Agricultural internet of things multi-source heterogeneous data preprocessing system and method

Abstract

The present application relates to a kind of agricultural internet of things multi-source heterogeneous data preprocessing system and method, comprising the following steps: step one, the data required by agriculture is collected from multiple sources, and multi-source heterogeneous data is established;RGB video image data and meteorological environment data of crop growth are obtained;Step two, the outlier correction of the collected data, the noise reduction of data, the filling of missing data and smoothing processing;Step three, the index fusion is carried out to the cleaned data.The system integrates the multi-source heterogeneous data processing flow into a module in the process of collecting different indexes by different devices with different collection intervals;For the type of data missing, the system selects different filling means;Meanwhile, the data of different structures and different collection intervals are aligned in time in the present application, to retain the time correlation between indexes.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of agricultural information engineering, and particularly relates to a preprocessing system and method for multi-source heterogeneous data of an agricultural Internet of Things. BACKGROUND

[0002] With the continuous development of the agricultural Internet of Things, satellite, unmanned aerial vehicle remote sensing technology, optical temperature and humidity sensors, picture and video technologies are applied to the data collection process of the crop growth system; with the development of data-driven digital agriculture and smart agriculture, the collected data will be used in data-driven decision models to provide auxiliary decision-making for agricultural production.

[0003] At present, the preprocessing of data in smart agriculture mainly includes cleaning of abnormal data and missing data, and index fusion of multi-source heterogeneous data. Due to the single agricultural data collection, the existing agricultural data fusion technology is only limited to the fusion of multi-sensors in the spatial range of ground climate index data collection (such as air temperature, air humidity, soil temperature, and soil humidity); however, with the development of collection technology, remote sensing image data, video data, and image data are also used for data collection in the front end of the agricultural system, and how to fuse the above image, video, and structured climate data at the crop index level is particularly important for the next agricultural decision-making. SUMMARY

[0004] The application is proposed to solve the above problems, and provides a preprocessing system and method for multi-source heterogeneous data of an agricultural Internet of Things.

[0005] The application relates to a preprocessing method for multi-source heterogeneous data of an agricultural Internet of Things, comprising the following steps:

[0006] Step one, multi-source collection of data required by agriculture is performed to establish multi-source heterogeneous data; RGB video image data and meteorological environment data of crop growth are acquired;

[0007] Step two, abnormal value correction, data denoising, missing data filling, and smoothing processing of the collected data are performed;

[0008] Step three, index fusion is performed on the cleaned data.

[0009] Step two comprises the following operations:

[0010] Whether the structural attribute of the data is remote sensing image data is identified, if yes, special processing of the remote sensing image data is performed, and if not, a time series data processing mode is performed;

[0011] For remote sensing image data acquired by satellite and UAV remote sensing, atmospheric correction, orthogonal correction, geometric correction, as well as mosaic processing and color balancing processing are performed on the remote sensing image data through a remote sensing image processing platform.

[0012] The ground sensor data is time-corrected according to the preset acquisition interval 'a'. This mainly involves calculating the time required for a time length N based on the interval 'a'. Data points, data volume per month Data volume per day Data volume per hour Determine if data with the same interval has the same length. If they are the same, match them according to the same time index. If they are different, rebuild the time index according to the position of the missing value.

[0013] Step two also includes the following operations:

[0014] For meteorological data with n attributes collected by ground sensors at fixed time intervals, for a single-attribute dataset {x1, x2, ..., x...} N}, by setting a threshold δ, through To identify and remove abnormal data, treating it as missing values;

[0015] Determine whether the missing values ​​are discrete point missing values ​​or continuous block missing values ​​over time. For point missing values, imputation is performed using existing linear interpolation methods. For continuous block missing values, a forward dataset of the missing blocks {x1, x2, ..., x} is constructed. m}, and the backward dataset {x m+t+1 x m+t+2 , ..., x N}, construct an RNN network from the forward and backward directions to fill in the missing data;

[0016] The populated meteorological attribute dataset is used to separate noisy data and information data through wavelet transform. After removing the noisy data, a complete and clean time series data is finally obtained.

[0017] For video image data, images are captured from the video data according to a preset acquisition time interval, forming image data in the time series dimension.

[0018] Step three includes the following operations:

[0019] Generate topographic and terrain image maps of crop growing areas based on remote sensing image data;

[0020] According to the different reflection characteristics of the chlorophyll content of crops in different wave bands, the visible light band and the near-infrared band of the remote sensing spectrum image are selected, and a linear regression model or a machine learning model is selected to obtain the chlorophyll content of crop growth.

[0021] According to the propagation efficiency of light in the crown layer of different vegetation crown structures, the characteristics of the crown layer reflectance spectrum are used to obtain the leaf area index at the field scale, and the calculation formula of the leaf area index is:

[0022] The normalized vegetation index is calculated by using the existing formula: The reflectivity is represented by, and is calculated according to the near-infrared band and the infrared band.

[0023] Step three further includes the following operations:

[0024] Crop index data is generated, a multiple regression model is constructed by the correlation between soil organic matter, salt content and spectral reflectance, soil organic matter content, salt content = f (x, R), and the correlation coefficient R is obtained by the existing soil type inversion model;

[0025] For the index fusion process of the ground meteorological sensor, the terrain type of the growth area is marked by the terrain information, {0, 1, 2} respectively represents the transition from the lowland to the highland, and the ground multi-position sensor is matched with the terrain category;

[0026] The terrain and the ground index attribute are used for statistical weighted average of the single index multi-sensor, and the meteorological index fusion results of air temperature, humidity, light, wind speed and direction are obtained; and the index fusion results of soil temperature and humidity are obtained;

[0027] Soil moisture and soil index data are generated, and multi-source heterogeneous data are stored according to time dimension and category data for subsequent data mining and analysis.

[0028] The application also relates to a system for implementing the preprocessing method of multi-source heterogeneous data of an agricultural Internet of Things.

[0029] The agricultural Internet of Things data acquisition module uses multiple devices to collect agricultural data from multiple sources, uses satellites, unmanned aerial vehicles to collect radar sensor data and remote sensing image data, establishes multi-spectral, hyperspectral, thermal infrared, SAR and chlorophyll fluorescence multi-source heterogeneous data of crop growth areas, uses RGB video image data obtained by video monitoring of crop growth areas, and uses wireless sensor networks arranged on the ground of crop growth to obtain meteorological environment data of crop growth.

[0030] The data cleaning module of the agricultural Internet of Things corrects the outliers of the collected data, reduces the noise of the data, fills the missing data and performs smoothing processing, and the multispectral image data generated by satellite and unmanned aerial vehicle remote sensing is subjected to atmospheric correction, radiation correction, geometric correction and splicing processing, and uniform color processing; for time series data, noise points are detected through autocorrelation of the time series data and are deleted, and different types of missing values are used to achieve the purpose of filling the missing data.

[0031] The multi-source heterogeneous data index fusion module of the agricultural sensor fuses the indexes of the cleaned data according to crop growth element indexes, climate indexes and soil moisture and soil composition.

[0032] Beneficial effects

[0033] 1. The agricultural Internet of Things multi-source heterogeneous data can be classified and cleaned

[0034] In the process of collecting different indexes by different devices at different collection intervals, the system integrates the multi-source heterogeneous data processing flow into a module, selects different data processing methods for different data structures, selects different filling methods for the types of data missing, and aligns the data of different structures and different collection intervals in time to retain the time correlation between indexes.

[0035] 2. The data index fusion processing in different dimensions can be realized

[0036] In the index fusion process, the fusion of meteorological indexes needs to be based on the topographic index generated by remote sensing image to perform spatial fusion, and in the fusion of remote sensing spectral indexes, the system can generate crop, soil and other index data through the spectral fusion method integrated in the module. Data users only need to select the indexes they are interested in for the next step of analysis and visualization.

[0037] 3. The agricultural Internet of Things data mining and agricultural decision-making implementation are guaranteed

[0038] In the present application, after the original collected data is processed by the index fusion module, meteorological indexes, soil moisture and soil nutrient indexes, crop growth indexes, image data and non-image data are generated in the time dimension, and developers only need to extract data when performing agricultural data mining and crop model construction; after the corresponding decision model is generated, the index fused data can also be directly put into the model to generate a decision result. BRIEF DESCRIPTION OF DRAWINGS

[0039] Fig. 1 It is a structural schematic view of the agricultural Internet of Things multi-source heterogeneous data preprocessing system and method.

[0040] Fig. 2 It is a flowchart of the agricultural Internet of Things multi-source heterogeneous data preprocessing system and method of the application. DETAILED DESCRIPTION

[0041] The following will be combined Figs. 1-2 The present embodiment will be specifically described.

[0042] The agricultural Internet of Things multi-source heterogeneous data preprocessing system of the application comprises an agricultural Internet of Things data acquisition module, an agricultural Internet of Things data cleaning module and an agricultural sensor multi-source heterogeneous data index fusion module.

[0043] The agricultural Internet of Things data acquisition module collects data required by agriculture from multiple sources by using multiple devices, collects radar sensor data and remote sensing image data by using satellites and unmanned aerial vehicles, establishes multi-spectral, hyperspectral, thermal infrared, SAR and chlorophyll fluorescence multi-source heterogeneous data of crop growth areas, acquires RGB video image data by using video monitoring of crop growth areas, and acquires meteorological environment data of crop growth by using wireless sensor networks arranged on the ground of crop growth.

[0044] The agricultural Internet of Things data cleaning module corrects outliers of collected data, reduces noise of data, fills in missing data and performs smoothing processing, performs atmospheric correction, radiation correction, geometric correction, splicing processing and color uniformity processing on multi-spectral image data generated by satellite and unmanned aerial vehicle remote sensing, detects noise points through autocorrelation of time series data and deletes them, and fills in missing data by different types of missing values.

[0045] The agricultural sensor multi-source heterogeneous data index fusion module fuses indexes according to crop growth element indexes, climate indexes and soil moisture and soil composition of three main aspects after cleaning the data.

[0046] The data cleaning module corrects outliers of collected data, reduces noise of data and fills in missing values, and the specific method comprises:

[0047] Step 1, identify whether the structural attribute of the data is remote sensing image data, if so, perform special processing of remote sensing image data, if not, perform time series data processing mode.

[0048] Step 2, for remote sensing image data acquired by satellite and unmanned aerial vehicle remote sensing, perform atmospheric correction, orthographic correction, geometric correction, splicing processing and color uniformity processing operations on the remote sensing image data by using professional remote sensing image processing platforms such as eCognition, ENVI and ERDAS.

[0049] Step 3: Perform time correction on the ground sensor data collected according to the preset acquisition interval 'a'. This mainly involves calculating the time required for time length N based on interval 'a'. Data points, data volume per month Data volume per day Data volume per hour Determine if data with the same interval has the same length. If they are the same, match them according to the same time index. If they are different, rebuild the time index according to the position of the missing value.

[0050] Step 4: For meteorological data with n attributes collected by ground sensors at fixed time intervals, for a single-attribute dataset {x1, x2, ..., x...} N}, by setting a threshold δ, through This is used to identify and remove abnormal data, treating it as missing values.

[0051] Step 5: Determine whether the missing values ​​are discrete point missing values ​​or continuous block missing values ​​in the time series. For point missing values, impute them using existing linear interpolation methods. For continuous block missing values, construct a forward dataset of the missing blocks {x1, x2, ..., x...}. m}, and the backward dataset (x) m+t+1 x m+t+2 , ..., x N We construct an RNN network from both forward and backward directions to fill in the missing data.

[0052] Step 6: Separate the noise data and information data from the populated meteorological attribute dataset using wavelet transform, delete the noise data, and finally obtain complete and clean time series data.

[0053] Step 7: For video image data, capture images of the video data according to the preset acquisition time interval t to form image data in the time series dimension.

[0054] In the aforementioned preprocessing system and method for multi-source heterogeneous data in agricultural IoT, the multi-source heterogeneous data index fusion module of agricultural sensors generates data indicators according to the following methods based on crop growth indicators, meteorological environment indicators, soil moisture indicators, and soil composition indicators:

[0055] Step 1: Using the remote sensing image data, generate a topographic image map of the crop growing area using the software employed in Step 1 of the data cleaning module.

[0056] Step 2, according to the different reflection characteristics of crop chlorophyll content in different wave bands, the visible light band and near infrared band images of remote sensing spectrum are selected, and a linear regression model or a machine learning model is selected to obtain the chlorophyll content of crop growth.

[0057] Step 3, according to the propagation efficiency of light in the canopy of different vegetation canopy structure images, the characteristics of the canopy reflectance spectrum are used, and the software used in step 1 in the data cleaning module or the existing inversion model or manual annotation is used to obtain the leaf area index at the field scale. The formula for calculating the leaf area index is:

[0058] Step 4, normalized vegetation index calculation, using the existing formula: ρ represents reflectivity, which is calculated according to the near-infrared band and infrared band.

[0059] Step 5, generate crop index data through steps 1, 2, 3, and 4.

[0060] Step 6, build a multiple regression model based on the correlation between soil organic matter and salt content and spectral reflectance, soil organic matter content and salt content = f(x, R), and the correlation coefficient R is obtained through the existing soil type inversion model;

[0061] Step 7, for the index fusion process of the ground meteorological sensor, the terrain type of the growth area is marked through the terrain information, {0, 1, 2} respectively represent the transition from low land to high land, and the ground multi-position sensor is matched with the terrain category;

[0062] Step 8, through the terrain and ground index attributes, the single index multi-sensor is statistically weighted and averaged to obtain the meteorological index fusion results of air temperature, humidity, light, wind speed and direction, and the index fusion results of soil temperature and humidity.

[0063] Step 9, generate soil moisture and soil index data through steps 6 and 8.

[0064] Step 10, store multi-source heterogeneous data according to time dimension and category data for subsequent data mining and analysis.

[0065] The above is only a preferred embodiment of the present application, not for limiting the implementation of the present application, those skilled in the art can easily make corresponding changes or modifications according to the main idea and spirit of the present application, therefore, the protection scope of the present application should be limited by the protection scope required by the claims.

Claims

1. A method for preprocessing multi-source heterogeneous data of an agricultural Internet of Things, characterized in that, Comprise the following steps: Step one, multi-source collection of required data, establishment of multi-source heterogeneous data, and acquisition of video image data and meteorological environment data for crop growth; Step two, outlier correction, data denoising, missing data filling and smoothing processing of the collected data; Step three, index fusion of the cleaned data; Identify whether the structural attribute of the data is remote sensing image data, if so, perform special processing of the remote sensing image data, if not, perform time series data processing mode; For remote sensing image data, perform atmospheric correction, orthographic correction, geometric correction, splicing processing, and uniform color processing operation through a remote sensing image processing platform; According to the preset collection interval The collection data of the ground sensor data is time corrected according to the interval Respectively calculate the time Length under the should have Data, data amount under the month unit , data amount under the day unit , data amount under the hour unit , judge whether the data length with the same interval is same, if same, match according to the unified time index, if different, reconstruct the time index according to the missing value position; The ground sensors collect n-attribute weather data at fixed time intervals, and for a single attribute data set , by setting a threshold , by to identify abnormal data and remove it as a missing value; Confirm whether the missing value category is a discrete point missing on a time series or a continuous block missing, for the point missing, fill in by the existing linear interpolation method, for the continuous block missing, construct a forward data set of the missing block and a backward data set , construct an RNN network from the forward and backward to fill in the missing data; Separate noise data and information data through wavelet transform of the filled meteorological attribute data set, delete noise data, and finally obtain complete and clean time series data; For video image data, image capture is performed on the video data according to the preset collection time interval to form picture data in the time series dimension; The crop index data is generated, a multiple regression model is constructed by correlation of soil organic matter, salt content and spectral reflection, soil organic matter content, salt content , correlation coefficient is obtained by an existing soil type inversion model; For the index fusion process of the ground meteorological sensor, the terrain type of the growth region is marked by the terrain information, respectively represent the transition from the lowland to the highland, and the ground multi-position sensor is matched with the terrain category; Through topography and ground index attributes, statistical weighted average is performed on single-index multi-sensor collection to obtain meteorological index fusion results such as air temperature, humidity, illumination, wind speed and direction, and soil temperature and humidity index fusion results; Soil moisture content and soil index data are generated, and multi-source heterogeneous data are stored according to time dimension and category data for subsequent data mining and analysis.

2. The pre-processing method of agricultural Internet of Things multi-source heterogeneous data according to claim 1, characterized in that, Step three includes the following operations: For remote sensing image data, generate topography and terrain image of the crop growth area; Crop chlorophyll content has different reflection characteristics in different wave bands, and visible light wave band and near-infrared wave band images of remote sensing spectrum are obtained through linear regression model or machine learning model to obtain crop growth chlorophyll content. The propagation efficiency of light in the canopy was imaged for different vegetation canopy structures, and the leaf area index at the field scale was obtained using the characteristics of the canopy reflectance spectrum. The formula for calculating the leaf area index is: ; The normalized difference vegetation index is calculated using the following formula: Rnir = reflectance in the near infrared band, Rnir = reflectance in the near infrared band, Rnir = reflectance in the near infrared band, 3. A system for implementing the pre-processing method of the agricultural Internet of Things multi-source heterogeneous data according to any one of claims 1 to 2, characterized in that, The system comprises an agricultural Internet of Things data collection module, an agricultural Internet of Things data cleaning module, and an agricultural sensor multi-source heterogeneous data index fusion module.

4. The pre-processing system of agricultural Internet of Things multi-source heterogeneous data according to claim 3, characterized in that, The agricultural Internet of Things data collection module collects agricultural required data from multiple sources using multiple devices, collects radar sensor data and remote sensing image data using satellites and unmanned aerial vehicles, establishes multi-spectral, hyperspectral, thermal infrared, SAR, and chlorophyll fluorescence multi-source heterogeneous data of the crop growth area, uses RGB video image data obtained by video monitoring of the crop growth area, and uses wireless sensor networks arranged on the ground of the crop growth area to obtain meteorological environment data for crop growth.

5. The pre-processing system of agricultural Internet of Things multi-source heterogeneous data according to claim 4, characterized in that, The agricultural Internet of Things data cleaning module corrects outliers, denoises data, fills missing data, and smoothes the collected data, uses special software to perform atmospheric correction, radiation correction, geometric correction, splicing processing, and uniform color processing on multi-spectral image data generated by satellites and unmanned aerial vehicles, detects noise points through autocorrelation of time series data, deletes the noise points, and fills missing data by different types of missing values.

6. The pre-processing system of agricultural Internet of Things multi-source heterogeneous data according to claim 5, characterized in that, The agricultural sensor multi-source heterogeneous data index fusion module fuses the cleaned data according to crop growth element indexes, climate indexes, and soil moisture content and soil composition.

Images