CN114529838B - Soil nitrogen content inversion model construction method and system based on convolutional neural network - Google Patents
Soil nitrogen content inversion model construction method and system based on convolutional neural network Download PDFInfo
- Publication number
- CN114529838B CN114529838B CN202210432884.2A CN202210432884A CN114529838B CN 114529838 B CN114529838 B CN 114529838B CN 202210432884 A CN202210432884 A CN 202210432884A CN 114529838 B CN114529838 B CN 114529838B
- Authority
- CN
- China
- Prior art keywords
- soil
- spectrum
- model
- nitrogen content
- total nitrogen
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a soil nitrogen content inversion model construction method and system based on a convolutional neural network, wherein the method comprises the following steps: acquiring remote sensing image data to acquire a remote sensing image range and a land utilization type sample, and performing model training on the land utilization type sample to acquire a first target model; combining the first target model with a remote sensing image range to obtain a land use type spatial distribution map, so that an image spectrum is converted into an indoor spectrum; acquiring soil spectral data, performing spectral pretreatment on the soil spectral data to acquire the total nitrogen content of the soil, establishing a spectrum and soil total nitrogen content target model by combining an indoor spectrum, and constructing a soil total nitrogen content spatial distribution map by combining a remote sensing image range. According to the soil nitrogen content inversion model construction method based on the convolutional neural network, the image spectrum is converted into the indoor spectrum, so that soil total nitrogen space mapping in southern mountainous and hilly areas is realized, and the soil total nitrogen space mapping precision in a vegetation coverage area is improved.
Description
Technical Field
The invention relates to the technical field of crop growth monitoring, in particular to a soil nitrogen content inversion model construction method and system based on a convolutional neural network.
Background
With the rapid development of satellite remote sensing at home and abroad, the variety of satellite images of multispectral data and hyperspectral data is increased, and effective data support is provided for researching inversion of soil alkaline-hydrolyzable nitrogen content based on the multispectral and hyperspectral data.
The soil nitrogen is used as an essential nutrient in soil, plays an important role in the growth and development process of plants, and the content of the nitrogen has an important influence on the yield and the quality of crops. The hyperspectral remote sensing technology is rapidly developed by the advantages of high spectral resolution and rich waveband information, has strong advantages in rapidly estimating soil and crop information, can provide monitoring basis for implementation of precision agriculture, and has wide research and application prospects.
At present, the research on soil ground hyperspectrum is mature, but the research is only the modeling of indoor sampling point spectrum data, the spectrum information of airborne or satellite-borne remote sensing data is more complex, an indoor spectrum inversion soil total nitrogen model is not suitable when being popularized to a remote sensing image spectrum, and certain limitation exists in the practical application process. Therefore, the currently feasible method mainly focuses on the exposed area of the earth surface in northeast China, the obtained spectral information is mainly soil information, the total nitrogen content of soil can be successfully inverted, and the accuracy is generally lower than that of the ground soil hyperspectral data inversion. In south China, most of the plants are in mountainous and hilly areas, vegetation coverage is luxuriant, remote sensing image data are comprehensive reflection of vegetation, soil and environment elements, and much interference information exists when soil total nitrogen is inverted. Therefore, how to perform spatial mapping of soil total nitrogen in southern mountainous and hilly areas by using remote sensing image data is a difficult point to try and overcome urgently at present.
Disclosure of Invention
Based on the above, the invention aims to provide a soil nitrogen content inversion model construction method and system based on a convolutional neural network, which are used for solving the technical problem of how to perform space mapping of soil total nitrogen in southern mountainous and hilly areas by using remote sensing image data.
The invention provides a soil nitrogen content inversion model construction method based on a convolutional neural network, which comprises the following steps:
obtaining model training data, wherein the model training data comprises remote sensing image data and field soil samples, obtaining a remote sensing image range and a plurality of land utilization type samples according to the remote sensing image data, and performing model training on the land utilization type samples by combining a convolutional neural network model to obtain a first target model;
obtaining a land use type space distribution map according to the first target model and the remote sensing image range, obtaining a land use classification result according to the land use type space distribution map, and analyzing image spectral features and indoor spectral features of different land use types according to the land use classification result so as to convert an image spectrum into an indoor spectrum;
acquiring soil spectral data through the field soil sample, performing spectral pretreatment on the soil spectral data to acquire target soil spectral data, acquiring the total nitrogen content of soil through the target soil spectral data, combining the total nitrogen content of the soil and the indoor spectrum with the convolutional neural network model to establish a spectrum and total nitrogen content model of the soil, performing model training on the spectrum and total nitrogen content model of the soil through the target soil spectral data to acquire a spectrum and total nitrogen content target model of the soil, and combining the spectrum and the total nitrogen content target model of the soil with the remote sensing image range to construct a soil total nitrogen content spatial distribution map.
According to the soil nitrogen content inversion model construction method based on the convolutional neural network, model training is carried out on the convolutional neural network model through a land utilization type sample to obtain a first target model, so that a land utilization type space distribution diagram and a land utilization classification result are obtained, and then image spectral features and indoor spectral features of different land utilization types are analyzed according to the land utilization classification result, so that the image spectrum is converted into an indoor spectrum; meanwhile, target soil spectral data are obtained through a field soil sample so as to obtain the total nitrogen content of the soil, and a spectrum and soil total nitrogen content model is established through the total nitrogen content of the soil and an indoor spectrum, so that a spectrum and soil total nitrogen content target model is obtained, and a soil total nitrogen content space distribution map is constructed by combining a remote sensing image range; convert the image spectrum into indoor spectrum, and then realized south mountain hilly area soil total nitrogen space drawing, improved and carried out the space drawing precision of soil total nitrogen in the vegetation coverage area, solved how to utilize remote sensing image data to carry out the technical problem of the space drawing of soil total nitrogen in south mountain hilly area.
In addition, the soil nitrogen content inversion model construction method based on the convolutional neural network can also have the following additional technical characteristics:
further, the step of performing model training on the land use type samples in combination with a convolutional neural network model to obtain a first target model comprises:
performing model training on the land use type sample by combining a convolutional neural network model;
and performing precision evaluation on the trained convolutional neural network model by combining precision evaluation indexes to obtain a first target model, wherein the precision evaluation indexes comprise overall classification precision, Kappa coefficient, producer precision and user precision.
Further, the step of analyzing the image spectrum characteristics and the indoor spectrum characteristics of different land use types according to the land use classification result so as to convert the image spectrum into the indoor spectrum comprises the following steps:
analyzing the correlation of the image spectrum, the indoor spectrum and the soil total nitrogen content by a correlation coefficient method;
establishing a conversion relation between the image spectrum and the indoor spectrum by combining the correlation with a convolutional neural network model;
and converting the image spectrum into an indoor spectrum through the conversion relation.
Further, the step of spectrally preprocessing the soil spectral data to obtain target soil spectral data is followed by:
screening the total nitrogen content of the soil by a correlation coefficient method to determine a characteristic wave band of the total nitrogen of the soil;
and taking the characteristic wave band as a model parameter for establishing a spectrum and soil total nitrogen content model.
Further, the land use type samples comprise modeling samples and testing samples, and the step of performing model training on the land use type samples in combination with the convolutional neural network model to obtain the first target model comprises the following steps of:
and combining the modeling sample with a convolutional neural network model to establish a first model, and testing the first model through the test sample so as to correct the classification precision of the model to obtain a first target model.
Further, the target soil spectrum data comprises a modeling sample and a test sample, and the step of performing model training on the spectrum and soil total nitrogen content model through the target soil spectrum data to obtain a spectrum and soil total nitrogen content target model comprises the following steps:
and carrying out model training on the spectrum and soil total nitrogen content model through the modeling sample, and carrying out precision testing on the trained spectrum and soil total nitrogen content model through the testing sample so as to obtain a spectrum and soil total nitrogen content target model.
In another aspect, the present invention provides a soil nitrogen content inversion model building system based on a convolutional neural network, the system comprising:
the system comprises an acquisition module, a data acquisition module and a data processing module, wherein the acquisition module is used for acquiring model training data, the model training data comprises remote sensing image data and field soil samples, a remote sensing image range and a plurality of land utilization type samples are acquired according to the remote sensing image data, and the land utilization type samples are combined with a convolutional neural network model for model training to acquire a first target model;
the conversion module is used for obtaining a land utilization type space distribution map according to the first target model and the remote sensing image range, obtaining a land utilization classification result according to the land utilization type space distribution map, and analyzing image spectrum characteristics and indoor spectrum characteristics of different land utilization types according to the land utilization classification result so as to convert an image spectrum into an indoor spectrum;
the building module is used for obtaining soil spectral data through the field soil sample, performing spectral preprocessing on the soil spectral data to obtain target soil spectral data, obtaining soil total nitrogen content through the target soil spectral data, combining the soil total nitrogen content and the indoor spectrum with the convolutional neural network model to build a spectrum and soil total nitrogen content model, performing model training on the spectrum and soil total nitrogen content model through the target soil spectral data to obtain a spectrum and soil total nitrogen content target model, and combining the spectrum and soil total nitrogen content target model with the remote sensing image range to build a soil total nitrogen content spatial distribution map.
Drawings
FIG. 1 is a model diagram of a convolutional neural network model of the present invention;
FIG. 2 is a schematic diagram of a soil nitrogen content inversion model construction method based on a convolutional neural network in the invention;
FIG. 3 is a flowchart of a soil nitrogen content inversion model construction method based on a convolutional neural network according to a first embodiment of the present invention;
FIG. 4 is a flowchart of a soil nitrogen content inversion model construction method based on a convolutional neural network according to a second embodiment of the present invention;
fig. 5 is a system block diagram of a soil nitrogen content inversion model construction system based on a convolutional neural network in a third embodiment of the present invention.
The following detailed description will further illustrate the invention in conjunction with the above-described figures.
Detailed Description
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Several embodiments of the invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
Convolutional Neural Network (CNN) is a deep learning model which is most widely used, and its Network structure includes an input layer, a Convolutional layer, a pooling layer, a full-link layer, and an output layer. The convolutional layer performs feature extraction on input data according to a certain step length through a plurality of convolutional kernels with certain sizes; the pooling layer is also called a down-sampling layer, and the maximum value or the average value of a sampling range with a certain size is used for replacing the value of the original range, so that the purposes of reducing the data processing amount and retaining important characteristic information are achieved; the fully connected layer is located after the convolutional layer and the pooling layer, and the function of the fully connected layer is to perform nonlinear combination on the extracted features to obtain an output result.
According to the flexibility of a deep learning model structure and strong feature extraction and autonomous learning capabilities, firstly, a soil nitrogen content inversion model construction method based on a convolutional neural network is provided, as shown in fig. 1, the model comprises 7 layers (a pooling layer does not need to update weight parameters and is not calculated in the number of layers), wherein the number of convolution kernels in the convolutional layer is respectively 24, 48 and 48, a maximum pooling layer is adopted, when modeling is performed by 1D, 2D and 3D, the sizes of the convolution kernels are respectively (3), (3, 3) and (3, 3 and 3), the pooling ranges are respectively (2), (2, 2) and (2, 2 and 2), and the number of neurons in a full connection layer is respectively 1000 and 200. Based on the model, the functions of land use type classification, spectrum conversion, soil total nitrogen inversion modeling and the like with different dimensions can be realized only by adjusting partial parameters in the model. The technical steps of applying the model to soil total nitrogen content inversion in southern mountain and hilly areas can be understood as including 5 stages, namely data collection, data preprocessing, sample selection, a modeling method, precision evaluation and result output, which are specifically shown in fig. 2.
Because southern mountain and hilly area receives the influence of topography fluctuation and land type change difference, the spatial heterogeneity of soil total nitrogen is higher for the modeling of soil total nitrogen is comparatively difficult, and this scheme provides a CNN model of continuous convolution, can realize the soil total nitrogen spatial mapping of high accuracy in southern mountain and hilly area.
Specifically, the application provides a continuous convolution CNN model, and particularly provides a CNN spectrum conversion method which is used for converting an image spectrum into an indoor spectrum, so that soil total nitrogen space mapping in southern mountainous and hilly areas is realized, and the difficulty of low soil total nitrogen space mapping precision in a vegetation coverage area is solved. And the accuracy of land use classification, spectrum conversion and soil total nitrogen inversion is improved by applying the continuous convolution CNN model. The method is beneficial to adjustment of regional agricultural management measures and sustainable utilization of land, provides basic data for real-time quantitative monitoring of soil total nitrogen content change, soil quality evaluation, grain estimation and the like in southern mountain and hilly areas, and has important significance for realizing accurate agricultural targets.
Example one
Referring to fig. 3, a method for constructing a soil nitrogen content inversion model based on a convolutional neural network according to a first embodiment of the present invention is shown, and the method includes steps S101-S103:
s101, obtaining model training data, wherein the model training data comprise remote sensing image data and field soil samples, obtaining a remote sensing image range and a plurality of land utilization type samples according to the remote sensing image data, and performing model training on the land utilization type samples by combining a convolutional neural network model to obtain a first target model.
In the steps, a land use type sample is combined with a convolutional neural network model for model training; and performing precision evaluation on the trained convolutional neural network model by combining precision evaluation indexes to obtain a first target model, wherein the precision evaluation indexes comprise overall classification precision, Kappa coefficient, producer precision and user precision. Specifically, the pre-selection module may be a CNN model.
Specifically, the land use type samples comprise modeling samples and testing samples, the modeling samples are combined with the convolutional neural network model to establish a first model, and the first model is tested through the testing samples, so that the classification accuracy of the model is corrected to obtain a first target model.
S102, obtaining a land use type space distribution map according to the first target model and the remote sensing image range, obtaining a land use classification result according to the land use type space distribution map, and analyzing image spectral features and indoor spectral features of different land use types according to the land use classification result, so that the image spectrum is converted into the indoor spectrum.
Specifically, the correlation of the image spectrum, the indoor spectrum and the total nitrogen content of the soil is analyzed by a correlation coefficient method; establishing a conversion relation between the image spectrum and the indoor spectrum by combining correlation with a convolutional neural network model; and converting the image spectrum into an indoor spectrum through a conversion relation.
Specifically, the spectrum preprocessing method adopted by the scheme comprises 3 spectrum preprocessing methods, namely a Savitzky-Golay smoothing method, a differential transformation method and a standard normal transformation method. Then removing the edge wave bands 350-399 nm and 2451-2500 nm with low signal-to-noise ratio; finally, to reduce the data dimension and redundancy, the spectral data is resampled by averaging every 10 nm interval, each sample yielding 206 bands for later analysis.
S103, acquiring soil spectral data through a field soil sample, performing spectral preprocessing on the soil spectral data to acquire target soil spectral data, acquiring soil total nitrogen content through the target soil spectral data, combining the soil total nitrogen content and an indoor spectrum with a convolutional neural network model to establish a spectrum and soil total nitrogen content model, performing model training on the spectrum and soil total nitrogen content model through the target soil spectral data to acquire a spectrum and soil total nitrogen content target model, and combining the spectrum and soil total nitrogen content target model with a remote sensing image range to establish a soil total nitrogen content spatial distribution map.
As a specific example, the step of spectrally preprocessing the soil spectral data to obtain target soil spectral data is followed by: screening the total nitrogen content of the soil by a correlation coefficient method to determine a characteristic wave band of the total nitrogen of the soil; and taking the characteristic wave band as a model parameter for establishing a spectrum and soil total nitrogen content model.
Further, the target soil spectrum data comprises a modeling sample and a testing sample, and the step of performing model training on the spectrum and soil total nitrogen content model through the target soil spectrum data to obtain the spectrum and soil total nitrogen content target model comprises the following steps:
and carrying out model training on the spectrum and soil total nitrogen content model through the modeling sample, and carrying out precision testing on the trained spectrum and soil total nitrogen content model through the testing sample so as to obtain a spectrum and soil total nitrogen content target model.
In summary, in the soil nitrogen content inversion model construction method based on the convolutional neural network in the above embodiment of the present invention, model training is performed on the convolutional neural network model through the land use type sample to obtain the first target model, so as to obtain a land use type spatial distribution map and a land use classification result, and then image spectral features and indoor spectral features of different land use types are analyzed according to the land use classification result, so as to convert the image spectrum into an indoor spectrum; meanwhile, target soil spectrum data are obtained through a field soil sample so as to obtain the total nitrogen content of the soil, and a spectrum and soil total nitrogen content model is established through the total nitrogen content of the soil and an indoor spectrum, so that a spectrum and soil total nitrogen content target model is obtained, and a soil total nitrogen content space distribution map is constructed by combining a remote sensing image range; convert the image spectrum into indoor spectrum, and then realized south mountain hilly area soil total nitrogen space drawing, improved and carried out the space drawing precision of soil total nitrogen in the vegetation coverage area, solved how to utilize remote sensing image data to carry out the technical problem of the space drawing of soil total nitrogen in south mountain hilly area.
Example two
Referring to fig. 4, a method for constructing a soil nitrogen content inversion model based on a convolutional neural network in a second embodiment of the present invention is shown, where the method includes steps S201 to S203:
s201, obtaining model training data, wherein the model training data comprise remote sensing image data and field soil samples, obtaining a remote sensing image range and a plurality of land utilization type samples according to the remote sensing image data, and combining the land utilization type samples with a convolutional neural network model to carry out model training so as to obtain a first target model.
The data collection stage comprises remote sensing image data, related picture data and field soil sample collection. The method comprises the following steps of naturally air-drying, grinding and sieving a soil sample in a laboratory, uniformly dividing the soil sample into two parts, wherein one part is used for measuring the total nitrogen content of the soil, and the other part adopts a surface feature spectrometer to obtain soil spectrum data in a range of 350-2500 nm.
In ArcGIS software, according to Sentinel-2A image characteristics, with the help of a third national state survey result and partial field survey data, samples of different land utilization types are created in an image area, then the samples are input into a CNN model for modeling, model classification Accuracy is known through establishing a confusion matrix of classification results, evaluation indexes comprise Overall classification Accuracy (OA) (formula 1), a Kappa coefficient (formula 2), Producer Accuracy (Producer Accuracy, PA) (formula 3) and User Accuracy (User Accuracy, UA) (formula 4), and finally the trained model is applied to the whole remote sensing image range to obtain a spatial distribution map of the land utilization types.
In the formula (I), the compound is shown in the specification,S d the number of correctly classified pixels;zthe total number of pixels of the test sample;X jj representjThe number of pixels with the land use types correctly classified;X j* representing realityjTotal number of land utilization type pixels;X *j the representation is classified asjTotal number of land utilization type pixels.
S202, obtaining a land use type space distribution map according to the first target model and the remote sensing image range, obtaining a land use classification result according to the land use type space distribution map, and analyzing image spectral features and indoor spectral features of different land use types according to the land use classification result, so that the image spectrum is converted into the indoor spectrum.
The classification result of land utilization is utilized to analyze the image spectrum characteristics and the indoor spectrum characteristics of different land utilization types, the correlation among the image spectrum, the indoor spectrum and the STN content (namely the total nitrogen content of soil) is analyzed by a Pearson correlation coefficient method, on the basis, the conversion relation between the image spectrum and the indoor spectrum is established by a CNN model, the image spectrum is converted into the indoor spectrum, and the precision evaluation index comprises the following steps: root Mean Square Error (RMSE) (equation 5), Spectral Angular Distance (SAD) (equation 6), mean of absolute values of correlation coefficients of the converted spectrum and the target spectrum (mean of absolute values of rband), and mean of absolute values of correlation coefficients of the converted spectrum and the total nitrogen content of the soil (mean of absolute values of rSTN).
In the formula (I), the compound is shown in the specification,bis the number of wave bands;x mi is the second of the target spectrumiA band value;x pi for testing the spectrumiValue of a band, xmIs a target spectrum; x is a radical of a fluorine atompTo test the spectrum.
S203, acquiring soil spectrum data through a field soil sample, performing spectrum preprocessing on the soil spectrum data to acquire target soil spectrum data, acquiring soil total nitrogen content through the target soil spectrum data, combining the soil total nitrogen content and an indoor spectrum with a convolutional neural network model to establish a spectrum and soil total nitrogen content model, performing model training on the spectrum and soil total nitrogen content model through the target soil spectrum data to acquire a spectrum and soil total nitrogen content target model, and combining the spectrum and soil total nitrogen content target model with a remote sensing image range to establish a soil total nitrogen content spatial distribution map.
The soil spectrum pretreatment has an important role in spectral analysis and soil property content estimation. In the process of collecting soil spectrum information by using a spectrometer, the collected soil spectrum information can be interfered by other external information besides the spectrum information of a soil sample, such as the influence of noise of an instrument, a measuring light source and measuring personnel, and other noises can be doped in the obtained spectrum, so that the characteristics of the soil sample can not be accurately reflected. In order to ensure the accuracy of the spectral information, certain spectral preprocessing needs to be performed on the soil spectral data. The spectrum preprocessing method adopted by the scheme comprises a Savitzky-Golay smoothing method, a differential transformation method and a standard normal transformation method, the three preprocessing methods can be combined according to actual use conditions so as to preprocess soil spectrum data, and then edge wave bands with low signal-to-noise ratio are removed, wherein the edge wave bands are 350-399 nm and 2451-2500 nm; finally, to reduce the data dimension and redundancy, the spectral data is resampled by averaging every 10 nm interval, and each sample is given 206 bands for later analysis.
Specifically, the Savitzky-Golay smoothing method performs fitting operation on the smoothing points and data of the left and right area ranges of the smoothing points through a polynomial, so that the purposes of smoothing a spectral curve and reducing noise are achieved.
The differential transformation method adopts a differential algorithm to obtain an approximate value of a function through derivation, can effectively extract and amplify implicit information in the spectrum, and simultaneously reduces redundancy.
The standard normal transformation method is mainly used for surface scattering, eliminating the influence of solid particle size and optical path transformation on the spectrum.
The soil spectral data is influenced by various external interference information when being collected by a spectrometer, so that a large amount of redundant information exists in the collected spectral data; and the spectral data itself has many wave bands, and the spectral information is easy to overlap. If full-wave-band data are used when the estimation model is built, the modeling process is large in calculation amount, low in modeling efficiency and long in time consumption, modeling accuracy is influenced to a certain extent, and even the modeling accuracy is reduced. Therefore, it is necessary to screen sensitive bands of soil properties for modeling. According to the scheme, the characteristic wave band of the total nitrogen of the soil is determined by a correlation coefficient method, and the higher the correlation is, the more sensitive the wave band response is. Therefore, Pearson correlation coefficient analysis is carried out on the spectrum after pretreatment and the total nitrogen content of the soil, and the wave band which passes significance test and has the absolute value of the correlation coefficient larger than 0.5 is used as a characteristic wave band and is used as an input parameter of the inversion model.
Further, a sample data subset is generated by utilizing a ground statistic module in ArcGIS software, the collected soil sample is divided into a modeling sample and a testing sample, then the converted spectrum is input into a CNN model, the relation between the spectrum and the total nitrogen content of the soil is established, and a coefficient (R) is determined2) And (formula 7), the root mean square error RMSE (formula 8), the quartile range (IQ) and the RMSE Ratio (RPIQ) (formula 9) 3 indexes are subjected to precision evaluation, and finally the trained model is applied to the whole remote sensing image range to obtain the spatial distribution map of the total nitrogen content of the soil.
In the formula (I), the compound is shown in the specification,nis the number of samples;y mi andy pi respectively an actual measured value and a predicted value of total nitrogen;
is the average value of the measured values of total nitrogen; IQ is the difference between the third quartile (Q3) and the first quartile (Q1) of the actual measured values of the samples.
It should be noted that, the method provided by the second embodiment of the present invention, which has the same implementation principle and produces some technical effects as the first embodiment, for the sake of brief description, the corresponding contents in the first embodiment may be referred to where this embodiment is not mentioned.
In summary, in the soil nitrogen content inversion model construction method based on the convolutional neural network in the above embodiment of the present invention, model training is performed on the convolutional neural network model through the land use type sample to obtain the first target model, so as to obtain a land use type spatial distribution map and a land use classification result, and then image spectral features and indoor spectral features of different land use types are analyzed according to the land use classification result, so as to convert the image spectrum into an indoor spectrum; meanwhile, target soil spectral data are obtained through a field soil sample so as to obtain the total nitrogen content of the soil, and a spectrum and soil total nitrogen content model is established through the total nitrogen content of the soil and an indoor spectrum, so that a spectrum and soil total nitrogen content target model is obtained, and a soil total nitrogen content space distribution map is constructed by combining a remote sensing image range; convert image spectral into indoor spectrum, and then realized southern mountain region hills area soil total nitrogen space drawing, improved and carried out the space drawing precision of soil total nitrogen in the vegetation coverage area, solved the technical problem that how to utilize remote sensing image data to carry out the space drawing of soil total nitrogen in southern mountain region hills area.
EXAMPLE III
Referring to fig. 5, a system for constructing a soil nitrogen content inversion model based on a convolutional neural network according to a third embodiment of the present invention is shown, where the system includes:
the system comprises an acquisition module, a data acquisition module and a data processing module, wherein the acquisition module is used for acquiring model training data, the model training data comprises remote sensing image data and field soil samples, a remote sensing image range and a plurality of land utilization type samples are acquired according to the remote sensing image data, and the land utilization type samples are combined with a convolutional neural network model for model training to acquire a first target model;
the conversion module is used for obtaining a land utilization type space distribution map according to the first target model and the remote sensing image range, obtaining a land utilization classification result according to the land utilization type space distribution map, and analyzing image spectral features and indoor spectral features of different land utilization types according to the land utilization classification result so as to convert the image spectrum into an indoor spectrum;
the building module is used for obtaining soil spectral data through the field soil sample, performing spectral preprocessing on the soil spectral data to obtain target soil spectral data, obtaining soil total nitrogen content through the target soil spectral data, combining the soil total nitrogen content and the indoor spectrum with the convolutional neural network model to build a spectrum and soil total nitrogen content model, performing model training on the spectrum and soil total nitrogen content model through the target soil spectral data to obtain a spectrum and soil total nitrogen content target model, and combining the spectrum and soil total nitrogen content target model with the remote sensing image range to build a soil total nitrogen content spatial distribution map.
In summary, in the soil nitrogen content inversion model construction system based on the convolutional neural network in the above embodiment of the present invention, model training is performed on the convolutional neural network model through the land use type sample to obtain the first target model, so as to obtain a land use type spatial distribution map and a land use classification result, and then image spectral features and indoor spectral features of different land use types are analyzed according to the land use classification result, so as to convert the image spectrum into an indoor spectrum; meanwhile, target soil spectrum data are obtained through a field soil sample so as to obtain the total nitrogen content of the soil, and a spectrum and soil total nitrogen content model is established through the total nitrogen content of the soil and an indoor spectrum, so that a spectrum and soil total nitrogen content target model is obtained, and a soil total nitrogen content space distribution map is constructed by combining a remote sensing image range; convert image spectral into indoor spectrum, and then realized southern mountain region hills area soil total nitrogen space drawing, improved and carried out the space drawing precision of soil total nitrogen in the vegetation coverage area, solved the technical problem that how to utilize remote sensing image data to carry out the space drawing of soil total nitrogen in southern mountain region hills area.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following technologies, which are well known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims (6)
1. A soil nitrogen content inversion model construction method based on a convolutional neural network is characterized by comprising the following steps:
obtaining model training data, wherein the model training data comprises remote sensing image data and field soil samples, obtaining a remote sensing image range and a plurality of land utilization type samples according to the remote sensing image data, and performing model training on the land utilization type samples by combining a convolutional neural network model to obtain a first target model;
obtaining a land use type space distribution map according to the first target model and the remote sensing image range, obtaining a land use classification result according to the land use type space distribution map, and analyzing image spectral features and indoor spectral features of different land use types according to the land use classification result so as to convert an image spectrum into an indoor spectrum;
acquiring soil spectral data through the field soil sample, performing spectral pretreatment on the soil spectral data to acquire target soil spectral data, acquiring the total nitrogen content of soil through the target soil spectral data, combining the total nitrogen content of the soil and the indoor spectrum with the convolutional neural network model to establish a spectrum and total nitrogen content model of the soil, performing model training on the spectrum and total nitrogen content model of the soil through the target soil spectral data to acquire a spectrum and total nitrogen content target model of the soil, and combining the spectrum and total nitrogen content target model of the soil with the remote sensing image range to construct a soil total nitrogen content spatial distribution map;
the method comprises the following steps of analyzing image spectral features and indoor spectral features of different land use types according to the land use classification result, and converting the image spectrum into an indoor spectrum, wherein the steps comprise:
analyzing the correlation of the image spectrum, the indoor spectrum and the soil total nitrogen content by a correlation coefficient method;
establishing a conversion relation between the image spectrum and the indoor spectrum by combining the correlation with a convolutional neural network model;
and converting the image spectrum into an indoor spectrum through the conversion relation.
2. The convolutional neural network-based soil nitrogen content inversion model building method as claimed in claim 1, wherein the step of performing model training on the land use type samples in combination with the convolutional neural network model to obtain a first target model comprises:
combining the land use type samples with a convolutional neural network model to perform model training;
and carrying out precision evaluation on the trained convolutional neural network model by combining precision evaluation indexes to obtain a first target model, wherein the precision evaluation indexes comprise overall classification precision, Kappa coefficient, producer precision and user precision.
3. The convolutional neural network-based soil nitrogen content inversion model building method as claimed in claim 1, wherein the step of performing spectrum preprocessing on the soil spectral data to obtain target soil spectral data is followed by:
screening the total nitrogen content of the soil by a correlation coefficient method to determine a characteristic wave band of the total nitrogen of the soil;
and taking the characteristic wave band as a model parameter for establishing a spectrum and soil total nitrogen content model.
4. The convolutional neural network-based soil nitrogen content inversion model building method according to claim 1, wherein the land use type samples comprise modeling samples and testing samples, and the step of performing model training on the land use type samples in combination with the convolutional neural network model to obtain a first target model comprises:
and combining the modeling sample with a convolutional neural network model to establish a first model, and testing the first model through the test sample so as to correct the classification precision of the model to obtain a first target model.
5. The soil nitrogen content inversion model building method based on the convolutional neural network as claimed in claim 1, wherein the target soil spectrum data comprises modeling samples and testing samples, and the step of performing model training on the spectrum and soil total nitrogen content model through the target soil spectrum data to obtain a spectrum and soil total nitrogen content target model comprises:
and carrying out model training on the spectrum and soil total nitrogen content model through the modeling sample, and carrying out precision testing on the trained spectrum and soil total nitrogen content model through the testing sample so as to obtain a spectrum and soil total nitrogen content target model.
6. A soil nitrogen content inversion model construction system based on a convolutional neural network is characterized by comprising the following steps:
the system comprises an acquisition module, a data acquisition module and a data processing module, wherein the acquisition module is used for acquiring model training data, the model training data comprises remote sensing image data and field soil samples, a remote sensing image range and a plurality of land utilization type samples are acquired according to the remote sensing image data, and the land utilization type samples are combined with a convolutional neural network model for model training to acquire a first target model;
the conversion module is used for obtaining a land utilization type space distribution map according to the first target model and the remote sensing image range, obtaining a land utilization classification result according to the land utilization type space distribution map, and analyzing image spectral features and indoor spectral features of different land utilization types according to the land utilization classification result so as to convert the image spectrum into an indoor spectrum;
the building module is used for obtaining soil spectral data through the field soil sample, performing spectral preprocessing on the soil spectral data to obtain target soil spectral data, obtaining soil total nitrogen content through the target soil spectral data, combining the soil total nitrogen content and the indoor spectrum with the convolutional neural network model to build a spectrum and soil total nitrogen content model, performing model training on the spectrum and soil total nitrogen content model through the target soil spectral data to obtain a spectrum and soil total nitrogen content target model, and combining the spectrum and soil total nitrogen content target model with the remote sensing image range to build a soil total nitrogen content spatial distribution map;
in the conversion module, analyzing image spectrum characteristics and indoor spectrum characteristics of different land use types according to the land use classification result, so that the step of converting the image spectrum into the indoor spectrum comprises the following steps:
analyzing the correlation of the image spectrum, the indoor spectrum and the soil total nitrogen content by a correlation coefficient method;
establishing a conversion relation between the image spectrum and the indoor spectrum by combining the correlation with a convolutional neural network model;
and converting the image spectrum into an indoor spectrum through the conversion relation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210432884.2A CN114529838B (en) | 2022-04-24 | 2022-04-24 | Soil nitrogen content inversion model construction method and system based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210432884.2A CN114529838B (en) | 2022-04-24 | 2022-04-24 | Soil nitrogen content inversion model construction method and system based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114529838A CN114529838A (en) | 2022-05-24 |
| CN114529838B true CN114529838B (en) | 2022-07-15 |
Family
ID=81627855
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210432884.2A Active CN114529838B (en) | 2022-04-24 | 2022-04-24 | Soil nitrogen content inversion model construction method and system based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114529838B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115410086A (en) * | 2022-08-26 | 2022-11-29 | 南方海洋科学与工程广东省实验室(广州) | Water quality inversion method, device and equipment based on remote sensing image |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103712923A (en) * | 2013-12-23 | 2014-04-09 | 浙江大学 | Method for eliminating moisture influence factor in field in-situ soil measurement spectrums |
| CN110263717A (en) * | 2019-06-21 | 2019-09-20 | 中国科学院地理科学与资源研究所 | It is a kind of incorporate streetscape image land used status determine method |
| CN112149827A (en) * | 2020-09-27 | 2020-12-29 | 哈尔滨航天恒星数据系统科技有限公司 | System and method for constructing soil quick-acting potassium analysis model based on satellite images |
| CN113435574A (en) * | 2021-06-22 | 2021-09-24 | 哈尔滨航天恒星数据系统科技有限公司 | Convolutional neural network soil quick-acting potassium analysis model construction system and method |
| CN113866102A (en) * | 2021-09-24 | 2021-12-31 | 中国地质调查局西安地质调查中心(西北地质科技创新中心) | Soil health investigation monitoring method based on spectrum |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104992068B (en) * | 2015-08-13 | 2018-10-26 | 四川农业大学 | A kind of prediction technique of topsoil nitrogen distribution |
| CN106650689B (en) * | 2016-12-30 | 2019-08-27 | 厦门理工学院 | A kind of coastal cities time series Land-use extracting method |
| CN108152235B (en) * | 2018-03-21 | 2020-09-22 | 中南大学 | Heavy metal content inversion method combining soil indoor and outdoor spectra |
| CN110020635A (en) * | 2019-04-15 | 2019-07-16 | 中国农业科学院农业资源与农业区划研究所 | Growing area crops sophisticated category method and system based on unmanned plane image and satellite image |
| CN111727443B (en) * | 2020-05-15 | 2023-10-31 | 安徽中科智能感知科技股份有限公司 | Soil available nutrient inversion method based on deep neural network |
| CN112557393B (en) * | 2020-11-19 | 2024-02-20 | 南京农业大学 | Wheat leaf layer nitrogen content estimation method based on hyperspectral image fusion map features |
| CN112836610B (en) * | 2021-01-26 | 2022-05-27 | 平衡机器科技(深圳)有限公司 | Land use change and carbon reserve quantitative estimation method based on remote sensing data |
| AU2021101780A4 (en) * | 2021-04-07 | 2021-05-27 | Beijing Normal University | Aboveground Biomass Estimation and Scale Conversion for Mean Regional Spectral Units |
| CN114359748A (en) * | 2021-12-30 | 2022-04-15 | 中国科学院空天信息创新研究院 | Target classification extraction method and device |
| CN114384023A (en) * | 2022-01-04 | 2022-04-22 | 清华大学 | Soil arsenic concentration spatial distribution inversion method and device and computer equipment |
-
2022
- 2022-04-24 CN CN202210432884.2A patent/CN114529838B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103712923A (en) * | 2013-12-23 | 2014-04-09 | 浙江大学 | Method for eliminating moisture influence factor in field in-situ soil measurement spectrums |
| CN110263717A (en) * | 2019-06-21 | 2019-09-20 | 中国科学院地理科学与资源研究所 | It is a kind of incorporate streetscape image land used status determine method |
| CN112149827A (en) * | 2020-09-27 | 2020-12-29 | 哈尔滨航天恒星数据系统科技有限公司 | System and method for constructing soil quick-acting potassium analysis model based on satellite images |
| CN113435574A (en) * | 2021-06-22 | 2021-09-24 | 哈尔滨航天恒星数据系统科技有限公司 | Convolutional neural network soil quick-acting potassium analysis model construction system and method |
| CN113866102A (en) * | 2021-09-24 | 2021-12-31 | 中国地质调查局西安地质调查中心(西北地质科技创新中心) | Soil health investigation monitoring method based on spectrum |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114529838A (en) | 2022-05-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106780091B (en) | Agricultural disaster information remote sensing extraction method based on vegetation index time-space statistical characteristics | |
| CN110427995B (en) | Bayesian soil moisture estimation method based on multi-source remote sensing data | |
| CN102324098B (en) | Relative radiation calibration method in combination with laboratory calibration and even landscape statistics | |
| CN103345707A (en) | Crop maturation stage remote sensing prediction method based on multi-source remote sensing data | |
| CN107607486B (en) | Soil total nitrogen detection method and device | |
| CN110687053B (en) | Regional organic matter content estimation method and device based on hyperspectral image | |
| CN112348812A (en) | Forest stand age information measuring method and device | |
| CN112711989A (en) | Corn straw coverage estimation method based on radar remote sensing and optical remote sensing | |
| CN113836779A (en) | CNN-based farmland surface soil moisture inversion method for Sentinel multi-source data | |
| CN109977991A (en) | Forest resourceies acquisition method based on high definition satellite remote sensing | |
| CN115235997A (en) | Soil texture inversion method based on satellite hyperspectral image | |
| CN113963260A (en) | Extraction method and device for winter wheat planting area and computer equipment | |
| CN113466143B (en) | Soil nutrient inversion method, device, equipment and medium | |
| CN114529838B (en) | Soil nitrogen content inversion model construction method and system based on convolutional neural network | |
| CN112836725A (en) | Weak supervision LSTM recurrent neural network rice field identification method based on time sequence remote sensing data | |
| CN114202675A (en) | Crop growth parameter determination method and system based on multispectral remote sensing image of unmanned aerial vehicle | |
| CN114819737B (en) | Method, system and storage medium for estimating carbon reserves of highway road vegetation | |
| GB2620469A (en) | Spatial prediction and evaluation method of soil organic matter content based on partition algorithm | |
| CN113553697A (en) | Long-time-sequence multi-source data-based vegetation disturbance analysis method for coal mining | |
| CN115641504A (en) | Automatic remote sensing extraction method for field boundary based on crop phenological characteristics and decision tree model | |
| CN111783288A (en) | Inversion method for soil salinity of yellow river delta based on Landsat8 | |
| AU2021101780A4 (en) | Aboveground Biomass Estimation and Scale Conversion for Mean Regional Spectral Units | |
| CN111579565B (en) | Agricultural drought monitoring method, system and storage medium | |
| Zhou et al. | Comparison of inversion method of maize leaf area index based on UAV hyperspectral remote sensing | |
| CN115508292A (en) | Soil profile nitrogen content high spectrum detection and visualization method based on machine learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |