CN-121999307-A - Soil salinity inversion method and device for collaborative sky earth observation data

CN121999307ACN 121999307 ACN121999307 ACN 121999307ACN-121999307-A

Abstract

The invention provides a soil salinity inversion method and device for collaborative sky-earth observation data, and relates to the technical field of soil salinity monitoring, wherein the method comprises the following steps: the method comprises the steps of inputting spectral features corresponding to all farmland pixels in an unmanned aerial vehicle image and hyperspectral data in a hyperspectral space distribution map into a field scale soil salinity inversion model, and outputting a field scale soil salinity distribution map; and sequentially inputting spectral features corresponding to all farmland pixels in each satellite image in a satellite image time sequence into a regional scale soil salinity inversion model respectively, and outputting the regional scale soil salinity distribution map of a long-time sequence. The invention can realize the space-time continuous accurate monitoring of the soil salinity in a large range.

Inventors

ZHANG XINING
LI HAO
ZHANG QIAN
LIU YU
LUAN RUPENG
ZHANG JINMENG
ZHAO RUIFANG
WANG WEI

Assignees

北京市农林科学院
北京智农天地网络技术有限公司

Dates

Publication Date: 20260508
Application Date: 20251205

Claims (10)

1. The soil salinity inversion method for collaborative sky earth observation data is characterized by comprising the following steps of: collecting a ground soil sample in a research area, hyperspectral data and salt content data of the ground soil sample, and acquiring an unmanned aerial vehicle image and a satellite image which cover the soil sample; inputting spectral features corresponding to all farmland pixels in the unmanned aerial vehicle image and hyperspectral data in a hyperspectral space distribution diagram into a trained field scale soil salinity inversion model, and outputting a field scale soil salinity distribution diagram, wherein the field scale soil salinity inversion model is obtained by training the hyperspectral features of pixels in the unmanned aerial vehicle image corresponding to the geographic position of the ground soil sample based on hyperspectral data and salt content data of the ground soil sample; upscaling the spatial resolution of the field scale soil salinity distribution map to be the same as the spatial resolution of the satellite image to obtain an upscaled soil salinity distribution map, and extracting the salinity of sample points from the upscaled soil salinity distribution map in a layering manner; The method comprises the steps of obtaining satellite image time sequences corresponding to time nodes to be predicted in a research area, sequentially inputting spectral features corresponding to all farmland pixels in each satellite image in the satellite image time sequences into a trained regional scale soil salinity inversion model respectively, and outputting a regional scale soil salinity distribution map of a long time sequence, wherein the regional scale soil salinity inversion model is obtained by training based on the salt content of sample points and the spectral features of pixels in the satellite images corresponding to the geographic positions of the sample points.
2. The method for inverting soil salinity of collaborative sky-earth observation data according to claim 1, wherein inputting the spectral features corresponding to all farmland pixels in the unmanned aerial vehicle image and hyperspectral data in a hyperspectral space distribution map into a trained field scale soil salinity inversion model to output a field scale soil salinity distribution map comprises: And executing the following steps for the spectral characteristics and the hyperspectral data corresponding to each farmland pixel in the unmanned aerial vehicle image: Converting the spectrum features corresponding to the current farmland pixels into two-dimensional images, inputting the two-dimensional images into a convolution block in the field scale soil salinity inversion model for feature extraction, and outputting unmanned aerial vehicle image features; Inputting the hyperspectral data corresponding to the current farmland pixels into a one-dimensional convolutional encoder in the farmland scale soil salinity inversion model for dimension reduction and feature extraction, and outputting soil hyperspectral features; inputting the unmanned aerial vehicle image features and the soil hyperspectral features into a fusion network in the field scale soil salinity inversion model to perform feature fusion, and outputting fusion features; Inputting the fusion characteristics into a regression calculation module in the field scale soil salinity inversion model to perform soil salinity estimation, and outputting a soil salinity estimation value corresponding to the current farmland pixels; And obtaining the soil salinity distribution map of the field scale according to the soil salinity estimation values corresponding to all the farmland pixels.
3. The soil salinity inversion method of collaborative sky-earth observation data according to claim 2, wherein the fusion network comprises N residual blocks and N fusion modules which are alternately connected in turn, one of the residual blocks and one of the fusion modules corresponding to one fusion stage; The unmanned aerial vehicle image feature and the soil hyperspectral feature are input into a fusion network in the field scale soil salinity inversion model to perform feature fusion, and fusion features are output, and the method comprises the following steps: At each fusion stage, the following steps are performed: Inputting the image characteristics of the current fusion stage into a residual block of the current fusion stage, and outputting the image enhancement characteristics of the current fusion stage, wherein the image characteristics of the first fusion stage are the image characteristics of the unmanned aerial vehicle; Inputting the image enhancement characteristic and the soil hyperspectral characteristic of the current fusion stage into a fusion module of the current fusion stage, and outputting the fusion characteristic of the current fusion stage; taking the fusion characteristics of the current fusion stage as the image characteristics of the next fusion stage; and taking the fusion characteristic of the Nth fusion stage as the fusion characteristic of the fusion network output.
4. The method for inverting the salinity of soil for collaborative sky-earth observation data according to claim 3, wherein the inputting the image enhancement feature of the current fusion stage and the soil hyperspectral feature into the fusion module of the current fusion stage and outputting the fusion feature of the current fusion stage comprises: inputting the image enhancement features and the soil hyperspectral features into a channel attention sub-module in the fusion module for the current fusion stage to obtain a channel attention map; multiplying the channel attention map by the image enhancement feature element by element to obtain a first improvement feature; Inputting the first improved feature into a spatial attention sub-module in the fusion module to obtain a spatial attention map; multiplying the spatial attention map by the channel attention modified feature element by element to obtain a second modified feature; And taking the second improved characteristic as a fusion characteristic of the current fusion stage.
5. The method of claim 4, wherein inputting the image enhancement features and the soil hyperspectral features into a channel attention sub-module in the fusion module to obtain a channel attention map comprises: respectively inputting the image enhancement features into an average pooling layer and a maximum pooling layer in the channel attention submodule to perform space dimension compression to obtain a first pooling feature and a second pooling feature; carrying out characteristic series connection on the first pooling characteristic and the second pooling characteristic and the soil hyperspectral characteristic respectively to obtain a first series characteristic and a second series characteristic; Inputting the first series characteristic and the second series characteristic into a multi-layer perceptron in the channel attention sub-module respectively for characteristic extraction to obtain a first extraction characteristic and a second extraction characteristic; adding the first extracted feature and the second extracted feature element by element to obtain a first fusion feature; Activating the first fusion feature to generate the channel attention map.
6. The method of soil salinity inversion of collaborative sky-earth observation data according to claim 4, wherein the inputting the first improvement feature into a spatial attention sub-module in the fusion module obtains a spatial attention map comprising: Respectively carrying out average value polymerization and maximum value polymerization on the first improved feature along the channel dimension to obtain a first polymerized feature and a second polymerized feature; Carrying out feature series connection on the first polymerization feature and the second polymerization feature in the channel dimension to obtain a third series feature; The third series characteristic is sequentially input into a convolution layer and an activation function, and the spatial attention map is generated.
7. Soil salinity inversion device of cooperation sky ground observation data, characterized by comprising: The acquisition module is used for acquiring a ground soil sample in a research area, hyperspectral data and salt content data of the ground soil sample, and acquiring unmanned aerial vehicle images and satellite images covering the soil sample; The field scale inversion module is used for inputting spectral features corresponding to all field pixels in the unmanned aerial vehicle image and hyperspectral data in a hyperspectral space distribution diagram into a trained field scale soil salinity inversion model and outputting a field scale soil salinity distribution diagram, wherein the field scale soil salinity inversion model is obtained by training the spectral features of pixels in the unmanned aerial vehicle image corresponding to the geographic position of the ground soil sample based on hyperspectral data and salt content data of the ground soil sample; The upscaling module is used for upscaling the spatial resolution of the field scale soil salinity distribution map to be the same as the spatial resolution of the satellite image to obtain an upscaling soil salinity distribution map, and extracting the salinity of sample points from the upscaling soil salinity distribution map in a layering manner; The regional scale inversion module is used for acquiring a satellite image time sequence corresponding to a time node to be predicted in the research area, sequentially inputting spectral features corresponding to all farmland pixels in each satellite image in the satellite image time sequence into a trained regional scale soil salinity inversion model respectively, and outputting a regional scale soil salinity distribution map of a long-time sequence, wherein the regional scale soil salinity inversion model is obtained by training based on the salinity of the sample points and the spectral features of the pixels in the satellite images corresponding to the geographic positions of the sample points.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor, when executing the computer program, implements a soil salinity inversion method of collaborative sky-earth observation data according to any one of claims 1-6.
9. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements a soil salinity inversion method of collaborative sky-earth observation data according to any one of claims 1 to 6.
10. A computer program product comprising a computer program which, when executed by a processor, implements a soil salinity inversion method of collaborative sky-earth observation data according to any one of claims 1 to 6.

Description

Soil salinity inversion method and device for collaborative sky earth observation data Technical Field The invention relates to the technical field of soil salinity monitoring, in particular to a soil salinity inversion method and device for collaborative sky-earth observation data. Background Salinization of farmland soil is one of the main reasons for causing the degradation of full-ball land, and forms a serious threat to sustainable development of agriculture and double safety of grains and ecology. The accurate mastering of the spatial distribution condition of saline-alkali soil in key areas is a primary premise for effectively treating saline-alkali soil in areas and guiding agricultural production activities in areas. The existing soil salinity research based on the remote sensing technology is mostly prone to adopting the remote sensing data of a single remote sensing observation platform of the sky, the air and the ground, and the utilization of cross-platform data is relatively fractured. Although some methods add environmental covariates such as terrains, soil, vegetation and the like related to soil salinity into a nonlinear inversion model to improve monitoring precision, remote sensing data of a single remote sensing observation platform is difficult to realize space-time continuous accurate monitoring of the soil salinity in a large range. Disclosure of Invention The invention provides a soil salinity inversion method and device for collaborative sky-earth observation data, which are used for solving the defect that the space-time continuous accurate monitoring of the soil salinity in a large range is difficult to realize in the prior art and realizing the space-time continuous accurate monitoring of the soil salinity in a large range. The invention provides a soil salinity inversion method for collaborative sky-earth observation data, which comprises the following steps: collecting a ground soil sample in a research area, hyperspectral data and salt content data of the ground soil sample, and acquiring an unmanned aerial vehicle image and a satellite image which cover the soil sample; inputting spectral features corresponding to all farmland pixels in the unmanned aerial vehicle image and hyperspectral data in a hyperspectral space distribution diagram into a trained field scale soil salinity inversion model, and outputting a field scale soil salinity distribution diagram, wherein the field scale soil salinity inversion model is obtained by training the hyperspectral features of pixels in the unmanned aerial vehicle image corresponding to the geographic position of the ground soil sample based on hyperspectral data and salt content data of the ground soil sample; upscaling the spatial resolution of the field scale soil salinity distribution map to be the same as the spatial resolution of the satellite image to obtain an upscaled soil salinity distribution map, and extracting the salinity of sample points from the upscaled soil salinity distribution map in a layering manner; The method comprises the steps of obtaining satellite image time sequences corresponding to time nodes to be predicted in a research area, sequentially inputting spectral features corresponding to all farmland pixels in each satellite image in the satellite image time sequences into a trained regional scale soil salinity inversion model respectively, and outputting a regional scale soil salinity distribution map of a long time sequence, wherein the regional scale soil salinity inversion model is obtained by training based on the salt content of sample points and the spectral features of pixels in the satellite images corresponding to the geographic positions of the sample points. In some embodiments, the inputting the spectral features corresponding to all farmland pixels in the unmanned aerial vehicle image and the hyperspectral data in the hyperspectral spatial distribution map into the trained field scale soil salinity inversion model, outputting a field scale soil salinity distribution map, includes: And executing the following steps for the spectral characteristics and the hyperspectral data corresponding to each farmland pixel in the unmanned aerial vehicle image: Converting the spectrum features corresponding to the current farmland pixels into two-dimensional images, inputting the two-dimensional images into a convolution block in the field scale soil salinity inversion model for feature extraction, and outputting unmanned aerial vehicle image features; Inputting the hyperspectral data corresponding to the current farmland pixels into a one-dimensional convolutional encoder in the farmland scale soil salinity inversion model for dimension reduction and feature extraction, and outputting soil hyperspectral features; inputting the unmanned aerial vehicle image features and the soil hyperspectral features into a fusion network in the field scale soil salinity inversion model to perform feature fusion, and outputting