CN116051821B

CN116051821B - Engineering-oriented single sliding window full-month landing zone selection method and system thereof

Info

Publication number: CN116051821B
Application number: CN202310184502.3A
Authority: CN
Inventors: 王永志; 刘亨玺; 曹亚琴; 王嘉翔; 温世博
Original assignee: Jilin University
Current assignee: Jilin University
Priority date: 2023-03-01
Filing date: 2023-03-01
Publication date: 2023-06-02
Anticipated expiration: 2043-03-01
Also published as: CN116051821A

Abstract

The invention discloses a single sliding window full-moon landing zone selection method and a system thereof for engineering, which relate to the field of intelligent evaluation, wherein a sliding window is used for traversing a full-moon surface gradient map to obtain a plurality of single window evaluation data, each single window evaluation data is processed by a deep neural network model to obtain a plurality of single window evaluation feature vectors, the single window evaluation feature vectors are processed by a context encoder to obtain global context association evaluation feature vectors, transfer matrixes of each single window evaluation feature vector and the global context association evaluation feature vector are calculated to obtain a plurality of classification feature matrixes, and finally, the classification feature matrixes are processed by a classifier to obtain a plurality of classification results for indicating whether a lunar surface area corresponding to a single window is suitable to land. In this way, the complexity of assessing the suitability of the landing zone can be reduced.

Description

Engineering-oriented single sliding window full-month landing zone selection method and system thereof

Technical Field

The application relates to the field of intelligent evaluation, and in particular relates to a method and a system for selecting a single sliding window full-month landing zone facing engineering.

Background

Moon as the first station for deep space exploration of humans, the development of lunar science is also a prerequisite for understanding solar systems. The direct method of recognizing moon is soft landing, the choice of landing zone has been the focus of lunar science, and the quantitative method of the suitability of moon landing is less.

The front topography of the moon is flatter than the back of the moon, the front topography is mainly flat moon sea, the area can occupy more than half of the front of the moon, so that the detector can safely land, the back of the moon is subjected to long-term direct meteorite bombardment, the impact pit density is larger, the topography is rugged and mainly high, and soft landing is difficult to carry out. In order to further advance the research of the key scientific concepts of the moon, in recent years, a plurality of domestic and foreign scholars have studied on landing area site selection of the moon from the aspects of engineering and science.

The target selection of the landing zone can be generalized to be a flat area and location preference in the scientific content set, however, this process lacks an efficient model for review quantification. The automatic landing zone addressing scheme has any of the following two points to be propelled: 1. the whole month grid segmentation lacks an overlapping part, so that the space at the grid boundary is omitted; 2. quantitative models of evidence weights and fractal methods are difficult to calculate landing zone suitability indexes of any area at any position.

Thus, an optimized engineering-oriented single sliding window full month landing zone selection scheme is desired.

Disclosure of Invention

The present application has been made in order to solve the above technical problems. The embodiment of the application provides a method and a system for selecting a single sliding window full-month landing zone facing engineering. Firstly traversing a full-moon surface gradient map by using a sliding window to obtain a plurality of single-window evaluation data, then passing each single-window evaluation data through a deep neural network model to obtain a plurality of single-window evaluation feature vectors, then passing the plurality of single-window evaluation feature vectors through a context encoder to obtain a global context correlation evaluation feature vector, then calculating a transfer matrix of each single-window evaluation feature vector and the global context correlation evaluation feature vector to obtain a plurality of classification feature matrices, and finally carrying out feature distribution correction on the plurality of classification feature matrices and then passing through a classifier to obtain a plurality of classification results for indicating whether a lunar surface area corresponding to a single window is suitable for landing. In this way, the complexity of assessing the suitability of the landing zone can be reduced.

According to one aspect of the application, a method for selecting an engineering-oriented single sliding window full-month landing zone is provided, which comprises the following steps:

Acquiring a full-moon surface gradient map;

traversing the full-moon surface gradient map by using a sliding window to obtain a plurality of single window evaluation data, wherein each single window evaluation data comprises a threshold value duty ratio, a mean value, a variation coefficient, a Morand index and a comprehensive index;

the single window evaluation data are passed through a deep neural network model comprising a plurality of full connection layers to obtain a plurality of single window evaluation feature vectors;

passing the plurality of single window evaluation feature vectors through a converter-based context encoder to obtain a global context-dependent evaluation feature vector;

taking each single window evaluation feature vector as a query feature vector, and calculating a transfer matrix of the single window evaluation feature vector and the global context association evaluation feature vector as a classification feature matrix to obtain a plurality of classification feature matrices;

based on the plurality of single window evaluation feature vectors and the global context correlation evaluation feature vector, performing feature distribution correction on the plurality of classification feature matrices to obtain a plurality of corrected classification feature matrices;

and passing each corrected classification characteristic matrix through a classifier to obtain a plurality of classification results, wherein each classification result is used for indicating whether the lunar surface area corresponding to the single window is suitable for landing.

In the engineering-oriented single sliding window full-month landing zone selection method, the step of obtaining a plurality of single window evaluation feature vectors by passing the single window evaluation data through a deep neural network model comprising a plurality of full-connection layers comprises the following steps:

arranging the single window evaluation data into an evaluation data input vector;

performing full-connection coding on the evaluation data input vector by using a plurality of full-connection layers of the deep neural network model according to the following formula to extract high-dimensional implicit features of feature values of all positions in the evaluation data input vector, wherein the formula is as follows:

, wherein />

Is the evaluation data input vector, +.>

Is the output vector, +.>

Is a weight matrix, < >>

Is a bias vector, ++>

Representing a matrix multiplication.

In the above engineering-oriented single sliding window full-month landing zone selection method, the step of obtaining the global context-associated evaluation feature vector by passing the plurality of single window evaluation feature vectors through a context encoder based on a converter includes:

inputting the plurality of single window evaluation feature vectors into the converter-based context encoder to obtain a plurality of single window evaluation semantic feature vectors;

and cascading the plurality of single window evaluation semantic feature vectors to obtain the global context-associated evaluation feature vector.

In the engineering-oriented single sliding window full-month landing zone selection method, the calculating the transfer matrix of the single window evaluation feature vector serving as the query feature vector and the global context associated evaluation feature vector serving as the classification feature matrix to obtain a plurality of classification feature matrices comprises the following steps:

calculating transfer matrixes of the single-window evaluation feature vectors and the global context association evaluation feature vectors as classification feature matrixes according to the following formula to obtain a plurality of classification feature matrixes;

wherein, the formula is:

wherein ,

representing said individual single window evaluation feature vectors, < >>

Representing the global context associated evaluation feature vector,/->

Representing the plurality of classification feature matrices, +.>

Representing vector multiplication.

In the engineering-oriented single sliding window full-month landing zone selection method, performing feature distribution correction on the multiple classification feature matrices based on the multiple single window evaluation feature vectors and the global context association evaluation feature vectors to obtain multiple corrected classification feature matrices, including:

the classification feature matrix is passed through an auxiliary convolution neural network serving as a feature extractor to obtain a weight feature map;

Modeling a feature correlation accumulation distinguishing mechanism in the weight feature map to obtain a weight feature vector;

multiplying the weight feature vector with each single window evaluation feature vector and the global context association evaluation feature vector according to position points to obtain a plurality of optimized single window evaluation feature vectors and optimized global context association evaluation feature vectors;

and calculating a transfer matrix of the optimized single-window evaluation feature vectors and the optimized global context associated evaluation feature vectors by taking the optimized single-window evaluation feature vectors as query feature vectors so as to obtain the corrected classification feature matrix.

In the above engineering-oriented single sliding window full-month landing zone selection method, the step of obtaining a weight feature map by passing the classification feature matrix through an auxiliary convolutional neural network serving as a feature extractor includes:

each layer of the auxiliary convolutional neural network serving as the feature extractor is respectively carried out in forward transfer of the layer:

carrying out convolution processing on input data to obtain a convolution characteristic diagram;

carrying out mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map;

non-linear activation is carried out on the pooled feature map so as to obtain an activated feature map;

The output of the last layer of the auxiliary convolution neural network serving as the feature extractor is the weight feature map, and the input of the first layer of the auxiliary convolution neural network serving as the feature extractor is the classification feature matrix.

In the above engineering-oriented single sliding window full-month landing zone selection method, modeling a feature correlation accumulation and differentiation mechanism in the weight feature map to obtain a weight feature vector, including:

modeling a feature correlation accumulation distinguishing mechanism in the weight feature map by using the following formula to obtain the weight feature vector;

wherein, the formula is:

wherein ,

representing the weight feature map, < >>

and />

Respectively representing the single-layer convolution operation based on different convolution kernels on the feature map,/and>

representation->

Activating function->

Representation->

Activate function, and->

Representing global pooling of each feature matrix of the feature map,/for each feature matrix>

Representing addition by position +.>

Representing the weighted feature vector.

In the above engineering-oriented single sliding window full-month landing zone selection method, the classifying feature matrix after correction is passed through a classifier to obtain a plurality of classifying results, where each classifying result is used to indicate whether the lunar surface area corresponding to the single window is suitable for landing, and the method includes:

Expanding each corrected classified feature matrix into a plurality of classified feature vectors according to row vectors or column vectors;

performing full-connection coding on the plurality of classification feature vectors by using a plurality of full-connection layers of the classifier to obtain a plurality of coding classification feature vectors;

and passing the plurality of coding classification feature vectors through a Softmax classification function of the classifier to obtain a plurality of classification results.

According to another aspect of the present application, there is provided an engineering-oriented single sliding window full month landing zone selection system, comprising:

the data acquisition module is used for acquiring a full-moon surface gradient map;

the traversing module is used for traversing the full-moon surface gradient map by using a sliding window to obtain a plurality of single window evaluation data, wherein each single window evaluation data comprises a threshold value duty ratio, a mean value, a variation coefficient, a Morand index and a comprehensive index;

the full-connection coding module is used for enabling each single window evaluation data to pass through a deep neural network model comprising a plurality of full-connection layers to obtain a plurality of single window evaluation feature vectors;

the context coding module is used for enabling the plurality of single window evaluation feature vectors to pass through a context coder based on a converter to obtain global context association evaluation feature vectors;

The classification feature matrix calculation module is used for calculating a transfer matrix of the single-window evaluation feature vector and the global context association evaluation feature vector to be used as a classification feature matrix by taking the single-window evaluation feature vector as a query feature vector so as to obtain a plurality of classification feature matrices;

the feature distribution correction module is used for carrying out feature distribution correction on the plurality of classification feature matrixes based on the plurality of single-window evaluation feature vectors and the global context association evaluation feature vectors so as to obtain a plurality of corrected classification feature matrixes;

and the classification module is used for passing each corrected classification characteristic matrix through a classifier to obtain a plurality of classification results, and each classification result is used for indicating whether the lunar surface area corresponding to the single window is suitable for landing.

In the engineering-oriented single sliding window full-month landing zone selection system, the full-connection coding module is used for:

inputting the evaluation data using a plurality of fully connected layers of the deep neural network model with the following formulaAnd carrying out full-connection coding on the vector to extract high-dimensional implicit features of feature values of each position in the evaluation data input vector, wherein the formula is as follows:

, wherein />

Is the evaluation data input vector, +.>

Is the output vector, +.>

Is a weight matrix, < >>

Is a bias vector, ++>

Representing a matrix multiplication.

Compared with the prior art, the engineering-oriented single sliding window full-month landing zone selection method and the engineering-oriented single sliding window full-month landing zone selection system are characterized in that firstly, a sliding window is used for traversing a full-month surface gradient map to obtain a plurality of single window evaluation data, then, each single window evaluation data is processed through a deep neural network model to obtain a plurality of single window evaluation feature vectors, then, the plurality of single window evaluation feature vectors are processed through a context encoder to obtain global context association evaluation feature vectors, then, transfer matrixes of each single window evaluation feature vector and the global context association evaluation feature vectors are calculated to obtain a plurality of classification feature matrixes, and finally, a classifier is used for obtaining a plurality of classification results for indicating whether a moon surface area corresponding to a single window is suitable to land after feature distribution correction is carried out on the plurality of classification feature matrixes. In this way, the complexity of assessing the suitability of the landing zone can be reduced.

Drawings

Fig. 1 is a schematic view of a scenario of an engineering-oriented single sliding window full-month landing zone selection method according to an embodiment of the present application.

Fig. 2 is a flowchart of a method for selecting an engineering-oriented single sliding window full-month landing zone according to an embodiment of the present application.

Fig. 3 is a schematic architecture diagram of an engineering-oriented single sliding window full-month landing zone selection method according to an embodiment of the present application.

Fig. 4 is a flowchart of sub-step S140 in the engineering-oriented single sliding window full-month landing zone selection method according to an embodiment of the present application.

Fig. 5 is a flowchart of sub-step S160 in the engineering-oriented single sliding window full-month landing zone selection method according to an embodiment of the present application.

Fig. 6 is a flowchart of a sub-step S170 in the engineering-oriented single sliding window full-month landing zone selection method according to an embodiment of the present application.

FIG. 7 is a block diagram of an engineering-oriented single sliding window full month landing zone selection system according to an embodiment of the present application.

Fig. 8 is a schematic illustration of a full moon surface grade according to an embodiment of the present application.

Fig. 9 is a schematic diagram of slope calculation according to an embodiment of the present application.

Fig. 10 is a schematic diagram of a sliding window algorithm according to an embodiment of the present application.

Fig. 11 is a schematic diagram of a spatial unit based on Queen proximity rules according to an embodiment of the present application.

Fig. 12 is a schematic diagram of a 1 ° by1 ° resolution single window visualization according to an embodiment of the present application.

Fig. 13 is a schematic diagram of a 0.125 ° x 0.125 ° resolution single window visualization according to an embodiment of the present application.

Fig. 14 is a schematic diagram of a 0.03125 deg. x 0.03125 deg. resolution single window visualization according to an embodiment of the present application.

Fig. 15 (a) - (f) show the results of quantization scoring of the full-month gradient threshold duty (20), threshold duty (8), coefficient of variation, average, molan index, and composite index, respectively.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application and not all of the embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

As described above, the target selection of the landing zone can be generalized to be a flat area and location preference in the scientific content set, however, this process lacks an efficient model for quantification of the comments. The automatic landing zone addressing scheme has any of the following two points to be propelled: the whole month grid segmentation lacks an overlapping part, so that the space at the grid boundary is omitted; quantitative models of evidence weights and fractal methods are difficult to calculate landing zone suitability indexes of any area at any position. Thus, an optimized engineering-oriented single sliding window full month landing zone selection scheme is desired.

Specifically, in the technical scheme of the application, the grid segmentation based on the sliding window is adopted, the sliding window overlapping and variable characteristics are adopted to solve the problem of space omission of the grid segmentation, and an evaluation index synthesized by a variation coefficient, a threshold value duty ratio and a Morlan index is established, so that the suitability degree of a single-window landing zone can be accurately detected, and the landing zone optimization method facing engineering constraint is realized.

Accordingly, the landing zone should provide a smooth and safe geographic area for soft landing of the probe and inspection of the probe, given the safety requirements of spacecraft and scientific instrumentation. The slope affects the fuel consumption during the descent of the probe and whether the impact on the lunar surface is slight. On the other hand, the gradient also affects the safety of the detector on the inspection route, and higher gradient fluctuation can harm the operation of the detector. Thus, grade is often the first factor to consider in the selection of a landing zone. Therefore, when evaluating whether the lunar surface area corresponding to the single window is suitable for landing or not, it is most important to analyze the full-moon surface slope map so as to select a suitable landing area. However, since the sliding window is used to traverse the full-moon surface gradient map, not only the correlation between the evaluation indexes of each single window data, but also the feature distribution information having the correlation between the single window data, the evaluation selection of the appropriate landing area cannot be performed based on the analysis of the evaluation index data of the single window. Therefore, in the process, the difficulty is how to fully and accurately mine the implicit association characteristic distribution information between the single window data and the evaluation indexes so as to accurately analyze the gradient of the full moon surface, and accurately select the safe area where the moon surface area is suitable for landing.

In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. In addition, deep learning and neural networks have also shown levels approaching and even exceeding humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like.

The development of deep learning and neural networks provides new solutions and schemes for mining implicit associated feature distribution information between each single window data and between each evaluation index.

Specifically, in the technical scheme of the application, firstly, a full-moon surface gradient map is obtained. It should be appreciated that the sliding window algorithm performs the required operations on the data in a specified direction for an array or data given a particular window size. When the method is applied to landing zone site selection, namely a single sliding window method is used, two-dimensional full-month gradient data are moved row by row and column by column, and analysis (remote sensing pictures of one gradient) is carried out, and finally different areas suitable for landing are obtained. Which enables full-month mesh segmentation without being limited to operations on full-month data, and thus, the complexity of evaluating the problem of the suitability of the landing zone can be reduced.

Based on the above, in the technical solution of the present application, a sliding window is further used to traverse the full-moon surface slope map to obtain a plurality of single window evaluation data, where each single window data includes a threshold duty ratio, a mean value, a variation coefficient, a molan index and a comprehensive index. In particular, here, the threshold duty ratio is defined as a ratio of the number of pixels to the total number of pixels that is smaller than a certain threshold, and the flatness of gradient data can be evaluated by introducing the threshold duty ratio; the degree of data dispersion can be quantified by introducing the variation coefficient; the moland index is used to evaluate the spatial distribution characteristics of the data, i.e., whether the data is spatially aggregated or discrete. Moreover, the fact that the pixel threshold value duty ratio cannot evaluate the discrete degree of the data is considered because each single index has respective disadvantages; the coefficient of variation does not take into account the spatial distribution characteristics of the data; the molan index can only evaluate the spatial aggregation degree of data, but for the case of data high-value and high-value aggregation (not conforming to flatter expectations), the molan index is higher than that of low-value and low-value aggregation (conforming to flatter expectations), and thus, the comprehensive index for a single window is further introduced.

Then, considering that the correlation relation exists between the evaluation data in each single window data, in order to extract the correlation characteristic information between the evaluation indexes in each single window data, so as to improve the accuracy of the appropriate landing evaluation of the lunar surface area, in the technical scheme of the application, the single window evaluation data are further processed through a deep neural network model comprising a plurality of full-connection layers to obtain a plurality of single window evaluation characteristic vectors. Here, the deep neural network model including a plurality of full connection layers can be used to extract high-dimensional implicit correlation characteristic information between the individual single window evaluation data.

Further, considering that the correlation relationship exists between the evaluation data in each single window data and the implicit correlation feature information exists between the single window data, in order to fully dig out the implicit correlation between the single window data and improve the detection precision of the landing area, in the technical scheme of the application, the single window evaluation feature vectors are further encoded in a context encoder based on a converter so as to extract the overall high-dimensional implicit context correlation feature information between the single window data and obtain the overall context correlation evaluation feature vector.

And then, after obtaining the global context associated evaluation feature vector with the implicit associated feature between the single window data and the implicit associated feature between the evaluation data in each single window data, calculating a transfer matrix of the global context associated evaluation feature vector by taking the single window evaluation feature vector as a query feature vector to obtain a plurality of transfer matrices, thereby obtaining implicit feature information about the lunar surface gradient of each single window based on the implicit associated feature distribution information between the single window data and the single window evaluation indexes, namely, high-dimensional hidden feature information about whether the lunar surface gradient is suitable for landing in the single window or not. And then, taking the transfer matrix as a classification characteristic matrix, and carrying out classification processing on the plurality of classification characteristic matrices in a classifier, thereby obtaining a classification result for indicating whether the lunar surface area corresponding to the single window is suitable for landing. Therefore, the gradient of the whole moon surface can be accurately analyzed, so that the safety area suitable for landing of the moon surface area is accurately selected, and the landing safety and stability of the detector are ensured.

In particular, in the technical solution of the present application, for the classification feature matrix that is the transfer matrix of the single window evaluation feature vector and the global context-associated evaluation feature vector, since the significance of the feature values of each position of the single window evaluation feature vector and the global context-associated evaluation feature vector in the transfer process is not differentiated, the expression effect of the classification feature matrix is affected.

Based on this, the applicant of the present application considered weighting the single-window evaluation feature vector and the global context-associated evaluation feature vector by the channel weights of the classification feature matrix, i.e., the global channel weights for expressing the local transfer features between the single-window evaluation feature vector and the global context-associated evaluation feature vector, respectively.

Specifically, firstly, the classification feature matrix is used for obtaining a weight feature map through an auxiliary convolution neural network serving as a feature extractor

And then +_for the weight feature map>

Modeling a feature correlation cumulative discrimination mechanism to obtain a weight feature vector +.>

Expressed as:

wherein ,

and />

Respectively, the weight characteristic diagram is subjected to single-layer convolution operation based on different convolution kernels, and +. >

Representing a global pooling operation on each feature matrix of the weight feature map.

Here, the feature correlation accumulating and distinguishing mechanism modeling firstly generates two new local association units of the weight feature map through convolution operation, then uses Sigmoid function and ReLU function to perform simple embedding, resetting and updating similar to a neural network architecture on the local association features, and then accumulates the correlation of the local features relative to the whole features through global average pooling operation, so that the feature importance sequence is explicitly modeled by using the feature distinguishing mechanism, and then the proper weighting factors in the channel dimension can be determined based on the feature accumulating and distinguishing mechanism of each feature matrix of the weight feature map. Then, the weight feature vector is used again

The single window evaluation feature vector and the global context associated evaluation feature vector are respectively subjected to point multiplication and weighting, so that the expression of the classification feature matrix can be improvedEffects. Therefore, the gradient of the whole moon surface can be accurately analyzed, so that the safety area suitable for landing of the moon surface area is accurately selected, and the landing safety and stability of the detector are ensured.

Fig. 1 is an application scenario diagram of an engineering-oriented single sliding window full-month landing zone selection method according to an embodiment of the present application. As shown in fig. 1, in this application scenario, firstly, a full-moon surface gradient map (for example, D illustrated in fig. 1) is acquired, and then, the full-moon surface gradient map is input into a server (for example, S illustrated in fig. 1) deployed with an engineering-oriented single-sliding-window full-month landing zone selection algorithm, where the server can process the full-moon surface gradient map by using the engineering-oriented single-sliding-window full-month landing zone selection algorithm to obtain a plurality of classification results for indicating whether a lunar surface area corresponding to a single window is suitable for landing.

Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings.

Fig. 2 is a flowchart of a method for selecting an engineering-oriented single sliding window full-month landing zone according to an embodiment of the present application. As shown in fig. 2, the method for selecting the engineering-oriented single sliding window full-month landing zone according to the embodiment of the application includes the following steps: s110, acquiring a full-moon surface gradient map; s120, traversing the full-moon surface gradient map by using a sliding window to obtain a plurality of single window evaluation data, wherein each single window evaluation data comprises a threshold value duty ratio, a mean value, a variation coefficient, a Morand index and a comprehensive index; s130, enabling each single window evaluation data to pass through a deep neural network model comprising a plurality of full connection layers to obtain a plurality of single window evaluation feature vectors; s140, the single window evaluation feature vectors pass through a context encoder based on a converter to obtain global context correlation evaluation feature vectors; s150, calculating a transfer matrix of the single-window evaluation feature vector and the global context associated evaluation feature vector to be used as a classification feature matrix by taking the single-window evaluation feature vector as a query feature vector so as to obtain a plurality of classification feature matrices; s160, carrying out feature distribution correction on the plurality of classification feature matrixes based on the plurality of single window evaluation feature vectors and the global context association evaluation feature vector to obtain a plurality of corrected classification feature matrixes; and S170, passing each corrected classification characteristic matrix through a classifier to obtain a plurality of classification results, wherein each classification result is used for indicating whether the lunar surface area corresponding to the single window is suitable for landing.

Fig. 3 is a schematic architecture diagram of an engineering-oriented single sliding window full-month landing zone selection method according to an embodiment of the present application. As shown in fig. 3, in the network architecture, first, a full-moon surface slope map is acquired; then traversing the full-moon surface gradient map by using a sliding window to obtain a plurality of single window evaluation data, wherein each single window evaluation data comprises a threshold value duty ratio, a mean value, a variation coefficient, a Morand index and a comprehensive index; then, the single window evaluation data are passed through a deep neural network model comprising a plurality of full connection layers to obtain a plurality of single window evaluation feature vectors; then, the single window evaluation feature vectors pass through a context encoder based on a converter to obtain global context correlation evaluation feature vectors; then, taking each single window evaluation feature vector as a query feature vector, and calculating a transfer matrix of the single window evaluation feature vector and the global context association evaluation feature vector as a classification feature matrix to obtain a plurality of classification feature matrices; then, based on the single window evaluation feature vectors and the global context correlation evaluation feature vectors, carrying out feature distribution correction on the classification feature matrices to obtain corrected classification feature matrices; and finally, passing each corrected classification characteristic matrix through a classifier to obtain a plurality of classification results, wherein each classification result is used for indicating whether the lunar surface area corresponding to the single window is suitable for landing.

More specifically, in step S110, a full-moon surface gradient map is acquired. The slope affects the fuel consumption during descent of the probe, whether the impact on the lunar surface is slight or not, and the safety of the probe on the tour, higher slope fluctuations jeopardizing the operation of the probe. Therefore, when evaluating whether the lunar surface area corresponding to the single window is suitable for landing or not, it is most important to analyze the full-moon surface slope map so as to select a suitable landing area. However, since the sliding window is used to traverse the full-moon surface gradient map, not only the correlation between the evaluation indexes of each single window data, but also the feature distribution information having the correlation between the single window data, the evaluation selection of the appropriate landing area cannot be performed based on the analysis of the evaluation index data of the single window.

Further, the landing zone should provide a smooth and safe geographic area for soft landing of the probe and inspection of the probe, given the safety requirements of spacecraft and scientific instrumentation. The slope affects the fuel consumption during the descent of the probe and whether the impact on the lunar surface is slight. On the other hand, the gradient also affects the safety of the detector on the inspection route, and higher gradient fluctuation can harm the operation of the detector. Thus, grade is often the first factor to consider in the selection of a landing zone. Fig. 8 is a schematic view of the slope of the whole moon surface, and fig. 9 is a schematic view of the slope calculation. As shown in fig. 9, the slope of the terrain represents the steepness of the terrain of the earth, and in DEM raster data, a certain pixel is usually combined with 8 adjacent pixels to form a slope, and the slope is calculated by dividing the east-west height difference by the north-south height difference. The equation for calculating the slope of pixel "e" is as follows:

， (1)

， (2)

， (3)

In equations (1) and (2), a-i represents the elevation of each pixel, L represents the size of the pixel, dz/dx represents the slope in the east-west direction (x), and dz/dy represents the slope in the north-south direction (y). The slope of e is the arctangent of the square root of the sum of the squares of the two slopes. S represents slope in degrees.

It should be appreciated that the sliding window algorithm performs the required operations on the data in a specified direction for an array or data given a particular window size. When the method is applied to landing zone site selection, namely a single sliding window method is used, two-dimensional full-month gradient data are moved row by row and column by column, and analysis (remote sensing pictures of one gradient) is carried out, and finally different areas suitable for landing are obtained. Which enables full-month mesh segmentation without being limited to operations on full-month data, and thus, the complexity of evaluating the problem of the suitability of the landing zone can be reduced.

Fig. 10 shows a schematic diagram of a sliding window algorithm, which includes four parameters, i.e., a step size and a window width and height, and the parameters can be freely defined. A sliding window algorithm is defined as performing the required operations on an array or data given a particular window size, traversing the loop for the data in a specified direction. When the method is applied to landing zone site selection, namely a single sliding window method is used, two-dimensional full-month gradient data are moved row by row and column by column, and analysis (remote sensing pictures of one gradient) is carried out, and finally different areas suitable for landing are obtained. The full-month mesh cut can be realized without being limited to the operation on the full-month data, and thus, the complexity of evaluating the problem of the suitability of the landing zone can be reduced.

More specifically, in step S120, the full-moon surface slope map is traversed using a sliding window to obtain a plurality of single-window evaluation data, wherein each single-window evaluation data includes a threshold duty ratio, a mean value, a coefficient of variation, a molan index, and a comprehensive index.

In particular, here, the threshold duty ratio is defined as a ratio of the number of pixels to the total number of pixels that is smaller than a certain threshold, and the flatness of gradient data can be evaluated by introducing the threshold duty ratio; the degree of data dispersion can be quantified by introducing the variation coefficient; the moland index is used to evaluate the spatial distribution characteristics of the data, i.e., whether the data is spatially aggregated or discrete. Moreover, the fact that the pixel threshold value duty ratio cannot evaluate the discrete degree of the data is considered because each single index has respective disadvantages; the coefficient of variation does not take into account the spatial distribution characteristics of the data; the molan index can only evaluate the spatial aggregation degree of data, but for the case of data high-value and high-value aggregation (not conforming to flatter expectations), the molan index is higher than that of low-value and low-value aggregation (conforming to flatter expectations), and thus, the comprehensive index for a single window is further introduced.

Further, regarding the threshold duty ratio, in order to evaluate the flatness of gradient data, the threshold duty ratio is introduced, which is defined as the ratio of the number of pixels smaller than a certain threshold to the total number of pixels.

wherein ,Srepresenting the threshold duty cycle of the signal,

representing that the window is less than +.>

The number of pixels (threshold value),Prepresenting the total number of pixels in the window. In the selection of landing points for the Chinese lunar landing task, the average terrain gradient of the landing zone should not exceed 8 degrees, and the area with the gradient smaller than 8 degrees should occupy a relatively large proportion, so that the threshold value is set to be 8 in the subsequent calculation.

Further, regarding the binarized variation coefficient, in order to quantify the degree of data dispersion, a variation coefficient is introduced

Standard deviation>

Mean>

. The corresponding formula is as follows:

because the excessively high average gradient does not meet the safe landing condition of the landing zone, the average gradient can be higher, so that the variation coefficient is lower, namely, the index for quantifying the fluctuation of the data is lower. Therefore, the original data of the single window is binarized according to the threshold value, so that the data smaller than the threshold value (flatter) is set to 1, and the other data are set to 0.

Further, regarding the Moran index, in order to evaluate the spatial distribution characteristics of data, i.e. whether the data is spatially aggregated or scattered, moran' I index is used as a measure of spatial autocorrelation, moran index values are distributed in [ -1,1], [0,1] indicating that positive correlation exists between geographic entities, and [ -1,0] indicating that negative correlation exists, while 0 value has no correlation, the expression is as follows:

wherein ,

is an element->

Property of (2) and its average value->

Deviation of->

Is an element->

and />

Spatial weight between->

Equal to the total number of elements->

Is the aggregation of all spatial weights: />

For space matrix

A spatial weight matrix based on the Queen adjacency is introduced, as shown in FIG. 11, the Queen adjacency unit of the A unit is the B unit, and the B unit comprises common vertex connection and common adjacent edge connection.

For assumption

To represent space unit +.>

Then the definition of the spatial weight matrix based on the Queen proximity is as follows:

further, regarding the single window evaluation comprehensive index, since each single index has respective disadvantages, the pixel threshold ratio cannot evaluate the discrete degree of the data; the coefficient of variation does not take into account the spatial distribution characteristics of the data; the Morgan index can only evaluate the spatial aggregation degree of data, but for the case of data with high value and high value aggregation (not conforming to flatter expectations), the Morgan index is higher than that of data with low value and low value aggregation (conforming to flatter expectations); thus, the introduction of the composite index for a single window is defined as:

wherein ,

as a comprehensive index, ->

For a threshold duty cycle, a pixel duty cycle conforming to a flat definition is measured; / >

The fluctuation of the data is quantized for the variation coefficient, and the larger the fluctuation is, the larger the variation coefficient is; />

For the Morgan index, the spatial distribution characteristics of the data are quantized, and the more data with similar values are gathered, the larger the Morgan index is. Thus, the comprehensive index effectively reflects the suitability of the landing zone of a single window.

After traversing a sliding window and evaluating a single window based on multiple indexes such as a threshold duty ratio, an average value, a variation coefficient, a Morgan index, a comprehensive index and the like, the full-month gradient obtained by utilizing DEM data processing can present an automatic landing zone optimization result in the following 3 modes: 1. quantification results of the landing suitability degree of the full-month grid and a corresponding database; 2. a local multi-resolution single window evaluation result; 3. and (5) quantifying and evaluating the regional gradient.

The single window visual part comprises a locally refined window evaluation and window evaluations with different resolutions, the single window visual result comprises a plurality of indexes such as a threshold value duty ratio, an average value, a variation coefficient, a Morgan index and the like, the window sizes are respectively 1 degree by 1 degree, 0.125 degrees by 0.125 degrees, 0.03125 degrees by 0.03125 degrees, and converted kilometers are respectively 30.3km, 3.79 3.79 and 0.95.95. Fig. 12, 13 and 14 show schematic diagrams of a 1 ° by 1 ° resolution single window visualization, a 0.125 ° by 0.125 ° resolution single window visualization and a 0.03125 ° by 0.03125 ° resolution single window visualization, respectively.

Fig.12, 13 and 14 show that the gradient data of a single window can be intuitively displayed according to the descending order of the comprehensive indexes under different window sizes. Along with the reduction of the comprehensive index, the gradient data of the single window gradually shows complex characteristics. As can be seen from the single window visualization results of fig.12, 13 and 14, the more suitable landing zone presents a high data ratio of low values on the data, has smaller overall volatility of the data, and presents the characteristic of high approximate aggregation degree, and corresponds to the characteristics of three indexes such as threshold ratio, variation coefficient, and molan index. To achieve the goal of quantifying the suitability of the landing zone gradient, the evaluation index corresponding to each window of fig.12, 13 and 14 was counted.

/>

/>

According to the above and fig. 12-14, the relationship between the suitability of the landing grade and the evaluation index is represented: 1. overall, the average slope is substantially within 3 ° for a combination index > 0.6. Therefore, a lower average grade is often a precondition for a more suitable landing, but the two do not constitute a strict proportional relationship. For example, FIG.12 (1, one row by one row), FIG.13 (4, one row by four rows; 11, two rows by five rows) FIG.14 (1, one row by one row; 11, two rows by five rows; 12, two rows by six rows) has an average slope higher than that of the latter window and a composite index higher than that of the latter window. This is due to their data being numerically stable and the data being more spatially concentrated, having a higher coefficient of variation with the morlan index, and secondly, on the threshold duty index, the higher threshold setting also weakens the contribution of the threshold duty to the composite index between the average slope windows within the threshold, since their threshold duty is often 1, which is one of the reasons that the average slope is not proportional to the composite index. Therefore, the comprehensive index is suitable for representing the suitability degree of the landing zone compared with the average gradient, and also accords with the principle that the gradient of the landing zone is relatively flat. 2. Under different window sizes, the comprehensive indexes corresponding to the occurrence of the impact pits are different. Excluding small impact pits, fig.12 (20, four rows and two columns), fig.13 (24, four rows and six columns), fig.14 (23, four rows and five columns), obvious impact pits appear, and the corresponding comprehensive index is 0.68,0.49,0.34. It can be obtained that the occurrence of the impact pit can obtain a lower comprehensive index score to a certain extent, and meets the subjective evaluation standard of the suitability of the landing zone. And the lower window size has lower comprehensive index when the collision pit appears, and is more suitable for being used as the window size selection of landing zone site selection.

In the whole month evaluation part, in the landing zone site selection process considering gradient in whole month, the single window size is set to be 1 degree multiplied by 1 degree (actual range: 30.3cos phi km multiplied by 30.3km, phi is displacement) and the two-direction movement step length is set to be 0.5 degrees, so that 359 multiplied by 719 windows with overlapping and a corresponding window evaluation index database are obtained. The quantitative scores for each window were summarized for the whole month to obtain fig. 15 (a) - (f), showing the whole month gradient evaluation results based on different indexes.

Fig. 15 (a) - (f) show the results of quantization scoring of the full-month gradient threshold duty (20), threshold duty (8), coefficient of variation, average, molan index, and composite index, respectively. The composite scoring results of FIG. 15 (f) superimpose previously successful soft landing sites (pentagram) and expert pre-selected landing sites (snowflakes). Considering that the lower the average value representing the overall flatness of the grade and the coefficient of variation representing the volatility of the grade data, the more suitable the landing, the colorbar of the results of fig. 15 (c) and 15 (d) is flipped to ensure the consistency of the results presentation.

More specifically, in step S130, the respective single-window evaluation data are passed through a deep neural network model including a plurality of full connection layers to obtain a plurality of single-window evaluation feature vectors. In order to extract the relevance characteristic information between the evaluation indexes in each single window data, so as to improve the accuracy of the appropriate landing evaluation of the moon surface area, in the technical scheme of the application, the single window evaluation data are further processed through a deep neural network model comprising a plurality of full-connection layers to obtain a plurality of single window evaluation characteristic vectors. Here, the deep neural network model including a plurality of full connection layers can be used to extract high-dimensional implicit correlation characteristic information between the individual single window evaluation data.

It should be appreciated that convolutional neural network (Convolutional Neural Network, CNN) is an artificial neural network and has wide application in the fields of image recognition and the like. The convolutional neural network may include an input layer, a hidden layer, and an output layer, where the hidden layer may include a convolutional layer, a pooling layer, an activation layer, a full connection layer, etc., where the previous layer performs a corresponding operation according to input data, outputs an operation result to the next layer, and obtains a final result after the input initial data is subjected to a multi-layer operation. The convolutional neural network model has excellent performance in the aspect of image local feature extraction by taking a convolutional kernel as a feature filtering factor, and has stronger feature extraction generalization capability and fitting capability compared with the traditional image feature extraction algorithm based on statistics or feature engineering.

Accordingly, in one specific example, the step of passing the single window evaluation data through a deep neural network model including a plurality of fully connected layers to obtain a plurality of single window evaluation feature vectors includes: arranging the single window evaluation data into an evaluation data input vector; performing full-connection coding on the evaluation data input vector by using a plurality of full-connection layers of the deep neural network model according to the following formula to extract high-dimensional implicit features of feature values of all positions in the evaluation data input vector, wherein the formula is as follows:

, wherein />

Is the evaluation data input vector, +.>

Is the output vector, +.>

Is a weight matrix, < >>

Is a bias vector, ++>

Representing a matrix multiplication.

More specifically, in step S140, the plurality of single window evaluation feature vectors are passed through a converter-based context encoder to obtain a global context-associated evaluation feature vector. Because each evaluation data in each single window data has an association relationship, and each single window data has implicit association characteristic information, in order to fully dig out the implicit association between each single window data and improve the detection precision of a landing area, in the technical scheme of the application, the plurality of single window evaluation characteristic vectors are further encoded in a context encoder based on a converter so as to extract the overall high-dimensional implicit context association characteristic information based on the single window data.

It should be appreciated that by the context encoder, the relationship between a certain word segment and other word segments in the vector representation sequence may be analyzed to obtain corresponding feature information. The context encoder aims to mine for hidden patterns between contexts in the word sequence, optionally the encoder comprises: CNN (Convolutional Neural Network ), reconvenn (RecursiveNeural Network, recurrent neural network), language Model (Language Model), and the like. The CNN-based method has a better extraction effect on local features, but has a poor effect on Long-Term Dependency (Long-Term Dependency) problems in sentences, so Bi-LSTM (Long Short-Term Memory) based encoders are widely used. The repetitive NN processes sentences as a tree structure rather than a sequence, has stronger representation capability in theory, but has the weaknesses of high sample marking difficulty, deep gradient disappearance, difficulty in parallel calculation and the like, so that the repetitive NN is less in practical application. The transducer has a network structure with wide application, has the characteristics of CNN and RNN, has a better extraction effect on global characteristics, and has a certain advantage in parallel calculation compared with RNN (RecurrentNeural Network ).

Accordingly, in one specific example, as shown in fig. 4, passing the plurality of single window evaluation feature vectors through a converter-based context encoder to obtain a global context-associated evaluation feature vector includes: s141, inputting the plurality of single-window evaluation feature vectors into the context encoder based on the converter to obtain a plurality of single-window evaluation semantic feature vectors; and S142, cascading the plurality of single window evaluation semantic feature vectors to obtain the global context association evaluation feature vector.

More specifically, in step S150, with the respective single-window evaluation feature vectors as query feature vectors, a transition matrix thereof with the global context-associated evaluation feature vector is calculated as a classification feature matrix to obtain a plurality of classification feature matrices. And obtaining implicit characteristic information about the lunar surface gradient of each single window based on the implicit associated characteristic distribution information between the single window data and the single window evaluation indexes, namely high-dimensional hidden characteristic information about whether the lunar surface gradient is suitable for landing in the single window.

Accordingly, in a specific example, with the single window evaluation feature vector as a query feature vector, calculating a transition matrix of the single window evaluation feature vector and the global context association evaluation feature vector as a classification feature matrix to obtain a plurality of classification feature matrices, including: calculating transfer matrixes of the single-window evaluation feature vectors and the global context association evaluation feature vectors as classification feature matrixes according to the following formula to obtain a plurality of classification feature matrixes; wherein, the formula is:

wherein ,

representing said individual single window evaluation feature vectors, < >>

Representing the global context associated evaluation feature vector,/->

Representing the plurality of classification feature matrices, +.>

Representing vector multiplication.

More specifically, in step S160, feature distribution correction is performed on the plurality of classification feature matrices based on the plurality of single window evaluation feature vectors and the global context-associated evaluation feature vector to obtain a plurality of corrected classification feature matrices.

Accordingly, in one specific example, as shown in fig. 5, performing feature distribution correction on the plurality of classification feature matrices to obtain a plurality of corrected classification feature matrices based on the plurality of single window evaluation feature vectors and the global context association evaluation feature vector, including: s161, the classification feature matrix is passed through an auxiliary convolution neural network serving as a feature extractor to obtain a weight feature map; s162, modeling a feature correlation accumulation distinguishing mechanism in the weight feature map to obtain a weight feature vector; s163, multiplying the weight feature vector with each single window evaluation feature vector and the global context associated evaluation feature vector according to position points to obtain a plurality of optimized single window evaluation feature vectors and optimized global context associated evaluation feature vectors; and S164, calculating a transfer matrix of the optimized single window evaluation feature vector and the optimized global context association evaluation feature vector by taking the optimized single window evaluation feature vectors as query feature vectors so as to obtain the corrected classification feature matrix.

Accordingly, in one specific example, passing the classification feature matrix through an auxiliary convolutional neural network as a feature extractor to obtain a weighted feature map includes: each layer of the auxiliary convolutional neural network serving as the feature extractor is respectively carried out in forward transfer of the layer: carrying out convolution processing on input data to obtain a convolution characteristic diagram; carrying out mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the auxiliary convolution neural network serving as the feature extractor is the weight feature map, and the input of the first layer of the auxiliary convolution neural network serving as the feature extractor is the classification feature matrix.

Accordingly, in a specific example, modeling the feature correlation cumulative distinguishing mechanism in the weight feature map to obtain a weight feature vector includes: modeling a feature correlation accumulation distinguishing mechanism in the weight feature map by using the following formula to obtain the weight feature vector; wherein, the formula is:

wherein ,

representing the weight feature map, < >>

and />

representation->

Activating function->

Representation->

Activate function, and->

Representing addition by position +.>

Representing the weighted feature vector.

Respectively carrying out dot multiplication with the single window evaluation feature vector and the global context association evaluation feature vectorAnd weighting is carried out, so that the expression effect of the classification characteristic matrix can be improved. Therefore, the gradient of the whole moon surface can be accurately analyzed, so that the safety area suitable for landing of the moon surface area is accurately selected, and the landing safety and stability of the detector are ensured.

More specifically, in step S170, the respective corrected classification feature matrices are passed through a classifier to obtain a plurality of classification results, where each classification result is used to indicate whether the lunar surface area corresponding to the single window is suitable for landing. Therefore, the gradient of the whole moon surface can be accurately analyzed, so that the safety area suitable for landing of the moon surface area is accurately selected, and the landing safety and stability of the detector are ensured.

It should be appreciated that the role of the classifier is to learn the classification rules and classifier using a given class, known training data, and then classify (or predict) the unknown data. Logistic regression (logistics), SVM, etc. are commonly used to solve the classification problem, and for multi-classification problems (multi-class classification), logistic regression or SVM can be used as well, but multiple bi-classifications are required to compose multiple classifications, but this is error-prone and inefficient, and the commonly used multi-classification method is the Softmax classification function.

Accordingly, in one specific example, as shown in fig. 6, the passing the corrected classification feature matrices through a classifier to obtain a plurality of classification results, where each classification result is used to indicate whether the lunar surface area corresponding to the single window is suitable for landing, includes: s171, expanding each corrected classification characteristic matrix into a plurality of classification characteristic vectors according to row vectors or column vectors; s172, performing full-connection coding on the plurality of classification feature vectors by using a plurality of full-connection layers of the classifier to obtain a plurality of coding classification feature vectors; and S171, passing the plurality of coding classification feature vectors through a Softmax classification function of the classifier to obtain a plurality of classification results.

In summary, according to the engineering-oriented single sliding window full-month landing zone selection method of the embodiment of the application, firstly, a sliding window is used for traversing a full-month surface gradient map to obtain a plurality of single window evaluation data, then, each single window evaluation data is processed through a deep neural network model to obtain a plurality of single window evaluation feature vectors, then, the plurality of single window evaluation feature vectors are processed through a context encoder to obtain global context association evaluation feature vectors, then, transfer matrixes of each single window evaluation feature vector and the global context association evaluation feature vectors are calculated to obtain a plurality of classification feature matrixes, and finally, a classifier is used for obtaining a plurality of classification results for indicating whether a moon surface area corresponding to a single window is suitable to land after feature distribution correction is carried out on the plurality of classification feature matrixes. In this way, the complexity of assessing the suitability of the landing zone can be reduced.

Fig. 7 is a block diagram of an engineering-oriented single sliding window full month landing zone selection system 100 according to an embodiment of the present application. As shown in fig. 7, the engineering-oriented single sliding window full month landing zone selection system 100 according to the embodiment of the present application includes: a data acquisition module 110, configured to acquire a full moon surface slope map; the traversing module 120 is configured to traverse the full-moon surface slope map using a sliding window to obtain a plurality of single-window evaluation data, where each single-window evaluation data includes a threshold duty ratio, a mean value, a variation coefficient, a molan index and a comprehensive index; the full-connection encoding module 130 is configured to pass the single-window evaluation data through a deep neural network model including a plurality of full-connection layers to obtain a plurality of single-window evaluation feature vectors; a context encoding module 140, configured to pass the plurality of single window evaluation feature vectors through a context encoder based on a converter to obtain a global context-associated evaluation feature vector; the classification feature matrix calculation module 150 is configured to calculate a transition matrix of the single-window evaluation feature vector and the global context-associated evaluation feature vector as a classification feature matrix by using the single-window evaluation feature vector as a query feature vector, so as to obtain a plurality of classification feature matrices; a feature distribution correction module 160, configured to perform feature distribution correction on the plurality of classification feature matrices based on the plurality of single window evaluation feature vectors and the global context association evaluation feature vector to obtain a plurality of corrected classification feature matrices; and a classification module 170, configured to pass the respective corrected classification feature matrices through a classifier to obtain a plurality of classification results, where each classification result is used to indicate whether the lunar surface area corresponding to the single window is suitable for landing.

In one example, in the engineering-oriented single sliding window full month landing zone selection system 100, the full connection encoding module 130 is configured to: arranging the single window evaluation data into an evaluation data input vector; performing full-connection coding on the evaluation data input vector by using a plurality of full-connection layers of the deep neural network model according to the following formula to extract high-dimensional implicit features of feature values of all positions in the evaluation data input vector, wherein the formula is as follows:

, wherein />

Is the evaluation data input vector, +.>

Is the output vector, +.>

Is a weight matrix, < >>

Is a bias vector, ++>

Representing a matrix multiplication.

In one example, in the engineering-oriented single sliding window full month landing zone selection system 100 described above, the context encoding module 140 is configured to: inputting the plurality of single window evaluation feature vectors into the converter-based context encoder to obtain a plurality of single window evaluation semantic feature vectors; and cascading the plurality of single window evaluation semantic feature vectors to obtain the global context associated evaluation feature vector.

In one example, in the engineering-oriented single sliding window full month landing zone selection system 100, the classification feature matrix calculation module 150 is configured to: calculating transfer matrixes of the single-window evaluation feature vectors and the global context association evaluation feature vectors as classification feature matrixes according to the following formula to obtain a plurality of classification feature matrixes; wherein, the formula is:

wherein ,

representing said individual single window evaluation feature vectors, < >>

Representing the global context associated evaluation feature vector,/->

Representing the plurality of classification feature matrices, +.>

Representing vector multiplication.

In one example, in the engineering-oriented single sliding window full month landing zone selection system 100 described above, the feature distribution correction module 160 is configured to: the classification feature matrix is passed through an auxiliary convolution neural network serving as a feature extractor to obtain a weight feature map; modeling a feature correlation accumulation distinguishing mechanism in the weight feature map to obtain a weight feature vector; multiplying the weight feature vector with each single window evaluation feature vector and the global context association evaluation feature vector according to position points to obtain a plurality of optimized single window evaluation feature vectors and optimized global context association evaluation feature vectors; and calculating a transfer matrix of the optimized single window evaluation feature vector and the optimized global context association evaluation feature vector by taking the optimized single window evaluation feature vector as a query feature vector so as to obtain the corrected classification feature matrix.

In one example, in the engineering-oriented single sliding window full month landing zone selection system 100, the step of passing the classification feature matrix through an auxiliary convolutional neural network as a feature extractor to obtain a weighted feature map includes: each layer of the auxiliary convolutional neural network serving as the feature extractor is respectively carried out in forward transfer of the layer: carrying out convolution processing on input data to obtain a convolution characteristic diagram; carrying out mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the auxiliary convolution neural network serving as the feature extractor is the weight feature map, and the input of the first layer of the auxiliary convolution neural network serving as the feature extractor is the classification feature matrix.

In one example, in the engineering-oriented single sliding window full month landing zone selection system 100, modeling a feature correlation accumulation and differentiation mechanism in the weight feature map to obtain a weight feature vector includes: modeling a feature correlation accumulation distinguishing mechanism in the weight feature map by using the following formula to obtain the weight feature vector; wherein, the formula is:

wherein ,

representing the weight feature map, < >>

and />

representation->

Activating function->

Representation->

Activate function, and->

Representing addition by position +.>

Representing the weighted feature vector.

In one example, in the engineering-oriented single sliding window full month landing zone selection system 100 described above, the classification module 170 is configured to: expanding each corrected classified feature matrix into a plurality of classified feature vectors according to row vectors or column vectors; performing full-connection coding on the plurality of classification feature vectors by using a plurality of full-connection layers of the classifier to obtain a plurality of coding classification feature vectors; and passing the plurality of encoded classification feature vectors through a Softmax classification function of the classifier to obtain the plurality of classification results.

Here, it will be appreciated by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described engineering-oriented single sliding window full month landing zone selection system 100 have been described in detail in the above description of the engineering-oriented single sliding window full month landing zone selection method with reference to fig. 1 to 6, and thus, repetitive descriptions thereof will be omitted.

As described above, the engineering-oriented single sliding window full month landing zone selection system 100 according to the embodiments of the present application may be implemented in various wireless terminals, such as a server or the like having an engineering-oriented single sliding window full month landing zone selection algorithm. In one example, the engineering-oriented single sliding window full month landing zone selection system 100 according to embodiments of the present application may be integrated into a wireless terminal as one software module and/or hardware module. For example, the engineering-oriented single sliding window full month landing zone selection system 100 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the engineering-oriented single sliding window full month landing zone selection system 100 can also be one of a number of hardware modules of the wireless terminal.

Alternatively, in another example, the engineering-oriented single sliding window full month landing zone selection system 100 and the wireless terminal may be separate devices, and the engineering-oriented single sliding window full month landing zone selection system 100 may be connected to the wireless terminal via a wired and/or wireless network and communicate interactive information in accordance with an agreed data format.

According to another aspect of the present application, there is also provided a non-volatile computer-readable storage medium having stored thereon computer-readable instructions which, when executed by a computer, can perform a method as described above.

Program portions of the technology may be considered to be "products" or "articles of manufacture" in the form of executable code and/or associated data, embodied or carried out by a computer readable medium. A tangible, persistent storage medium may include any memory or storage used by a computer, processor, or similar device or related module. Such as various semiconductor memories, tape drives, disk drives, or the like, capable of providing storage functionality for software.

All or a portion of the software may sometimes communicate over a network, such as the internet or other communication network. Such communication may load software from one computer device or processor to another. For example: a hardware platform loaded from a server or host computer of the video object detection device to a computer environment, or other computer environment implementing the system, or similar functioning system related to providing information needed for object detection. Thus, another medium capable of carrying software elements may also be used as a physical connection between local devices, such as optical, electrical, electromagnetic, etc., propagating through cable, optical cable, air, etc. Physical media used for carrier waves, such as electrical, wireless, or optical, may also be considered to be software-bearing media. Unless limited to a tangible "storage" medium, other terms used herein to refer to a computer or machine "readable medium" mean any medium that participates in the execution of any instructions by a processor.

Claims

1. The engineering-oriented single sliding window full-month landing zone selection method is characterized by comprising the following steps of:

acquiring a full-moon surface gradient map;

2. The engineering-oriented single sliding window full month landing zone selection method according to claim 1, wherein the step of obtaining a plurality of single window evaluation feature vectors by passing each single window evaluation data through a deep neural network model including a plurality of full connection layers comprises:

, wherein />

Is the evaluation data input vector, +.>

Is the output vector, +.>

Is a weight matrix, < >>

Is a bias vector, ++>

Representing a matrix multiplication.

3. The engineering-oriented single sliding window full month landing zone selection method of claim 2, wherein passing the plurality of single window evaluation feature vectors through a converter-based context encoder to obtain a global context-associated evaluation feature vector comprises:

4. The engineering-oriented single sliding window full month landing zone selection method according to claim 3, wherein calculating a transfer matrix of each single window evaluation feature vector as a classification feature matrix with the global context-associated evaluation feature vector to obtain a plurality of classification feature matrices comprises:

wherein, the formula is:

wherein ,

representing said individual single window evaluation feature vectors, < >>

Representing the global context associated evaluation feature vector,/->

Representing the plurality of classification feature matrices, +.>

Representing vector multiplication. />

5. The engineering-oriented single sliding window full moon landing zone selection method according to claim 4, wherein performing feature distribution correction on the plurality of classification feature matrices to obtain a plurality of corrected classification feature matrices based on the plurality of single window evaluation feature vectors and the global context-associated evaluation feature vector, comprises:

6. The engineering-oriented single sliding window full month landing zone selection method according to claim 5, wherein the step of obtaining a weight feature map by passing the classification feature matrix through an auxiliary convolutional neural network serving as a feature extractor comprises the steps of:

7. The engineering-oriented single sliding window full month landing zone selection method of claim 6, wherein modeling a feature correlation accumulation and differentiation mechanism in the weight feature map to obtain a weight feature vector comprises:

wherein, the formula is:

wherein ,

representing the weight feature map, < >>

and />

representation->

Activating function->

Representation->

Activate function, and->

The representation is added by location, />

Representing the weighted feature vector.

8. The method for selecting the engineering-oriented single sliding window full-month landing zone according to claim 7, wherein the step of passing each corrected classification feature matrix through a classifier to obtain a plurality of classification results, each classification result being used for indicating whether the corresponding lunar surface area of the single window is suitable for landing or not, comprises the steps of:

9. An engineering-oriented single sliding window full-month landing zone selection system is characterized by comprising:

10. The engineering-oriented single sliding window full moon landing zone selection system of claim 9, wherein the fully connected encoding module is configured to:

, wherein />

Is the evaluation data input vector, +.>

Is the output vector, +.>

Is a weight matrix, < >>

Is a bias vector, ++>

Representing a matrix multiplication. />