquired remote sensing data of a rice region, and one historical fusion feature corresponds to an actual historical yield of the region.

[0076] The historical fusion features are input into the 3D-CNN network architecture. 3D-CNN is also called the spatiotemporal method. By adding a time dimension to the convolution kernel, the convolution kernel can operate on a specific time window. Compared with 2D-CNN, in 2D-CNN, all spatiotemporal data are stacked in the input. After the first convolution, the time information will be completely folded, resulting in low model accuracy or unusable model.

[0077] Therefore, the 3D-CNN network architecture uses three-dimensional convolutions, applying three-dimensional kernels sliding across the width, height, and time image dimensions, with each kernel generating a three-dimensional feature map for each spatial-spectral and temporal data volume.

[0078] In addition, when predicting rice yield, considering multi-time series data can help better understand seasonal crop growth. The 3D-CNN network architecture embeds different time observations into a new spatial dimension, allowing the 3D convolution kernel to have additional degrees of freedom to better detect temporal correlations and extract dynamic features through continuous time periods. In addition, 3D-CNN can also solve the long-term time constraints of the data while performing comprehensive spatiotemporal spectral feature extraction, thereby improving the accuracy of yield prediction.

[0079] The historical fusion features of the input 3D-CNN network architecture are passed through the convolution layer-batch normalization... fully connected layer (please refer to Figure 4 ), the historical predicted output corresponding to the historical fusion features can be obtained, and the error loss can be obtained by comparing the historical predicted output with the historical actual output.

[0080] If the error loss is within the allowable range of the preset error, it means that the preset yield prediction model has high accuracy and no further training is required. If the error loss is not within the allowable range of the preset error, it means that the accuracy of the preset yield prediction model is not high and the yield prediction is inaccurate. You can continue to adjust the operating parameters related to the preset yield prediction model and repeat the above steps until the error loss is within the allowable range of the preset error.

[0081] In summary, the present application provides a rice yield prediction method, including: obtaining multispectral data of a rice yield prediction area, and performing data preprocessing to obtain an RGB image of the rice yield prediction area and an infrared image of the rice yield prediction area; fusing the color features of the RGB image of the rice yield prediction area and the infrared temperature features of the infrared image of the rice yield prediction area to obtain a fusion feature; inputting the fusion feature into a preset yield prediction model to obtain a predicted yield of the rice yield prediction area. Compared with traditional methods, the present application combines multispectral images with deep learning technology to collect data in a large range and with high precision. At the same time, VAE is used to fuse the RGB color features and infrared temperature features of multispectral images, extract deep features through latent space, learn and use the relationship between different features, and more comprehensively evaluate the growth status and health level of crops. In the prediction model, 3D-CNN is used to analyze the patterns and trends in multi-time series remote sensing data, without the need for complex preprocessing of input variables, and has higher prediction accuracy and better modeling convenience. In addition, an additional time dimension is introduced to capture multi-time series features that are not based solely on spectral information, so that seasonal data can be better distinguished from spectral data, providing more accurate feature indicators for the prediction of annual rice yield. The method provided in this application can effectively use the spatiotemporal spectral features of images to accurately predict rice yield, and has stronger applicability and higher recognition accuracy.

[0082] Based on the same inventive concept, the present application provides a rice yield prediction device, comprising:

[0083] An acquisition module 51 is used to acquire multispectral data of the area to be predicted for rice yield, and perform data preprocessing to obtain an RGB image of the area to be predicted for rice yield and an infrared image of the area to be predicted for rice yield;

[0084] A fusion module 52 is used to fuse the color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area to obtain a fusion feature;

[0085] The prediction module 53 is used to input the fusion features into the preset yield prediction model to obtain the predicted yield of the rice yield in the area to be predicted.

[0086] Based on the same inventive concept, the present application also provides Figure 6 An electronic device as shown includes:

[0087] Processor 61;

[0088] A memory 62 for storing instructions executable by the processor 61;

[0089] The processor 61 is configured to execute to implement a rice yield prediction method as provided above.

[0090] Based on the same inventive concept, the present application also provides a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by the processor 61 of the electronic device, the electronic device can implement a rice yield prediction method as provided above.

[0091] Since the electronic device introduced in this embodiment is an electronic device used to implement the information processing method in the embodiment of the present application, based on the information processing method introduced in the embodiment of the present application, a person skilled in the art can understand the specific implementation of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as a person skilled in the art implements the electronic device used by the information processing method in the embodiment of the present application, it belongs to the scope of protection of this application.

[0092] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0093] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0094] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0096] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0097] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A rice yield prediction method, It is characterized in that The method comprises: The method comprises: obtaining multispectral data of a rice yield to be predicted area, and performing data preprocessing to obtain an RGB image of the rice yield to be predicted area and an infrared image of the rice yield to be predicted area, comprising: performing radiation calibration, atmospheric correction and geometric correction processing on the multispectral data of the rice yield to be predicted area; determining the spectral reflectance of the multispectral data of the rice yield to be predicted area after the processing; performing spectral feature statistical analysis according to the spectral reflectance to obtain color features of the RGB image of the rice yield to be predicted area and infrared temperature features of the infrared image of rice in the to-be-predicted area; outputting the RGB image of the rice yield to be predicted area and the infrared image of the rice yield to be predicted area, and performing spectral feature statistical analysis according to the spectral reflectance to obtain the color features and the infrared temperature features, further comprising: performing feature enhancement on the color features and the infrared temperature features; The color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area are fused to obtain fused features, including: inputting the color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area, and performing pre-processing; performing re-parameterization and KL divergence minimization in a preset encoder, and decoding through a preset decoder to obtain the fused features; The fusion feature is input into a preset yield prediction model to obtain the predicted yield of the rice yield to be predicted area, wherein the preset yield prediction model is trained, including: obtaining historical fusion features and historical actual yields corresponding to the historical fusion features; inputting the historical fusion features into a 3D-CNN network architecture to obtain historical predicted yields corresponding to the historical fusion features; comparing the historical predicted yields corresponding to the historical fusion features with the historical actual yields corresponding to the historical fusion features to determine the error loss of the 3D-CNN network architecture; and adjusting the operating parameters of the 3D-CNN network architecture according to the error loss of the 3D-CNN network architecture.

2. A rice yield prediction method as claimed in claim 1, It is characterized in that include: The 3D-CNN network architecture uses three-dimensional convolution, including width dimension, height dimension and time dimension.

3. A rice yield prediction method as claimed in claim 1, It is characterized in that KL divergence minimization includes: Mean and log variance.

4. A rice yield prediction device, It is characterized in that A rice yield prediction method according to any one of claims 1 to 3, wherein the device comprises: An acquisition module is used to acquire multispectral data of the area to be predicted for rice yield, and perform data preprocessing to obtain an RGB image of the area to be predicted for rice yield and an infrared image of the area to be predicted for rice yield; A fusion module, used for fusing the color features of the RGB image of the rice yield to be predicted area and the infrared temperature features of the infrared image of the rice yield to be predicted area to obtain a fusion feature; The prediction module is used to input the fusion features into a preset yield prediction model to obtain the predicted yield of the rice yield in the area to be predicted.

5. An electronic device, It is characterized in that include: processor; a memory for storing instructions executable by the processor; Wherein, the processor is configured to execute to implement a rice yield prediction method as described in any one of claims 1 to 3.

6. A non-transitory computer-readable storage medium, when the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to implement a rice yield prediction method as described in any one of claims 1 to 3.

Images