Winter wheat LAI SAR data genetic neural network model inversion method

Abstract

The invention discloses a winter wheat LAI SAR data genetic neural network model inversion method. The method comprises the following steps: 1) based on GF-1WFV and GF-3SAR data, analyzing the correlation between an optical and microwave polarization decomposition vegetation index model and a leaf area index; 2) based on GF-3SAR data, analyzing the correlation between a sigma mod backscattering coefficient obtained by the improved cloud water model and an actually measured leaf area index; 3) using a backscattering coefficient obtained by the improved cloud water model, a leaf area index obtained by inversion of the improved cloud water model, and an optical and microwave polarization decomposition vegetation index model as input factors of a genetic neural network model; the stability of the BP neural network and the fitness proportion, real number crossover and variation of the genetic algorithm are utilized to enhance training and optimize search capability, and a genetic neural network model is constructed to obtain better leaf area index inversion precision.

Description

Technical Field

[0001] The present invention relates to a SAR data genetic neural network model inversion method for winter wheat LAI, in particular to a method for leaf area index detection over a large area using radar and multi-spectral satellite data collaboration. Background Art

[0002] Leaf area index (LAI) is one of the important vegetation structural parameters in the biogeochemical cycle, and is closely related to the biophysical processes of vegetation such as respiration, photosynthesis and transpiration, as well as the nitrogen, potassium and water cycles of the earth's ecosystem. In this paper, leaf area index is defined as half of the sum of all leaf areas per unit surface area. Leaf area index is an important indicator for estimating vegetation coverage, monitoring and forecasting crop growth, biomass, yield, etc. Therefore, rapid and accurate inversion of leaf area index is of great research significance for agricultural monitoring, biogeochemical cycles, etc.

[0003] Although traditional ground measurement of leaf area index can obtain leaf area index values ​​of local specific surfaces, it is difficult to monitor leaf area index over a large range. Remote sensing technology, with its advantages of large coverage area and high timeliness, provides an effective way to monitor crop leaf area index in a regional or global scope. Based on optical remote sensing images, the main means of remote sensing inversion of leaf area index are optical vegetation index model and physical model, but the inversion accuracy of the two models is related to whether the optical remote sensing image is blocked by clouds and fog, which makes the reliability and universality poor.

[0004] However, synthetic aperture radar (SAR) has the advantages of all-day and all-weather operation, can penetrate clouds and fog, and has a certain penetration effect on dense vegetation canopies, and has great application potential in vegetation parameter inversion and monitoring. Based on SAR remote sensing images, the main means of remote sensing inversion of leaf area index are empirical models and cloud water models. The empirical model uses the backscatter coefficients of three polarization modes, HH, HV, and VV, to establish a correlation with the measured leaf area index. Inoué et al. measured the full polarization backscatter coefficient of rice throughout the growth period using five bands, Ka, Ku, X, C, and L, with incident angles of 25°, 35°, 45°, and 55°. The measurement results show that there is a good correlation between the backscatter coefficient and the rice leaf area index, and that HH polarization is significantly higher than VV polarization in the middle and late stages of rice growth.

[0005] The cloud water model is a semi-empirical parameter inversion model first proposed by Attema and U l aby in 1978. Due to the simplicity and practicality of the water cloud model, a large number of subsequent studies on soil moisture and vegetation growth parameter inversion have used the water cloud model. U l aby et al. proposed a multi-layer vegetation scattering model to consider the scattering components from various components of the vegetation canopy (stems, leaves and spikes).

[0006] At present, the inversion of winter wheat leaf area index at home and abroad mostly uses a single image data source, a single growth period, and a single model for research. However, there are few studies that use the method of multi-source data, multi-growth period, and multi-model fusion to invert the leaf area index.

[0007] Taking the above problems into consideration, based on GF-3SAR images and GF-1WFV images, the optical and microwave polarization decomposition vegetation index model, the backscattering coefficient extracted by the improved cloud water model (MWCM), and the leaf area index obtained by inversion of the improved cloud water model are used as input factors, and a genetic neural network model is proposed to invert the leaf area index of winter wheat in five growth stages. This method combines optical and radar remote sensing and makes full use of their respective advantages to invert surface parameters, which is very meaningful and can improve the inversion accuracy. Summary of the invention

[0008] In view of the problems of insufficient research on winter wheat leaf area index inversion methods due to single data source, single growth period and single model, and low precision, the present invention combines multiple data sources, multiple growth periods and multiple models to establish a winter wheat leaf area index inversion method based on SAR data using a genetic neural network model.

[0009] The purpose of the present invention is achieved through the following technical steps:

[0010] Step 1) Based on GF-1WFV and GF-3SAR remote sensing image data, an optical and microwave polarization decomposition vegetation index model is constructed to obtain a fitting accuracy map between the model and the measured leaf area index;

[0011] Step 2) Based on the GF-3SAR remote sensing image data, the backscattering coefficients HV, VV and VV are obtained by using the improved cloud water model, and the fitting accuracy diagram of the three backscattering coefficients and the measured leaf area index is obtained;

[0012] Step 3) Based on GF-3SAR remote sensing image data, the improved cloud-water model is used to predict the leaf area index, and a fitting accuracy diagram of the predicted leaf area index and the measured leaf area index is obtained;

[0013] Step 4) Based on the GF-3SAR remote sensing image data, the backscattering coefficient obtained by the improved cloud water model, the leaf area index obtained by the improved cloud water model inversion, and the optical and microwave polarization decomposition vegetation index model are used as input factors of the neural network model, and the BP neural network model is used to predict the leaf area index to obtain a fitting accuracy map of the predicted leaf area index and the measured leaf area index;

[0014] Step 5) Based on the GF-3SAR remote sensing image data, the backscattering coefficient obtained by the improved cloud water model, the leaf area index obtained by inverting the improved cloud water model, and the optical and microwave polarization decomposition vegetation index model are used as input factors of the neural network model. The stability of the BP neural network and the fitness ratio, real number crossover and mutation of the genetic algorithm are used to enhance the training and optimize the search capabilities, and a genetic neural network model is constructed to obtain the fitting accuracy map of the predicted leaf area index and the measured leaf area index and the leaf area index inversion result map. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is the fitting accuracy diagram of the optical and microwave polarization decomposition vegetation index model and the measured leaf area index of winter wheat in five growth stages;

[0016] Figure 2 To obtain the fitting accuracy diagrams of three backscattering coefficient models and measured leaf area index of winter wheat at five growth stages;

[0017] Figure 3 The fitting accuracy diagram of leaf area index predicted by improved cloud water model and measured leaf area index for winter wheat in five growth stages;

[0018] Figure 4 The fitting accuracy diagram of leaf area index predicted by BP neural network model and genetic neural network model and measured leaf area index of winter wheat in five growth stages;

[0019] Figure 5 This is the inversion result map of leaf area index of winter wheat in five growth stages in Xiangfu District, Kaifeng City, Henan Province; DETAILED DESCRIPTION

[0020] The present invention "a winter wheat leaf area index inversion method based on SAR data and genetic neural network model" is further explained below in conjunction with the accompanying drawings.

[0021] The "construction of optical and microwave polarization decomposition vegetation index model based on GF-1WFV and GF-3SAR remote sensing images", "improved backscattering coefficient model obtained by cloud-water model based on GF-3SAR remote sensing images" and "genetic neural network model based on GF-3SAR remote sensing images" involved in the present invention are important methodological innovations for leaf area index. Among them, "construction of optical and microwave polarization decomposition vegetation index model based on GF-1WFV and GF-3SAR remote sensing images" combines the optical vegetation index model with the microwave polarization decomposition model to establish a leaf area index inversion method for the optical and microwave polarization decomposition vegetation index model, and applies it to Xiangfu District, Kaifeng City, Henan Province. This method achieves the effect of multi-model combination and multi-growth period experiment of winter wheat, while improving the operation efficiency of the algorithm, laying a good foundation for establishing the inversion of winter wheat leaf area index; "improved backscattering coefficient model obtained by cloud-water model based on GF-3SAR remote sensing images" establishes a leaf area index inversion method for the backscattering coefficient model based on the improved cloud-water model theory, and applies it to the river Xiangfu District, Kaifeng City, Nanning Province, this method achieved the effect of multi-growth period experiment of winter wheat, and laid a good foundation for establishing the leaf area index inversion of winter wheat; "Genetic neural network model based on GF-3SAR remote sensing image", the backscattering coefficient model obtained by improving the cloud water model based on GF-3SAR remote sensing image" and "optical and microwave polarization decomposition vegetation index model based on GF-1WFV and GF-3SAR remote sensing images" were combined as the input factors of the genetic neural network model to establish the leaf area index inversion method of the genetic neural network model. The effects of multi-source data coordination, multi-model combination and multi-growth period experiment were achieved, so that the inversion results of winter wheat leaf area index have a certain reliability.

[0022] (I) Constructing the optical and microwave polarization decomposition vegetation index model based on GF-1WFV and GF-3SAR remote sensing images

[0023] Based on GF-1WFV and GF-3SAR remote sensing images, an optical and microwave polarization decomposition vegetation index model inversion method was constructed and applied to Xiangfu District, Kaifeng City, Henan Province. First, based on GF-1WFV remote sensing images, five optical vegetation index models were constructed; then, based on GF-3SAR remote sensing images, a polarization decomposition index model (RVI) was constructed. Yamaguchi ); Finally, the five optical vegetation index models were compared with RVI Yamaguchi The models were combined to establish five optical and microwave polarization decomposition vegetation index models ( Figure 1 ).

[0024] (II) Backscattering coefficient model obtained by improving the cloud water model based on GF-3SAR remote sensing images

[0025] Based on GF-3SAR remote sensing images, an improved cloud-water model is constructed to extract the backscattering coefficient model inversion method, and it is applied to Xiangfu District, Kaifeng City, Henan Province. First, the optimal optical vegetation index model and the measured leaf area index are integrated into the cloud-water model; then, the improved cloud-water model is used to extract the backscattering coefficient; finally, the fitting accuracy of the backscattering coefficient model and the leaf area index is analyzed ( Figure 2 ), and the improved cloud-water model was used to obtain the predicted leaf area index, and the fitting accuracy between the predicted leaf area index and the measured leaf area index was analyzed ( Figure 3 ).

[0026] (III) Genetic neural network model based on GF-3SAR remote sensing images

[0027] Based on GF-3SAR remote sensing images, the backscattering coefficient model obtained by improving the cloud water model is combined with the optical and microwave polarization decomposition vegetation index model based on GF-1WFV and GF-3SAR remote sensing images. The leaf area index inversion method of the genetic neural network model and the BP neural network model is established. Figure 4 ), extract leaf area index information of winter wheat in five growth stages in Xiangfu District, Kaifeng City, Henan Province ( Figure 5 ).

Claims

1. A SAR data genetic neural network model inversion method for winter wheat LAI, the measurement method comprises the following steps: Step 1) Based on GF-1WFV and GF-3SAR remote sensing images, an optical and microwave polarization decomposition vegetation index model is constructed to obtain a fitting accuracy map between the model and the measured leaf area index; Step 2) Based on GF-3SAR remote sensing images, the improved cloud water model is used to extract the backscattering coefficients HV, VV, and VV, and the fitting accuracy of the three backscattering coefficient models and the measured leaf area index is obtained; Step 3) Based on GF-3SAR remote sensing images, the improved cloud-water model is used to predict the leaf area index, and a fitting accuracy diagram of the predicted leaf area index and the measured leaf area index is obtained; Step 4) Based on the GF-3SAR remote sensing image, the backscattering coefficient obtained by the improved cloud water model, the leaf area index obtained by inversion of the improved cloud water model, and the optical and microwave polarization decomposition vegetation index model are used as input factors of the neural network model, and the BP neural network model is used to predict the leaf area index to obtain a fitting accuracy map of the predicted leaf area index and the measured leaf area index; Step 5) Based on GF-3SAR remote sensing images, the backscattering coefficient obtained by the improved cloud water model, the leaf area index obtained by inverting the improved cloud water model, and the optical and microwave polarization decomposition vegetation index model are used as input factors of the neural network model. The stability of the BP neural network and the fitness ratio, real number crossover and mutation of the genetic algorithm are used to enhance the training and optimize the search capabilities, and a genetic neural network model is constructed to obtain the fitting accuracy map of the predicted leaf area index and the measured leaf area index, as well as the inversion result map of the leaf area index of winter wheat in 5 growth periods.

2. The method according to claim 1, characterized in that The step 1): proposes an empirical model between the vegetation index model and the leaf area index based on optical and microwave polarization decomposition.

3. The method according to claim 1, characterized in that The step 2): proposes a backscattering coefficient model HV, VV and an empirical model between VV and leaf area index based on an improved cloud water model.

4. The method according to claim 1, characterized in that The step 3): proposes an improved cloud water model based on the traditional cloud water model theory.

5. The method according to claim 1, characterized in that The step 4): using the backscattering coefficient obtained by the improved cloud water model, the leaf area index obtained by inverting the improved cloud water model, and the optical and microwave polarization decomposition vegetation index model as input factors of the BP neural network model, a leaf area index inversion model based on the BP neural network is proposed.

6. The method according to claim 1, characterized in that The step 5): using the backscattering coefficient obtained by the improved cloud water model, the leaf area index obtained by inverting the improved cloud water model, and the optical and microwave polarization decomposition vegetation index model as input factors of the genetic neural network model, a leaf area index inversion model based on the genetic neural network is proposed.

Images