CN112819794A - Small celestial body meteorite crater detection method based on deep learning - Google Patents
Small celestial body meteorite crater detection method based on deep learning Download PDFInfo
- Publication number
- CN112819794A CN112819794A CN202110156137.6A CN202110156137A CN112819794A CN 112819794 A CN112819794 A CN 112819794A CN 202110156137 A CN202110156137 A CN 202110156137A CN 112819794 A CN112819794 A CN 112819794A
- Authority
- CN
- China
- Prior art keywords
- image
- celestial body
- crater
- small celestial
- small
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 81
- 238000013135 deep learning Methods 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000000034 method Methods 0.000 claims abstract description 24
- 238000001914 filtration Methods 0.000 claims abstract description 4
- 230000000694 effects Effects 0.000 claims description 12
- 230000011218 segmentation Effects 0.000 claims description 7
- 230000000717 retained effect Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 claims description 4
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 238000003709 image segmentation Methods 0.000 claims description 4
- 238000005728 strengthening Methods 0.000 claims description 4
- 230000016776 visual perception Effects 0.000 claims description 4
- 101710200331 Cytochrome b-245 chaperone 1 Proteins 0.000 claims description 3
- 102100037186 Cytochrome b-245 chaperone 1 Human genes 0.000 claims description 3
- 101710119396 Cytochrome b-245 chaperone 1 homolog Proteins 0.000 claims description 3
- 241001505295 Eros Species 0.000 claims description 3
- ULGZDMOVFRHVEP-RWJQBGPGSA-N Erythromycin Chemical compound O([C@@H]1[C@@H](C)C(=O)O[C@@H]([C@@]([C@H](O)[C@@H](C)C(=O)[C@H](C)C[C@@](C)(O)[C@H](O[C@H]2[C@@H]([C@H](C[C@@H](C)O2)N(C)C)O)[C@H]1C)(C)O)CC)[C@H]1C[C@@](C)(OC)[C@@H](O)[C@H](C)O1 ULGZDMOVFRHVEP-RWJQBGPGSA-N 0.000 claims description 3
- 241000341910 Vesta Species 0.000 claims description 3
- 238000003384 imaging method Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 230000002776 aggregation Effects 0.000 claims description 2
- 238000004220 aggregation Methods 0.000 claims description 2
- 238000011176 pooling Methods 0.000 claims description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 abstract 1
- 230000005764 inhibitory process Effects 0.000 abstract 1
- 230000009466 transformation Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 4
- 238000005286 illumination Methods 0.000 description 4
- 230000001788 irregular Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000009987 spinning Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a small celestial body meteorite crater detection method based on deep learning, and belongs to the field of deep space exploration. The implementation method of the invention comprises the following steps: firstly, in order to effectively enhance and retain meteor crater characteristics, a local variance equalization algorithm is adopted to optimize a data set, the data set is enhanced by affine transformation, mean filtering and other methods, and a meteor crater detection model in a small celestial body environment is obtained through deep learning network training; secondly, for the problem of missing detection of small meteorite craters in the high-resolution image, adaptively cutting the predicted image into a plurality of sub-images with overlapping regions and respectively sending the sub-images into a detection network; and finally, removing redundant frames by using a non-maximum value inhibition method, and merging prediction results. The method overcomes the problems of low speed, low recognition rate and the like of the traditional target detection method, is superior to the current mainstream target detection network, and can better complete the task of detecting the small celestial body meteorite crater.
Description
Technical Field
The invention belongs to the field of deep space exploration, is particularly suitable for meteor crater detection in a small celestial body exploration landing process, and particularly relates to a small celestial body meteor crater detection method based on deep learning.
Background
The small celestial body detection is helpful for people to know the formation and evolution of the solar system and explore the origin of the earth life and water, and is one of the key research problems of the deep space detection tasks of the main aerospace countries at present and in the future. The meteorite crater is used as the most abundant topographic feature of the surface of the small celestial body, has the characteristics of regular geometric shape, small influence of illumination and the like, is often used as an ideal navigation landmark, and is also one of main objects for obstacle detection and avoidance in the landing process of the small celestial body.
Aiming at the requirement of a deep space exploration task, researchers at home and abroad carry out a plurality of related research works on the detection of the meteor crater. According to different meteorite crater detection principles, the current detection algorithm can be divided into a detection method based on meteorite crater morphological characteristics and a meteorite crater detection method based on deep learning. In the prior art [1] (Cheng Y, Johnson A E, Matthies L H, et al, optical land detection for space in the scientific [ J ]. Ad-lance in the analytical Sciences,2003,114(3):1785 and 1803.), Cheng et al combine the illumination direction and meteor crater edge information, select the elliptic arc by locating point detection and edge grouping, finally realize the detection of meteor crater by elliptic fitting; in the prior art [2] (Zuo W, Li C L, Yu L J, et al. Shadow-high height feature creating automatic small cratter registration using high-resolution digital orthogonal map from Chang' E Missions [ J ]. Acta geochemica, 2019,38(04): 541) 554.), Zuo et al, based on IGF (the Image Gray frequency) algorithm, segment and extract highlight and shadow information areas of meteor craters under illumination conditions, thereby realizing automatic recognition of meteor craters. The above algorithms all belong to meteorite crater detection algorithms based on meteorite crater morphological characteristics, although the detection effect is good, the detection speed is slow, certain requirements are made on the shape and the size of the meteorite crater, and small meteorite craters with irregular shapes cannot be effectively detected. In addition, the problem of shading caused by sunlight illumination change and rugged ground surface shape is easily identified by mistake, so that the identification rate is low.
In recent years, with the continuous development of big data and computing power, deep learning is rapidly advanced, and convolutional neural networks are widely applied in various fields of computer vision. In the prior art [3] (Xin X, Di K, Wang Y, et al. automated detection of new impact sites on Martian surface from HiRISE im-agents [ J ]. Advances in Space Research,2017,60(7): 1557) and Xin et al combine dark area extraction and AdaBoost machine learning algorithm on a Mars image, and the average detection rate of meteor pits reaches 84.5%; in The prior art [4] (Emami E, Ahmad T, Bebis G, et al, Lunar Crater Detec-tion via Region-based volumetric Neural Network-works [ C ]. The 49th Lunar and plant Science Con-reference, Texas, USA, March 19-23,2018.), Emami et al performed meteorite-Crater detection using The fast-RCNN (Faster-Region volumetric Neural Network) Network, and The meteorite-Crater recognition rate reached 92% on images taken by Lunar rail surveys. The meteorite crater detection algorithm based on deep learning realizes automatic identification of meteorite craters, and has the advantages of high accuracy and robustness, high detection speed and huge development potential.
However, the small celestial object environment has the characteristics of weak and irregular darkness and self spinning, and compared with large celestial bodies such as moon and mars, the detection difficulty is higher. Meanwhile, the meteorite crater is easily affected by factors such as erosion and weathering, the situations of shape crushing, overlapping, burying and the like of the meteorite crater are caused, the difficulty of detection of the meteorite crater is further increased, and meanwhile, higher requirements are provided for the stability and accuracy of the detection.
Disclosure of Invention
The invention aims to solve the problems of dim and weak small celestial body images, scarcity, missing detection of small meteorite craters under high resolution and the like, and provides a small celestial body meteorite crater detection method based on deep learning. By data enhancement and local variance equalization algorithm, the generalization capability of the training model is improved, and the meteor crater characteristic is effectively enhanced and retained; aiming at the problem of missing detection of small meteorite craters in high-resolution images, the self-adaptive segmentation high-resolution image detection method is provided, compared with the prior art, the detection speed is high, the recognition rate is high, and therefore the task of detecting the small meteorite craters can be better completed.
The invention is realized by adopting the following technical scheme: a small celestial body meteorite crater detection method based on deep learning comprises the following steps:
step A, constructing a small celestial meteorite crater reference data set;
most of the meteorite crater data sets disclosed at present are meteorite crater images of large celestial bodies such as the moon and the mars, and therefore meteorite crater reference data sets belonging to small celestial bodies need to be constructed. The image data of the small celestial body with the meteor crater collected by the invention is derived from a small planet image library of the NASA official website, and the image mainly covers 730 images containing the meteor crater of the small celestial bodies such as the Aids star (Eros), the Ida (Ida), the Range star (Vesta) and the like. And marking meteor craters and position information thereof in the image by using labelImg software according to the labeling mode of the Pascal VOC data set.
Because the number of the acquired small celestial body images cannot meet the requirement of deep learning network training easily, the method adopts various image processing methods of randomly rotating the images by 0-180 degrees, filtering the mean value, adding a small amount of Gaussian noise and the like to enhance the data of the images. And expanding the collected small celestial body image to 5000 pieces according to a certain proportion, and establishing a small celestial body meteorite crater reference data set.
Step B, strengthening the visual perception of the small celestial meteorite crater characteristics, comprising the following steps:
b1, the meteor crater characteristic is effectively enhanced and retained through a local variance equalization algorithm;
b2, aiming at the small celestial body image with high resolution, providing a self-adaptive segmentation image detection method to improve the detection effect of the small meteorite crater;
c, performing deep learning network training on the small celestial body meteorite crater data set processed in the step A, B, setting training parameters, and obtaining a meteorite crater detection model in the small celestial body environment after training for 10000 generations;
d, detecting self-adaptive segmented sub-images by using a deep learning network according to the meteor crater training model under the small celestial body environment obtained in the step C, and merging detection results to realize the representation and identification of the meteor craters on the surface of the small celestial body;
further, in the step B1, when the meteor crater feature is enhanced and retained, the following method is specifically adopted:
defining a sliding window with the size of (2k +1) × (2k +1) and taking the pixel point (i, j) as the center, wherein k is an integer:
wherein m (i, j), σ2(i, j) are respectively the local mean-square error and the variance based on the window center pixel point (i, j); f (x, y) represents the gray value corresponding to the central pixel point (i, j) of the window.
In order to avoid the phenomenon of over-enhancement of the image, the small celestial body image under the dark and weak environment is correspondingly enhanced by using the following formula, and q (i, j) is the gray value of the enhanced central pixel point (i, j) of the sliding window.
And selecting a sliding window with proper size to move the image pixel by pixel, replacing the gray value at the center of the original window with the enhanced gray value, and traversing the whole image so as to achieve the effect of integrally enhancing the image.
Further, in the step B2, when the high-resolution small celestial body image is targeted, the following method is specifically adopted:
when the deep learning network prediction is input, the images need to be uniformly reduced to a specified size, so that the small meteorite crater in the high-resolution image is easy to generate missing detection. In order to solve the above problems, the present invention adaptively divides the predicted image after the step B1 into a plurality of sub-images with overlapping regions, sends the sub-images into a deep learning detection network, and merges and displays the prediction result in the predicted image.
In image segmentation, in order to ensure that the maximum meteorite crater appears in at least any one segmented subimage, a certain overlap region should be reserved between the segmented subimages. Assuming that the detected image is an image shot by a camera at the height H, the focal length of the camera is f, the maximum meteorite crater diameter of the small celestial body is R, and p is the pixel size occupied by the maximum meteorite crater diameter in the image, according to the small hole imaging principle, namely:
because the resolution of the predicted image is not fixed, the edge part is segmented by adopting a black area complementing method.
Wherein c is the height and width of the sub-image to be sliced; t is the width of the overlapping area, and the numerical value is the same as the number p of the pixels imaged in the image by the maximum meteorite crater diameter; m and n are the number of sub-image blocks divided in the horizontal direction and the vertical direction of the image; and the delta L and the delta M are the number of pixels complemented in the horizontal direction and the vertical direction.
Compared with the prior art, the invention has the advantages and positive effects that:
(1) the invention provides a small celestial body meteorite hole detection method based on deep learning, which can still effectively detect meteorite holes with large scale difference and irregular shape only by utilizing an optical image shot by a navigation camera on the surface of a small celestial body, has high detection speed and high recognition rate compared with the prior art, and can better complete the task of detecting the small celestial body meteorite holes;
(2) the invention provides a small celestial body meteorite crater detection method based on deep learning, which can fully excavate deep characteristics of the small celestial body meteorite crater from limited data through a deep learning network, thereby greatly improving the meteorite crater detection effect; the local variance equalization algorithm can keep stable detection under the influence of darkness and weakness of a small celestial body environment, and the representation and identification of the meteorite crater on the surface of the small celestial body are realized.
(3) The invention provides a small celestial body meteorite crater detection method based on deep learning, which reduces missing detection of the small meteorite crater by a method of adaptively segmenting a prediction image, is beneficial to improving the identification rate of the meteorite crater, and has great significance for visual navigation, obstacle detection, avoidance and the like in the small celestial body landing process.
Drawings
FIG. 1 is a flow chart of the small celestial meteorite crater detection according to an embodiment of the invention;
FIG. 2 is a diagram illustrating a result of data enhancement according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating the result of image enhancement according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of image segmentation according to an embodiment of the present invention;
FIG. 5 is a flowchart illustrating deep learning network training in accordance with an embodiment of the present invention;
FIG. 6 is a diagram of the detection result of the small celestial body merle crate detection method based on deep learning in the embodiment of the present invention.
Detailed Description
In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be further described with reference to the accompanying drawings and examples. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those described herein, and thus, the present invention is not limited to the specific embodiments disclosed below.
A method for detecting small celestial body meteorite crater based on deep learning is shown in figure 1 and comprises the following steps:
and 4, detecting the sub-images of the self-adaptive segmentation by using a deep learning network according to the meteor crater training model under the small celestial body environment obtained in the step 3, and merging the detection results to realize the representation and identification of the meteor craters on the surface of the small celestial body.
Specifically, the method comprises the following steps:
most of the meteorite crater data sets disclosed at present are meteorite crater images of large celestial bodies such as the moon and the mars, and therefore meteorite crater reference data sets belonging to small celestial bodies need to be constructed.
In this embodiment, the collected image data of the small celestial body with the meteor crater is derived from the asteroid image library of the NASA official website, and the image mainly covers 730 images of the small celestial bodies with the meteor crater, such as the love star (Eros), the Ida (Ida), the Range star (Vesta), and the like. The number of specific images of each small celestial body is shown in table 1.
TABLE 1 Small celestial body image quantity statistics
And marking meteor craters and position information thereof in the image by using labelImg software according to the labeling mode of the Pascal VOC data set.
Because the number of the acquired small celestial body images cannot meet the requirement of deep learning network training easily, the method adopts various image processing methods of randomly rotating the images by 0-180 degrees, filtering the mean value, adding a small amount of Gaussian noise and the like to enhance the data of the images. The addition of zero mean gaussian noise with too large variance easily causes the image to be too fuzzy, seriously affects the image quality and is not beneficial to the feature extraction of the subsequent meteor crater. Experiments show that the addition of zero-mean Gaussian noise with the variance of 0.01 to the image is most beneficial to model training. The small celestial body images after image preprocessing are expanded to 5000 pieces according to a certain proportion, and the number and the effect of the samples after image data enhancement are shown in table 2 and fig. 2.
TABLE 2 number of samples after enhancement of various types of data
The method for enhancing the data not only can greatly increase the number of the data set samples, but also can increase the diversity of the data set samples and improve the generalization capability of the training model.
2.1, the meteor crater characteristic is effectively enhanced and retained through a local variance equalization algorithm;
in this embodiment, a sliding window with a size of (2k +1) × (2k +1) and centered on a pixel point (i, j) is defined, where k is an integer:
wherein m (i, j), σ2(i, j) are respectively the local mean-square error and the variance based on the window center pixel point (i, j); f (x, y) represents the gray value corresponding to the central pixel point (i, j) of the window.
In order to avoid the phenomenon of over-enhancement of the image, the small celestial body image under the dark and weak environment is correspondingly enhanced by using the following formula, and q (i, j) is the gray value of the enhanced central pixel point (i, j) of the sliding window.
And selecting a sliding window with proper size to move the image pixel by pixel, replacing the gray value at the center of the original window with the enhanced gray value, and traversing the whole image so as to achieve the effect of integrally enhancing the image. Experiments show that the sliding window with the size of 3 multiplied by 3 can effectively inhibit noise influence and enable meteorite crater edge details to be more obvious. The enhancement effect is shown in fig. 3.
2.2, aiming at the small celestial body image with high resolution, a self-adaptive segmentation image detection method is provided, and the detection effect of the small meteorite crater is improved;
when the deep learning network prediction is input, the images need to be uniformly reduced to a specified size, so that the small meteorite crater in the high-resolution image is easy to generate missing detection. In order to solve the above problems, the present invention adaptively divides the predicted image after the step B1 into a plurality of sub-images with overlapping regions, sends the sub-images into a deep learning detection network, and merges and displays the prediction result in the predicted image.
In image segmentation, in order to ensure that the maximum meteorite crater appears in at least any one segmented subimage, a certain overlap region should be reserved between the segmented subimages. Assuming that the detected image is an image shot by a camera at the height H, the focal length of the camera is f, the maximum meteorite crater diameter of the small celestial body is R, and p is the pixel size occupied by the maximum meteorite crater diameter in the image, according to the small hole imaging principle, namely:
because the resolution of the predicted image is not fixed, the edge part is segmented by adopting a black area complementing method. The splitting effect is shown in fig. 4.
Wherein c is the height and width of the sub-image to be sliced; t is the width of the overlapping area, and the numerical value is the same as the number p of the pixels imaged in the image by the maximum meteorite crater diameter; m and n are the number of sub-image blocks divided in the horizontal direction and the vertical direction of the image; and the delta L and the delta M are the number of pixels complemented in the horizontal direction and the vertical direction.
the deep learning network in the example greatly enhances the detection efficiency through the use of a residual error structure and an anchor box, and simultaneously, the detection performance of the small meteor crater is correspondingly enhanced by adding a space pyramid pooling network and a path aggregation network. In addition, in the training process of the deep learning network, the model is enhanced by randomly shielding and erasing a part of image, so that the problem of incomplete meteor craters is solved to a certain extent.
A flow chart for deep learning network training is shown in fig. 5.
In this embodiment, the data set is randomly divided into a training set, a test set, and a verification set according to a ratio of 7:2:1, where specific training parameters are shown in table 3.
TABLE 3 training parameters
And 4, detecting the sub-images of the self-adaptive segmentation by using a deep learning network according to the meteor crater training model under the small celestial body environment obtained in the step 3, and merging the detection results to realize the representation and identification of the meteor craters on the surface of the small celestial body.
Compared with the prior art, the detection effect of the meteorite crater and the small meteorite crater with irregular, overlapped and broken shapes is obviously improved, the detection accuracy rate of the meteorite crater is up to 98.44%, and the recall rate is 82.65%.
FIG. 6 is the result of meteor crater detection of the Range-Star and love-Star small celestial images.
The above description is only a preferred embodiment of the present invention, and not intended to limit the present invention in other forms, and any person skilled in the art may apply the above modifications or changes to the equivalent embodiments with equivalent changes, without departing from the technical spirit of the present invention, and any simple modification, equivalent change and change made to the above embodiments according to the technical spirit of the present invention still belong to the protection scope of the technical spirit of the present invention.
Claims (4)
1. A small celestial body meteorite crater detection method based on deep learning is characterized by comprising the following steps:
step A, constructing a small celestial meteorite crater reference data set;
most of the meteorite crater data sets disclosed at present are meteorite crater images of large celestial bodies such as the moon and the mars, and therefore meteorite crater reference data sets belonging to small celestial bodies need to be constructed. The image data of the small celestial body with the meteor crater collected by the invention is derived from a small planet image library of the NASA official website, and the image mainly covers 730 images containing the meteor crater of the small celestial bodies such as the Aids star (Eros), the Ida (Ida), the Range star (Vesta) and the like. And marking meteor craters and position information thereof in the image by using labelImg software according to the labeling mode of the Pascal VOC data set. However, because the number of the acquired small celestial body images cannot meet the requirement of deep learning network training easily, the method adopts various image processing methods of randomly rotating the images by 0-180 degrees, performing mean filtering, adding a small amount of Gaussian noise and the like to enhance the data of the images. And expanding the collected small celestial body image to 5000 pieces according to a certain proportion, and establishing a small celestial body meteorite crater reference data set.
Step B, strengthening the visual perception of the small celestial meteorite crater characteristics, comprising the following steps:
b1, the meteor crater characteristic is effectively enhanced and retained through a local variance equalization algorithm;
b2, aiming at the small celestial body image with high resolution, providing a self-adaptive segmentation image detection method to improve the detection effect of the small meteorite crater;
c, performing deep learning network training on the small celestial body meteorite crater data set processed in the step A, B, setting training parameters, and obtaining a meteorite crater detection model in the small celestial body environment after training for 10000 generations;
and D, detecting the sub-images of the self-adaptive segmentation by using a deep learning network according to the meteor crater training model under the small celestial body environment obtained in the step C, and merging the detection results to realize the representation and identification of the meteor craters on the surface of the small celestial body.
2. The small celestial body merle detection method based on deep learning of claim 1, characterized in that: in step B1, when the meteor crater feature is enhanced and retained, the following method is specifically adopted:
defining a sliding window with the size of (2k +1) × (2k +1) and taking the pixel point (i, j) as the center, wherein k is an integer:
wherein m (i, j), σ2(i, j) are respectively the local mean-square error and the variance based on the window center pixel point (i, j); f (x, y) represents the gray value corresponding to the window center pixel point (i, j);
in order to avoid the phenomenon of over-enhancement of the image, the small celestial body image under the dark and weak environment is correspondingly enhanced by using the following formula, and q (i, j) is the gray value of the enhanced central pixel point (i, j) of the sliding window;
and selecting a sliding window with proper size to move the image pixel by pixel, replacing the gray value at the center of the original window with the enhanced gray value, and traversing the whole image so as to achieve the effect of integrally enhancing the image.
3. The small celestial body merle detection method based on deep learning according to claim 2, characterized in that: in the step B2, when the high-resolution small celestial body image is targeted, the following method is specifically adopted:
when the deep learning network prediction is input, the images need to be uniformly reduced to a specified size, so that the small meteorite crater in the high-resolution image is easy to generate missing detection. In order to solve the problems, the invention adaptively divides the prediction image subjected to the step B1 into a plurality of sub-images with overlapping areas, sends the sub-images into a deep learning detection network, and merges and displays the prediction result in the prediction image;
in image segmentation, in order to ensure that the maximum meteorite crater appears in at least any one segmented subimage, a certain overlap region should be reserved between the segmented subimages. Assuming that the detected image is an image shot by a camera at the height H, the focal length of the camera is f, the maximum meteorite crater diameter of the small celestial body is R, and p is the pixel size occupied by the maximum meteorite crater diameter in the image, according to the small hole imaging principle, namely:
as the resolution of the predicted image is not fixed, the edge part is segmented by adopting a black area complementing method;
wherein c is the height and width of the sub-image to be sliced; t is the width of the overlapping area, and the numerical value is the same as the number p of the pixels imaged in the image by the maximum meteorite crater diameter; m and n are the number of sub-image blocks divided in the horizontal direction and the vertical direction of the image; and the delta L and the delta M are the number of pixels complemented in the horizontal direction and the vertical direction.
4. The small celestial body merle detection method based on deep learning of claim 3, wherein: in the step C, in the deep learning network training, the following method is specifically adopted:
the deep learning network greatly enhances the detection efficiency by using a residual error structure and an anchor box, and simultaneously, the detection performance of the small meteorite crater is correspondingly enhanced by adding a space pyramid pooling network and a path aggregation network. In addition, in the training process of the deep learning network, the model is enhanced by randomly shielding and erasing a part of image, so that the problem of incomplete meteor craters is solved to a certain extent.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110156137.6A CN112819794B (en) | 2021-02-04 | 2021-02-04 | Small celestial body meteorite crater detection method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110156137.6A CN112819794B (en) | 2021-02-04 | 2021-02-04 | Small celestial body meteorite crater detection method based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112819794A true CN112819794A (en) | 2021-05-18 |
CN112819794B CN112819794B (en) | 2022-12-30 |
Family
ID=75861427
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110156137.6A Active CN112819794B (en) | 2021-02-04 | 2021-02-04 | Small celestial body meteorite crater detection method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112819794B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113375676A (en) * | 2021-05-26 | 2021-09-10 | 南京航空航天大学 | Detector landing point positioning method based on impulse neural network |
CN117437491A (en) * | 2023-12-11 | 2024-01-23 | 昆明理工大学 | Diffusion model fusion active learning-based deep space probe few-sample geological image category identification method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103093463A (en) * | 2013-01-15 | 2013-05-08 | 南京航空航天大学 | Meteor crater detecting method based on gray level image |
CN108871349A (en) * | 2018-07-13 | 2018-11-23 | 北京理工大学 | A kind of deep space probe optical guidance pose weight determination method |
CN109064478A (en) * | 2018-07-17 | 2018-12-21 | 暨南大学 | A kind of astronomical image contour extraction method based on extreme learning machine |
CN110095123A (en) * | 2019-05-31 | 2019-08-06 | 北京理工大学 | Irregular small feature loss surface road sign observation information evaluates optimization method |
CN111652896A (en) * | 2020-05-29 | 2020-09-11 | 北京理工大学 | Inertial navigation auxiliary meteorite crater coarse-to-fine detection method |
CN111738236A (en) * | 2020-08-14 | 2020-10-02 | 之江实验室 | Image segmentation identification method, device and system of self-adaptive hierarchy |
-
2021
- 2021-02-04 CN CN202110156137.6A patent/CN112819794B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103093463A (en) * | 2013-01-15 | 2013-05-08 | 南京航空航天大学 | Meteor crater detecting method based on gray level image |
CN108871349A (en) * | 2018-07-13 | 2018-11-23 | 北京理工大学 | A kind of deep space probe optical guidance pose weight determination method |
CN109064478A (en) * | 2018-07-17 | 2018-12-21 | 暨南大学 | A kind of astronomical image contour extraction method based on extreme learning machine |
CN110095123A (en) * | 2019-05-31 | 2019-08-06 | 北京理工大学 | Irregular small feature loss surface road sign observation information evaluates optimization method |
CN111652896A (en) * | 2020-05-29 | 2020-09-11 | 北京理工大学 | Inertial navigation auxiliary meteorite crater coarse-to-fine detection method |
CN111738236A (en) * | 2020-08-14 | 2020-10-02 | 之江实验室 | Image segmentation identification method, device and system of self-adaptive hierarchy |
Non-Patent Citations (7)
Title |
---|
ANTHONY N,EMAMI M R: "Asteroid engineering: The state-of-the-art of near-earth asteroids science and technology", 《PROGRESS IN AEROSPACE SCIENCE》 * |
于正湜等: "行星表面非规则陨石坑检测与识别方法", 《宇航学报》 * |
冯军华等: "行星表面陨石坑检测与匹配方法", 《航空学报》 * |
崔平远等: "小天体光学导航特征识别与提取研究进展", 《宇航学报》 * |
董增寿等: "卫星遥感图像增强处理方法研究", 《计算机仿真》 * |
郑磊等: "基于深度学习的大型陨石坑识别方法研究", 《北京航空航天大学学报》 * |
陈建清等: "应用灰度特征的行星陨石坑自主检测方法与着陆导航研究", 《宇航学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113375676A (en) * | 2021-05-26 | 2021-09-10 | 南京航空航天大学 | Detector landing point positioning method based on impulse neural network |
CN113375676B (en) * | 2021-05-26 | 2024-02-20 | 南京航空航天大学 | Detector landing site positioning method based on impulse neural network |
CN117437491A (en) * | 2023-12-11 | 2024-01-23 | 昆明理工大学 | Diffusion model fusion active learning-based deep space probe few-sample geological image category identification method |
CN117437491B (en) * | 2023-12-11 | 2024-06-14 | 昆明理工大学 | Diffusion model fusion active learning-based deep space probe few-sample geological image category identification method |
Also Published As
Publication number | Publication date |
---|---|
CN112819794B (en) | 2022-12-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108921799B (en) | Remote sensing image thin cloud removing method based on multi-scale collaborative learning convolutional neural network | |
CN111046768B (en) | Deep learning method for simultaneously extracting road pavement and center line of remote sensing image | |
CN112819794B (en) | Small celestial body meteorite crater detection method based on deep learning | |
CN111797712B (en) | Remote sensing image cloud and cloud shadow detection method based on multi-scale feature fusion network | |
CN103955710B (en) | Method for monocular vision space recognition in quasi-earth gravitational field environment | |
CN108229342B (en) | Automatic sea surface ship target detection method | |
CN109886159B (en) | Face detection method under non-limited condition | |
CN111951226A (en) | Multi-scale contraband detection system and method based on X-ray image | |
Chen et al. | Object detection of optical remote sensing image based on improved faster RCNN | |
CN113610024B (en) | Multi-strategy deep learning remote sensing image small target detection method | |
Zhang et al. | Nearshore vessel detection based on Scene-mask R-CNN in remote sensing image | |
Wang et al. | Vehicle license plate recognition based on wavelet transform and vertical edge matching | |
CN114743103A (en) | Island reef remote sensing image geological classification method based on Deeplabv3+ network model | |
CN114266947A (en) | Classification method and device based on fusion of laser point cloud and visible light image | |
CN110853058B (en) | High-resolution remote sensing image road extraction method based on visual saliency detection | |
CN112561808A (en) | Road boundary line restoration method based on vehicle-mounted laser point cloud and satellite image | |
CN117475145A (en) | Multi-scale remote sensing image semantic segmentation method and system integrating multiple attention mechanisms | |
CN114445726B (en) | Sample library establishing method and device based on deep learning | |
CN116486273A (en) | Method for extracting water body information of small sample remote sensing image | |
CN116665071A (en) | Remote sensing image building parallel extraction method based on multi-scale attention | |
CN110751150A (en) | FPGA-based binary neural network license plate recognition method and system | |
CN111553222B (en) | Remote sensing ground feature classification post-processing method based on iteration superpixel segmentation | |
Zhang et al. | An unsupervised change detection based on clustering combined with multiscale and region growing | |
CN108596088B (en) | Building detection method for panchromatic remote sensing image | |
CN116052019B (en) | High-quality detection method suitable for built-up area of large-area high-resolution satellite image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |