A probe type soil nutrient rapid testing method based on eight detection index calibration

Abstract

This invention relates to the field of soil testing technology and discloses a probe-based rapid soil nutrient testing method based on calibration of eight testing indicators. The method uses a built-in standard signal source to automatically calibrate sensors corresponding to eight indicators: soil temperature, moisture content, electrical conductivity, pH, available nitrogen, available phosphorus, available potassium, and water-soluble organic matter, generating compensation coefficients and calibration reliability scores. After the probe is inserted into the soil to collect raw data, the data is corrected and temperature compensated using the compensation coefficients. The corrected data from the eight indicators, along with crop type, growth stage, and meteorological data, are input into a multimodal data fusion model to output a comprehensive soil nutrient evaluation result. A testing report is generated, containing the eight test values, calibration reliability scores, and recommended agricultural strategies. This invention achieves simultaneous automatic calibration of the eight indicators without the need for external reagents, has high testing efficiency, quantifiable calibration reliability, and can directly output decision-making schemes adapted to AI-driven planting.

Description

Technical Field

[0001] This invention relates to the field of soil testing technology, specifically to a probe-based rapid soil nutrient testing method calibrated based on eight testing indicators. Background Technology

[0002] With the rapid development of precision agriculture, rapid, in-situ detection of soil nutrients has become a key technological link in achieving scientific fertilization, precision irrigation, and crop health management. Probe-type soil analyzers are increasingly widely used in agricultural production due to their advantages such as portability, ease of operation, and ability to perform real-time on-site testing. These devices typically integrate multiple sensors to measure soil temperature, moisture content, electrical conductivity (EC), pH, and key nutrient indicators such as available nitrogen, phosphorus, and potassium.

[0003] However, existing probe-based soil testing technologies generally rely on manual calibration using external standard solutions or reagents. Operators must carry multiple reagents and complete a cumbersome calibration process on-site, which is not only time-consuming and labor-intensive but also requires high operational skills, making it difficult to meet the rapid testing needs of large-scale field operations. Furthermore, existing calibration mechanisms are mostly based on independent calibration of single indicators, failing to achieve simultaneous and coordinated calibration of eight core indicators: soil temperature, moisture content, EC value, pH value, available nitrogen, available phosphorus, available potassium, and water-soluble organic matter. Due to the complex interactive coupling relationships among various physicochemical parameters in the soil, independent calibration of single indicators cannot eliminate cross-interference between sensors, resulting in significant deviations between the calibrated test data and the actual soil conditions, making it difficult to effectively control detection errors.

[0004] Furthermore, existing technologies typically only record the final calibration result after calibration, lacking the recording and quantitative evaluation of calibration process data, and thus failing to generate professional testing reports that include calibration credibility. This results in a lack of objective evaluation criteria for the reliability of test results, making it difficult for users to judge the credibility of the test data and limiting the reference value of the test reports in scientific decision-making. Moreover, existing soil testing methods often stop at the data output level, failing to deeply integrate and analyze the calibrated multi-parameter test data with factors such as crop type, growth stage, and meteorological conditions. Their data formats and output methods also cannot effectively interface with AI precision planting decision-making systems, making it difficult to directly use them to generate intelligent agricultural management solutions such as fertilization, irrigation, and pest and disease control, creating a break between data collection and decision-making applications.

[0005] In summary, existing probe-based soil nutrient testing technologies have significant shortcomings in terms of the convenience of calibration methods, multi-parameter collaborative calibration capabilities, detection accuracy, report professionalism, and compatibility with AI decision-making systems. There is an urgent need to provide a rapid soil nutrient testing method that can achieve simultaneous automatic calibration of eight indicators, has reliable quantification capabilities, and can seamlessly integrate with AI precision planting decision-making systems. To this end, a probe-based rapid soil nutrient testing method based on calibration of eight detection indicators is proposed. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a probe-based rapid soil nutrient testing method calibrated based on eight detection indicators, thereby resolving the problems in the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a probe-based rapid soil nutrient testing method calibrated based on eight detection indicators, comprising: Calibration steps: Before testing, the sensors corresponding to the eight test indicators of the probe-type soil tester are automatically calibrated using the built-in standard signal source to generate compensation coefficients and calibration reliability scores for each sensor. The eight test indicators include soil temperature, soil moisture content, electrical conductivity, pH, available nitrogen, available phosphorus, available potassium and water-soluble organic matter. Preferably, the calibration step specifically includes: Calibration signal output sub-step: Control the built-in calibration module to sequentially output multiple standard excitation signals corresponding to the eight detection indicators to the corresponding sensor channels; Deviation calculation sub-step: Collect the actual response signals of each sensor under the standard excitation signal, and calculate the deviation value between the actual response signal and the preset theoretical response signal; Compensation generation sub-step: Generate dynamic compensation coefficients or compensation curves for each sensor within each detection range based on the deviation value; The credibility scoring sub-step involves generating the calibration credibility score based on the deviation of each sensor, the consistency of calibration history, and the sensor lifetime model.

[0008] Preferably, the built-in standard signal source is a multi-channel standard signal generation circuit with temperature self-compensation function integrated inside the probe-type soil analyzer. The multi-channel standard signal generation circuit can independently output standard electrical signals corresponding to temperature, conductivity, pH and nutrient simulation signals.

[0009] Preferably, in the calibration step, the fully automatic calibration of the sensors corresponding to the eight detection indicators is performed in the following order: first, the soil temperature sensor is calibrated; then, the conductivity and pH sensors are calibrated based on the calibrated temperature values; and finally, the sensors corresponding to available nitrogen, available phosphorus, available potassium, and water-soluble organic matter are calibrated in sequence.

[0010] Data collection steps: Insert the probe into the soil to be tested and collect the raw sensor data of the eight detection indicators; Data processing steps: Correct the original sensor data according to the compensation coefficient, and perform temperature compensation processing in combination with soil temperature data to generate eight corrected detection data. Preferably, the data processing step of correcting the original sensor data specifically includes: The original sensing data is corrected for linear or nonlinear errors using the compensation coefficients generated in the calibration step. Temperature drift compensation was performed on the measured values ​​of electrical conductivity, pH and available nutrients using synchronously collected soil temperature data. The compensated data is digitally filtered to eliminate high-frequency noise during the acquisition process.

[0011] Fusion analysis steps: The corrected eight detection data points, along with externally input crop type, growth period, and meteorological data, are input into a preset multimodal data fusion model to output a comprehensive evaluation result of soil nutrients; Preferably, the multimodal data fusion model in the fusion analysis step is a model built based on a deep neural network. Its input layer includes a first input terminal and a second input terminal. The first input terminal is used to receive a vector composed of the corrected eight detection data. The second input terminal is used to receive an auxiliary feature vector composed of crop type code, growth stage code and meteorological parameters. The model output layer outputs the comprehensive evaluation result of soil nutrients.

[0012] Preferably, the multimodal data fusion model uses an attention mechanism to adaptively learn the correlation weights among the eight detection indicators, and uses a multi-head attention mechanism to model the interaction relationship between the detection data and external input features.

[0013] Report generation steps: Based on the comprehensive evaluation results and the calibration reliability score, generate a test report that includes agricultural recommendation strategies.

[0014] Preferably, in the report generation step, the detection report includes at least the following information fields: Corrected values ​​and corresponding normal reference ranges for the eight testing indicators; The calibration reliability score and the error correction record for this calibration; Suitability assessment based on comprehensive soil nutrient evaluation results; The agricultural recommendation strategy includes at least fertilizer formula recommendations, irrigation amount recommendations, and pest and disease early warning prompts.

[0015] Preferably, the method further includes an AI decision-making and feedback optimization step: pushing the detection report to the user terminal, receiving crop growth effect data and soil retest data returned by the user after performing agricultural operations according to the agricultural recommendation strategy, and using the crop growth effect data and soil retest data to perform incremental training and parameter optimization on the multimodal data fusion model.

[0016] Preferably, the agricultural recommendation strategy is generated by inputting the comprehensive evaluation results of soil nutrients into a preset crop precision management decision tree model. The decision tree model combines crop type, growth period and target yield, dynamically matches agricultural operation plans in a preset expert knowledge base, and outputs quantifiable suggestions on fertilizer application amount, fertilizer ratio, irrigation cycle and pesticide application timing.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention employs a built-in standard signal source and a fully automatic collaborative calibration mechanism to achieve simultaneous automatic calibration of eight indicators: soil temperature, moisture content, electrical conductivity, pH, available nitrogen, available phosphorus, available potassium, and water-soluble organic matter. It eliminates the dependence on external reagents, requires no manual intervention during the calibration process, simplifies the operation steps, and reduces detection errors and improves detection accuracy through dynamic compensation and cross-interference elimination technology.

[0018] 2. This invention automatically generates a calibration reliability score after calibration and incorporates this score, along with eight test values ​​and error correction records, into the test report. This enables quantifiable evaluation of the calibration process and professional output of test results, providing the test report with an objective and reliable reference basis and effectively compensating for the shortcomings of existing technologies in providing calibration quality evaluation.

[0019] 3. This invention constructs a multimodal data fusion model to deeply integrate and analyze the calibrated eight detection data with crop type, growth period, and meteorological information, and directly outputs agricultural recommendation strategies including fertilizer formula, irrigation amount, and pest and disease early warning. This achieves seamless integration from soil detection to precision planting decision-making, and provides a data interface and decision basis that can be directly called for the AI ​​precision planting management system.

[0020] 4. By setting up an AI decision-making and feedback optimization process, this invention feeds back the effect data of the user's implementation of agricultural strategies to the model for incremental training, forming a closed-loop system of detection-decision-execution-feedback-optimization. This enables the model's prediction accuracy to continuously improve with the frequency of use, achieving the synergistic evolution of detection methods and application effects.

[0021] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0022] Figure 1 This is a flowchart of the overall method of the present invention; Figure 2 This is a flowchart of the overall method of the present invention; Figure 3 This is a structural diagram of the multimodal data fusion model of the present invention; Figure 4 This is a flowchart of the AI ​​decision-making and feedback optimization process of this invention; Figure 5 This is a calibration sequence diagram for the eight indicators of this invention; Figure 6 This is a flowchart illustrating the process of generating the agricultural recommendation strategy of this invention. Detailed Implementation

[0023] Please see Figures 1-6 This invention discloses a probe-based rapid soil nutrient testing method based on eight detection indicators. This method is implemented through a hardware module integrated within a probe-based soil analyzer and an embedded algorithm. The technical solution of this embodiment is described in detail below with reference to the specific execution flow.

[0024] I. Equipment Initialization and Calibration Preparation Before performing soil testing, the probe-type soil analyzer is first powered on and initialized. After powering on, the main control chip initiates a self-test program, sequentially checking the electrical connectivity of each sensor channel, the operating status of the analog-to-digital conversion module, and the readability of the storage module. Upon successful self-test, the system reads the preset calibration database from its built-in non-volatile memory. This database stores the factory calibration parameters of each sensor, historical calibration records, and sensor lifespan degradation models. After initialization, the system enters a calibration-ready state, awaiting the trigger of an automatic calibration command.

[0025] The purpose of this step is to ensure that the equipment hardware is functioning properly and that the calibration reference data is readable, laying the foundation for subsequent high-precision calibration. Through a self-test mechanism, hardware faults such as open circuits, short circuits, or abnormal analog-to-digital conversion in sensors can be detected in advance, avoiding calibration or testing under invalid conditions and thus ensuring the reliability of the entire testing process.

[0026] II. Fully Automated Calibration of Eight Indicators This step involves synchronous and collaborative calibration of the sensors corresponding to the eight detection indicators using a built-in standard signal source, and specifically includes the following sub-steps: 1. Standard signal output The main control chip controls the built-in standard signal source module to sequentially output multiple standard excitation signals corresponding to eight detection indicators. These eight indicators include: soil temperature (T), soil moisture content (MC), electrical conductivity (EC), pH, available nitrogen (N), available phosphorus (P), available potassium (K), and water-soluble organic matter (SOM). The built-in standard signal source is a multi-channel standard signal generation circuit integrated within the instrument, capable of independently outputting standard electrical signals corresponding to different physical quantities. Specifically, for the temperature sensor, a standard signal source outputs a set of precision resistors with known resistance values ​​(e.g., standard resistance values ​​corresponding to 0℃, 25℃, and 50℃); for the conductivity sensor, a set of standard resistance network signals with known conductivity values ​​is output (e.g., standard signals corresponding to 0.5 mS / cm, 1.0 mS / cm, and 2.0 mS / cm); for the pH sensor, a set of standard voltage signals with known potential differences is output (e.g., standard voltage values ​​corresponding to pH 4.00, 6.86, and 9.18); and for the four nutrient sensors, a set of simulated electrochemical signals with known concentrations is output (e.g., standard simulated signals corresponding to available nitrogen concentrations of 10 mg / kg, 50 mg / kg, and 100 mg / kg).

[0027] The calibration sequence follows the principle of benchmark first, then correlation: first, the soil temperature sensor is calibrated; then, based on the calibrated temperature values, the conductivity and pH sensors are calibrated using temperature correlation; finally, the sensors corresponding to available nitrogen, available phosphorus, available potassium, and water-soluble organic matter are calibrated sequentially. Let the first... The standard value of the built-in standard signal source output is [value missing]. ,in These correspond to the eight indicators mentioned above.

[0028] This step automates the calibration process by replacing traditional external standard solutions with a built-in standard signal source. It eliminates the need for operators to carry reagents or manually prepare solutions, reducing the complexity of on-site operations and the risk of human error. The calibration sequence of temperature first, followed by other indicators, ensures that subsequent calibrations of conductivity, pH, and nutrient indicators are based on an accurate temperature reference, effectively eliminating the interference of temperature factors on the calibration results.

[0029] 2. Sensor response acquisition and deviation calculation While outputting the standard signal, the main control chip acquires the actual response signals of each sensor under the standard excitation signal through an analog-to-digital converter. To ensure data acquisition accuracy, each standard signal point is sampled multiple times and the arithmetic mean is calculated. For example, for a temperature sensor, under a standard resistance signal at 25℃, the response value is continuously acquired 10 times, and the arithmetic mean is calculated as the actual response at that point.

[0030] For linear sensors, a two-point or multi-point calibration method is used: standard signals are applied at the low and high ends of the measurement range, and the corresponding response values ​​are collected. In this embodiment, for the conductivity sensor, response values ​​are collected at two standard points, 0.5 mS / cm and 2.0 mS / cm, to establish a linear calibration curve.

[0031] For the i-th indicator, the deviation between its actual response and the standard value is... Defined as: , in, For the first The theoretical transfer function of the sensor is determined by the sensor's factory calibration data. For a temperature sensor, the theoretical transfer function is a linear function. ,in This is the sensitivity coefficient. This represents the zero-point offset. For pH sensors, the theoretical transfer function follows the Nernst equation: , in This is the standard electrode potential. Let be the ideal gas constant. Absolute temperature is Faraday's constant.

[0032] This step quantifies the deviation between the sensor's current state and its theoretical state at the factory. By averaging multiple samples, random errors from single sampling are eliminated; deviation calculation provides a precise input basis for generating subsequent compensation coefficients. This process ensures that calibration no longer relies on operator experience but is based on objective electrical measurement data.

[0033] 3. Generation of compensation coefficients Based on the calculated deviation value The main control chip generates dynamic compensation coefficients for each sensor within its respective detection range. For linear sensors, a linear compensation model is used: , Taking a conductivity sensor as an example, the actual response was collected at a standard point of 0.5 mS / cm. The actual response was collected at the standard point of 2.0 mS / cm. The compensation coefficient is then determined through linear interpolation, so that the measured value... ,in and It is calculated from two-point calibration data.

[0034] For nonlinear sensors, piecewise linear compensation or polynomial fitting is used to generate compensation curves. Taking a pH sensor as an example, a three-point calibration method is used, collecting response values ​​at three standard points with pH values ​​of 4.00, 6.86, and 9.18 respectively. The compensated slope is obtained by fitting using the least squares method. and zero offset : , in, This is the pH reading corresponding to the sensor's original output. This is the corrected pH value.

[0035] For fast-acting nutrient sensors, since the sensor response is affected by multiple factors such as temperature and pH, the compensation coefficient is generated using a multidimensional regression model. Taking a fast-acting nitrogen sensor as an example, during calibration, the built-in standard signal source sequentially outputs three standard concentration points: 10 mg / kg, 50 mg / kg, and 100 mg / kg. At each concentration point, different temperature conditions (20℃, 25℃, 30℃) and different pH conditions (6.0, 7.0, 8.0) are simultaneously simulated, and sensor response data is collected. Through multiple linear regression, the following compensation model is obtained: , in, This is the original output value of the fast-acting nitrogen sensor. This is a reference temperature (taken as 25℃). For reference pH value (7.0), The temperature compensation coefficients are obtained from the regression during the calibration phase. The pH compensation coefficient is obtained from the regression during the calibration phase. This is the intercept correction term (e.g., -0.5 mg / kg). Similarly, the compensation coefficients for available phosphorus, available potassium, and water-soluble organic matter sensors are generated using their respective regression models.

[0036] This step quantifies sensor bias into compensation parameters applicable to actual measurements. Through a multidimensional regression model, the compensation coefficients of the nutrient sensor can simultaneously correct for the effects of coupled factors such as temperature and pH, improving detection accuracy compared to traditional single-dimensional calibration. Taking available nitrogen as an example, traditional methods calibrating at 25℃ may result in a bias of 5-8 mg / kg when measured in soil at 30℃. However, using the multidimensional compensation coefficients generated in this step, this bias can be reduced to less than 2 mg / kg, making the detection results closer to the true soil nutrient content.

[0037] 4. Calibration Reliability Score Generation After calculating the compensation coefficients for each sensor, the system generates a calibration reliability score. The rating is based on a comprehensive quantification of three dimensions: calibration deviation values ​​of each indicator, consistency of calibration history, and sensor lifetime model.

[0038] Definition of the first Calibration deviation weighting factor of the indicator for: ;in, For the first The maximum allowable deviation threshold for the sensor. For example, for a temperature sensor, =0.5∘C, if the measured deviation =0.1∘C, then If the deviation reaches 0.5℃, then The larger the factor value, the smaller the current calibration deviation and the higher the reliability.

[0039] Definition of the first Historical consistency factor of the indicator for: ;in, This is the historical average of the compensation coefficients generated from the most recent n calibrations of the sensor (in this embodiment, n=5). To prevent division by zero, a small constant (taken as 0.001) is used. For example, if the historical average compensation coefficient of a sensor is 1.02 and the current calibration compensation coefficient is 1.03, then the historical consistency factor is... This factor reflects the consistency between the current calibration results and historical calibration records. The higher the consistency, the higher the reliability. If the compensation coefficient changes abruptly, it indicates that the sensor may have experienced performance degradation or abnormal interference, and the reliability will decrease accordingly.

[0040] Define the lifetime decay factor for the i-th index. for: ,

[0041] in, For the first The sensor has been used for a cumulative period of time (in hours). This represents the estimated total lifespan of the sensor. For example, if a pH sensor has an estimated lifespan of 500 hours and has already been used for 300 hours, then... This indicates that the sensor is nearing the end of its lifespan and has low reliability.

[0042] Taking all the above factors into account, the first Individual credibility score of each indicator for: , in, , , Let be the weighting coefficient, satisfying + + In this embodiment, we take... This reflects the dominant role of the bias factor in credibility evaluation. (Calibrating credibility score) The weighted average of the individual scores for the eight indicators: , in, The weighting of each indicator to the overall test results can be dynamically adjusted based on crop type and testing scenario. For example, in a corn planting scenario, the weighting of available nitrogen... The value can be set to 0.25, available phosphorus and available potassium can be set to 0.15 respectively, pH can be set to 0.15, and the remaining weights of other indicators can be evenly distributed.

[0043] After calibration, the system will store the compensation coefficient, calibration reliability score, and calibration timestamp generated during the calibration into the built-in calibration database as historical records for reference in subsequent calibrations.

[0044] This step quantifies calibration quality into comparable numerical indicators, allowing users to intuitively assess the reliability of the test data. For example, a reliability score of 0.92 indicates high reliability of the calibration result, and users can confidently use the test data; a score of 0.65 suggests that the sensor may require maintenance or replacement. This mechanism fills the gap in existing technologies that cannot provide calibration quality evaluation, giving the test report objective professional reference value.

[0045] III. Soil Parameter Collection and Data Processing After automatic calibration, the equipment enters the testing state. The operator inserts the multi-sensor probe of the probe-type detector into the soil to be tested, ensuring full contact between the probe and the soil. Taking a corn planting base as an example, the operator selects five representative locations on the target plot and inserts the probes at each location, repeating the measurement three times to obtain statistically representative soil data. The system simultaneously collects raw sensor data for eight detection indicators, with a sampling frequency set to 10 Hz, continuously collecting data for 5 seconds, obtaining a total of 50 sets of raw data.

[0046] The collected raw data first enters the data processing flow. For the first... The original sampling sequence of the indicator is denoted as . ,in =50. Data processing includes the following sub-steps:

[0047] 1. Digital filtering for noise reduction The moving average filtering method is used to smooth the original data and eliminate high-frequency noise interference. The filtered values... for: ,

[0048] in, In this embodiment, M=5 is used as the sliding window size. For sampling sequences containing obvious outliers, median filtering is first used to remove outliers, followed by moving average filtering. For example, in a conductivity sampling sequence, if a sample value deviates significantly from the preceding and following values ​​(e.g., exceeding three times the standard deviation of the mean), that point is replaced with the median of the adjacent sample points, and then a moving average is performed.

[0049] This step eliminates random errors introduced by factors such as poor soil particle contact and circuit noise. Through filtering, high-frequency fluctuations in the raw data are effectively suppressed, making the final detection values ​​more stable. Experimental results show that the standard deviation of the unfiltered raw data is approximately 0.15 mS / cm, while after filtering, the standard deviation is reduced to 0.03 mS / cm, improving data stability.

[0050] 2. Error Correction The compensation coefficients generated during the calibration process are used to correct errors in the filtered data. For linear sensors, the corrected values ​​are... for: ; Taking a conductivity sensor as an example, if the compensation coefficient... Filtered response value Then the corrected conductivity .

[0051] For pH sensors, a slope and zero-point compensation model is used: ; For example, if Filtered response value Then the corrected pH value .

[0052] For fast-acting nutrient sensors, the calibrated values ​​are directly output by the multidimensional regression model.

[0053] This step eliminates individual sensor variations and performance drift caused by long-term use. After error correction, sensors from different devices and batches can output consistent detection results, providing a basis for data comparison across devices and regions.

[0054] 3. Temperature compensation Because the measured values ​​of electrical conductivity, pH, and available nutrients are affected by soil temperature, temperature drift compensation is required based on synchronously collected soil temperature data. Taking electrical conductivity (EC) as an example, the temperature compensation formula is: , in, The conductivity value is the one measured at the current temperature. This is the temperature compensation coefficient for conductivity, typically taken as 0.02 / ℃. The soil temperature (unit: °C) was measured synchronously. This is to convert the conductivity value to a standard temperature of 25°C. For example, if the conductivity measured in soil at 30°C... ,but .

[0055] For the readily available nutrient index, a linear correction model is used for temperature compensation: ,

[0056] For example, the corrected value of available nitrogen. Measured temperature ,but .

[0057] For pH value, the temperature compensation formula is: , in, To obtain the measured pH value, This represents the asymmetric potential of the electrode. It is the gas constant (8.314 J / (mol·K)). Absolute temperature (unit: K). This is the Faraday constant (96485 C / mol). For example, it was measured at 30 °C. , The calculated correction term is approximately 0.03. .

[0058] After the above processing, the final eight-item detection data vector is generated: ,

[0059] This step eliminates the interference of temperature on various test indicators, ensuring the comparability of soil data collected in different seasons and at different times. For example, without temperature compensation, 1.36 mS / cm measured at 25℃ might appear as 1.50 mS / cm at 30℃. Directly using the original value could lead to misjudgments of soil salinity. After temperature compensation, all test values ​​are normalized to standard temperature conditions, ensuring temporal consistency and spatial comparability of the data.

[0060] IV. Multi-parameter data fusion and AI analysis The corrected eight-item detection data vector The auxiliary feature vector, along with externally input auxiliary feature vectors, is fed into a pre-defined multimodal data fusion model. It includes crop type codes, growth stage codes, and meteorological parameters, including the average temperature of the previous 7 days, the cumulative precipitation of the previous 7 days, and the average sunshine duration of the previous 7 days.

[0061] Taking a corn planting base as an example, before testing, operators select "corn" as the crop type and "jointing stage" as the growth stage via a mobile app. The system automatically obtains historical soil data for the plot and meteorological data for the past 7 days provided by the local weather station to construct an auxiliary feature vector: the crop type is encoded using a unique thermal coding method, with the first dimension of the corn vector being 1 and the others being 0; in the growth stage coding, the jointing stage is coded as 3; among the meteorological parameters, the average temperature for the past 7 days was 24.5℃, the cumulative precipitation was 12.3 mm, and the average sunshine duration was 6.8 hours. This yields the auxiliary feature vector. .

[0062] The multimodal data fusion model employs a multi-input fusion architecture built upon deep neural networks. The model's input layer contains two input terminals: the first input terminal receives eight detection data vectors. The second input terminal receives the auxiliary feature vector. ,in For auxiliary feature dimensions (in this embodiment) (Including crop type coding dimension, growth period coding dimension, and meteorological parameter dimension). The two input vectors are processed by independent fully connected layers for feature extraction to obtain hidden features. and : , in, , This is the weight matrix. , For bias vectors, For activation function ( ).

[0063] This step involves capturing the correlations among the eight detection metrics, and the model incorporates a multi-head attention mechanism. This includes hiding features. It is considered as a sequence containing 8 feature vectors, and the interaction weights between the indicators are calculated through a multi-head self-attention layer. For example, the attention mechanism can automatically learn implicit correlations such as "there is a positive correlation between available nitrogen content and soil organic matter content" and "there is a coupling relationship between electrical conductivity and water content", and incorporate these correlation information into the feature representation.

[0064] The calculation process for multi-head self-attention is as follows: The query matrix is ​​obtained through three linear transformations. Key matrix Value matrix , in The weight matrix is ​​a learnable matrix. The dimension of the key vector (in this embodiment, we take...) Attention weights are calculated by the dot product of the query and the key, and then normalized using softmax. .

[0065] The multi-head attention mechanism computes h attention heads in parallel (h=4 in this embodiment), concatenates the results from each head, and outputs them through a linear transformation. , in This is for outputting the projection matrix.

[0066] To model the interaction between the detection data and external input features, the model further employs a cross-attention mechanism. As a query, with As keys and values, calculate the fused features: , After the fused features are processed through two fully connected network layers, the output layer generates a comprehensive soil nutrient evaluation result. The output layer contains multiple output nodes, corresponding to the abundance / deficiency level of each nutrient, the soil health index, and the crop suitability score. In this embodiment, the output result... Defined as: , in, This is the output layer weight matrix. Using the bias vector, the softmax function converts the output into a probability distribution. The five output nodes correspond to the probability values ​​of five levels: "severe deficiency", "deficient", "adequate", "sufficient", and "excess".

[0067] This step deeply integrates soil testing data with crop characteristics and environmental conditions, making the evaluation results more targeted and dynamically adaptable. For example, for maize during the jointing stage, the model assigns a higher weight to available nitrogen because maize has the greatest nitrogen requirement at this stage; while for tomatoes during the fruiting stage, the model increases the weight of available potassium. Through an attention mechanism, the model can automatically identify the synergistic and antagonistic relationships between different indicators, avoiding the subjectivity of manually setting weights in traditional methods.

[0068] V. Test Report Generation Based on the comprehensive evaluation results output by the multimodal data fusion model, the system automatically generates an inspection report. Taking the inspection of a corn planting base as an example, the generated inspection report includes the following information fields: 1. Corrected values ​​and corresponding normal reference ranges for eight testing indicators. The report shows: soil temperature 24.8℃ (reference range 15-30℃), moisture content 18.2% (reference range 12-25%), electrical conductivity 1.36 mS / cm (reference range 0.5-2.0 mS / cm), pH value 7.21 (reference range 6.5-7.5), available nitrogen 45.6 mg / kg (reference range 40-80 mg / kg), available phosphorus 8.3 mg / kg (reference range 15-30 mg / kg), available potassium 112 mg / kg (reference range 80-150 mg / kg), and water-soluble organic matter 1.8% (reference range 1.5-3.5%). The numerical comparison shows that the available phosphorus content in this plot is relatively low, while other indicators are within the appropriate range.

[0069] 2. Calibration reliability score and error correction record for this calibration. The report shows that the reliability score for this calibration is 0.91, indicating good calibration quality. The error correction record lists the calculated deviation values ​​for each sensor during the calibration process in tabular form. and compensation coefficient This allows users to intuitively understand the quality of the calibration. For example, the pH sensor's deviation value is 0.02, the compensation coefficient slope is 0.98, and the zero-point offset is 0.12. These data indicate that the sensor performance is stable and the calibration results are reliable.

[0070] 3. Comprehensive Soil Nutrient Evaluation Results. Based on the model output, the report classifies each indicator as either sufficient or deficient: available nitrogen is rated "suitable," available phosphorus is rated "deficient," available potassium is rated "sufficient," and the soil health index is 78 points (out of 100). The suitability evaluation is described in natural language: "The soil pH in this plot is suitable, with good fertilizer retention capacity and sufficient nitrogen and potassium supply. However, the phosphorus level is low, which may affect maize root development and ear formation. It is recommended to apply phosphorus fertilizer during the jointing stage."

[0071] 4. Recommended Agricultural Strategies. The report generates the following specific strategies: The recommended fertilizer formula is "15 kg / mu of diammonium phosphate, combined with 20 kg / mu of compound fertilizer." This recommendation is based on the difference in phosphorus requirements during the corn jointing stage (requirement of 20 mg / kg minus the soil supply of 8.3 mg / kg, converted to fertilizer amount after considering phosphorus utilization rate). The recommended irrigation amount is "Current soil moisture content is 18.2%, which is at the lower limit of the suitable range. It is recommended to irrigate 20 m³ / mu to maintain soil moisture at 20-25%." The pest and disease warning is "Low phosphorus levels may lead to decreased corn resistance. Considering the recent weather forecast (continuous rain), it is recommended to spray potassium dihydrogen phosphate foliar fertilizer and pay attention to preventing stem rot."

[0072] This step transforms complex testing data and model outputs into agricultural guidance information that users can directly understand and implement. By incorporating calibration reliability into the report, users gain a clear understanding of the reliability of the testing data. By generating specific fertilization, irrigation, and pest control strategies, a complete closed loop from data collection to decision-making application is achieved, solving the pain point of existing technologies that only detect data but do not provide decision-making support.

[0073] VI. AI Decision-Making and Feedback Optimization After the test report is generated, the system pushes the report to the user's mobile app via a wireless communication module. The user can then view the report at the corn planting base and perform corresponding fertilization and irrigation operations according to the agricultural recommendations in the report.

[0074] After users complete agricultural operations, the system receives feedback data from them. For example, after applying phosphate fertilizer at the jointing stage, users submitted plant growth data (plant height increased by an average of 12 cm and stem diameter increased by 0.3 cm compared to unfertilized plots) and soil retest data (available phosphorus content increased from 8.3 mg / kg to 16.5 mg / kg) during the corn tasseling stage. This feedback data serves as supervisory signals for incremental training and parameter optimization of the multimodal data fusion model.

[0075] Specifically, the feedback data is combined with the input data from the current detection to form new training sample pairs. The input for the current detection includes the detection data vector. and auxiliary feature vectors In the feedback data, soil remeasurement values ​​serve as the true labels, and crop growth data are used for effect verification. Mini-batch gradient descent is employed to update model parameters, and the loss function... Defined as the mean square error between the predicted output and the actual feedback: , in, To provide the number of feedback samples, For the model to the first The predicted output for each sample (including the probability of nutrient abundance or deficiency, health index, etc.). This represents the actual result vector corresponding to the feedback (derived from soil re-measurement values ​​and crop growth data). The network weights are updated using the backpropagation algorithm to make the model more accurate in subsequent predictions.

[0076] This step establishes a closed-loop optimization mechanism between detection and production. As user frequency increases, the model continuously absorbs feedback data from different regions, crops, and management conditions, gradually improving prediction accuracy. For example, the initial model's threshold for phosphorus deficiency was 15 mg / kg. After multiple feedback learning iterations, the model discovered that applying phosphate fertilizer to corn at a phosphorus level of 12 mg / kg still resulted in increased yields. Therefore, it automatically adjusted the phosphorus deficiency threshold to 12 mg / kg, making the recommended strategy more aligned with actual production. This mechanism transforms the detection method from a one-off, static tool into a dynamic, intelligent system that continuously evolves with data accumulation.

[0077] This invention achieves fully automated collaborative calibration of eight indicators through a built-in standard signal source, requiring no external reagents. The calibration process is fast and reliable, improving on-site testing efficiency. Through a multi-dimensional compensation model and temperature compensation algorithm, it eliminates sensor cross-interference and the influence of environmental factors, ensuring that the test data accurately reflects the actual soil conditions. Through a multi-modal data fusion model and attention mechanism, it deeply integrates test data with crop characteristics and environmental conditions, generating targeted agricultural recommendation strategies. Through calibration reliability scoring and professional test reports, it provides users with quantifiable quality evaluation criteria. Through a feedback optimization mechanism, it achieves adaptive evolution of the model, providing efficient and accurate soil testing and decision support tools for precision agriculture.

[0078] The technical solution of this embodiment can be widely applied to various scenarios such as field crops, facility agriculture, and orchard planting. It can effectively guide farmers to fertilize scientifically and irrigate precisely, ensuring crop yield while reducing excessive fertilizer input, thus achieving both economic and environmental benefits.

Claims

1. A method for rapid determination of soil nutrients based on eight detection indicators, characterized in that, include: Calibration steps: Before testing, the sensors corresponding to the eight test indicators of the probe-type soil tester are automatically calibrated using the built-in standard signal source to generate compensation coefficients and calibration reliability scores for each sensor. The eight test indicators include soil temperature, soil moisture content, electrical conductivity, pH, available nitrogen, available phosphorus, available potassium and water-soluble organic matter. Data collection steps: Insert the probe into the soil to be tested and collect the raw sensor data of the eight detection indicators; Data processing steps: Correct the original sensor data according to the compensation coefficient, and perform temperature compensation processing in combination with soil temperature data to generate eight corrected detection data. Fusion analysis steps: The corrected eight detection data points, along with externally input crop type, growth period, and meteorological data, are input into a preset multimodal data fusion model to output a comprehensive evaluation result of soil nutrients; Report generation steps: Based on the comprehensive evaluation results and the calibration reliability score, generate a test report that includes agricultural recommendation strategies.

2. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 1, characterized in that, The calibration steps specifically include: Calibration signal output sub-step: Control the built-in calibration module to sequentially output multiple standard excitation signals corresponding to the eight detection indicators to the corresponding sensor channels; Deviation calculation sub-step: Collect the actual response signals of each sensor under the standard excitation signal, and calculate the deviation value between the actual response signal and the preset theoretical response signal; Compensation generation sub-step: Generate dynamic compensation coefficients or compensation curves for each sensor within each detection range based on the deviation value; The credibility scoring sub-step involves generating the calibration credibility score based on the deviation of each sensor, the consistency of calibration history, and the sensor lifetime model.

3. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 1, characterized in that, The data processing step of correcting the original sensor data specifically includes: The original sensing data is corrected for linear or nonlinear errors using the compensation coefficients generated in the calibration step. Temperature drift compensation was performed on the measured values ​​of electrical conductivity, pH and available nutrients using synchronously collected soil temperature data. The compensated data is digitally filtered to eliminate high-frequency noise during the acquisition process.

4. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 1, characterized in that, The multimodal data fusion model in the fusion analysis step is a model built on a deep neural network. Its input layer includes a first input terminal and a second input terminal. The first input terminal is used to receive a vector composed of the eight corrected detection data. The second input terminal is used to receive an auxiliary feature vector composed of crop type code, growth stage code and meteorological parameters. The model output layer outputs the comprehensive evaluation result of soil nutrients.

5. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 4, characterized in that, The multimodal data fusion model uses an attention mechanism to adaptively learn the correlation weights among the eight detection indicators, and employs a multi-head attention mechanism to model the interaction relationship between the detection data and external input features.

6. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 1, characterized in that, In the report generation step, the detection report includes at least the following information fields: Corrected values ​​and corresponding normal reference ranges for the eight testing indicators; The calibration reliability score and the error correction record for this calibration; Suitability assessment based on comprehensive soil nutrient evaluation results; The agricultural recommendation strategy includes at least fertilizer formula recommendations, irrigation amount recommendations, and pest and disease early warning prompts.

7. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 1, characterized in that, It also includes AI decision-making and feedback optimization steps: pushing the detection report to the user terminal, receiving crop growth effect data and soil retest data returned by the user after performing agricultural operations according to the agricultural recommendation strategy, and using the crop growth effect data and soil retest data to incrementally train and optimize the parameters of the multimodal data fusion model.

8. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 1, characterized in that, The built-in standard signal source is a multi-channel standard signal generation circuit with temperature self-compensation function integrated inside the probe-type soil analyzer. The multi-channel standard signal generation circuit can independently output standard electrical signals corresponding to temperature, conductivity, pH and nutrient simulation signals.

9. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 1, characterized in that, In the calibration steps, the fully automatic calibration of the sensors corresponding to the eight detection indicators is performed in the following order: first, the soil temperature sensor is calibrated; then, the conductivity and pH sensors are calibrated based on the calibrated temperature values; and finally, the sensors corresponding to available nitrogen, available phosphorus, available potassium, and water-soluble organic matter are calibrated in sequence.

10. The probe-based rapid soil nutrient testing method based on eight detection indicators as described in claim 1, characterized in that, The agricultural recommendation strategy is generated by inputting the comprehensive evaluation results of soil nutrients into a preset crop precision management decision tree model. The decision tree model combines crop type, growth period and target yield, dynamically matches the agricultural operation plan in the preset expert knowledge base, and outputs quantifiable suggestions on fertilizer application amount, fertilizer ratio, irrigation cycle and pesticide application timing.

Images