Intelligent asteroid albedo estimation method based on visible light and near infrared spectrum data

Abstract

The invention discloses an asteroid albedo intelligent estimation method based on visible light and near infrared spectrum data, and belongs to the technical field of asteroid spectrum analysis. The method comprises the following steps: S1, collecting original spectral data; s2, finding out corresponding asteroid albedo data from the original spectral data, and labeling; s3, dividing the data according to 8: 1: 1 to obtain a spectrum training set, a spectrum verification set and a spectrum test set; s4, constructing a deep learning albedo estimation model AAE-Net, substituting the spectrum training set into the constructed model for training, and stopping training when the performance on the verification set is no longer improved; and S5, calculating the estimation error of the trained model in each category on the test set, and evaluating the model estimation performance. According to the method, a deep learning albedo estimation model AAE-Net based on the combination of a convolutional neural network and a Transform network is used for giving an asteroid albedo estimation value only by means of visible light and near infrared spectrum data when data is insufficient.

Description

Technical Field

[0001] The present invention belongs to the technical field of asteroid spectral analysis, and specifically relates to an asteroid albedo intelligent estimation method based on visible light and near-infrared spectral data. Background Art

[0002] The exploration of small celestial bodies in the solar system has always been the preface and hot topic of deep space exploration. The main significance of asteroid exploration is threefold: studying the secrets of the evolution of the solar system based on the materials retained in asteroids, monitoring asteroids that may threaten the Earth, and developing asteroids rich in mineral resources. The study of the physical properties of asteroids is an important part of asteroid research.

[0003] The reflectivity of a planet's surface is an important basis for determining the type of surface minerals, inverting mineral content, inverting roughness, and correcting terrain. Albedo is an important parameter for characterizing the reflectivity of a planet's surface and a key parameter for establishing a thermal model of the planet's surface. Albedo is also related to other important physical characteristics such as the type and content of surface soil particles and the mass of the planet. By observing changes in the albedo of a planet's surface, the physical structure of the asteroid surface can be analyzed. The trend of changes in surface albedo is strongly correlated with the space weathering of the asteroid's surface. The study of surface albedo is considered to be one of the important links in scientific research on asteroids.

[0004] In general, the albedo of an asteroid is estimated by measuring the intensity of the asteroid's thermal radiation in the infrared band, which can estimate its surface temperature. Combined with the diameter of the asteroid, the albedo can be derived. Or the albedo can be estimated by measuring the brightness of the asteroid's visible light. Another method is to infer the composition of the asteroid's surface material by measuring the spectrum of the asteroid, and the albedo of the asteroid can be estimated by combining the reflective properties of known materials. A large amount of existing asteroid albedo data comes from the observation results of space missions and model inferences combined with ground observations.

[0005] It can be seen that the calculation of albedo usually requires complex measurements and calculations, which usually requires a combination of multiple observational data and physical models to estimate. However, when data is insufficient, how to give a rough estimate of albedo based on the small amount of existing data is a very valuable research topic. Summary of the invention

[0006] In view of the above technical problems, since the albedo of an asteroid has a certain correlation with its category, this application uses a neural network with strong feature extraction and generalization capabilities to estimate the albedo of an asteroid through spectral data of visible light + near-infrared light bands. The present invention constructs a deep learning albedo estimation model AAE-Net based on a combination of a convolutional neural network (CNN) and a Transformer network. With the help of this model and combined with the spectral data of the visible light and near-infrared light bands of an asteroid, the albedo of an asteroid is intelligently estimated.

[0007] The present invention provides the following technical solution: an intelligent estimation method for asteroid albedo based on visible light and near-infrared spectral data, comprising the following steps:

[0008] S1. Collect raw data: collect various types of spectral data of asteroids in visible and near-infrared bands;

[0009] S2. Find the albedo data of the corresponding asteroid for the collected original spectral data and mark it as the actual albedo;

[0010] S3, divide the data set into training set, validation set and test set according to 8:1:1;

[0011] S4. Construct a deep learning albedo estimation model AAE-Net based on the combination of convolutional neural network CNN and Transformer network, bring the spectral training set into the constructed model for training, and stop training when the performance on the validation set no longer improves;

[0012] S5. Calculate the error between the estimated albedo and the actual albedo of each category of the trained model on the test set, and compare it with the albedo results estimated by other learning algorithms (such as SVR, DT, AdB, MLP, etc.) to evaluate the model estimation performance.

[0013] Furthermore, the deep learning albedo estimation model AAE-Net is an encoding-decoding structure, the encoder includes three one-dimensional convolutional layers, the front end of the decoder includes three fully connected layers, and the back end includes a transformer module and four fully connected layers; the encoder encodes the input information and inputs the encoded feature information into the decoder, which further estimates the albedo.

[0014] In the deep learning albedo estimation model AAE-Net, the differential module Diff is introduced to expand the input information, and the differential information and the original spectral data are input into the subsequent encoder as different feature channels to enhance the accuracy and robustness of the network; the differential module is calculated as follows:

[0015] diff i =ri -r i-1

[0016] Where diff i is the reflectance value r corresponding to the i-th band of the spectral sequence i Subtract the reflectance value r corresponding to the i-1th band i-1 .

[0017] Furthermore, the activation function of the encoder is LeakyReLu, which is expressed as follows:

[0018]

[0019] In the formula, α is a manually set hyperparameter, and α is set to 0.01 in this model.

[0020] By adopting the above technology, compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] The present invention combines the convolutional neural network CNN with the Transformer network to construct a neural network for estimating the albedo of an asteroid with a deeper number of layers and a more sophisticated design. The network can extract higher-dimensional semantic features, which helps to improve the accuracy of the network in estimating the albedo of the spectrum. A Diff difference module is added to the input end to extract additional input information, which helps the network to extract spectral gradient information and enhance the network's sensitivity to spectral peaks and troughs. Since there is a certain relationship between albedo and asteroid type, and the type can be determined by the spectrum, this network can be used to give an estimate of the albedo of an asteroid when data is insufficient, relying solely on visible light and near-infrared spectral data. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of the method of the present invention;

[0023] Figure 2 AAE-Net network structure diagram of the present invention;

[0024] Figure 3 This is a graph showing the relationship between the spectral shape of the X-type asteroid with the largest MAE and its corresponding albedo. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments of the specification. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0026] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below. Those skilled in the art can fully understand the present invention without the description of these details.

[0027] See also Figure 1-Figure 2 , a deep learning-based asteroid albedo estimation model, which combines convolutional neural network (CNN) with Transformer network to construct a deep learning albedo estimation model AAE-Net.

[0028] The network structure of AAE-Net is as follows Figure 2 shown.

[0029] Specifically, AAE-Net is designed as an encoder-decoder structure. The encoder includes three one-dimensional convolutional layers, the front end of the decoder includes three fully connected layers, and the back end includes a transformer module and four fully connected layers. The encoder encodes the input information and inputs the encoded feature information into the decoder, which further estimates the albedo.

[0030] In the data input part, this embodiment introduces a difference module Diff to expand the input information. The difference module is calculated as follows:

[0031] diff i =r i -r i-1

[0032] Where diff i is the reflectance value r corresponding to the i-th band of the spectral sequence i Subtract the reflectance value r corresponding to the i-1th band i-1 .

[0033] The differential information and the original spectrum are input into the subsequent encoder as different feature channels to enhance the accuracy and robustness of the network. The encoder consists of three one-dimensional convolutional layers, and the activation function is LeakyReLu, which is expressed as follows:

[0034]

[0035] The encoder encodes the input information and inputs the encoded feature information into the decoder, which further estimates the albedo. The decoder is designed to be similar to the U-Net structure. The front end consists of three fully connected layers with 128, 256, and 512 neurons, which further extract and refine the input feature information. The second half includes a transformer module and four fully connected layers. The number of neurons in the first three fully connected layers is 512, 256, and 11, respectively. The number of neurons in the last layer is 1, and the estimated value of the albedo is output.

[0036] The specific process of the intelligent estimation method is as follows:

[0037] Step 1: The original spectral data used in the test comes from A. The VisNIR dataset shared by the team includes 591 individual asteroid spectra; each spectrum covers wavelengths from 0.45 to 2.45 μm with a step size of 0.01 μm.

[0038] Step 2: The obtained spectral data are annotated with the corresponding asteroid albedo information obtained from the Solar System Open Database Network (SsODNet, Berthier et al.), which constitutes the dataset used in our experiment.

[0039] Step 3: Divide the dataset into training set, validation set and test set according to the ratio of 8:1:1.

[0040] Step 4: Calculate the error between the estimated albedo and the actual albedo of each category of the trained model on the test set, and compare it with other algorithms to evaluate the model performance.

[0041] In step 4, the mean absolute error of each category of the model finally obtained, as well as the comparison results with other algorithms are shown in Table 1 and Table 2. The mean absolute error (MAE) is used as the evaluation indicator, and the specific calculation method is as follows:

[0042]

[0043] Where: n is the number of asteroids, is the albedo predicted by the network, y i is the true value of albedo.

[0044] Table 1. Albedo accuracy test results estimated by AAE-Net

[0045]

[0046] Table 2 Comparison results of AAE-Net and other algorithms

[0047]

[0048] From the results, we can see that in the task of asteroid albedo estimation, AAE-Net has obvious advantages over traditional machine learning methods. The mean absolute error is reduced by about 30% compared with other algorithms. The MAE for X-type asteroids is somewhat large. Although the MAE for C-type asteroids is small, the albedo of C-type asteroids is mostly less than 0.1, so the relative error may be large. However, the albedo estimation of S-type asteroids is more feasible, and an albedo can be estimated through the visible light and near-infrared spectra of the asteroid, which proves the practicality and effectiveness of the present invention.

[0049] At the same time, the relationship between the spectrum shape of the X-type asteroid with the largest MAE and the corresponding albedo is given as follows: Figure 3 As shown, it can be seen that the albedo of some asteroids with similar spectral shapes is significantly different, such as 1999HF1 (137170) and Kenos (2449) in (b). Their spectral shapes are similar, and the SAD (spectral angular distance, a parameter used to evaluate spectral similarity, the smaller the SAD, the more similar the spectra are.) is only 0.053, and the albedo difference is 0.333; however, the reflectivity difference of other asteroids with significantly different spectral shapes is relatively small, such as 1999HF1 (137170) and 1998WT24 (33342) in (c). There is a significant difference in the spectral shape of these two asteroids, with an SAD of 0.128, but the reflectivity difference is only 0.003. The reason may be that the spectral data we use is normalized, which may lose some relevant information, or it may be that only relying on this part of the spectral data can only obtain the current results, and further characteristic data is needed.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent estimation method for asteroid albedo based on visible light and near-infrared spectral data, characterized in that: The following steps are involved: S1. Collect raw data: collect various types of spectral data of asteroids in visible and near-infrared bands; S2. Find the albedo data of the corresponding asteroid for the collected original spectral data and mark it as the actual albedo; S3, divide the data set into training set, validation set and test set according to 8:1:1; S4. Construct a deep learning albedo estimation model AAE-Net based on the combination of convolutional neural network CNN and Transformer network, bring the spectral training set into the constructed model for training, and stop training when the performance on the validation set no longer improves; S5. Calculate the error between the estimated albedo and the actual albedo of each category of the trained model on the test set, and compare it with the albedo results estimated by other algorithms to evaluate the model performance.

2. The method for intelligently estimating the albedo of an asteroid based on visible light and near-infrared spectral data according to claim 1, characterized in that: The deep learning spectral albedo estimation model AAE-Net is an encoder-decoder structure. The encoder includes three one-dimensional convolutional layers, the front end of the decoder includes three fully connected layers, and the back end includes a transformer module and four fully connected layers. The encoder encodes the input information and inputs the encoded feature information into the decoder, which outputs the corresponding asteroid albedo estimate.

3. The method for intelligently estimating the albedo of an asteroid based on visible light and near-infrared spectral data according to claim 2, characterized in that: In the deep learning albedo estimation model AAE-Net, the differential module Diff is introduced to expand the input information, and the differential information and the original spectral data are input into the subsequent encoder as different feature channels to enhance the accuracy and robustness of the network; the differential module is calculated as follows: diff i =r i -r i-1 Where diff i is the reflectance value r corresponding to the i-th band of the spectral sequence i Subtract the reflectance value r corresponding to the i-1th band i-1 .

4. The method for intelligently estimating asteroid albedo based on visible light and near-infrared spectral data according to claim 3, characterized in that: The activation function of the encoder is LeakyReLu, which is expressed as follows: In the formula, α is a manually set hyperparameter, and α is set to 0.01 in this model.

5. The method for intelligently estimating asteroid albedo based on visible light and near-infrared spectral data according to claim 1, characterized in that: In step S5, the mean absolute error (MAE) is used as the evaluation index, and the specific calculation method is as follows: Where: n is the number of asteroids, is the albedo predicted by the network, y i is the true value of albedo, i.e. the actual albedo.

Images