CN118038223A - AI-based remote sensing data sample information fusion and multi-scale reconstruction method and system - Google Patents

Abstract

The invention discloses an AI-based remote sensing data sample information fusion and multi-scale reconstruction method and system, belonging to the technical field of remote sensing interpretation, wherein the method comprises the following steps: acquiring multi-source sample true value data for remote sensing interpretation in a preset area; constructing a multisource sample fusion and multiscale reconstruction model based on ANN and CNN; training and optimizing an ANN module in the multi-source sample fusion and multi-scale reconstruction model by using the multi-source sample true value data; and carrying out remote sensing data sample information fusion and multi-scale reconstruction by using the trained multi-source sample fusion and multi-scale reconstruction model, and outputting a true value of the multi-scale sample. The method can be used for fusing the useful information of the multisource samples, is convenient for rapidly and intelligently acquiring high-quality sample true values of different scales, solves the problems of insufficient sample verification and low sample quality of the existing large-scale remote sensing intelligent interpretation products, and is beneficial to carrying out authenticity verification and comprehensive evaluation on the accuracy of the remote sensing interpretation products to be verified.

Description

AI-based remote sensing data sample information fusion and multi-scale reconstruction method and system

Technical Field

The invention relates to the technical field of remote sensing interpretation, in particular to a remote sensing data sample information fusion and multi-scale reconstruction method and system based on AI.

Background

At present, with the continuous development of satellite remote sensing technology, the variety of sensors is more and more, and multimode and multi-scale high-resolution remote sensing interpretation data gradually become important information sources in aspects of resources, ecological environment, forestry, agriculture and the like. The development of Artificial Intelligence (AI) technology, especially the rapid development of machine learning/deep learning models and the wide application thereof in the remote sensing technical field, provides an effective solution for the rapid and intelligent interpretation of multi-mode remote sensing big data. However, the accuracy of the corresponding multi-mode high-resolution remote sensing intelligent interpretation product is also urgently required to be improved, and particularly the interpretation of the remote sensing image in a complex natural scene. Therefore, to improve the precision of the intelligent interpretation product of the complex scene remote sensing and improve the quality of the product, expand the application range of the product, the authenticity of the product must be verified.

The core problem of the verification of the authenticity of the remote sensing interpretation product is a problem of the dimension, and the problem that the dimension of a sample is matched with the dimension of the interpretation product to be verified is always a pending problem in the verification research of the authenticity. The absolute true value of a remote sensing intelligent interpretation sample is mainly obtained by a small-range ground sampling investigation method in the industry, the supply-demand contradiction between the measurement capability of a ground actual measurement 'point' and the demand of a remote sensing pixel scale 'surface' is obvious, the ground observation cannot directly obtain the true value of the pixel scale, and only the optimal approximation can be realized; the acquisition of the true value of the sample in the area which cannot be reached by human is mainly realized by using the verified high-quality remote sensing interpretation product, however, the remote sensing interpretation product is influenced by various factors, so that the error of the surface coverage remote sensing data product itself exists to a considerable extent; for multi-scale sample acquisition, the method mainly utilizes a mature scale conversion method (such as a double-line interpolation method and the like) to acquire, so that the requirements of interpretation models and verification of mass samples in the precision evaluation of the existing remote sensing intelligent interpretation products are hardly met, a large amount of useful information is lost, and high-quality sample true values cannot be acquired rapidly and intelligently.

At present, models such as machine learning/deep learning and the like are widely applied to the authenticity verification research of quantitative remote sensing, and the research of the authenticity verification of a remote sensing intelligent interpretation product is rarely reported. In order to solve the above-mentioned problems, a set of multi-scale sample scale conversion and reconstruction technology for obtaining the true value useful information of the fusion multi-source sample rapidly and intelligently is needed to be established.

Disclosure of Invention

In view of this, the present invention provides a remote sensing data sample information fusion and multiscale reconstruction method and system based on AI, which at least solves the above part of the technical problems, so as to facilitate rapid and intelligent acquisition of high quality sample true values with different scales, and facilitate solving the problems of insufficient verification samples and low sample quality of the current large-scale remote sensing intelligent interpretation products, and facilitate carrying out authenticity verification and comprehensive evaluation on the accuracy of the remote sensing interpretation products to be verified.

In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:

In a first aspect, an embodiment of the present invention provides a method for information fusion and multi-scale reconstruction of remote sensing data samples based on AI, where the method includes:

acquiring multi-source sample true value data for remote sensing interpretation in a preset area;

constructing a multisource sample fusion and multiscale reconstruction model based on ANN and CNN;

training and optimizing an ANN module in the multi-source sample fusion and multi-scale reconstruction model by using the multi-source sample true value data;

And carrying out remote sensing data sample information fusion and multi-scale reconstruction by using the trained multi-source sample fusion and multi-scale reconstruction model, and outputting a true value of the multi-scale sample.

In an alternative embodiment, the multi-source sample truth data comprises; ground sample data, historical data, and human interpretation data.

In an optional implementation manner, hierarchical optimization sampling strategies based on association of different spatial hierarchical complex natural scenes and observation targets are adopted to preprocess the multi-source sample truth value data; the optimized sampling strategy comprises the following steps: determining the total sample size, hierarchical sampling, spatial correlation analysis, topography and geography analysis and distribution sample density of the region.

In an alternative embodiment, the total sample size of the region is determined by calculation using an optimization model; the optimization model is as follows:

Wherein n represents the sample size to be calculated; mu _1-α/2 represents a standard normal distribution critical value when the confidence coefficient is 1-alpha/2; alpha is taken to be 5%; p represents the estimated error accuracy; r is the sampling error; n is the total sample amount.

In an optional embodiment, in the multi-source sample fusion and multi-scale reconstruction model, the ANN module is utilized to perform information fusion on the multi-source sample truth value data, a high-precision target sample truth value is output through optimization and iterative training, the output target sample truth value is input into the CNN module, parameters are set in different convolution layers according to the scale of an interpretation product to be verified, and the multi-scale sample truth value is output.

In an alternative embodiment, the parameters are set in different convolution layers according to the scale of the interpretation product to be verified, and the parameters include: number of layers, number of channels, convolution kernel size, and offset value.

In an alternative embodiment, the sample true scale is scaled up and down by setting a move step, adding pooling layers, or filling each convolution operation.

In a second aspect, the embodiment of the invention further provides an AI-based remote sensing data sample information fusion and multi-scale reconstruction system, and the AI-based remote sensing data sample information fusion and multi-scale reconstruction method is applied to intelligently generate a multi-scale sample true value; the system comprises:

the data acquisition module is used for acquiring multi-source sample true value data for remote sensing interpretation in a preset area;

the model construction module is used for constructing a multisource sample fusion and multiscale reconstruction model based on the ANN and the CNN;

The model training module is used for training and optimizing the ANN module in the multi-source sample fusion and multi-scale reconstruction model by using the multi-source sample true value data;

The model application module is used for carrying out remote sensing data sample information fusion and multi-scale reconstruction by using the trained multi-source sample fusion and multi-scale reconstruction model and outputting a true value of the multi-scale sample.

Compared with the prior art, the invention has at least the following beneficial effects:

The invention provides an AI-based remote sensing data sample information fusion and multi-scale reconstruction method and system. The invention can solve the problems of sample quality and scale in the verification of the authenticity of the large-scale interpretation product. The generated multi-scale sample is beneficial to providing a large number of high-quality sample true values for the subsequent authenticity verification and comprehensive evaluation of the accuracy of the interpretation product to be verified.

Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and drawings.

The technical scheme of the invention is further described in detail through the drawings and the embodiments.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.

Fig. 1 is a schematic flow chart of a remote sensing data sample information fusion and multi-scale reconstruction method based on AI according to an embodiment of the present invention.

Fig. 2 is a schematic flow chart of an optimized sampling strategy according to an embodiment of the present invention.

Fig. 3 is a schematic diagram of an operation principle of an ANN module according to an embodiment of the present invention.

Fig. 4 is a schematic diagram of a CNN module according to an embodiment of the present invention.

Fig. 5 is a schematic diagram of sample collection of a test area according to different complex natural scenes according to an embodiment of the present invention.

Fig. 6 is a schematic diagram of three sample truth data of a model input according to an embodiment of the present invention.

Fig. 7 is a schematic diagram of truth values of 5 scale samples of a model output according to an embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. Moreover, various numbers and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.

Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Referring to fig. 1, the invention provides a scale conversion and reconstruction technology for constructing three truth samples of ground actual measurement, historical data and high-resolution remote sensing visual interpretation products based on an information fusion and artificial intelligence technology, which can fuse useful information of a multi-source sample, and can rapidly and intelligently acquire a multi-scale-plane truth sample matched with the interpretation products to be verified so as to verify and evaluate the accuracy of the different-scale remote sensing interpretation products, aiming at the problem that the current verification multi-scale remote sensing interpretation products lack rapid intelligent acquisition of high-quality multi-scale sample truth values. The technology is convenient for solving the problems of insufficient sample verification and sample quality of the current large-scale remote sensing intelligent interpretation product. The method overcomes the limitation of the current ground sampling (small-range sampling) and reduces the errors and uncertainties generated in the algorithm and data acquisition of the sample value acquired by the verified high-quality remote sensing product.

Referring to fig. 1, the method for fusing and reconstructing information of remote sensing data samples based on AI provided by the invention mainly comprises the following steps:

acquiring multi-source sample true value data for remote sensing interpretation in a preset area;

constructing a multisource sample fusion and multiscale reconstruction model based on ANN and CNN;

training and optimizing an ANN module in the multi-source sample fusion and multi-scale reconstruction model by using the multi-source sample true value data;

And carrying out remote sensing data sample information fusion and multi-scale reconstruction by using the trained multi-source sample fusion and multi-scale reconstruction model, and outputting a true value of the multi-scale sample.

The following describes in detail the specific embodiments of the technical scheme of the present invention:

(1) Optimized sampling scheme

In this embodiment, the present invention proposes a hierarchical optimized sampling strategy based on association of different spatially layered complex natural scenes with an observation target. The optimized sampling strategy flow is shown in fig. 2: the optimized sampling strategy comprises the following steps: determining the total sample size of the area, hierarchical sampling, spatial correlation analysis, topography and geography analysis, distribution of sample point density and the like.

In the embodiment, the remote sensing data is covered on the surface of the existing research area (demonstration area), the research area is pre-classified, and then the research area is sampled in a layered mode according to the type of the pre-classifying result, so that confusion among layers can be avoided, intra-layer difference is small, inter-layer difference is large, and the type representativeness principle is met. Determining the total sample quantity of the region, calculating the total sample quantity of each surface type to be extracted by using an optimization model, wherein the optimization model formula is as follows:

Wherein n represents the sample size to be calculated; mu _1-α/2 represents a standard normal distribution critical value when the confidence coefficient is 1-alpha/2; alpha is taken to be 5%; p represents the estimated error accuracy; r is the sampling error; n is the total sample amount. When the sample size is calculated, only sampling error and pre-estimated error precision need to be given.

In this embodiment, each floor in the hierarchical sampling is a layer, and the sample size is allocated from the sample population according to the area ratio occupied by each layer. And (3) calculating and analyzing the correlation degree layout samples of each space unit and the adjacent units by using a local correlation index (such as a local Morlan index), and reducing the correlation among the samples. The sample with space representation is properly selected in the sample area with low topography fluctuation by referring to DEM data, and the sample can be properly encrypted in the area with high topography fluctuation.

(2) Multisource sample fusion and multiscale reconstruction model based on ANN model and CNN model

In the embodiment, the remote sensing sample information fusion and multi-scale conversion and reconstruction model of the ANN combined CNN model based on the AI technology under the deep learning framework is built. According to the model, information fusion is carried out on three data input samples by using an ANN module, high-quality target sample true values can be output through optimization and iterative training, the output high-quality target sample true values are input into a CNN model, parameters are set on different convolution layers according to the scale of an interpretation product to be verified, sample true values of different scales are output, the requirement of intelligently acquiring the sample true values of different scales is met, and technical support and a novel method are provided for quickly and intelligently acquiring the sample true values under the condition of large-scale and ground-free verification.

In a specific embodiment, the invention combines the Artificial Neural Network (ANN) based simulation with multi-source sample truth value data to synthesize a sample truth value with higher precision. In the high-quality sample true value generation process, firstly, a trained ANN is initialized, then three sample true values (ground sampling, historical data and manual interpretation) are input into the ANN, and through a series of iterative operation and parameter optimization, errors are minimized, so that the purposes of fusing three data useful information and outputting the high-quality interpretation sample true values are achieved. The working principle of the ANN module in the invention is shown in figure 3: the left neuron is an input neuron that receives different sample truth data, i.e., three different product data (X ₁、X₂ and X ₃) are input. The intermediate neuron is a hidden layer, and through synthesizing input data information and transmitting the information to the next layer, finally, the true value y of the interpretation target sample with higher accuracy is output. In this embodiment, the main parameters of the ANN module are the number of hidden layers and the number of neurons per layer.

Furthermore, the true value data of the target sample output by the ANN module is output based on different CNN convolution kernel operations, so that the lifting scale expansion of the data product is realized.

In a specific embodiment, different convolution kernel parameters are set in different convolution layers of the CNN module, and the convolution kernel operation synthesizes local information, so that scale lifting is realized, and the number and the size of convolution kernels required by different scale lifting can be different.

Referring to fig. 4, the principle of the CNN module is that the input layer gradually highlights the image features through multiple convolution layers, and in the convolution layers, the features of the previous layer extract new features through convolution kernel operation and output the new features to the next layer. And sliding in the image window through the convolution kernel, calculating the product of the corresponding positions each time, summing the products, and outputting the products to the next layer. The main parameters are the number of layers, the number of channels, the length-width scale of the convolution kernel, the paranoid value and the like. And for each convolution operation, the target of sample true value scale lifting is achieved by setting a moving step length, adding a pooling layer or conducting padding.

The method of the invention is described below by taking multi-scale sample reference truth value acquisition of remote sensing interpretation products developed in improve literature county of Guiyang city as an example:

the Guizhou Guiyang city improve literature county is located in the southwest mountain area of China, and has various topography, complex meteorological conditions, severe topography change, finely divided ground surface structures, staggered multiple places and other complex scenes. Under such complex terrain conditions, it is difficult to implement to accurately obtain ground sampling points of different ground types over a large range. The method of the invention is implemented as follows:

(1) Optimized sampling of samples in complex scenarios

In this embodiment, the improve literature county partial block is intercepted as the test area, and three truth values are adopted for the multi-source sample data: the method comprises the steps of selecting a same area from three kinds of data, wherein the area is provided with a ground sampling point, a historical ground surface coverage classification product and a visual interpretation chart of 0.8m domestic GF-2 optical remote sensing data fused in 2023 years, and selecting a plurality of samples with the size of 1000 pixels as input data of a model network according to a sampling strategy. Where ground samples were taken as absolute true values, 70% as model training dataset, 15% as model test dataset, and 15% as validation dataset. The sample collection of improve literature county test areas according to different complex natural scenes is shown in fig. 5, wherein the left part in fig. 5 is a remote sensing image, and the right part is DEM data.

(2) Multi-source sample fusion and multi-scale reconstruction model based on ANN and CNN in research area

In this embodiment, the present invention is described with respect to modeling a sample sampling area of three input data sources. Three sample truth data input by the model are shown in fig. 6, which is ground sampling data, visual interpretation data and historical data: after three samples are input into an ANN and CNN-based model, referring to FIG. 7, the model outputs multi-scale (5 scales, 2m, 8m, 10m, 16m and 30 m) sample truth results; the samples may also be vector/trellis converted.

In the embodiment, 70% of the model is trained by using the absolute true value of ground sampling, 15% of model parameters are corrected, and 15% of model output multi-scale sample results are evaluated in precision; the accuracy evaluation index is Kappa coefficient, and the Kappa coefficient is over 80% as a result, so that the model constructed by the method is feasible, and the multi-scale sample true value with high quality can be obtained rapidly and intelligently by utilizing the multi-source sample fusion and multi-scale reconstruction model constructed by the ANN and CNN modules. The sample true value generated by the model in an intelligent way is a quick, effective and feasible method for generating the sample true value. In the future, the method can provide new technical support for generating high-quality surface truth value samples for areas with few or missing ground sampling points.

From the description of the above embodiments, those skilled in the art can appreciate that the present invention provides an AI-based remote sensing data sample information fusion and multi-scale reconstruction method, in which a multi-source sample information fusion and scale reconstruction model is built based on an Artificial Neural Network (ANN) model and a Convolutional Neural Network (CNN) model. The useful information of the three truth samples is fused in the ANN module, the sample error is reduced through continuous iteration, the accuracy of the sample truth is improved, and the high-quality face truth samples with different scales matched with the space of the remote sensing interpretation product to be verified are obtained through rapid and intelligent setting parameters of different convolution layers in the CNN module. The invention can solve the problems of sample quality and scale in the verification of the authenticity of the large-scale interpretation product. And providing a large number of high-quality sample true values for the subsequent authenticity verification and comprehensive evaluation of the accuracy of the interpretation product to be verified by using the generated multi-scale samples.

Further, the invention also provides an AI-based remote sensing data sample information fusion and multi-scale reconstruction system, which is applied to the AI-based remote sensing data sample information fusion and multi-scale reconstruction method in the above embodiment, and intelligently generates a multi-scale sample true value, and the system comprises:

the data acquisition module is used for acquiring multi-source sample true value data for remote sensing interpretation in a preset area;

the model construction module is used for constructing a multisource sample fusion and multiscale reconstruction model based on the ANN and the CNN;

The model training module is used for training and optimizing the ANN module in the multi-source sample fusion and multi-scale reconstruction model by using the multi-source sample true value data;

The model application module is used for carrying out remote sensing data sample information fusion and multi-scale reconstruction by using the trained multi-source sample fusion and multi-scale reconstruction model and outputting a true value of the multi-scale sample.

The system provided by the embodiment of the present invention has the same implementation principle and technical effects as those of the foregoing method embodiment, and for brevity description, the corresponding contents in the foregoing method embodiment may be referred to for the parts of the system embodiment that are not mentioned, and will not be described herein again.

In addition, an embodiment of the present invention further provides a storage medium having stored thereon one or more programs readable by a computing device, the one or more programs including instructions, which when executed by the computing device, cause the computing device to perform a remote sensing data sample information fusion and multiscale reconstruction method based on AI in the above embodiment.

In an embodiment of the present invention, the storage medium may be, for example, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the storage medium include: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, and any suitable combination of the foregoing.

It will be appreciated by those skilled in the art that embodiments of the invention may be provided as a method, system, or computer program product, or the like. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects.

It is to be noticed that the term 'comprising', does not exclude the presence of elements or steps other than those listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer.

In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (8)

1. The AI-based remote sensing data sample information fusion and multi-scale reconstruction method is characterized by comprising the following steps:

acquiring multi-source sample true value data for remote sensing interpretation in a preset area;

constructing a multisource sample fusion and multiscale reconstruction model based on ANN and CNN;

training and optimizing an ANN module in the multi-source sample fusion and multi-scale reconstruction model by using the multi-source sample true value data;

And carrying out remote sensing data sample information fusion and multi-scale reconstruction by using the trained multi-source sample fusion and multi-scale reconstruction model, and outputting a true value of the multi-scale sample.

2. The AI-based remote sensing data sample information fusion and multiscale reconstruction method of claim 1, wherein the multisource sample truth data comprises; ground sample data, historical data, and human interpretation data.

3. The AI-based remote sensing data sample information fusion and multi-scale reconstruction method of claim 2, wherein hierarchical optimization sampling strategies associated with observation targets based on different spatial hierarchical complex natural scenes are adopted to preprocess the multi-source sample true value data; the optimized sampling strategy comprises the following steps: determining the total sample size, hierarchical sampling, spatial correlation analysis, topography and geography analysis and distribution sample density of the region.

4. The AI-based remote sensing data sample information fusion and multiscale reconstruction method of claim 3, wherein the overall sample size of the region is determined by calculation using an optimization model; the optimization model is as follows:

Wherein n represents the sample size to be calculated; mu _1-α/2 represents a standard normal distribution critical value when the confidence coefficient is 1-alpha/2; alpha is taken to be 5%; p represents the estimated error accuracy; r is the sampling error; n is the total sample amount.

5. The AI-based remote sensing data sample information fusion and multiscale reconstruction method according to claim 1, wherein in the multisource sample fusion and multiscale reconstruction model, an ANN module is utilized to conduct information fusion on multisource sample true value data, high-precision target sample true values are output through optimization and iterative training, the output target sample true values are input into a CNN module, parameters are set in different convolution layers according to the scale of an interpretation product to be verified, and multiscale sample true values are output.

6. The AI-based remote sensing data sample information fusion and multiscale reconstruction method of claim 5, wherein parameters are set at different convolution layers according to the scale of the interpreted product to be verified, and the parameters include: number of layers, number of channels, convolution kernel size, and offset value.

7. The AI-based remote sensing data sample information fusion and multiscale reconstruction method of claim 6, wherein the sample true scale is scaled up and down by setting a move step, adding a pooling layer, or filling each convolution operation.

8. An AI-based remote sensing data sample information fusion and multiscale reconstruction system, characterized in that an AI-based remote sensing data sample information fusion and multiscale reconstruction method according to any one of claims 1-7 is applied to intelligently generate a multiscale sample true value; the system comprises:

the data acquisition module is used for acquiring multi-source sample true value data for remote sensing interpretation in a preset area;

the model construction module is used for constructing a multisource sample fusion and multiscale reconstruction model based on the ANN and the CNN;

The model training module is used for training and optimizing the ANN module in the multi-source sample fusion and multi-scale reconstruction model by using the multi-source sample true value data;

The model application module is used for carrying out remote sensing data sample information fusion and multi-scale reconstruction by using the trained multi-source sample fusion and multi-scale reconstruction model and outputting a true value of the multi-scale sample.