A Distributed Method and System for Monitoring Soil Nitrogen, Phosphorus, and Potassium Status

Abstract

This invention discloses a distributed method and system for monitoring the state of nitrogen, phosphorus, and potassium (NPK) in soil. The method includes: adaptively calibrating the NPK sensors at each sensing node to collect soil NPK potential data in real time; performing improved adaptive wavelet threshold denoising; extracting features from the denoised feature data using an improved lightweight convolutional neural network; processing the multidimensional deep feature vectors of NPK using cross-node collaborative filtering and anomaly removal algorithms; retrieving NPK concentration values ​​from regional consistency feature data using an NPK inversion model based on nonlinear coupling of environmental factors; and generating a regional soil nutrient heat map. This application improves the convenience and accuracy of soil NPK state monitoring by deploying distributed sensing nodes and combining adaptive calibration, improved adaptive wavelet threshold denoising, improved lightweight convolutional neural networks, cross-node collaborative filtering, and an NPK inversion model based on nonlinear coupling of environmental factors.

Description

Technical Field

[0001] This application relates to the field of soil nitrogen, phosphorus and potassium monitoring technology, and in particular to a distributed soil nitrogen, phosphorus and potassium status monitoring method and system. Background Technology

[0002] Nitrogen, phosphorus, and potassium in soil are essential nutrients for crop growth and development. Accurate monitoring of their content is a core technological support for agricultural environmental monitoring, arable land quality protection, and the development of precision agriculture, and represents an important research direction at the intersection of agricultural engineering and environmental monitoring. With the acceleration of agricultural modernization and the widespread adoption of green agriculture and precision fertilization concepts, large-scale farmland and facility agriculture place higher demands on the timeliness and coverage of soil nutrient monitoring. Related monitoring technologies have become a crucial link between arable land quality control and high-quality, high-yield crops. Currently, soil nutrient monitoring technology is gradually iterating towards intelligence and large-scale operation. Its technological level directly affects the efficiency of agricultural resource utilization and the sustainable development capacity of arable land; therefore, research on soil nitrogen, phosphorus, and potassium monitoring technologies is necessary.

[0003] In the prior art, Chinese patent CN121231463A discloses a soil in-situ nitrogen, phosphorus, and potassium measurement system based on digital colorimetry. It includes a base and a host computer. A monitoring box is mounted on the base via a support column. The monitoring box contains a colorimetric component and a data acquisition component that acts on the colorimetric component. An energy component is mounted on the monitoring box. The data acquisition component is electrically connected to both the host computer and the energy component. A water supply component and an extraction component are mounted on the side wall of the monitoring box. The extraction component is connected to the colorimetric component. By introducing a colorimetric method through the water supply component, extraction component, and colorimetric component, and by using the energy component to support the data acquisition component and collect colorimetric data, which is then transmitted to the host computer for color information extraction, the system can quickly and accurately measure nitrogen, phosphorus, and potassium in the soil without the need for electrode sensors.

[0004] However, the aforementioned existing technologies rely on colorimetric reactions, which require the addition of colorimetric reagents. The operation is cumbersome and time-consuming. The monitoring boxes are deployed in a fixed manner, which cannot achieve large-area distributed in-situ monitoring. They are also susceptible to interference from ambient light, which can lead to measurement deviations. The convenience and accuracy of soil nitrogen, phosphorus and potassium status monitoring need to be improved. Summary of the Invention

[0005] This application provides a distributed method and system for monitoring the state of nitrogen, phosphorus, and potassium in soil, in order to address the problem that the convenience and accuracy of soil nitrogen, phosphorus, and potassium status monitoring in the prior art need to be improved.

[0006] On the one hand, this application provides a distributed method for monitoring the nitrogen, phosphorus, and potassium status of soil, comprising the following steps:

[0007] Step 1: Deploy distributed sensing nodes, perform adaptive calibration on the nitrogen, phosphorus, and potassium sensors of each sensing node, and use the adaptively calibrated distributed sensing nodes to collect soil nitrogen, phosphorus, and potassium potential data in real time.

[0008] Step 2: Improved adaptive wavelet threshold denoising is performed on the soil nitrogen, phosphorus and potassium potential data of each sensing node to obtain denoised feature data.

[0009] Step 3: An improved lightweight convolutional neural network is used to extract features from the noise reduction feature data of each sensing node to obtain a multidimensional deep feature vector of nitrogen, phosphorus and potassium.

[0010] Step four involves processing the nitrogen, phosphorus, and potassium multidimensional deep feature vectors of each sensing node using cross-node collaborative filtering and anomaly removal algorithms to obtain regional consistency feature data.

[0011] Step 5: The nitrogen, phosphorus and potassium concentration values ​​are inverted using a nitrogen, phosphorus and potassium inversion model based on nonlinear coupling of environmental factors, according to the regional consistency characteristic data of each sensing node.

[0012] Step 6: Generate a regional soil nutrient thermogram based on the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node.

[0013] In one possible implementation, in step one, the area to be monitored is divided into square grids of equal area, and a sensing node is set at the center of each grid. Each sensing node integrates a nitrogen, phosphorus, and potassium ion selective electrode, i.e., the nitrogen, phosphorus, and potassium sensor, and also integrates a temperature and humidity sensor and a conductivity sensor.

[0014] Each sensor node uses LoRa decentralized networking communication.

[0015] In one possible implementation, step one, the adaptive calibration includes:

[0016] A real-time correction coefficient is constructed based on the real-time temperature and real-time conductivity of each sensing node, and the corresponding nitrogen, phosphorus and potassium sensors are adaptively calibrated using the real-time correction coefficient.

[0017] In one possible implementation, step two of the improved adaptive wavelet threshold denoising includes:

[0018] An adaptive weighting factor is constructed, and the soil nitrogen, phosphorus, and potassium potential data after wavelet hard thresholding and the soil nitrogen, phosphorus, and potassium potential data after wavelet soft thresholding are weighted and summed based on the adaptive weighting factor to obtain the noise reduction feature data.

[0019] In one possible implementation, in step three, the improved lightweight convolutional neural network integrates depthwise separable convolutional layers and an improved channel attention mechanism layer.

[0020] In one possible implementation, the improved channel attention mechanism layer calculates channel feature weights by combining the Sigmoid activation function with fully connected layer operations, average pooling operations, and max pooling operations.

[0021] In one possible implementation, step four includes:

[0022] A cross-node collaborative filtering algorithm based on adjacent node correction is used to correct the multidimensional deep feature vectors of nitrogen, phosphorus and potassium of each sensing node, thus obtaining cross-node corrected features.

[0023] The 3σ criterion is used to remove outliers from the cross-node correction features to obtain regional consistency feature data.

[0024] In one possible implementation, in step five, the nitrogen, phosphorus, and potassium inversion model based on nonlinear coupling of environmental factors sequentially performs dimensional standardization, environmental-nitrogen, phosphorus, and potassium nonlinear coupling processing, and nitrogen, phosphorus, and potassium differential electrochemical inversion on the regional consistency feature data of each sensing node to obtain nitrogen, phosphorus, and potassium concentration values.

[0025] The environmental-nitrogen, phosphorus, and potassium nonlinear coupling treatment uses a natural exponential cross-coupling formula to couple the dimensionally standardized temperature, humidity, and electrical conductivity with soil nitrogen, phosphorus, and potassium data.

[0026] The differential electrochemical inversion of nitrogen, phosphorus, and potassium is designed with linear-logarithmic response inversion formulas for available nitrogen, nonlinear quadratic response inversion formulas for available phosphorus, and diffusion-type radical-logarithmic response inversion formulas for available potassium.

[0027] In one possible implementation, step six includes:

[0028] Based on the nitrogen, phosphorus, and potassium nutrient requirements of crops, nitrogen, phosphorus, and potassium early warning thresholds are set. Based on the nitrogen, phosphorus, and potassium early warning thresholds and the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node, the early warning level of each sensor node is obtained. A regional soil nutrient heat map is generated based on the early warning level of each sensor node.

[0029] On the one hand, this application provides a distributed soil nitrogen, phosphorus and potassium state monitoring system, which adopts the above-mentioned distributed soil nitrogen, phosphorus and potassium state monitoring method, including: distributed sensing nodes, data noise reduction module, feature extraction module, consistency processing module, concentration inversion module, and heat map generation module.

[0030] The distributed sensing nodes are configured to collect soil nitrogen, phosphorus, and potassium potential data in real time.

[0031] The data denoising module is configured to perform improved adaptive wavelet threshold denoising on the soil nitrogen, phosphorus, and potassium potential data of each sensing node to obtain denoised feature data.

[0032] The feature extraction module is configured to use an improved lightweight convolutional neural network to extract features from the noise-reduced feature data of each sensing node, thereby obtaining a multidimensional deep feature vector of nitrogen, phosphorus, and potassium.

[0033] The consistency processing module is configured to process the nitrogen, phosphorus, and potassium multidimensional deep feature vectors of each sensing node using cross-node collaborative filtering and anomaly removal algorithms to obtain regional consistency feature data.

[0034] The concentration inversion module is configured to use a nitrogen, phosphorus, and potassium inversion model based on nonlinear coupling of environmental factors to invert nitrogen, phosphorus, and potassium concentration values ​​based on the regional consistency characteristic data of each sensor node.

[0035] The heat map generation module is configured to generate a regional soil nutrient heat map based on the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node.

[0036] The distributed soil nitrogen, phosphorus, and potassium status monitoring method and system disclosed in this application have the following advantages:

[0037] By deploying distributed sensing nodes and combining adaptive calibration, improved adaptive wavelet threshold denoising, improved lightweight convolutional neural networks, cross-node collaborative filtering, and a nitrogen, phosphorus, and potassium inversion model based on nonlinear coupling of environmental factors, the convenience and accuracy of soil nitrogen, phosphorus, and potassium status monitoring have been improved.

[0038] By constructing real-time correction coefficients based on the real-time temperature and conductivity of each sensing node, the nitrogen, phosphorus, and potassium sensors are adaptively calibrated to dynamically correct sensor drift and ensure the accuracy of basic monitoring.

[0039] By constructing an adaptive weighting factor, the soil nitrogen, phosphorus, and potassium potential data after wavelet hard thresholding and the soil nitrogen, phosphorus, and potassium potential data after wavelet soft thresholding are weighted and summed based on the adaptive weighting factor, which takes into account both noise reduction effect and signal fidelity, adapts to complex soil environments, and improves the accuracy of the original data.

[0040] The proposed improved lightweight convolutional neural network integrates depthwise separable convolutional layers and an improved channel attention mechanism layer, which is adapted to edge computing terminals. It does not require cloud computing power and can complete feature extraction locally on the terminal, thus improving the convenience of terminal processing.

[0041] The proposed improved channel attention mechanism layer calculates channel feature weights by combining the Sigmoid activation function with fully connected layer operations, average pooling operations, and max pooling operations, accurately extracting the core features of nitrogen, phosphorus, and potassium, and improving feature effectiveness.

[0042] By employing a cross-node collaborative filtering algorithm based on adjacent node correction to correct the multidimensional deep feature vectors of nitrogen, phosphorus, and potassium for each sensing node, outlier removal is performed on the cross-node corrected features, and abnormal data is removed by utilizing regional nutrient correlation, thereby improving the accuracy of regional monitoring data.

[0043] By sequentially performing dimensional standardization, environmental-nitrogen-phosphorus-potassium nonlinear coupling processing, and nitrogen-phosphorus-potassium differential electrochemical inversion on the regional consistency characteristic data of each sensing node, the nitrogen-phosphorus-potassium concentration values ​​are obtained, thereby improving the reliability and environmental adaptability of the nitrogen-phosphorus-potassium concentration values. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating a distributed soil nitrogen, phosphorus, and potassium status monitoring method provided in an embodiment of this application. Detailed Implementation

[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0047] like Figure 1 As shown in the figure, this application provides a distributed soil nitrogen, phosphorus, and potassium status monitoring method, which includes the following steps:

[0048] Step 1: Deploy distributed sensing nodes, perform adaptive calibration on the nitrogen, phosphorus, and potassium sensors of each sensing node, and use the adaptively calibrated distributed sensing nodes to collect soil nitrogen, phosphorus, and potassium potential data in real time.

[0049] Step 2: Improved adaptive wavelet threshold denoising is performed on the soil nitrogen, phosphorus and potassium potential data of each sensing node to obtain denoised feature data.

[0050] Step 3: An improved lightweight convolutional neural network is used to extract features from the noise reduction feature data of each sensing node to obtain a multidimensional deep feature vector of nitrogen, phosphorus and potassium.

[0051] Step four involves processing the nitrogen, phosphorus, and potassium multidimensional deep feature vectors of each sensing node using cross-node collaborative filtering and anomaly removal algorithms to obtain regional consistency feature data.

[0052] Step 5: The nitrogen, phosphorus and potassium concentration values ​​are inverted using a nitrogen, phosphorus and potassium inversion model based on nonlinear coupling of environmental factors, according to the regional consistency characteristic data of each sensing node.

[0053] Step 6: Generate a regional soil nutrient thermogram based on the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node.

[0054] For example, in step one, the area to be monitored is divided into square grids of equal area, and a sensing node is set at the center of each grid. Each sensing node integrates a fast-acting nitrogen, fast-acting phosphorus, and fast-acting potassium ion selective electrode, i.e., the nitrogen, phosphorus, and potassium sensor, and also integrates a temperature and humidity sensor and a conductivity sensor.

[0055] Each sensor node uses LoRa decentralized networking communication.

[0056] Specifically, in this embodiment, in step one, the side length of the square grid is set to 5m. In other possible embodiments, it can be increased or decreased according to actual needs.

[0057] Each sensor node uses LoRa decentralized networking communication, eliminating the need for a central gateway. Adjacent nodes directly transmit data, adapting to wire-free scenarios and achieving seamless coverage of the area.

[0058] In step one, the temperature and humidity sensor and the conductivity sensor are used to collect real-time temperature, real-time humidity, and real-time conductivity.

[0059] For example, in step one, the adaptive calibration includes:

[0060] A real-time correction coefficient is constructed based on the real-time temperature and real-time conductivity of each sensing node, and the corresponding nitrogen, phosphorus and potassium sensors are adaptively calibrated using the real-time correction coefficient.

[0061] Specifically, in this embodiment, the real-time correction coefficient is constructed as follows:

[0062] .

[0063] in, This indicates the real-time correction factor. Indicates the temperature compensation coefficient. Indicates real-time temperature. Indicates the conductivity compensation coefficient. Indicates real-time conductivity. This represents the sensor reference offset coefficient. Temperature compensation coefficient. Conductivity compensation coefficient Sensor reference offset coefficient The calibration was performed in the laboratory beforehand.

[0064] The formula for adaptive calibration is as follows:

[0065] .

[0066] in, This represents the potential data after adaptive calibration (i.e., soil nitrogen, phosphorus, and potassium potential data). This represents the potential data before adaptive calibration. This indicates the electrode zero potential offset (calibrated with a standard solution before the sensor leaves the factory).

[0067] For example, in step two, the improved adaptive wavelet threshold denoising includes:

[0068] An adaptive weighting factor is constructed, and the soil nitrogen, phosphorus, and potassium potential data after wavelet hard thresholding and the soil nitrogen, phosphorus, and potassium potential data after wavelet soft thresholding are weighted and summed based on the adaptive weighting factor to obtain the noise reduction feature data.

[0069] Specifically, in this embodiment, the formula for the adaptive weighting factor in step two is as follows:

[0070] .

[0071] in, Indicates the adaptive weighting factor. This represents the maximum potential data of soil nitrogen, phosphorus, and potassium in a single period.

[0072] The noise reduction feature data is shown below:

[0073] .

[0074] in, This represents the noise reduction feature data. This represents the soil nitrogen, phosphorus, and potassium potential data after wavelet hard thresholding. This represents the soil nitrogen, phosphorus, and potassium potential data after wavelet soft thresholding.

[0075] In one possible embodiment, in step two, the noise reduction feature data is normalized to the (0,1) interval to eliminate the influence of dimensions.

[0076] For example, in step three, the improved lightweight convolutional neural network integrates depthwise separable convolutional layers and an improved channel attention mechanism layer.

[0077] Specifically, in this embodiment, in step three, the structure of the improved lightweight convolutional neural network is as follows: input layer - depthwise separable convolutional layer - improved channel attention mechanism layer - global average pooling layer - output layer.

[0078] The input layer is used to input the noise reduction feature data, the depthwise separable convolutional layer is used to perform depthwise separable convolution on the noise reduction feature data, the improved channel attention mechanism layer is used to dynamically assign channel feature weights to the noise reduction feature data after depthwise separable convolution, the global average pooling layer is used to aggregate the noise reduction feature data after assigning channel feature weights, and the output layer is used to map the aggregated noise reduction feature data into a fixed-dimensional nitrogen, phosphorus and potassium multidimensional depth feature vector.

[0079] For example, the improved channel attention mechanism layer calculates channel feature weights by combining the Sigmoid activation function with fully connected layer operations, average pooling operations, and max pooling operations.

[0080] Specifically, in this embodiment, the formula for the channel feature weights is as follows:

[0081] .

[0082] in, Indicates the channel feature weights. This represents the Sigmoid activation function. This indicates a fully connected layer operation. This indicates the average pooling operation. This indicates a max pooling operation. This represents the output of a depthwise separable convolutional layer. A larger channel feature weight indicates that the features of that channel (such as nitrogen ion channels) are more important to the monitoring results, thus automatically strengthening the effective feature weights.

[0083] For example, step four includes:

[0084] A cross-node collaborative filtering algorithm based on adjacent node correction is used to correct the multidimensional deep feature vectors of nitrogen, phosphorus and potassium of each sensing node, thus obtaining cross-node corrected features.

[0085] The 3σ criterion is used to remove outliers from the cross-node correction features to obtain regional consistency feature data.

[0086] Specifically, in this embodiment, the formula for cross-node feature correction in step four is as follows:

[0087] .

[0088] in, This represents the cross-node correction feature of the i-th sensor node. This represents the nitrogen, phosphorus, and potassium multidimensional depth feature vector of the i-th sensor node. Indicates the number of adjacent nodes. This represents the multidimensional deep feature vector of nitrogen, phosphorus, and potassium for the j-th neighboring node. The spatial weight factor of the i-th sensor node and its j-th neighbor is equal to the reciprocal of the Euclidean distance between the i-th sensor node and its j-th neighbor.

[0089] In this embodiment, the 3σ criterion is used to remove outliers from cross-node modified features as follows:

[0090] when At that time, determine the current These are outliers and are removed. This represents the mean of the cross-node correction features across all sensing nodes.

[0091] For example, in step five, the nitrogen, phosphorus and potassium inversion model based on nonlinear coupling of environmental factors sequentially performs dimensional standardization, environmental-nitrogen, phosphorus and potassium nonlinear coupling processing, and nitrogen, phosphorus and potassium differential electrochemical inversion on the regional consistency feature data of each sensing node to obtain nitrogen, phosphorus and potassium concentration values.

[0092] The environmental-nitrogen, phosphorus, and potassium nonlinear coupling treatment uses a natural exponential cross-coupling formula to couple the dimensionally standardized temperature, humidity, and electrical conductivity with soil nitrogen, phosphorus, and potassium data.

[0093] The differential electrochemical inversion of nitrogen, phosphorus, and potassium is designed with linear-logarithmic response inversion formulas for available nitrogen, nonlinear quadratic response inversion formulas for available phosphorus, and diffusion-type radical-logarithmic response inversion formulas for available potassium.

[0094] Specifically, in this embodiment, in step five, the dimensional standardization process maps the regional consistency feature data of each sensing node to the (0,1] interval, and simultaneously maps the collected real-time temperature, real-time humidity, and real-time conductivity to the (0,1] interval.

[0095] In step five, the natural exponential cross-coupling formula is set as follows:

[0096] .

[0097] in, This indicates the nonlinear coupling characteristics of the environment and nutrients (when X represents nitrogen, phosphorus, and potassium, respectively). These are the nonlinear coupling characteristics of the environment-nitrogen, environment-phosphorus, and environment-potassium, respectively. This represents the value of nutrient X in the regional consistency characteristic data after dimensional standardization. Represents the natural index. Represents the temperature response coefficient. This represents the real-time temperature after dimensional standardization. This represents the humidity-conductivity cross-coupling coefficient. This represents the real-time humidity after dimensional standardization. This represents the real-time conductivity after dimensional standardization.

[0098] Based on the nonlinear coupling characteristics of environment and nutrients Using nitrogen as the independent variable, the following inversion formulas are set for the linear-logarithmic response of available nitrogen, the nonlinear quadratic response of available phosphorus, and the diffusion-type radical-logarithmic response of available potassium:

[0099] Fast-acting nitrogen inversion:

[0100] .

[0101] Fast-acting phosphorus inversion:

[0102] .

[0103] Available potassium inversion:

[0104] .

[0105] in, This represents the nitrogen concentration obtained from the inversion. This indicates the phosphorus concentration obtained from the inversion. This indicates the potassium concentration obtained from the inversion. , , These represent the nonlinear coupling characteristics between the environment and nitrogen, the environment and phosphorus, and the environment and potassium, respectively. Represents the nitrogen linear response coefficient. Represents the logarithmic response coefficient of nitrogen. This represents the zero-value protection constant (0.001 in this embodiment). Represents the phosphorus quadratic response coefficient. This represents the primary response coefficient of phosphorus. This represents the potassium-based response coefficient. The values ​​represent the logarithmic response coefficients of potassium, and all response coefficients were obtained through laboratory calibration.

[0106] For example, step six includes:

[0107] Based on the nitrogen, phosphorus, and potassium nutrient requirements of crops, nitrogen, phosphorus, and potassium early warning thresholds are set. Based on the nitrogen, phosphorus, and potassium early warning thresholds and the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node, the early warning level of each sensor node is obtained. A regional soil nutrient heat map is generated based on the early warning level of each sensor node.

[0108] Specifically, in this embodiment, in step six, two sets of nitrogen, phosphorus, and potassium warning thresholds are set, including the upper limit of suitable nitrogen, phosphorus, and potassium concentrations and the lower limit of suitable nitrogen, phosphorus, and potassium concentrations.

[0109] When the nitrogen, phosphorus, and potassium concentrations at a certain sensor node are all between the upper and lower limits of the suitable concentration range, it indicates that the soil nutrients at that sensor node are suitable. The warning level is suitable.

[0110] When one or more of the nitrogen, phosphorus, and potassium concentration values ​​at a certain sensor node exceed the corresponding upper limit of the suitable concentration for nitrogen, phosphorus, and potassium (for example, the nitrogen concentration exceeds the upper limit of the suitable nitrogen concentration), it indicates that the soil nutrient level for that sensor node exceeds the standard for the corresponding value. The warning level is "the corresponding value exceeds the standard".

[0111] When one or more of the nitrogen, phosphorus, and potassium concentration values ​​at a certain sensor node are lower than the corresponding values ​​of the lower limits for nitrogen, phosphorus, and potassium concentrations (for example, nitrogen concentration is lower than the lower limit for nitrogen concentration), it indicates that the soil nutrients for that corresponding value at that sensor node are deficient. The warning level is deficiency of the corresponding value.

[0112] A regional soil nutrient heat map is generated based on the warning level of each sensor node, with different colors representing different warning levels.

[0113] In other possible embodiments, other nitrogen, phosphorus, and potassium warning thresholds and warning level systems may also be set.

[0114] This application also provides a distributed soil nitrogen, phosphorus, and potassium state monitoring system, which adopts the above-mentioned distributed soil nitrogen, phosphorus, and potassium state monitoring method, including: distributed sensing nodes, a data noise reduction module, a feature extraction module, a consistency processing module, a concentration inversion module, and a heat map generation module.

[0115] The distributed sensing nodes are configured to collect soil nitrogen, phosphorus, and potassium potential data in real time.

[0116] The data denoising module is configured to perform improved adaptive wavelet threshold denoising on the soil nitrogen, phosphorus, and potassium potential data of each sensing node to obtain denoised feature data.

[0117] The feature extraction module is configured to use an improved lightweight convolutional neural network to extract features from the noise-reduced feature data of each sensing node, thereby obtaining a multidimensional deep feature vector of nitrogen, phosphorus, and potassium.

[0118] The consistency processing module is configured to process the nitrogen, phosphorus, and potassium multidimensional deep feature vectors of each sensing node using cross-node collaborative filtering and anomaly removal algorithms to obtain regional consistency feature data.

[0119] The concentration inversion module is configured to use a nitrogen, phosphorus, and potassium inversion model based on nonlinear coupling of environmental factors to invert nitrogen, phosphorus, and potassium concentration values ​​based on the regional consistency characteristic data of each sensor node.

[0120] The heat map generation module is configured to generate a regional soil nutrient heat map based on the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node.

[0121] This application embodiment improves the convenience and accuracy of soil nitrogen, phosphorus, and potassium status monitoring by deploying distributed sensing nodes and combining adaptive calibration, improved adaptive wavelet threshold denoising, improved lightweight convolutional neural networks, cross-node collaborative filtering, and a nitrogen, phosphorus, and potassium inversion model based on nonlinear coupling of environmental factors.

[0122] By constructing real-time correction coefficients based on the real-time temperature and conductivity of each sensing node, the nitrogen, phosphorus, and potassium sensors are adaptively calibrated to dynamically correct sensor drift and ensure the accuracy of basic monitoring.

[0123] By constructing an adaptive weighting factor, the soil nitrogen, phosphorus, and potassium potential data after wavelet hard thresholding and the soil nitrogen, phosphorus, and potassium potential data after wavelet soft thresholding are weighted and summed based on the adaptive weighting factor, which takes into account both noise reduction effect and signal fidelity, adapts to complex soil environments, and improves the accuracy of the original data.

[0124] The proposed improved lightweight convolutional neural network integrates depthwise separable convolutional layers and an improved channel attention mechanism layer, which is adapted to edge computing terminals. It does not require cloud computing power and can complete feature extraction locally on the terminal, thus improving the convenience of terminal processing.

[0125] The proposed improved channel attention mechanism layer calculates channel feature weights by combining the Sigmoid activation function with fully connected layer operations, average pooling operations, and max pooling operations, accurately extracting the core features of nitrogen, phosphorus, and potassium, and improving feature effectiveness.

[0126] By employing a cross-node collaborative filtering algorithm based on adjacent node correction to correct the multidimensional deep feature vectors of nitrogen, phosphorus, and potassium for each sensing node, outlier removal is performed on the cross-node corrected features, and abnormal data is removed by utilizing regional nutrient correlation, thereby improving the accuracy of regional monitoring data.

[0127] By sequentially performing dimensional standardization, environmental-nitrogen-phosphorus-potassium nonlinear coupling processing, and nitrogen-phosphorus-potassium differential electrochemical inversion on the regional consistency characteristic data of each sensing node, the nitrogen-phosphorus-potassium concentration values ​​are obtained, thereby improving the reliability and environmental adaptability of the nitrogen-phosphorus-potassium concentration values.

[0128] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0129] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A distributed method for monitoring the nitrogen, phosphorus, and potassium status of soil, characterized in that, Includes the following steps: Step 1: Deploy distributed sensing nodes, perform adaptive calibration on the nitrogen, phosphorus and potassium sensors of each sensing node, and use the adaptively calibrated distributed sensing nodes to collect soil nitrogen, phosphorus and potassium potential data in real time. Step 2: Improved adaptive wavelet threshold denoising is performed on the soil nitrogen, phosphorus and potassium potential data of each sensing node to obtain denoised feature data. Step 3: An improved lightweight convolutional neural network is used to extract features from the noise-reduced feature data of each sensing node to obtain a multi-dimensional deep feature vector of nitrogen, phosphorus and potassium. Step 4: Cross-node collaborative filtering and anomaly removal algorithms are used to process the multi-dimensional deep feature vectors of nitrogen, phosphorus and potassium of each sensing node to obtain regional consistency feature data. Step 5: The nitrogen, phosphorus and potassium concentration values ​​are inverted using a nitrogen, phosphorus and potassium inversion model based on nonlinear coupling of environmental factors, according to the regional consistency characteristic data of each sensing node. Step 6: Generate a regional soil nutrient thermogram based on the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node.

2. The distributed soil nitrogen, phosphorus, and potassium status monitoring method according to claim 1, characterized in that, In step one, the area to be monitored is divided into square grids of equal area. A sensing node is set at the center of each grid. Each sensing node integrates a nitrogen, phosphorus and potassium ion selective electrode, i.e., the nitrogen, phosphorus and potassium sensor, and also integrates a temperature and humidity sensor and a conductivity sensor. Each sensor node uses LoRa decentralized networking communication.

3. The distributed soil nitrogen, phosphorus, and potassium status monitoring method according to claim 1, characterized in that, In step one, the adaptive calibration includes: A real-time correction coefficient is constructed based on the real-time temperature and real-time conductivity of each sensing node, and the corresponding nitrogen, phosphorus and potassium sensors are adaptively calibrated using the real-time correction coefficient.

4. The distributed soil nitrogen, phosphorus, and potassium status monitoring method according to claim 1, characterized in that, In step two, the improved adaptive wavelet threshold denoising includes: An adaptive weighting factor is constructed, and the soil nitrogen, phosphorus, and potassium potential data after wavelet hard thresholding and the soil nitrogen, phosphorus, and potassium potential data after wavelet soft thresholding are weighted and summed based on the adaptive weighting factor to obtain the noise reduction feature data.

5. The distributed soil nitrogen, phosphorus, and potassium status monitoring method according to claim 1, characterized in that, In step three, the improved lightweight convolutional neural network integrates depthwise separable convolutional layers and an improved channel attention mechanism layer.

6. The distributed soil nitrogen, phosphorus, and potassium status monitoring method according to claim 5, characterized in that, The improved channel attention mechanism layer calculates channel feature weights by combining the Sigmoid activation function with fully connected layer operations, average pooling operations, and max pooling operations.

7. The distributed soil nitrogen, phosphorus, and potassium status monitoring method according to claim 1, characterized in that, Step four includes: A cross-node collaborative filtering algorithm based on adjacent node correction is used to correct the nitrogen, phosphorus, and potassium multidimensional deep feature vectors of each sensing node, resulting in cross-node corrected features. The 3σ criterion is used to remove outliers from the cross-node correction features to obtain regional consistency feature data.

8. The distributed soil nitrogen, phosphorus, and potassium status monitoring method according to claim 1, characterized in that, In step five, the nitrogen, phosphorus and potassium inversion model based on nonlinear coupling of environmental factors performs dimensional standardization, environmental-nitrogen, phosphorus and potassium nonlinear coupling processing, and nitrogen, phosphorus and potassium differential electrochemical inversion on the regional consistency feature data of each sensing node in sequence to obtain nitrogen, phosphorus and potassium concentration values. The environmental-nitrogen-phosphorus-potassium nonlinear coupling treatment adopts a natural exponential cross-coupling formula to couple the dimensionally standardized temperature, humidity, and electrical conductivity with soil nitrogen, phosphorus, and potassium data. The differential electrochemical inversion of nitrogen, phosphorus, and potassium is designed with linear-logarithmic response inversion formulas for available nitrogen, nonlinear quadratic response inversion formulas for available phosphorus, and diffusion-type radical-logarithmic response inversion formulas for available potassium.

9. The distributed soil nitrogen, phosphorus, and potassium status monitoring method according to claim 1, characterized in that, Step six includes: Based on the nitrogen, phosphorus, and potassium nutrient requirements of crops, nitrogen, phosphorus, and potassium early warning thresholds are set. Based on the nitrogen, phosphorus, and potassium early warning thresholds and the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node, the early warning level of each sensor node is obtained. A regional soil nutrient heat map is generated based on the early warning level of each sensor node.

10. A distributed soil nitrogen, phosphorus, and potassium state monitoring system, employing a distributed soil nitrogen, phosphorus, and potassium state monitoring method as described in any one of claims 1 to 9, characterized in that, include: Distributed sensing nodes, data denoising module, feature extraction module, consistency processing module, concentration inversion module, and heat map generation module; The distributed sensing nodes are configured to collect soil nitrogen, phosphorus, and potassium potential data in real time. The data denoising module is configured to perform improved adaptive wavelet threshold denoising on the soil nitrogen, phosphorus, and potassium potential data of each sensing node to obtain denoised feature data. The feature extraction module is configured to: use an improved lightweight convolutional neural network to extract features from the noise-reduced feature data of each sensing node to obtain a multidimensional deep feature vector of nitrogen, phosphorus and potassium. The consistency processing module is configured to: process the nitrogen, phosphorus, and potassium multidimensional deep feature vectors of each sensing node using cross-node collaborative filtering and anomaly removal algorithms to obtain regional consistency feature data; The concentration inversion module is configured to: use a nitrogen, phosphorus, and potassium inversion model based on nonlinear coupling of environmental factors to invert nitrogen, phosphorus, and potassium concentration values ​​based on the regional consistency characteristic data of each sensing node; The heat map generation module is configured to generate a regional soil nutrient heat map based on the nitrogen, phosphorus, and potassium concentration values ​​of each sensor node.

Images