Rice growth period identification method and system based on edge calculation

Abstract

The invention discloses a rice growth period identification method and system based on edge calculation, and the method specifically comprises the following steps: S1, data collection: collecting various data related to a rice growth period through a multi-mode sensor disposed in a farmland, and the data comprises image data, meteorological data and soil data; according to the method, a deep learning model combining a multi-scale convolutional neural network (MSCNN) and a long short-term memory network (LSTM) is utilized, multi-source data such as images, weather and soil are fused, the rice growth period is recognized in real time, data processing is performed on edge nodes, data transmission time delay is reduced, and real-time recognition and feedback are achieved; by fusing data from different sources, more comprehensive rice growth information is obtained, and the accuracy of growth period identification is improved; and image features of different scales are extracted by adopting a multi-scale convolution kernel, and modeling is performed on time series data such as weather and the like in combination with an LSTM model, so that the generalization ability and the recognition precision of the model are further improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent agricultural monitoring, and in particular to a rice growth period identification method and system based on edge computing. Background Art

[0002] As one of the most important food crops in the world, precise management of rice during its growth cycle is of great significance for improving yield and quality. In modern agricultural production, understanding and mastering the information of rice growth period can provide an important basis for precision agricultural management, ensuring that agricultural activities such as fertilization, irrigation, pest and disease control are carried out at the right time, thereby improving rice yield and quality.

[0003] 1. Deficiencies of existing rice growth period identification technology

[0004] At present, the identification of rice growth period mainly relies on traditional manual observation and statistical methods, which usually require agricultural technicians to enter the farmland regularly and judge the growth period by visually observing the growth status of rice. This method is not only inefficient, but also easily affected by human resources, experience level and weather conditions. In addition, traditional observation methods usually cannot reflect subtle changes in the crop growth process in a timely manner, resulting in a certain lag and subjectivity in agricultural management decisions.

[0005] In the information age, with the rapid development of remote sensing technology, image processing technology and deep learning technology, crop growth period recognition technology based on image recognition has gradually attracted attention. UAVs, satellite remote sensing and other means can conduct extensive monitoring of large-scale farmland, and automatically identify the growth status and growth period of crops through high-resolution image data analysis. However, due to the influence of environment, lighting and imaging equipment resolution, the recognition accuracy of remote sensing images still needs to be improved in complex farmland environments. In addition, traditional remote sensing image processing relies on cloud computing, which takes a long time to process and is difficult to meet real-time requirements.

[0006] 2. Advantages of edge computing

[0007] As a new computing model, edge computing technology sinks data processing and storage to edge nodes close to data sources, greatly reducing data transmission delays and bandwidth pressure. In farmland environments, edge computing can be deployed directly in the fields to process large amounts of sensor data on-site to achieve real-time analysis and rapid response. This model can effectively compensate for the latency problem of cloud computing while reducing the reliance of data transmission on network bandwidth.

[0008] Through edge computing, the rice growth period recognition system can pre-process and analyze a large amount of image and sensor data locally to improve the system's response speed. In addition, by combining multimodal data (such as images, meteorological data, soil data, etc.), the growth status of rice can be captured more comprehensively, thereby improving recognition accuracy and real-time performance.

[0009] 3. Application of deep learning in crop identification

[0010] The successful application of deep learning technology in image processing, time series analysis and other fields has greatly promoted the development of intelligent agricultural management. Convolutional neural networks (CNNs) perform well in image feature extraction and recognition, and can automatically extract rice features at different growth stages from farmland images, such as leaf morphology, color, density and other features. Long short-term memory networks (LSTMs) have advantages in time series data modeling, and can capture the dynamic effects of environmental factors such as temperature and humidity on rice growth from continuous meteorological data.

[0011] At present, many studies have begun to try to apply convolutional neural networks and other deep learning models to crop identification and growth monitoring. However, traditional single-mode recognition methods (such as relying only on images or meteorological data) have certain limitations. In the complex environment of farmland, a single data source is difficult to fully reflect the growth status of rice. In order to improve the accuracy and reliability of recognition, it is urgent to fuse multimodal data and use deep learning models to simultaneously process image and time series data to obtain more accurate growth period recognition results.

[0012] 4. Necessity of multimodal data fusion

[0013] In a farmland environment, the growth process of rice is affected by many factors, including climate, soil, irrigation, etc. A single data source (such as image or meteorological data) is difficult to fully reflect the combined effects of these complex factors. Multimodal data fusion technology can effectively make up for the shortcomings of a single data source by fusing different types of data (such as image, meteorological, soil data, etc.), making the identification of the growth period more accurate.

[0014] For example, image data can reflect the growth morphology of rice, while meteorological data can provide background information about the crop growth environment. By weighted fusion of the features of these different data sources, a more comprehensive feature description can be obtained, thereby improving the recognition ability of the deep learning model.

[0015] 5. Problems

[0016] Although edge computing, deep learning, and multimodal data fusion technologies have shown great potential in the agricultural field, the following problems still exist in current technology applications:

[0017] Lack of real-time performance. The traditional cloud computing model is limited by the delay of data transmission and processing, and it is difficult to meet the needs of large-scale real-time monitoring of farmland. The recognition accuracy is limited. The deep learning model that only relies on a single data source still has insufficient recognition accuracy in complex farmland environments, especially in conditions of uneven lighting and changes in vegetation coverage. Data fusion challenges. In the process of fusing multimodal data, the feature extraction and weighted processing of different types of data are complex. How to effectively fuse multi-source data and use them in deep learning models is a technical difficulty. Summary of the invention

[0018] The present invention provides a rice growth period recognition method and system based on edge computing, which utilizes a deep learning model combining a multi-scale convolutional neural network (MSCNN) and a long short-term memory network (LSTM) to fuse multi-source data such as images, meteorology, and soil to recognize the rice growth period in real time.

[0019] In order to achieve the above object, the technical solution adopted by the present invention is: a rice growth period identification method based on edge computing, which specifically includes the following steps:

[0020] Step S1, data collection: collecting various data related to the rice growth period, including image data, meteorological data and soil data, through multimodal sensors deployed in the farmland;

[0021] Step S2, data preprocessing: preprocessing the collected multimodal data on the edge node, specifically including image denoising, data filtering and outlier processing to ensure data quality;

[0022] Step S3, feature extraction: a multi-scale convolutional neural network (MSCNN) is used to perform multi-level feature extraction on the image data, and a long short-term memory network (LSTM) is used to perform feature modeling on the time series data to obtain preliminary features;

[0023] Step S4, multimodal feature fusion: the features extracted from different data sources are fused by weighted fusion to generate a final comprehensive feature vector;

[0024] Step S5, growth period identification: based on the feature vector of multimodal feature fusion, input into the deep learning model to infer the growth period of rice;

[0025] Step S6, growth period classification output: the model generates classification results of rice growth periods and feeds the results back to the remote backend management system.

[0026] The multimodal sensor module in step S1 includes:

[0027] Step S1.1, an image sensor is used to obtain the morphology and color characteristics of rice leaves;

[0028] Step S1.2, meteorological sensor, used to collect environmental data such as temperature, humidity, and light;

[0029] Step S1.3: Soil sensor is used to monitor soil moisture, nutrient content and other root growth environment parameters.

[0030] The multi-scale convolutional neural network (MSCNN) model used in step S3 extracts image features of rice at different growth stages, including leaf morphology, color and distribution, through convolution kernels of different scales. The calculation formula of the convolution layer is:

[0031]

[0032] in, is the convolution output of the lth layer, is the convolution kernel weight, is the input image feature, b l For bias.

[0033] The multimodal feature fusion in step S4 adopts a weighted fusion strategy to fuse the feature representations of different modal data, and the formula is:

[0034] F fused =α·F image +β·F time

[0035] Among them, F fused is the fused feature vector, F image is the image feature, F time is the time series data feature, α and β are the weighted fusion coefficients.

[0036] The deep learning model in step S5 combines a multi-scale convolutional neural network (MSCNN) and a long short-term memory network (LSTM). The MSCNN is used to extract the spatial features of rice images, and the LSTM is used to process the temporal features of time series data. The loss function of the model is a cross entropy loss function, and the formula is:

[0037]

[0038] Among them, y i is the true label, is the probability predicted by the model.

[0039] A rice growth period recognition system based on edge computing includes the following steps:

[0040] Step 1: Edge nodes are deployed in farmland environments and have data processing and storage functions. They are responsible for data collection, preprocessing, and reasoning analysis.

[0041] Step 2: The multimodal sensor module is installed in the farmland to collect multi-source data related to the rice growth period in real time and transmit it to the edge node;

[0042] Step 3: Data processing module, deployed in edge nodes, for preprocessing and feature extraction of sensor data;

[0043] Step 4: The deep learning model runs in the edge node and combines the multi-scale convolutional neural network (MSCNN) and the long short-term memory network (LSTM) to perform real-time recognition of the rice growth period;

[0044] Step 5: Backend management platform, connects edge nodes through wireless communication, remotely monitors, stores data and provides decision support.

[0045] The edge nodes in the system use embedded processors, have local data storage and processing capabilities, and support real-time reasoning and remote communication.

[0046] The multimodal sensor module in the system includes a spectral imaging sensor for acquiring spectral reflectance information of rice to assist in the identification of the growth period.

[0047] The background management platform supports online model updates and regularly updates weights and adjusts parameters of deep learning models to improve recognition accuracy and adaptability.

[0048] The beneficial effects of adopting the above technical solution are:

[0049] The present invention uses a deep learning model combining a multi-scale convolutional neural network (MSCNN) and a long short-term memory network (LSTM) to integrate multi-source data such as images, meteorology, and soil to identify the rice growth period in real time. The method has the following advantages:

[0050] (1) Edge computing improves real-time performance: By processing data on edge nodes, data transmission delay is reduced, and real-time identification and feedback are achieved.

[0051] (2) Multimodal data fusion improves recognition accuracy: By fusing data from different sources, more comprehensive rice growth information can be obtained, thereby improving the accuracy of growth period recognition.

[0052] (3) Improvement of deep learning models: Multi-scale convolution kernels are used to extract image features of different scales, and the LSTM model is combined to model time series data such as meteorology, further improving the generalization ability and recognition accuracy of the model.

[0053] The invention aims to solve the shortcomings of the existing rice growth period recognition technology in terms of accuracy and real-time performance, and to provide more efficient decision support for smart agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is the overall control flow chart of the rice growth period identification method based on edge computing;

[0055] Figure 2 This is the overall control flow chart of the rice growth period recognition system based on edge computing;

[0056] Figure 3 It is the basic flow chart of the rice growth period identification method and system based on edge computing;

[0057] Figure 4 It is a statistical graph of observed values ​​for model training of the present invention;

[0058] Figure 5 This is a diagram of the growth period identification results of the present invention. DETAILED DESCRIPTION

[0059] The specific implementation methods of the present invention are further explained in detail below by describing the embodiments with reference to the accompanying drawings, with the aim of helping those skilled in the art to have a more complete, accurate and in-depth understanding of the concept and technical solution of the present invention and facilitating its implementation.

[0060] Embodiment 1

[0061] like Figures 1 to 5 As shown, the present invention is a rice growth period recognition method and system based on edge computing, which utilizes a deep learning model combining a multi-scale convolutional neural network (MSCNN) and a long short-term memory network (LSTM) to integrate multi-source data such as images, meteorology, and soil to recognize the rice growth period in real time.

[0062] Specifically, Figures 1 to 5 As shown, the specific steps include:

[0063] Step S1, data collection: collecting various data related to the rice growth period, including image data, meteorological data and soil data, through multimodal sensors deployed in the farmland;

[0064] Step S2, data preprocessing: preprocessing the collected multimodal data on the edge node, specifically including image denoising, data filtering and outlier processing to ensure data quality;

[0065] Step S3, feature extraction: a multi-scale convolutional neural network (MSCNN) is used to perform multi-level feature extraction on the image data, and a long short-term memory network (LSTM) is used to perform feature modeling on the time series data to obtain preliminary features;

[0066] Step S4, multimodal feature fusion: the features extracted from different data sources are fused by weighted fusion to generate a final comprehensive feature vector;

[0067] Step S5, growth period identification: based on the feature vector of multimodal feature fusion, input into the deep learning model to infer the growth period of rice;

[0068] Step S6, growth period classification output: the model generates classification results of rice growth periods and feeds the results back to the remote backend management system.

[0069] The multimodal sensor module in step S1 includes:

[0070] Step S1.1, an image sensor is used to obtain the morphology and color characteristics of rice leaves;

[0071] Step S1.2, meteorological sensor, used to collect environmental data such as temperature, humidity, and light;

[0072] Step S1.3: Soil sensor is used to monitor soil moisture, nutrient content and other root growth environment parameters.

[0073] The multi-scale convolutional neural network (MSCNN) model used in step S3 extracts image features of rice at different growth stages, including leaf morphology, color, and distribution, through convolution kernels of different scales. The calculation formula of the convolution layer is:

[0074]

[0075] in, is the convolution output of the lth layer, is the convolution kernel weight, is the input image feature, b l For bias.

[0076] The multimodal feature fusion in step S4 adopts a weighted fusion strategy to fuse the feature representations of different modal data. The formula is:

[0077] F fused =α·F image +β·F time

[0078] Among them, F fused is the fused feature vector, F image is the image feature, F time is the time series data feature, α and β are the weighted fusion coefficients.

[0079] The deep learning model in step S5 combines a multi-scale convolutional neural network (MSCNN) and a long short-term memory network (LSTM). MSCNN is used to extract the spatial features of rice images, and LSTM is used to process the temporal features of time series data. The loss function of the model is a cross entropy loss function, and the formula is:

[0080]

[0081] Among them, y i is the true label, is the probability predicted by the model.

[0082] In step S1, high-definition image data, meteorological data, and soil data of different rice growth stages are collected according to the divided rice growth stages, and the collected data are transmitted to the remote platform at the same time:

[0083] a1. Image data acquisition. Image data is mainly obtained by smart cameras regularly to obtain high-resolution images of rice fields for subsequent analysis of the morphology, color, density and other characteristics of rice growth. Smart cameras are installed on the ridges or monitoring points in a fixed area for all-weather monitoring and data collection. When collecting, ensure that the angle and height are consistent to reduce the impact of image distortion on subsequent identification. When collecting images, attention should be paid to the frequency and time period of collection. During the tillering period, the images are collected once a week to capture the density of rice tillers and leaf growth; from the jointing stage to the heading stage: collect once a day, because the rice grows faster at this time, and the changes in plant height and morphology need to be closely monitored; from the filling stage to the maturity stage: collect once every 3 days to observe the color changes and maturity status of the rice panicle. The resolution of the collected images is not less than 1080p, and images with too much blur or noise are avoided. It is preferred to collect images in the morning or afternoon when the light is relatively stable to avoid strong light or shadows.

[0084] a2. Meteorological data collection. Meteorological data can help determine the suitable growth environment for rice at each growth stage. The main meteorological factors collected include temperature, humidity, light intensity, rainfall, wind speed, etc. Meteorological monitoring stations should be installed in the fields or near the ridges. The meteorological stations should be away from rice plants to avoid affecting the measurement accuracy. The equipment needs to be calibrated regularly and equipped with a backup power supply to ensure uninterrupted operation. Meteorological data collection needs to pay attention to the collection frequency. It should be collected once every hour to ensure the real-time and continuity of the data, and to be able to adapt to the rapid environmental changes in different growth stages of rice. If certain weather factors (such as rainfall and wind speed) change dramatically in the short term, the sampling frequency can be increased as needed.

[0085] a3. Soil data collection. Soil data is crucial for water and nutrient management during the rice growth period. It mainly collects indicators such as soil moisture, temperature and nutrient content. Deploy soil moisture and temperature sensors at different locations in the field and collect data regularly. The collection points need to be evenly distributed to represent the soil conditions of the entire field. The depth of the soil sensor buried in the soil must be consistent with the root distribution characteristics of rice, generally 10-15 cm, to avoid interference with the results caused by changes in the surface soil. When collecting soil data, attention should be paid to the frequency of collection. During the tillering, jointing and filling stages when rice requires more water, soil moisture and temperature data should be collected once an hour to manage irrigation and fertilization operations in real time. During other growth periods, the data collection frequency can be adjusted to once every 6 hours to ensure the suitability of soil moisture and temperature.

[0086] a4. Data synchronization and upload. Different types of data collection need to be aligned according to the corresponding rice growth period to ensure the synchronization of meteorological, soil, and image data for subsequent multimodal data fusion. The processed data is transmitted to the remote management platform through the edge computing node. The management platform can store and call the data in batches according to the collection time and geographical location.

[0087] In step S2, the collected data is processed to construct a rice growth period data set. Rice growth period identification, data labeling and data set generation are key links in model training. High-quality data sets can significantly improve the recognition accuracy and stability of the model. The following are the detailed steps:

[0088] b1. Data cleaning: The collected images, meteorological data and soil data need to be cleaned before annotation, and the blurred, incomplete and noisy data are deleted. For image data, the image quality (such as resolution, clarity, brightness) is checked and standardized cropping is performed to ensure that the size of all images is consistent. At the same time, anomaly detection is performed on meteorological and soil data to remove mutations or erroneous data points.

[0089] b2. Data classification: The data are preliminarily classified according to the growth stage of rice, including seedling stage, tillering stage, jointing stage, booting stage, heading stage, filling stage, maturity stage, and other 8 categories, as shown in the following table; the data of each category of the data set should be evenly distributed to ensure that the number of each growth stage in the data set is roughly equal to avoid the impact of data imbalance on the model.

[0090]

[0091]

[0092] Table 1

[0093] b3. Data labeling: The goal of image data labeling is to give accurate labels to each growth period image to guide the model to learn the visual characteristics of rice in different growth periods, assign corresponding growth period category labels to each image (such as tillering stage, jointing stage, etc.), and perform detailed labeling based on the morphological characteristics of rice. In some growth period intersection stages (such as the late tillering stage and the early jointing stage), multi-label labeling can be added to the image to reflect the transition characteristics of the growth period. After the labeling is completed, manual inspection is carried out to ensure the accuracy and consistency of the labels.

[0094] b4. Data enhancement: Data enhancement is used to expand the size of the image dataset and improve the generalization ability of the model, which is especially important when the amount of data is insufficient in certain growth periods. Image enhancement methods include rotation and flipping, which randomly rotates the image by a certain angle or flips it horizontally to adapt the model to different shooting angles; brightness and contrast adjustment, which simulates rice images under different lighting conditions by adjusting the brightness and contrast of the image; scaling and cropping, which randomly scales or crops the image to increase sample diversity; and noise addition, which adds slight Gaussian noise or salt and pepper noise to the image to enhance the model's robustness to noise.

[0095] b5. Dataset format and storage: Save the annotated and enhanced dataset in a standard format to facilitate subsequent model training and testing. Save the image data in JPEG or PNG format, and save the label information in JSON or XML files (can be saved in COCO or Pascal VOC format); save the meteorological and soil data in CSV or JSON files, and add growth period labels to each data record to form a standard time series dataset. Finally, establish a folder structure by growth period category. Each folder contains all images and time series data of the corresponding growth period. At the same time, record the collection time, collection equipment, location and other information of each data sample to trace the data source.

[0096] The training and optimization of the model in step S3 includes model selection, model training, hyperparameter adjustment, and model evaluation. The following are the detailed steps:

[0097] c1. Model selection: Rice growth period recognition involves multimodal data (image, meteorological, soil data), which requires the use of convolutional neural networks (CNN) to process image features and long short-term memory networks (LSTM) to process time series data. Based on this, the following model architecture is selected: image feature extraction: use the multi-scale convolutional neural network (MSCNN) model to extract rice image features; time series feature extraction: use LSTM to process time series features such as meteorological and soil data; feature fusion: fuse image and time series features in the fully connected layer to make full use of multi-source data features and improve model recognition accuracy.

[0098] c2. Model training: The cross-entropy loss function is used, which is suitable for multi-category classification tasks of the growth period. Weights are applied to different growth period categories to alleviate the problem of category imbalance. A small initial learning rate (such as 0.001) is used, and a learning rate decay strategy (such as step decay or cosine decay) is used to control the model convergence speed.

[0099] Choose an appropriate batch size (such as 32 or 64) based on hardware resources to increase the model training speed while ensuring reasonable memory usage. Use the Adam optimizer to optimize the model. The Adam optimizer has an adaptive learning rate adjustment function that can converge faster. During the training process, the loss and accuracy of the validation set are calculated in real time to observe whether the model is overfitting. When the accuracy of the validation set no longer improves, stop training to prevent the model from overfitting.

[0100] c3. Hyperparameter adjustment: The Dropout layer can be applied to the CNN part, and the regularization weight can be applied to the LSTM part to prevent overfitting. At the same time, the Batch Normalization layer is added to the CNN model to stabilize the training process and improve the generalization ability of the model.

[0101] c4. Model evaluation: In the case of class imbalance, the F1 score reflects the model effect, and the calculation formula is as follows:

[0102]

[0103] in,

[0104]

[0105] In step S4, model conversion and deployment is performed. This step deploys the trained rice growth period recognition model on the edge computing device, which can improve data processing efficiency and reduce latency, thereby achieving real-time intelligent agricultural management. The following are the detailed steps:

[0106] d1. Format conversion: Export the trained model to a format supported by the edge computing device (such as TensorFlowLite, ONNX or OpenVINO format). The present invention adopts the onnx format so that the model can run efficiently on devices with limited resources. The resources of edge computing devices are limited. The model is compressed through quantization (such as 8-bit quantization) to reduce the amount of calculation while ensuring the model accuracy.

[0107] d2. Model deployment: Deploy the converted model on the edge computing product, install the necessary software dependencies and model inference framework to support efficient model inference on the edge device, write the model inference code on the edge device, input the input data (image and time series data) into the model, and generate the recognition results of the rice growth period.

[0108] In step S5, the recognition result push and decision making are carried out. This step builds a system from data processing to user feedback. Through edge devices to collect and analyze data, the recognition results are pushed to farmers in a timely manner to guide the corresponding agricultural management measures. The following are the detailed steps:

[0109] e1. Processing and formatting of growth period identification results. The identified rice growth period, predicted time, meteorological data, geographic location and other information are uniformly formatted to ensure that the results are clear and easy to read.

[0110] e2. Farming advice generation and push: Based on the growth period identification results, refer to the farming operation suggestions of the corresponding growth period in the database to generate current farming decisions. Specific suggestions such as fertilization, irrigation, and pest control can be made for different growth periods. According to the equipment conditions of farmers, the most appropriate push method is selected, such as mobile phone SMS push, app application push, and email push.

[0111] e3. Monitoring and feedback: Provide farmers with a real-time monitoring interface on the remote platform, display the current growth period, historical data, trend charts, etc., so that farmers can understand the crop growth more clearly. Farmers can provide feedback on the decision suggestions received, record the adoption status and provide opinions, such as whether fertilization has been carried out, whether the irrigation frequency has been modified, etc. The feedback results are used to optimize the decision suggestions. By tracking the execution results of farmers, such as yield and crop health, and analyzing the effects of agricultural suggestions, the decision-making algorithm and growth period identification model are continuously optimized.

[0112] Embodiment 1

[0113] A rice growth period recognition system based on edge computing, comprising:

[0114] Step 1: Edge nodes are deployed in farmland environments and have data processing and storage functions. They are responsible for data collection, preprocessing, and reasoning analysis.

[0115] Step 2: The multimodal sensor module is installed in the farmland to collect multi-source data related to the rice growth period in real time and transmit it to the edge node;

[0116] Step 3: Data processing module, deployed in edge nodes, for preprocessing and feature extraction of sensor data;

[0117] Step 4: The deep learning model runs in the edge node and combines the multi-scale convolutional neural network (MSCNN) and the long short-term memory network (LSTM) to perform real-time recognition of the rice growth period;

[0118] Step 5: Backend management platform, connects edge nodes through wireless communication, remotely monitors, stores data and provides decision support.

[0119] The edge nodes in the system use embedded processors, have local data storage and processing capabilities, and support real-time reasoning and remote communication.

[0120] The multimodal sensor module in the system includes a spectral imaging sensor, which is used to obtain the spectral reflectance information of rice to assist in the identification of the growth period.

[0121] The backend management platform supports online model updates and regularly updates weights and adjusts parameters of deep learning models to improve recognition accuracy and adaptability.

[0122] The present invention is described above by way of example in conjunction with the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as various non-substantial improvements are made using the method concept and technical solution of the present invention; or the above-mentioned concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A rice growth period identification method based on edge computing, characterized in that: The specific steps include: Step S1, data collection: collecting various data related to the rice growth period, including image data, meteorological data and soil data, through multimodal sensors deployed in the farmland; Step S2, data preprocessing: preprocessing the collected multimodal data on the edge node, specifically including image denoising, data filtering and outlier processing to ensure data quality; Step S3, feature extraction: a multi-scale convolutional neural network (MSCNN) is used to perform multi-level feature extraction on the image data, and a long short-term memory network (LSTM) is used to perform feature modeling on the time series data to obtain preliminary features; Step S4, multimodal feature fusion: the features extracted from different data sources are fused by weighted fusion to generate a final comprehensive feature vector; Step S5, growth period identification: based on the feature vector of multimodal feature fusion, input into the deep learning model to infer the growth period of rice; Step S6, growth period classification output: the model generates classification results of rice growth periods and feeds the results back to the remote backend management system.

2. The rice growth period identification method based on edge computing according to claim 1, characterized in that: The multimodal sensor module in step S1 includes: Step S1.1, an image sensor is used to obtain the morphology and color characteristics of rice leaves; Step S1.2, meteorological sensor, used to collect environmental data such as temperature, humidity, and light; Step S1.3: Soil sensor is used to monitor soil moisture, nutrient content and other root growth environment parameters.

3. The rice growth period identification method based on edge computing according to claim 1 is characterized in that: The multi-scale convolutional neural network (MSCNN) model used in step S3 extracts image features of rice at different growth stages, including leaf morphology, color and distribution, through convolution kernels of different scales. The calculation formula of the convolution layer is: in, is the convolution output of the lth layer, is the convolution kernel weight, is the input image feature, b l For bias.

4. The rice growth period identification method based on edge computing according to claim 1, characterized in that: The multimodal feature fusion in step S4 adopts a weighted fusion strategy to fuse the feature representations of different modal data, and the formula is: F fused =α·F image +β·F time Among them, F fused is the fused feature vector, F image is the image feature, F time is the time series data feature, α and β are the weighted fusion coefficients.

5. The rice growth period identification method based on edge computing according to claim 1, characterized in that: The deep learning model in step S5 combines a multi-scale convolutional neural network (MSCNN) and a long short-term memory network (LSTM). The MSCNN is used to extract the spatial features of rice images, and the LSTM is used to process the temporal features of time series data. The loss function of the model is a cross entropy loss function, and the formula is: Among them, y i is the true label, is the probability predicted by the model.

6. A rice growth period recognition system based on edge computing, characterized in that: include Step 1: Edge nodes are deployed in farmland environments and have data processing and storage functions. They are responsible for data collection, preprocessing, and reasoning analysis. Step 2: The multimodal sensor module is installed in the farmland to collect multi-source data related to the rice growth period in real time and transmit it to the edge node; Step 3: Data processing module, deployed in edge nodes, for preprocessing and feature extraction of sensor data; Step 4: The deep learning model runs in the edge node and combines the multi-scale convolutional neural network (MSCNN) and the long short-term memory network (LSTM) to perform real-time recognition of the rice growth period; Step 5: Backend management platform, connects edge nodes through wireless communication, remotely monitors, stores data and provides decision support.

7. The method and system for identifying rice growth period based on edge computing according to claim 6, characterized in that: The edge nodes in the system use embedded processors, have local data storage and processing capabilities, and support real-time reasoning and remote communication.

8. The rice growth period identification method and system based on edge computing according to claim 6, characterized in that: The multimodal sensor module in the system includes a spectral imaging sensor for acquiring spectral reflectance information of rice to assist in the identification of the growth period.

9. The rice growth period identification method and system based on edge computing according to claim 6, characterized in that: The background management platform supports online model updates and regularly updates weights and adjusts parameters of deep learning models to improve recognition accuracy and adaptability.

Images