A Smart Agriculture Crop Growth Analysis System Based on Big Data

Abstract

The application discloses a kind of big data-based wisdom agricultural crop growth analysis systems, and initial crop data are acquired by sensor array;Wavelet denoising-Kalman filter coupling algorithm is used to carry out noise reduction processing to initial crop data, image data is extracted based on the improved MobileNetV3 network, and crop morphological feature vector is output;Through LSTM, the correlation analysis of environment and physiology is carried out to time sequence network, the dynamic relationship between learning parameter and photosynthetic efficiency is learned, and crop state prediction model is established;The denoising crop data are input into prediction model for prediction, and growth health score is output;According to growth health score, crop health grade is judged, and crop growth intervention strategy is generated according to the judgment result.Can more accurately predict the growth health score of crop, improve resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of crop growth analysis technology, and in particular to a smart agricultural crop growth analysis system based on big data. Background Technology

[0002] Currently, data collection in agricultural production faces numerous challenges. Traditional feature extraction and prediction models have limited ability to uncover crop growth patterns and cannot effectively handle the integration of image data with environmental and physiological time-series data. Using convolutional neural networks alone is insufficient to capture the dynamic impact of environmental factors on crop growth, while relying solely on time-series models fails to fully utilize crop morphological features. Furthermore, existing crop health assessment and intervention strategies lack precision, often employing a one-size-fits-all management approach that fails to meet the personalized and refined growth needs of crops, leading to resource waste and yield losses. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a smart agricultural crop growth analysis system based on big data.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned intelligent agricultural crop growth analysis system based on big data:

[0005] The data acquisition and processing module is used to acquire image data, environmental data and physiological data of crops through a sensor array to obtain multi-dimensional crop data, and to perform data preprocessing on the multi-dimensional crop data to obtain initial crop data.

[0006] The crop data denoising module is used to denoise the initial crop data using a wavelet denoising-Kalman filter coupled algorithm to obtain denoised crop data.

[0007] The prediction model building module is used to extract features from image data based on the improved Mobil eNetV3 network and output crop morphological feature vectors; a temporal network is constructed through LSTM to perform correlation analysis between environment and physiology, learn the dynamic relationship between parameters and photosynthetic efficiency, and establish a Mobil eNetV3-LSTM crop state prediction model.

[0008] The crop status recognition module is used to input the denoised crop data into the Mobil eNetV3-LSTM crop status prediction model for prediction and output a growth health score.

[0009] The crop growth intervention module is used to determine the crop health level based on the growth health score and generate crop growth intervention strategies based on the determination results.

[0010] Furthermore, in the aforementioned smart agricultural crop growth analysis system based on big data, the data acquisition and processing module includes the following sub-modules:

[0011] The conversion submodule is used to perform grayscale conversion on image data in multidimensional crop data, converting multispectral images into grayscale images.

[0012] The enhancement submodule is used to enhance the contrast of image data using histogram equalization to obtain processed image data.

[0013] The unified submodule is used to convert the format of environmental and physiological data, converting signals output from different sensors into a unified digital format.

[0014] The resulting submodule is used to detect outliers using the Z-score method and to repair outliers exceeding three times the standard deviation using adjacent data interpolation, thus obtaining the initial crop data.

[0015] Furthermore, in the aforementioned smart agricultural crop growth analysis system based on big data, the crop data noise reduction module includes the following sub-modules:

[0016] The alignment submodule is used to divide the continuous initial crop data stream into data blocks according to the data type and timestamp of the initial crop data, and then align the divided data blocks by time to obtain aligned crop data.

[0017] The decomposition submodule is used to perform wavelet transform on each data block in the aligned crop data using the Daubechies wavelet function, decomposing the original signal into different frequency levels to obtain the decomposed crop data.

[0018] The filtering submodule is used to analyze the amplitude distribution of the decomposed crop data, estimate the noise level based on statistical characteristics, and filter the data according to a threshold to obtain filtered crop data.

[0019] A submodule is obtained to synthesize the approximation coefficients and detail coefficients of the screened crop data through inverse wavelet transform to obtain a signal block after wavelet denoising, thus obtaining filtered crop data.

[0020] Furthermore, in the aforementioned smart agricultural crop growth analysis system based on big data, the crop data noise reduction module further includes the following sub-modules:

[0021] Define a submodule for building a dynamic model based on environmental parameters such as ambient temperature, humidity, and light intensity, and physiological parameters such as soil moisture and stem flow rate, and define process noise and observation noise;

[0022] The prediction submodule is used to predict the state value of the parameter and its uncertainty at the current moment based on the best estimate at the previous moment and the defined dynamic model.

[0023] The fusion submodule is used to obtain the actual observation value of the corresponding data channel of the filtered crop data at the current time, and to fuse the predicted value and the observed value according to the covariance to obtain the optimal estimate.

[0024] The output submodule is used to take the optimal estimate output of the Kalman filter at the current time as the final denoising result of the model, and obtain denoised crop data.

[0025] Furthermore, in the aforementioned smart agricultural crop growth analysis system based on big data, the prediction model building module includes the following sub-modules:

[0026] The analysis submodule is used to analyze the input image based on the MobileNetV3 network and discard redundant visual information that is irrelevant to the crop state.

[0027] The adjustment submodule is used to adjust the feature extraction layer in the MobileNetV3 network to extract crop growth characteristics including at least leaf size, color, texture uniformity, plant density, disease and pest spots, and chlorophyll expression.

[0028] The extraction submodule is used to extract high-dimensional crop morphological feature vectors that characterize the current crop appearance, growth status, potential health problems, and other aspects of the crop.

[0029] Furthermore, in the aforementioned smart agricultural crop growth analysis system based on big data, the prediction model building module further includes the following sub-modules:

[0030] The input submodule is used to input environmental and physiological time-series data from the denoised crop data into the LSTM to construct the time-series network;

[0031] The learning submodule is used to construct a temporal network using LSTM to analyze the characteristic parameters that are directly related to photosynthetic efficiency, such as light intensity, CO2, stomatal conductance index, and chlorophyll level. It learns the evolution of these parameters over time and how they interact with each other.

[0032] The integration submodule is used to integrate and encode the environmental physiological history information and dynamic correlation information within the entire time window into a context state vector containing time-series information.

[0033] Furthermore, in the aforementioned smart agricultural crop growth analysis system based on big data, the crop growth intervention module includes the following sub-modules:

[0034] The matching submodule is used to determine the health level of crops based on the growth health score, establish an agricultural expert knowledge base, and match corresponding intervention strategies from the knowledge base based on the crop health level determination results and current environmental and physiological data.

[0035] Furthermore, in a method for implementing a smart agricultural crop growth analysis system based on big data, the method includes the following steps:

[0036] Image data, environmental data, and physiological data of crops are acquired through a sensor array to obtain multi-dimensional crop data. The multi-dimensional crop data is then preprocessed to obtain initial crop data.

[0037] The initial crop data was denoised using a wavelet denoising-Kalman filter coupled algorithm to obtain denoised crop data.

[0038] Based on the improved MobileNetV3 network, feature extraction is performed on image data to output crop morphology feature vectors; a temporal network is constructed using LSTM to perform correlation analysis between environment and physiology, learn the dynamic relationship between parameters and photosynthetic efficiency, and establish a MobileNetV3-LSTM crop status prediction model.

[0039] The denoised crop data is input into the MobileNetV3-LSTM crop status prediction model for prediction, and a growth health score is output.

[0040] The crop health level is determined based on the growth health score, and crop growth intervention strategies are generated based on the determination results.

[0041] Furthermore, in a method for implementing a smart agricultural crop growth analysis system based on big data, the method includes the following steps:

[0042] Grayscale conversion is performed on image data in multidimensional crop data to convert multispectral images into grayscale images;

[0043] Histogram equalization is used to enhance the contrast of image data, resulting in processed image data.

[0044] Perform format conversion on environmental and physiological data, converting signals output from different sensors into a unified digital format;

[0045] The Z-score method was used for outlier detection, and adjacent data interpolation was used to repair outliers exceeding three standard deviations to obtain initial crop data.

[0046] Furthermore, in a method for implementing a smart agricultural crop growth analysis system based on big data, the method includes the following steps:

[0047] Based on the data type and timestamp of the initial crop data, the continuous initial crop data stream is divided into data blocks, and the divided data blocks are time-aligned to obtain aligned crop data.

[0048] The Daubechies wavelet function is used to perform wavelet transform on each data block in the aligned crop data to decompose the original signal into different frequency levels, thus obtaining the decomposed crop data.

[0049] Analyze the amplitude distribution of the decomposed crop data, estimate the noise level based on statistical characteristics, and filter the data according to the threshold to obtain filtered crop data.

[0050] The approximation coefficients and detail coefficients of the screened crop data are synthesized by inverse wavelet transform to obtain a signal block after wavelet denoising, thus obtaining the filtered crop data.

[0051] Its beneficial effects are as follows: 1. Compared with traditional single-sensor data acquisition and simple data cleaning methods, this solution collects more comprehensive and accurate data, providing a reliable data foundation for subsequent analysis and enabling the data analysis results to more accurately reflect the growth status of crops. 2. Compared with a single model, it has a significant improvement in prediction accuracy, and can more accurately predict the growth health score of crops, providing strong support for agricultural production decisions. 3. It can provide personalized management plans based on the actual growth status of crops, achieving precision irrigation, precision fertilization, and precision pest and disease control, effectively improving resource utilization efficiency, reducing production costs, reducing environmental pollution, and simultaneously improving crop yield and quality. Attached Figure Description

[0052] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0053] Figure 1 This is a schematic diagram of the first embodiment of a smart agricultural crop growth analysis system based on big data according to the present invention;

[0054] Figure 2 This is a schematic diagram of a second embodiment of a smart agricultural crop growth analysis system based on big data according to the present invention;

[0055] Figure 3 This is a schematic diagram of a third embodiment of a smart agricultural crop growth analysis system based on big data, as described in this invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0057] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0058] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, a smart agricultural crop growth analysis system based on big data includes the following modules:

[0059] The data acquisition and processing module is used to acquire image data, environmental data and physiological data of crops through a sensor array to obtain multi-dimensional crop data, and to perform data preprocessing on the multi-dimensional crop data to obtain initial crop data.

[0060] Specifically, this embodiment also includes a conversion submodule, which is used to perform grayscale conversion on the image data in the multidimensional crop data, converting the multispectral image into a grayscale image;

[0061] The enhancement submodule is used to enhance the contrast of image data using histogram equalization to obtain processed image data.

[0062] The unified submodule is used to convert the format of environmental and physiological data, converting signals output from different sensors into a unified digital format.

[0063] The resulting submodule is used to detect outliers using the Z-score method and to repair outliers exceeding three times the standard deviation using adjacent data interpolation, thus obtaining the initial crop data.

[0064] Specifically,

[0065] I. Sensor Array Setup and Data Acquisition

[0066] 1. Sensor Types and Deployment

[0067] Image data acquisition: Deploy multispectral drone cameras over farmland, equipped with RGB, near-infrared and thermal infrared sensors, to acquire image data of crops under different spectra.

[0068] Environmental data collection:

[0069] Soil data: Soil moisture sensor, soil temperature sensor, soil pH sensor and soil nutrient (nitrogen, phosphorus and potassium) sensor are installed.

[0070] Meteorological data: Meteorological stations are installed around farmland, equipped with wind speed sensors, wind direction sensors, air temperature sensors, air humidity sensors, light intensity sensors, and rainfall sensors.

[0071] Physiological data acquisition: Chlorophyll fluorescence sensors and stem diameter sensors were used for crop leaves.

[0072] 1. Data Preprocessing

[0073] Image data preprocessing: First, grayscale conversion is performed to transform the multispectral image into a grayscale image for easier subsequent processing. Then, contrast adjustment is applied using histogram equalization to enhance image contrast and make crop details more apparent. Next, median filtering is used to remove noise from the image while preserving edge information. Finally, image cropping and normalization are performed to unify the image size to 224×224 for easier input into the subsequent neural network model.

[0074] Environmental and physiological data preprocessing: The collected data underwent format conversion, transforming signals from different sensors into a unified digital format. Units were standardized, converting temperature to degrees Celsius, humidity to percentages, and light intensity to lux, etc. The Z-score method was used for outlier detection; data exceeding three standard deviations were considered outliers and corrected using adjacent data interpolation. Finally, the data was normalized, scaling the data range to [0,1] to improve model training efficiency and accuracy.

[0075] The crop data denoising module is used to denoise the initial crop data using a wavelet denoising-Kalman filter coupled algorithm to obtain denoised crop data.

[0076] Specifically, this embodiment also includes an alignment submodule, which is used to divide the continuous initial crop data stream into data blocks according to the data type and timestamp of the initial crop data, and to perform time alignment on the divided data blocks to obtain aligned crop data;

[0077] The decomposition submodule is used to perform wavelet transform on each data block in the aligned crop data using the Daubechies wavelet function, decomposing the original signal into different frequency levels to obtain the decomposed crop data.

[0078] The filtering submodule is used to analyze and decompose the amplitude distribution of crop data, estimate the noise level based on statistical characteristics, and filter the data according to the threshold to obtain filtered crop data.

[0079] The submodule is used to synthesize the approximation coefficients and detail coefficients of the filtered crop data through inverse wavelet transform, and obtain the signal block after wavelet denoising, thus obtaining the filtered crop data.

[0080] Define a submodule for building a dynamic model based on environmental parameters such as ambient temperature, humidity, and light intensity, and physiological parameters such as soil moisture and stem flow rate, and define process noise and observation noise;

[0081] The prediction submodule is used to predict the state value of the parameter and its uncertainty at the current moment based on the best estimate at the previous moment and the defined dynamic model.

[0082] The fusion submodule is used to obtain the actual observation value of the corresponding data channel of the filtered crop data at the current time, and to fuse the predicted value and the observed value according to the covariance to obtain the optimal estimate.

[0083] The output submodule is used to take the optimal estimate output of the Kalman filter at the current time as the final denoising result of the model, and obtain denoised crop data.

[0084] Specifically,

[0085] Preprocessing and data preparation:

[0086] Data chunking / grouping: Based on data type and timestamps, the continuous initial crop data stream is divided into logical blocks or windows that are more suitable for processing. For example, image data is divided by frame or time series, environmental data (temperature, humidity, light) is grouped by fixed time intervals, and physiological data (stem flow, leaf temperature) is also grouped by time series.

[0087] Data alignment (optional but recommended): Ensure that data from different sensors are synchronized at the same point in time, which facilitates subsequent correlation analysis and the implementation of coupling algorithms (although coupling is mainly performed within their respective channels).

[0088] Wavelet denoising (for non-stationary noise):

[0089] Choosing wavelet basis functions: Based on the characteristics of the data type (texture features of images, abrupt changes in environmental data), select appropriate wavelet functions (Daubechies, Haar, Symlets, etc.) as analysis tools. The principle is to choose a wavelet that can effectively capture the main features of the target signal.

[0090] Multi-scale analysis: Wavelet transform is performed on each data block (especially suitable for image data and non-stationary time-series data containing significant instantaneous changes). This process decomposes the original signal into different levels of "resolution" or "frequency" (approximation coefficients - representing the overall signal profile; detail coefficients - representing details and noise at different scales).

[0091] Noise threshold estimation and filtering: The amplitude distribution is analyzed at each detailed level obtained from the decomposition. The noise level is estimated based on specific criteria (statistical methods), and an appropriate threshold is set. Coefficients below the threshold (considered noise-dominant) are significantly attenuated or zeroed, while coefficients above the threshold (considered signal-dominant) are preserved or reduced.

[0092] Signal reconstruction: The processed approximation coefficients and detail coefficients are synthesized using inverse wavelet transform to obtain a signal block after wavelet denoising. At this point, granular noise and isolated outliers in the image are effectively suppressed, and instantaneous spike interference in the time series data is also smoothed.

[0093] Kalman filtering (for time-series noise and model prediction):

[0094] Establish a dynamic model: Define a simplified dynamic model for the time-series data channels (environmental parameters such as ambient temperature, humidity, and light intensity; physiological parameters such as soil moisture and stem flow rate). This model describes how the signal (temperature) changes over time (and defines the statistical characteristics of process noise (model inaccuracy) and observation noise (sensor error). It does not predict the crop state itself, but rather describes the evolutionary trend of the parameter itself.

[0095] Prediction phase: Based on the best estimate (state value and uncertainty) from the previous moment and the defined dynamic model, predict the state value and uncertainty of the parameter at the current moment.

[0096] Update phase: Receive the actual observation value (i.e. the part of the data that has been pre-filtered by wavelet) of the corresponding data channel from step 2 (after initial wavelet denoising) at the current time.

[0097] Optimal fusion: The predicted and observed values ​​are weighted and fused according to their respective uncertainties (covariance). Sources with lower uncertainty are given higher weights. This fusion produces the optimal estimate of the parameter's state at the current moment and updates the assessment of the uncertainty of that estimate.

[0098] Output: The optimal estimate from the Kalman filter at the current time step is used as the final denoising result for this time-series channel. This process is particularly effective for parameters that change slowly over time, effectively filtering out stationary random noise and providing a smooth and stable estimate.

[0099] Coupling mechanism (key point):

[0100] Sequence and Interface: Coupling is reflected in the processing flow. Step 2 (wavelet denoising), as the front-end processing module, first operates on all initial data, and is particularly good at quickly removing transient and high-frequency noise, providing relatively "clean" observations for Kalman filtering.

[0101] Processing division of labor: For image data and non-stationary components in time-series data (sudden disturbances in environmental data), wavelet denoising is the main method, while Kalman filtering may only be fine-tuned or not involved at all (depending on the specific implementation).

[0102] Adaptive / Feedback (Advanced): In advanced implementations, the residuals of the Kalman filter output (the difference between the prediction and the final estimate) or the covariance of the estimation error may be indirectly fed back to adjust the threshold selection strategy in subsequent wavelet denoising (increase the threshold, trust the model; or decrease the threshold, rely more on new observations), enabling the entire coupled algorithm to adapt more intelligently to different noise conditions.

[0103] Output integration: For the data in each channel:

[0104] If the channel (image) primarily benefits from wavelet denoising, the output of wavelet denoising may be used directly as the final denoised output of that channel, or it may undergo slight Kalman smoothing.

[0105] If it is a core timing channel (temperature, humidity, specific physiological parameters), then the output of the Kalman filter in step 3 is the final noise reduction output of that channel.

[0106] After all channels are processed, they are integrated to form the final denoised crop dataset, which can be used by subsequent MobileNetV3-LSTM models.

[0107] The prediction model building module is used to extract features from image data based on the improved Mobil eNetV3 network and output crop morphological feature vectors; a temporal network is constructed through LSTM to perform correlation analysis between environment and physiology, learn the dynamic relationship between parameters and photosynthetic efficiency, and establish a Mobil eNetV3-LSTM crop state prediction model.

[0108] Specifically, this embodiment also includes an analysis submodule, which is used to analyze the input image based on the Mobil eNetV3 network and discard redundant visual information that is irrelevant to the crop state;

[0109] The adjustment submodule is used to adjust the feature extraction layer in the MobileNetV3 network to extract crop growth characteristics including at least leaf size, color, texture uniformity, plant density, disease and pest spots, and chlorophyll expression.

[0110] The extraction submodule is used to extract high-dimensional crop morphological feature vectors that characterize the current crop appearance, growth status, potential health problems, and other aspects of the crop.

[0111] The input submodule is used to input environmental and physiological time-series data from the denoised crop data into the LSTM to construct the time-series network;

[0112] The learning submodule is used to construct a temporal network using LSTM to analyze the characteristic parameters that are directly related to photosynthetic efficiency, such as light intensity, CO2, stomatal conductance index, and chlorophyll level. It learns the evolution of these parameters over time and how they interact with each other.

[0113] The integration submodule is used to integrate and encode the environmental physiological history information and dynamic correlation information within the entire time window into a context state vector containing time-series information.

[0114] Specifically,

[0115] Image morphological feature depth extraction (based on an improved Mobil eNetV3 network):

[0116] Input: Denoising-reduced crop image data (high-quality crop images from multiple angles and time periods).

[0117] Model selection and improvement philosophy: The MobileNetV3 network is adopted as the backbone. This model, due to its advantages of high computational efficiency (lightweight design) and small model size while maintaining high accuracy, is very suitable for deployment on edge devices or smart agriculture IoT nodes where resources may be limited. Key improvements may include:

[0118] Input adaptation for agricultural images: The input layer may be adjusted (to suit a specific resolution or number of channels).

[0119] Feature layer optimization: Adjust or enhance the deep feature extraction part of the network to better focus on morphological features closely related to crop growth status (visible structures such as leaf size, color, texture uniformity, plant density, disease and pest spots, chlorophyll expression, etc.).

[0120] Attention mechanism integration: A lightweight attention module is introduced, enabling the model to more intelligently "focus" on areas in the image that are more critical to the health status (new leaves, diseased areas, leaf veins, etc.) and suppress background interference.

[0121] Function: The Mobil eNetV3 network acts as a powerful "morphological information miner." It automatically analyzes input images, discards redundant visual information irrelevant to crop status, and extracts and refines high-dimensional crop morphological feature vectors that best represent the current crop appearance, growth status, and potential health problems.

[0122] Dynamic correlation analysis between temporal environment and physiological parameters (based on LSTM temporal network):

[0123] Input: Noise-reduced environmental time-series data (historical and current temperature, humidity, light intensity, CO2 concentration, etc.) and physiological time-series data (soil moisture, stem sap flow, leaf temperature, nutrient solution indicators, etc., if applicable) from step 2. This data is a data stream organized in time series format.

[0124] Model selection: A Long Short-Term Memory (LSTM) network was used to construct the time series model. The strength of LSTM lies in its unique "gating mechanism" structure (forget gate, input gate, output gate), which gives it a powerful ability to capture long-term dependencies and effectively learn patterns and trends in sequence data that span long time intervals.

[0125] Function and Analysis Objectives: The LSTM network acts as a "temporal pattern and correlation interpreter." Its core task is:

[0126] Modeling complex dynamic relationships: Learning the complex, nonlinear dynamic relationships between environmental parameters (light, temperature), physiological parameters, and crucial environmental parameters and key physiological responses (especially photosynthesis-related parameters).

[0127] Learning the dynamics of photosynthetic efficiency: Focusing on physiological / environmental parameters (light intensity, CO2, stomatal conductance index, chlorophyll level indicators, etc.) that are directly or indirectly related to photosynthetic efficiency, LSTM learns how these parameters evolve over time and how they interact to implicitly or explicitly model the dynamic response of the core processes of photosynthesis. This is the cornerstone of assessing crop health and productivity potential.

[0128] Temporal context compression: The environmental and physiological history information and their dynamic relationships within the entire time window are condensed and encoded into a context state vector rich in temporal information (which is the output of the LSTM hidden layer).

[0129] Model fusion and state prediction model building (MobileNetV3-LSTM integration):

[0130] Feature fusion: The high-dimensional static morphological feature vector output from step 1 (representing the current "look") is fused with the context state vector rich in spatiotemporal correlation information output from step 2 (representing "the environment experienced" and "physiological response"). Fusion is performed in the later layers of the model.

[0131] Integration Strategy: The design of the integration method is crucial. Possible approaches include:

[0132] Concatenation: Simply concatenates two vectors into a longer composite feature vector.

[0133] Prediction Layer Construction: On top of the fused features, the final prediction layer (a fully connected network) is constructed. This layer is responsible for learning and inferring a growth health score (a learnable quantitative value) that best reflects the overall health status and growth momentum of the crop based on the fused comprehensive information (current morphological status + environmental physiological evolution process and dynamic relationships).

[0134] The crop status recognition module is used to input denoised crop data into the MobileNetV3-LSTM crop status prediction model for prediction and output a growth health score.

[0135] Specifically, this embodiment also includes inputting image data from the denoised crop data into an improved MobileNetV3 network to extract crop morphological feature vectors; inputting environmental and physiological data into an LSTM temporal network to learn the dynamic relationship between parameters and photosynthetic efficiency, and outputting a predicted value of photosynthetic efficiency. Then, the outputs of both are concatenated and input into a fully connected layer to calculate a crop growth health score, ranging from 0 to 100 points, with higher scores indicating healthier crop growth.

[0136] The crop growth intervention module is used to determine the health level of crops based on the growth health score and to generate crop growth intervention strategies based on the determination results.

[0137] Specifically, this embodiment also includes a matching submodule, which is used to determine the health level of crops based on growth health scores, establish an agricultural expert knowledge base, and match corresponding intervention strategies from the knowledge base based on the crop health level determination results and current environmental and physiological data.

[0138] Specifically,

[0139] I. Health Level Assessment Criteria

[0140] Based on the growth health score, the health level of crops is divided into four levels:

[0141] Health Level 1 (Excellent, 90-100 points): Crops are growing well, all indicators are normal, and no special intervention is required.

[0142] Health Level 2 (Good, 70-89 points): There are minor problems with crop growth. Close attention should be paid to changes in environmental and physiological indicators, and minor adjustments to management measures should be made as needed.

[0143] Health Level 3 (Medium, 50-69 points): There are some problems with crop growth, which may be affected by environmental factors or pests and diseases. Targeted intervention measures are needed, such as adjusting irrigation and fertilization, and spraying pesticides.

[0144] Health Level 4 (Poor, 0-49 points): Crop growth is severely damaged and may face the risk of reduced yield or even crop failure. Immediate emergency intervention measures are required, such as pest and disease control, soil improvement, and adjustment of planting density.

[0145] II. Intervention Strategy Generation

[0146] An agricultural expert knowledge base is established, containing information on optimal environmental conditions for different crop growth stages, ranges of physiological indicators, symptoms and control methods for common pests and diseases, and optimal fertilization and irrigation plans. Based on the crop's health level assessment, combined with current environmental and physiological data, corresponding intervention strategies are matched from the knowledge base. For example, when the health level is 3, if soil moisture is detected to be below the optimal range, a strategy to increase irrigation is generated; if pest and disease symptoms are found on the leaves, a strategy to spray appropriate pesticides is generated. Intervention strategies are presented in a visual manner, such as generating irrigation plans, fertilization schemes, and pest and disease control schedules, facilitating operation and implementation by agricultural managers. Simultaneously, the effectiveness of the intervention strategies is tracked and evaluated, and the knowledge base is updated and optimized based on feedback results to improve the effectiveness and relevance of the intervention strategies.

[0147] Its beneficial effects are as follows: 1. Compared with traditional single-sensor data acquisition and simple data cleaning methods, this solution collects more comprehensive and accurate data, providing a reliable data foundation for subsequent analysis and enabling the data analysis results to more accurately reflect the growth status of crops. 2. Compared with a single model, it has a significant improvement in prediction accuracy, and can more accurately predict the growth health score of crops, providing strong support for agricultural production decisions. 3. It can provide personalized management plans based on the actual growth status of crops, achieving precision irrigation, precision fertilization, and precision pest and disease control, effectively improving resource utilization efficiency, reducing production costs, reducing environmental pollution, and simultaneously improving crop yield and quality.

[0148] Please see Figure 2 In a smart agricultural crop growth analysis system based on big data, the data acquisition and processing module includes the following sub-modules:

[0149] The conversion submodule is used to perform grayscale conversion on image data in multidimensional crop data, converting multispectral images into grayscale images.

[0150] The enhancement submodule is used to enhance the contrast of image data using histogram equalization to obtain processed image data.

[0151] The unified submodule is used to convert the format of environmental and physiological data, converting signals output from different sensors into a unified digital format.

[0152] The resulting submodule is used to detect outliers using the Z-score method and to repair outliers exceeding three times the standard deviation using adjacent data interpolation, thus obtaining the initial crop data.

[0153] Please see Figure 3 In a smart agricultural crop growth analysis system based on big data, the crop data noise reduction module includes the following sub-modules:

[0154] The alignment submodule is used to divide the continuous initial crop data stream into data blocks according to the data type and timestamp of the initial crop data, and then align the divided data blocks by time to obtain aligned crop data.

[0155] The decomposition submodule is used to perform wavelet transform on each data block in the aligned crop data using the Daubechies wavelet function, decomposing the original signal into different frequency levels to obtain the decomposed crop data.

[0156] The filtering submodule is used to analyze and decompose the amplitude distribution of crop data, estimate the noise level based on statistical characteristics, and filter the data according to the threshold to obtain filtered crop data.

[0157] The submodule is used to synthesize the approximation coefficients and detail coefficients of the filtered crop data through inverse wavelet transform, and obtain the signal block after wavelet denoising, thus obtaining the filtered crop data.

[0158] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A smart agricultural crop growth analysis system based on big data, characterized in that, The intelligent agricultural crop growth analysis system includes the following modules: The data acquisition and processing module is used to acquire image data, environmental data and physiological data of crops through a sensor array to obtain multi-dimensional crop data, and to perform data preprocessing on the multi-dimensional crop data to obtain initial crop data. The crop data noise reduction module is used to divide the continuous initial crop data stream into data blocks according to the data type and timestamp of the initial crop data, and to perform time alignment on the divided data blocks to obtain aligned crop data. The Daubechies wavelet function is used to perform wavelet transform on each data block in the aligned crop data to decompose the original signal into different frequency levels, thus obtaining the decomposed crop data. The amplitude distribution of the decomposed crop data is analyzed, the noise level is estimated based on statistical characteristics, and the data is filtered according to the threshold to obtain filtered crop data. The approximation coefficients and detail coefficients of the filtered crop data are synthesized by wavelet inverse transform to obtain a signal block after wavelet denoising, thus obtaining filtered crop data. A dynamic model is established based on environmental parameters such as ambient temperature, humidity, and light intensity, and physiological parameters such as soil moisture and stem flow rate, and process noise and observation noise are defined. Based on the best estimate from the previous moment and the well-defined dynamic model, predict the state value of the parameter at the current moment and its uncertainty. Obtain the actual observation value of the corresponding data channel of the filtered crop data at the current time, and then weight and fuse the predicted value and the observed value according to the covariance to obtain the optimal estimate; use the optimal estimate output of the Kalman filter at the current time as the final denoising result of the model to obtain the denoised crop data; The prediction model building module is used to extract features from image data based on the improved MobileNetV3 network and output crop morphology feature vectors; a temporal network is constructed through LSTM to perform correlation analysis between environment and physiology, learn the dynamic relationship between parameters and photosynthetic efficiency, and establish a MobileNetV3-LSTM crop status prediction model. The crop status recognition module is used to input the denoised crop data into the MobileNetV3-LSTM crop status prediction model for prediction and output a growth health score. The crop growth intervention module is used to determine the crop health level based on the growth health score and generate crop growth intervention strategies based on the determination results.

2. The smart agricultural crop growth analysis system based on big data as described in claim 1, characterized in that, The data acquisition and processing module includes the following sub-modules: The conversion submodule is used to perform grayscale conversion on image data in multidimensional crop data, converting multispectral images into grayscale images. The enhancement submodule is used to enhance the contrast of image data in grayscale images using histogram equalization to obtain processed image data. The unified submodule is used to convert the format of environmental and physiological data, converting signals output from different sensors into a unified digital format. The resulting submodule is used to detect outliers using the Z-score method and to repair outliers exceeding three times the standard deviation using adjacent data interpolation, thus obtaining the initial crop data.

3. The smart agricultural crop growth analysis system based on big data as described in claim 1, characterized in that, The prediction model building module includes the following sub-modules: The analysis submodule is used to analyze the input image based on the MobileNetV3 network and discard redundant visual information that is irrelevant to the crop state. The adjustment submodule is used to adjust the feature extraction layer in the MobileNetV3 network to extract crop growth characteristics including at least leaf size, color, texture uniformity, plant density, disease and pest spots, and chlorophyll expression. The extraction submodule is used to extract high-dimensional crop morphological feature vectors that characterize the current crop appearance, growth status, potential health problems, and other aspects of the crop.

4. The smart agricultural crop growth analysis system based on big data as described in claim 1, characterized in that, The prediction model building module also includes the following sub-modules: The input submodule is used to input environmental and physiological time-series data from the denoised crop data into the LSTM to construct the time-series network; The learning submodule is used to construct a temporal network using LSTM to analyze the characteristic parameters that are directly related to photosynthetic efficiency, such as light intensity, CO2, stomatal conductance index, and chlorophyll level. It learns the evolution of these parameters over time and how they interact with each other. The integration submodule is used to integrate and encode the environmental physiological history information and dynamic correlation information within the entire time window into a context state vector containing time-series information.

5. The smart agricultural crop growth analysis system based on big data as described in claim 1, characterized in that, The crop growth intervention module includes the following sub-modules: The matching submodule is used to determine the health level of crops based on the growth health score, establish an agricultural expert knowledge base, and match corresponding intervention strategies from the knowledge base based on the crop health level determination results and current environmental and physiological data.

6. A method for implementing a smart agricultural crop growth analysis system based on big data as described in claim 1, characterized in that, The method includes the following steps: Image data, environmental data, and physiological data of crops are acquired through a sensor array to obtain multi-dimensional crop data. The multi-dimensional crop data is then preprocessed to obtain initial crop data. Based on the data type and timestamp of the initial crop data, the continuous initial crop data stream is divided into data blocks. These blocks are then time-aligned to obtain aligned crop data. The Daubechies wavelet function is used to perform a wavelet transform on each data block in the aligned crop data, decomposing the original signal into different frequency levels to obtain decomposed crop data. The amplitude distribution of the decomposed crop data is analyzed, and the noise level is estimated based on statistical characteristics. Data is then filtered according to a threshold to obtain filtered crop data. The approximation coefficients and detail coefficients of the filtered crop data are synthesized using inverse wavelet transform to obtain... After wavelet denoising, the signal block is used to obtain filtered crop data. A dynamic model is established based on environmental parameters (temperature, humidity, light intensity) and physiological parameters (soil moisture, stem flow rate), and process noise and observation noise are defined. Based on the best estimate from the previous moment and the defined dynamic model, the state value and uncertainty of the parameter at the current moment are predicted. The actual observation value of the corresponding data channel of the filtered crop data at the current moment is obtained, and the predicted value and the observed value are weighted and fused according to covariance to obtain the optimal estimate. The optimal estimate output of the Kalman filter at the current moment is used as the final denoising result of the model, resulting in denoised crop data. Based on the improved MobileNetV3 network, feature extraction is performed on image data to output crop morphology feature vectors; a temporal network is constructed using LSTM to perform correlation analysis between environment and physiology, learn the dynamic relationship between parameters and photosynthetic efficiency, and establish a MobileNetV3-LSTM crop status prediction model. The denoised crop data is input into the MobileNetV3-LSTM crop status prediction model for prediction, and a growth health score is output. The crop health level is determined based on the growth health score, and crop growth intervention strategies are generated based on the determination results.

7. A method for implementing a smart agricultural crop growth analysis system based on big data as described in claim 6, characterized in that, The method includes the following steps: Grayscale conversion is performed on image data in multidimensional crop data to convert multispectral images into grayscale images; Histogram equalization is used to enhance the contrast of image data, resulting in processed image data. Perform format conversion on environmental and physiological data, converting signals output from different sensors into a unified digital format; The Z-score method was used for outlier detection, and adjacent data interpolation was used to repair outliers exceeding three standard deviations to obtain initial crop data.

8. A method for implementing a smart agricultural crop growth analysis system based on big data as described in claim 7, characterized in that, The method includes the following steps: Based on the data type and timestamp of the initial crop data, the continuous initial crop data stream is divided into data blocks, and the divided data blocks are time-aligned to obtain aligned crop data. The Daubechies wavelet function is used to perform wavelet transform on each data block in the aligned crop data to decompose the original signal into different frequency levels, thus obtaining the decomposed crop data. Analyze the amplitude distribution of the decomposed crop data, estimate the noise level based on statistical characteristics, and filter the data according to a threshold to obtain filtered crop data. The approximation coefficients and detail coefficients of the screened crop data are synthesized by inverse wavelet transform to obtain a signal block after wavelet denoising, thus obtaining the filtered crop data.

Images