Intelligent planting decision-making method and system based on multi-source farmland data fusion

Abstract

The invention relates to the technical field of agricultural planting, in particular to an intelligent planting decision-making method and system based on multi-source farmland data fusion, and the system comprises a data aggregation module which carries out the collection of farmland data through a data collection device, carries out the centralized aggregation of the farmland data, and transmits the collected farmland data to a central platform. Soil, weather, crop growth and operation multi-source data are integrated through the data aggregation module, different data sources are fused through the data extraction module, a data foundation is laid for follow-up feature extraction, a planting decision scheme generated by the planting decision scheme is output through the strategy output module, and meanwhile language instruction form output easy to understand is generated. Farmers can easily obtain operation guidance in a text or voice mode, the farmers are helped to optimize planting management, the resource utilization rate is increased, and agricultural development is promoted.

Description

Technical Field

[0001] This invention relates to the field of agricultural planting technology, specifically to an intelligent planting decision-making method and system based on the fusion of multi-source farmland data. Background Technology

[0002] With the continuous development of modern agriculture, traditional planting decision-making methods can no longer meet the growing agricultural production needs.

[0003] Agricultural planting methods mainly rely on farmers' accumulated experience. This experience-based judgment is highly subjective. Different farmers have different understandings of soil conditions and crop growth, which can easily lead to inaccurate control of the timing and amount of irrigation, fertilization, and pesticide application, resulting in resource waste or inhibited crop growth.

[0004] In recent years, with the rapid development of technology, more and more data sources have been used in agricultural production to optimize agricultural decision-making processes. However, farmland data is multi-source. For example, soil data such as soil moisture, temperature, and pH value; meteorological data such as air temperature, precipitation, and light intensity; growth data such as crop leaf color and stem diameter; and operational data such as irrigation and fertilization are often collected independently by different devices. Although these data sources can provide a lot of valuable information for agricultural decision-making, the inconsistent data formats and varying collection frequencies make it difficult to achieve effective integration, which affects the accuracy of decision-making.

[0005] In conclusion, developing an intelligent planting decision-making method and system based on multi-source farmland data fusion, which can achieve efficient fusion of multi-source farmland data and generate easily understandable and executable planting decision schemes, is key to promoting the modernization and upgrading of agriculture. Summary of the Invention

[0006] This invention provides an intelligent planting decision-making method and system based on multi-source farmland data fusion.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] This invention provides an intelligent planting decision-making system based on multi-source farmland data fusion, the system comprising:

[0009] The data aggregation module collects farmland data through data collection equipment, centrally aggregates and transmits it to the central platform;

[0010] The data extraction module processes the farmland data and merges different data sources to extract key features.

[0011] The strategy output module generates a strategy based on the extracted key features and outputs a planting decision plan.

[0012] The decision generation module structures planting decisions and transforms them into understandable language instructions.

[0013] Furthermore, the data aggregation module includes a data acquisition unit, a data transmission unit, and a data aggregation unit. The data acquisition unit collects farmland data through the collection device. The collected data includes soil data, meteorological data, crop growth data, and operational data. The soil data includes soil moisture, soil temperature, and soil pH. The meteorological data includes air temperature, precipitation, and light intensity. The crop growth data includes crop leaf color and crop stem diameter. The operational data includes irrigation amount, fertilizer amount, and pesticide spraying amount.

[0014] Furthermore, the data extraction module includes a data processing unit, a data fusion unit, and a data extraction unit. The data processing unit converts the raw data into clean data suitable for fusion through data cleaning, standardization, and format unification. The data processing unit processes soil data, meteorological data, crop growth data, and operational data. The data cleaning includes outlier handling and missing value imputation. Outlier handling uses the 2σ rule to identify outliers; data values ​​exceeding the mean ± 2 standard deviations are considered outliers, and the outlier is replaced by the mean of valid data within 45 minutes before and after it. Missing value imputation includes short-term and long-term missing value imputation. Short-term missing data refers to missing values ​​less than 3 hours, and long-term missing data refers to missing values ​​greater than 48 hours. Short-term missing data imputation uses linear interpolation, expressed as:

[0015]

[0016] In the formula, t is the target time point where the data is missing, t0 is the nearest time point before the target time point where the data is missing, t1 is the nearest time point after the target time point where the data is missing, and X... t0 X represents the data value corresponding to time point t0. t1 The data value X corresponding to time point t1 t For short-term imputation of missing data at time point t, the long-term data imputation uses the historical average method, taking the average of historical data from the same period over the past 10 months as the imputation value. The calculation formula is as follows:

[0017]

[0018] In the formula, t is the target time point where the data is missing, and X t X represents the short-term imputation data value for the missing time point t. t-1 +X t-2 +…+X t-1010 represents the historical data value that is in the same time period as t in the past 10 months, and 10 represents the data for the same period in the past 10 months.

[0019] Furthermore, the standardization process includes standardization of continuous data and encoding conversion of discrete data. The continuous data standardization employs baseline deviation standardization and extreme value interval mapping standardization. The baseline deviation standardization includes processing data for soil moisture, soil temperature, air temperature, light intensity, and stem diameter. The extreme value interval mapping standardization includes processing data for soil pH, precipitation, irrigation, fertilization, and pesticide application. The formula for the baseline deviation standardization is as follows:

[0020]

[0021] In the formula, X i Let X be the data at a single moment, μ be the mean of all original data of the same type, σ be the standard deviation of the dataset, and X be the mean of the dataset. i ′ This represents the standardized value of the i-th sample after standardization.

[0022] The formula for the standardization of the reference value deviation is as follows:

[0023]

[0024] In the formula, X i Let X be the i-th standardized continuous data, min(x) be the minimum value among the data of the same type, max(x) be the maximum value among the data of the same type, and X be the maximum value among the data of the same type. i ′ The standardized value of the i-th sample obtained after standardization processing; the discrete data encoding is used to process the color of the crop leaves; the discrete data encoding transforms the intuitive features of leaf color into quantifiable data through four-level encoding, eliminating the ambiguity of discrete data and enabling it to participate in data layer fusion in collaboration with continuous data.

[0025] Furthermore, the four-level coding includes the following steps:

[0026] a. A four-level classification is based on the NDVI index, with the NDVI value ranging from -1 to 1, and a critical value is set in combination with the crop growth status;

[0027] b. Four levels of characteristics and corresponding NDVI ranges: Level 1: Leaves are yellowish, NDVI < 0.2, and the chlorophyll content of the leaves is extremely low; Level 2: Leaves are light green, 0.2 ≤ NDVI < 0.5, and the chlorophyll content is moderate; Level 3: Leaves are dark green, 0.5 ≤ NDVI < 0.8, and the crop is in good growth condition; Level 4: Leaves are dark green, and NDVI ≥ 0.8.

[0028] c. Map the four-level features to standardized digital codes for data fusion and model computation.

[0029] Furthermore, the data fusion unit employs data layer fusion and feature layer fusion. The data layer fusion calculates the fused value of the data using a weighted average method, the expression of which is F = ∑(W i ·X i In the formula, F represents the result after integrating the data, and W represents... i Let X be the weight value of the i-th data source. i Let represent the original data value collected from the i-th data source, and the weights of all data sources satisfy ∑W i =1, ensuring the fusion value is within a reasonable data range;

[0030] The feature layer fusion extraction of soil feature X, meteorological feature Y, crop feature Z, and operational feature O is specifically as follows: In the formula, αx, αY, αz, and αo are the attention weights for soil feature X, meteorological feature Y, crop feature Z, and operational feature O, respectively, satisfying αx + αY + αz + αo = 1. The data extraction unit performs key feature screening on the final fused feature vector after feature layer fusion. The data extraction unit uses the ReliefF algorithm to calculate the influence weight of the fused features on crop demand, retains the top 25% of features by weight, and then performs dimensionality reduction on the screened features through principal component analysis. The principal components with a cumulative contribution rate ≥90% are selected as the key features of the input strategy generation module.

[0031] Furthermore, the strategy output module generates a planting decision scheme through an intelligent model based on the key features output by the data extraction module. The intelligent model includes a random forest model trained on historical planting data and a BP neural network model. The random forest model learns the input key features through ensemble learning of multiple decision trees, and the BP neural network model fits the nonlinear relationship between the key features and the optimal planting operation through multi-layer nonlinear mapping.

[0032] Furthermore, the natural language instructions in the decision generation module include irrigation instructions, fertilization instructions, and pesticide spraying instructions. The irrigation instructions include irrigation time, irrigation amount, irrigation area, and irrigation method. The fertilization instructions include fertilizer type, fertilizer amount, fertilization time, and fertilization method. The pesticide spraying instructions include pesticide type, spraying time, spraying range, and safety interval. The natural language instructions are output in text or voice form.

[0033] On the other hand, the present invention also provides an intelligent planting decision-making method based on multi-source farmland data fusion, comprising the following steps:

[0034] S1. Data collection equipment is installed in farmland to collect data. The data aggregation module centrally aggregates the collected farmland data and transmits it to the central platform.

[0035] S2. The data extraction module processes the farmland data transmitted to the central platform.

[0036] S3. The strategy output module generates planting decision-making schemes based on the extracted key features through an intelligent model.

[0037] S4, the decision generation module processes the planting decision scheme, converts it into understandable natural language instructions, and outputs them.

[0038] Further, in step S1, the data collection device includes a soil moisture sensor, a soil temperature sensor, and a pH meter. The soil moisture sensor is a capacitive soil moisture sensor to measure the moisture content in the soil. The capacitive soil moisture sensor is vertically inserted into the soil to a depth of 20-30 cm. The soil temperature sensor is a digital temperature sensor to measure the temperature in the soil. It is installed in the same position as the soil moisture sensor to a depth of 10-20 cm. The pH meter is used to detect the soil pH value. During detection, it is vertically inserted into the soil at a depth of 20-30 cm in the crop root growth area. The data aggregation module includes a data acquisition unit, a data transmission unit, and a data aggregation unit. After the data acquisition unit collects farmland data through the collection device, the data transmission unit transmits the data to the data aggregation unit for centralized aggregation, and then transmits it to the central platform.

[0039] Beneficial effects

[0040] Compared with known public technologies, the technical solution provided by this invention has the following beneficial effects:

[0041] When in use, this invention integrates multi-source data such as soil, meteorology, crop growth, and operation through a data aggregation module, and then fuses different data sources through a data extraction module to lay the data foundation for subsequent feature extraction. The planting decision scheme generated by the strategy output module is output, and easy-to-understand language instructions are generated for output. Farmers can easily obtain operation guidance through text or voice, which helps farmers optimize planting management, improve resource utilization, and promote agricultural development. Attached Figure Description

[0042] Figure 1 This is a flowchart of an intelligent planting decision-making system based on multi-source farmland data fusion according to the present invention;

[0043] Figure 2 This is a system diagram of an intelligent planting decision-making method based on multi-source farmland data fusion according to the present invention. Detailed Implementation

[0044] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0045] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0046] The present invention will now be described in further detail with reference to the accompanying drawings:

[0047] Example 1:

[0048] like Figure 1 - Figure 2 As shown, this invention provides an intelligent planting decision-making system based on multi-source farmland data fusion, comprising:

[0049] The data aggregation module collects farmland data through data collection equipment, centrally aggregates and transmits it to the central platform;

[0050] Furthermore, the data aggregation module includes a data acquisition unit, a data transmission unit, and a data aggregation unit. The data acquisition unit collects farmland data through collection equipment. The collected data includes soil data, meteorological data, crop growth data, and operational data. Soil data includes soil moisture, soil temperature, and soil pH value. Meteorological data includes air temperature, precipitation, and light intensity. Crop growth data includes crop leaf color and crop stem diameter. Operational data includes irrigation amount, fertilizer amount, and pesticide spraying amount.

[0051] The data extraction module processes farmland data and merges data from different data sources to extract key features.

[0052] Furthermore, the data extraction module includes a data processing unit, a data fusion unit, and a data extraction unit.

[0053] Specifically, the data processing unit transforms raw data into clean data suitable for fusion through data cleaning, standardization, and format unification. The data processing unit handles soil data, meteorological data, crop growth data, and operational data. Data cleaning includes outlier handling and missing value imputation. Outlier handling uses the 2σ rule to identify outliers; data values ​​exceeding the mean ± 2 standard deviations are considered outliers, and the outlier is replaced by the mean of valid data within 45 minutes before and after it. Missing value imputation includes short-term and long-term imputation. Short-term data refers to missing values ​​less than 3 hours old, and long-term data refers to missing values ​​greater than 48 hours old. Short-term imputation uses linear interpolation, expressed as:

[0054]

[0055] In the formula, t is the target time point where the data is missing, t0 is the nearest time point before the target time point where the data is missing, t1 is the nearest time point after the target time point where the data is missing, and X... t0 X represents the data value corresponding to time point t0. t1 The data value X corresponding to time point t1 t For short-term missing data points at time point t, imputation is performed using the historical average method, taking the average of historical data from the same period over the past 10 months as the imputation value. The calculation formula is as follows:

[0056]

[0057] In the formula, t is the target time point where the data is missing, and X t X represents the short-term imputation data value for the missing time point t. t-1 +X t-2 +…+X t-10 10 represents the historical data value that is in the same time period as t in the past 10 months, and 10 represents the data for the same period in the past 10 months.

[0058] Furthermore, the standardization process includes standardization of continuous data and encoding transformation of discrete data. Continuous data standardization employs baseline deviation standardization and extreme value interval mapping standardization. Baseline deviation standardization includes processing data for soil moisture, soil temperature, air temperature, light intensity, and stem diameter. Extreme value interval mapping standardization includes processing data for soil pH, precipitation, irrigation, fertilization, and pesticide spraying. The formula for baseline deviation standardization is as follows:

[0059]

[0060] In the formula, X i Let X be the data at a single moment, μ be the mean of all original data of the same type, σ be the standard deviation of the dataset, and X be the mean of the dataset. i ′ This represents the standardized value of the i-th sample after standardization.

[0061] The formula for standardizing the baseline deviation is as follows:

[0062]

[0063] In the formula, X i Let X be the i-th standardized continuous data, min(x) be the minimum value among the data of the same type, max(x) be the maximum value among the data of the same type, and X be the maximum value among the data of the same type. i ′ The standardized value of the i-th sample obtained after standardization is used for crop leaf color processing. Discrete data encoding is used to process the color of crop leaves. Discrete data encoding transforms the intuitive features of leaf color into quantifiable data through four-level encoding, eliminating the ambiguity of discrete data and enabling it to participate in data layer fusion in collaboration with continuous data.

[0064] Furthermore, the four-level coding includes the following steps:

[0065] a. Based on the NDVI index, a four-level classification is made, with the NDVI value ranging from -1 to 1. Combined with the crop growth status, which includes chlorophyll content and photosynthetic intensity, a critical value is set, and the differences in the levels can accurately reflect the crop health level.

[0066] b. Four levels of characteristics and corresponding NDVI ranges: Level 1: Leaves are yellow, NDVI < 0.2, indicating extremely low chlorophyll content, suggesting stunted growth due to water shortage, fertilizer deficiency, or pests and diseases; Level 2: Leaves are light green, 0.2 ≤ NDVI < 0.5, indicating moderate chlorophyll content, suggesting average crop growth, in a critical nutritional stage; Level 3: Leaves are dark green, 0.5 ≤ NDVI < 0.8, indicating sufficient chlorophyll content, vigorous photosynthesis, and good crop growth; Level 4: Leaves are dark green, NDVI ≥ 0.8, indicating excessively high chlorophyll content, caused by excessive fertilization or insufficient light.

[0067] c. Map the four-level features to standardized digital codes for data fusion and model calculation, where level 1 corresponds to code "0", level 2 corresponds to code "1", level 3 corresponds to code "2", and level 4 corresponds to code "3".

[0068] The strategy output module generates strategies based on extracted key features and outputs planting decision schemes.

[0069] Furthermore, the standardized format converts data from different collection frequencies into a standard data time of one record every 20 minutes.

[0070] Furthermore, the data fusion unit employs data-layer fusion and feature-layer fusion. Data-layer fusion calculates the fused value of the data using a weighted average method, expressed as F = ∑(W i ·X i In the formula, F represents the result after integrating the data, and W represents... i Let X be the weight value of the i-th data source. i Let represent the original data value collected from the i-th data source, and the weights of all data sources satisfy ∑W i =1, ensuring the fusion value is within a reasonable data range; feature layer fusion extracts soil feature X, meteorological feature Y, crop feature Z, and operational feature O, specifically: In the formula, αx, αY, αz, and αo are the attention weights for soil feature X, meteorological feature Y, crop feature Z, and operational feature O, respectively, satisfying αx + αY + αz + αo = 1, F e‖ The final fused feature vector after feature layer fusion is obtained. The data extraction unit performs key feature screening on the final fused feature vector after feature layer fusion. The data extraction unit uses the ReliefF algorithm to calculate the influence weight of the fused features on crop demand, retains the top 25% of features by weight, and then performs dimensionality reduction on the screened features through principal component analysis. The principal components with a cumulative contribution rate of ≥90% are selected as the key features for input strategy generation module.

[0071] The strategy output module generates strategies based on extracted key features and outputs planting decision schemes.

[0072] Furthermore, based on the key features output by the data extraction module, a planting decision scheme is generated through an intelligent model. The intelligent model includes a random forest model trained on historical planting data and a BP neural network model. The random forest model learns the input key features through the ensemble learning of multiple decision trees, and the BP neural network model fits the nonlinear relationship between the key features and the optimal planting operation through multi-layer nonlinear mapping.

[0073] The decision generation module structures planting decisions and transforms them into understandable language instructions.

[0074] Furthermore, the natural language instructions in the decision generation module include irrigation instructions, fertilization instructions, and pesticide spraying instructions. The irrigation instructions include irrigation time, irrigation amount, irrigation area, and irrigation method. The fertilization instructions include fertilizer type, fertilizer amount, fertilization time, and fertilization method. The pesticide spraying instructions include pesticide type, spraying time, spraying range, and safety interval. The natural language instructions are output in text or voice form.

[0075] Example 2 provides an intelligent planting decision-making method based on multi-source farmland data fusion, including the following steps:

[0076] S1. Data collection equipment is installed in the farmland to collect data. The data aggregation module centrally aggregates the collected farmland data and transmits it to the central platform.

[0077] Furthermore, the data collection equipment includes a soil moisture sensor, a soil temperature sensor, and a pH meter. The soil moisture sensor uses a capacitive soil moisture sensor to measure the moisture content in the soil. In farmland, the capacitive soil moisture sensor is vertically inserted into the soil to a depth of 20-30 cm. The soil temperature uses a digital temperature sensor to measure the temperature in the soil, installed in the same location as the soil moisture sensor to a depth of 10-20 cm. The pH meter is used to detect the soil pH value. During detection, it is vertically inserted into the soil, 20-30 cm from the crop root growth area. At the same time, meteorological data such as air temperature, precipitation, and light intensity are collected through weather stations and other equipment. Crop leaf color and crop stem diameter are acquired through image acquisition equipment and other means. Operational data such as irrigation amount, fertilizer amount, and pesticide spraying amount are recorded. After the data collection unit collects these data, the data transmission unit transmits them to the data aggregation unit for centralized aggregation and then transmits them to the central platform.

[0078] S2, the data extraction module processes the farmland data transmitted to the central platform.

[0079] Furthermore, outliers are first removed and missing values ​​are filled through data cleaning, then standardization and format unification are performed, and the cleaned data is then fused at the data layer and feature layer to extract key features.

[0080] S3, the strategy output module generates planting decision schemes based on the extracted key features through an intelligent model.

[0081] Furthermore, the intelligent model includes a random forest model and a BP neural network model trained based on historical planting data, which together generate planting decision schemes and generate specific parameters for irrigation, fertilization and spraying.

[0082] S4, the decision generation module processes the planting decision scheme, converts it into understandable natural language instructions, and outputs them.

[0083] Furthermore, the specific values ​​and requirements related to irrigation, fertilization, and pesticide spraying are analyzed, and then these parameters are converted into understandable natural language instructions and output according to the preset language conversion rules. The natural language instructions include irrigation instructions, fertilization instructions, and pesticide spraying instructions. The irrigation instructions specify the irrigation time, irrigation amount, irrigation area, and irrigation method. The fertilization instructions specify the fertilizer type, fertilizer amount, fertilization time, and fertilization method. The pesticide spraying instructions specify the pesticide type, spraying time, spraying range, and safety interval. The natural language instructions are output in the form of text or voice.

[0084] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent planting decision system based on multi-source farmland data fusion, characterized in that, The system comprises: a data aggregation module for collecting and aggregating farmland data through a data collection device and transmitting the data to a central platform; a data extraction module for processing the farmland data and fusing different data sources to extract key features; a strategy output module for generating a strategy based on the extracted key features and outputting a planting decision scheme; a decision generation module for structurally processing the planting decision scheme and converting it into understandable language instructions.

2. The intelligent planting decision system based on multi-source farmland data fusion according to claim 1, characterized in that, The data aggregation module comprises a data collection unit, a data transmission unit and a data aggregation unit, wherein the data collection unit collects farmland data through the collection device, and the collected data includes soil data, weather data, crop growth data and operation data, wherein the soil data includes soil moisture, soil temperature and soil pH value, the weather data includes air temperature, precipitation and light intensity, the crop growth data includes crop leaf color and crop stem diameter, and the operation data includes irrigation amount, fertilizer amount and pesticide spraying amount.

3. The intelligent planting decision system based on multi-source farmland data fusion according to claim 2, characterized in that, The data extraction module comprises a data processing unit, a data fusion unit and a data extraction unit, the data processing unit converts raw data into neat data that can be used for fusion through data cleaning, standardization processing and format unification, the processing objects of the data processing unit include soil data, weather data, crop growth data and operation data, the data cleaning includes data outlier processing and data missing value filling, wherein the data outlier processing adopts 2σ rule to judge outliers, and when the data value exceeds the range of mean value ± 2 times standard deviation, it is considered as an outlier, and the mean value of effective data within 45 minutes before and after the outlier is used for replacement, the data missing value filling includes short-term data missing filling and long-term data missing filling, wherein short-term data refers to missing time less than 3 hours, and long-term data refers to missing time more than 48 hours, the short-term data missing filling adopts linear interpolation method for filling, and the expression is: In the formula, t is the target time point of data loss, t0 is the nearest time point before the target time point of loss, t1 is the nearest time point after the target time point of loss, X t0 is the data value corresponding to the time point t0, X t1 is the data value corresponding to the time point t1, X t is the short-term filling data value of the missing time point t, the long-term data loss filling adopts the historical same period mean method to fill, and the historical data mean of the same time period in the past 10 months is taken as the filling value, and the calculation formula is: In the formula, t is the target time point of data missing, X t is the short-term filling data value of the missing time point t, X t-1 +X t-2 +…+X t-10 is the historical data value in the same time period as t in the past 10 months, and 10 is the same period data in the past 10 months.

4. The intelligent planting decision system based on multi-source farmland data fusion according to claim 3, characterized in that, The standardization processing includes continuous data standardization processing and discrete data encoding conversion, the continuous data standardization adopts benchmark value deviation standardization and extreme value interval mapping standardization processing, wherein the benchmark value deviation standardization processing includes the processing of soil moisture, soil temperature, air temperature, light intensity and stem diameter data, the extreme value interval mapping standardization processing includes the processing of soil pH value, precipitation, irrigation amount, fertilizer amount and pesticide spraying amount, the expression formula of the benchmark value deviation standardization processing is, where X i is the data at a time, μ is the mean of all raw data of the same kind, σ is the standard deviation of the data set, X i ′ is the standardized value of the i-th sample after standardization. The expression formula of the benchmark value deviation standardization processing is: In the formula, X i is the ith normalized continuous data, min(x) is the minimum value in the same data, max(x) is the maximum value in the same data, X i ′ The normalized value of the ith sample obtained after normalization, the discrete data code is used for processing the crop leaf color, and the discrete data code is converted into quantifiable data by four-level coding, eliminating the ambiguity of discrete data, so that it can participate in data layer fusion with continuous data.

5. The intelligent planting decision system based on multi-source farmland data fusion according to claim 4, characterized in that, The four-level coding comprises the following steps: a. four-level division based on NDVI index, NDVI value range is -1 to 1, and critical value is set according to crop growth state; b. Four-level features and corresponding NDVI ranges, level 1: leaves are yellow, NDVI <0.2, leaf chlorophyll content is extremely low; level 2: leaves are light green, 0.2≤NDVI<0.5, chlorophyll content is medium; level 3: leaves are dark green, 0.5≤NDVI<0.8, crop growth state is good; level 4: leaves are dark green, NDVI≥0.8; c. The four-level features are mapped to standardized digital codes for data fusion and model calculation.

6. The intelligent planting decision system based on multi-source farmland data fusion according to claim 4, characterized in that, The format unification converts data of different collection frequencies into data standard time of one piece per 20 minutes, and the data fusion unit adopts data layer fusion and feature layer fusion, wherein the data layer fusion calculates a fusion value of data through a weighted average method, and an expression of the weighted average method is F=∑(W i ·X i ), wherein F is a result after data integration, W i is a weight value of the i-th data source, X i represents an original data value collected by the i-th data source, and weights of all data sources satisfy ∑W i =1, so as to ensure that the fusion value is within a reasonable data range. The feature layer fusion extracts soil features X, meteorological features Y, crop features Z and operation features O, and specifically F e‖ =αx·X+αY·Y+αz·z+αo·o, wherein αx, αY, αz and αo are attention weights of the soil features X, the meteorological features Y, the crop features Z and the operation features O, and satisfy αx+αY+αz+αo=1, F e‖ The data extraction unit performs key feature screening on the final fusion feature vector after the feature layer fusion, the data extraction unit calculates the influence weight of the fusion features on crop demand by using a ReliefF algorithm, retains the features with the top 25% weight, and then performs dimension reduction on the screened features by principal component analysis, and selects the principal components with a cumulative contribution rate of ≥90% as the key features of the input strategy generation module.

7. The intelligent planting decision system based on multi-source farmland data fusion according to claim 6, characterized in that, The strategy output module generates a planting decision scheme based on the key features output by the data extraction module, and the intelligent model includes a random forest model trained based on historical planting data and a BP neural network model. The random forest model integrates learning of input key features through multiple decision trees, and the BP neural network model fits the nonlinear relationship between the key features and the optimal planting operation through multiple layers of nonlinear mapping.

8. The intelligent planting decision system based on multi-source farmland data fusion according to claim 6, characterized in that, The natural language instructions in the decision generation module include irrigation instructions, fertilization instructions, and pesticide spraying instructions. The irrigation instructions include irrigation time, irrigation amount, irrigation area, and irrigation method. The fertilization instructions include fertilizer type, fertilization amount, fertilization time, and fertilization method. The pesticide spraying instructions include pesticide type, spraying time, spraying range, and safety interval. The natural language instructions are output in the form of text or voice.

9. An intelligent planting decision method based on multi-source farmland data fusion, based on the intelligent planting decision system based on multi-source farmland data fusion of any one of claims 1-8, characterized in that, The method comprises the following steps: S1, a data collection device is installed in a farmland to collect data, and a data aggregation module centrally aggregates the collected farmland data and transmits the data to a central platform; S2, a data extraction module processes the farmland data transmitted to the central platform; S3, a strategy output module generates a planting decision scheme based on the extracted key features through an intelligent model; S4, a decision generation module processes the planting decision scheme, converts it into understandable natural language instructions, and outputs the instructions.

10. The intelligent planting decision-making method based on multi-source farmland data fusion according to claim 9, characterized in that, In step S1, the data collection device includes a soil moisture sensor, a soil temperature sensor, and a pH tester. The soil moisture sensor measures the water content in the soil using a capacitive soil moisture sensor. The capacitive soil moisture sensor is vertically inserted into the soil to a depth of 20-30 cm. The soil temperature is measured using a digital temperature sensor, which is installed at the same position as the soil moisture sensor and has a depth of 10-20 cm. The pH tester is used to detect the pH value of the soil. The pH tester is vertically inserted into the soil to a depth of 20-30 cm in the crop root growth area. The data aggregation module includes a data collection unit, a data transmission unit, and a data aggregation unit. The data collection unit collects farmland data through the collection device, and then transmits the data to the data aggregation unit for centralized aggregation. The data is then transmitted to the central platform.

Images