CN114241296A - Method for detecting meteorite crater obstacle during lunar landing, storage medium and electronic device - Google Patents
Method for detecting meteorite crater obstacle during lunar landing, storage medium and electronic device Download PDFInfo
- Publication number
- CN114241296A CN114241296A CN202111333774.2A CN202111333774A CN114241296A CN 114241296 A CN114241296 A CN 114241296A CN 202111333774 A CN202111333774 A CN 202111333774A CN 114241296 A CN114241296 A CN 114241296A
- Authority
- CN
- China
- Prior art keywords
- merle
- lunar
- crater
- network model
- detecting
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000012545 processing Methods 0.000 claims abstract description 21
- 238000002372 labelling Methods 0.000 claims abstract description 4
- 238000001514 detection method Methods 0.000 claims description 41
- 238000004590 computer program Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 description 11
- 238000000605 extraction Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000002708 enhancing effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000005286 illumination Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 241000282414 Homo sapiens Species 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000007786 learning performance Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000012634 optical imaging Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- 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
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for detecting meteorite crater obstacle during lunar landing, a storage medium and electronic equipment, wherein the method comprises the following steps: acquiring lunar remote sensing image data with a preset resolution, and making the lunar remote sensing image data into a meteor crater data set; performing data enhancement processing on the image in the meteor crater data set, and labeling the meteor crater data set subjected to the data enhancement processing; establishing a dense connection network model based on a YOLO-V4 backbone network structure; inputting a data set containing a meteor crater label into the dense connection network model in a preset size, and setting a preset training parameter to train the dense connection network model; and inputting the meteor crater data set to be detected into the trained dense connection network model so as to detect the target meteor crater through the trained dense connection network model. The invention can autonomously identify meteorite crater obstacles with multi-scale changes, and is convenient for avoiding dangerous landing areas in the lunar landing rough obstacle avoidance stage.
Description
Technical Field
The invention relates to the technical field of aerospace, in particular to a method for detecting lunar landing merle crate obstacle, a storage medium and electronic equipment.
Background
With the development of aerospace technology, the exploration of space by human beings is more and more deep, especially the global lunar exploration hot tide brought by the united states re-lunar project. However, the lunar surface is rugged in terrain, is continuously bombarded by a large number of various meteorites, and lacks the protection of the atmospheric layer, and these impactors can reach the lunar surface, so that a new meteorite pit is formed, and in addition, the lunar surface has a lot of unknown obstacles such as rocks, slopes and the like, which brings a very big challenge to the accurate autonomous landing problem of the future unmanned or manned lunar surface, so that an obstacle detection method which can enable the spacecraft to accurately avoid meteorite pit obstacles and can meet very strict landing accuracy requirements is needed.
Detection of lunar landing merle pits typically faces the following challenges: firstly, when the lander just enters a landing state and is far away from the lunar surface, the imaging of the meteorite crater in the camera is usually small, the number of pixels is small, and the feature extraction is difficult. The second challenge is that the lunar surface image is often disturbed by shadows, light, and other external factors such as lunar dust. In addition, the size of the target also changes greatly during the landing process of the lander. In order to solve these problems, researchers have made diligent efforts. The early detection research of the lunar meteorite crater mainly utilizes the traditional image and geometric vision processing, for example, a learner uses a specific template to match a target for detection, although the template matching method is simple and effective, the overall robustness is poor, and the method is sensitive to the shape and geometric deformation of the target.
In the past few years, due to advances in parallel computing technologies such as Graphics Processing units, GPUs (Graphics Processing units), availability of large amounts of tagged data, and encouragement by breakthroughs in deep Neural Network understanding, scholars have employed deep learning CNNs (Convolutional Neural networks) to detect lunar meteorite pits, such as Network models MarsNet, CraterIDNet, and U-Net. Currently, the typical detection method is a two-step method represented by R-CNN (regional convolutional neural network), and then the upgraded versions of R-CNN such as Fast R-CNN, and Mask R-CNN, and other target detection methods developed on the basis of the two steps. Wherein, the two-step method is to divide the detection process into two steps. First, information of the meteor crater target is extracted using RPN (Region pro social Network), and then the position and category information of the meteor crater is predicted by the detection layer. Other Single-step methods such as target detection include SSD (Single Shot multi box Detector), DSSD (deconvolution Single Shot Detector), FSSD (Feature Fusion Single Shot multi box Detector), YOLO (a target detection algorithm), YOLO-v2, YOLO-v3, and YOLO-v 4. In the YOLO series algorithm, the RPN network is not used, but prediction information of a location and a category is directly obtained. Therefore, they are also called regression-based algorithms, and can generally obtain higher detection speed than the two-step method, and are more suitable for real-time monitoring during landing device descending. However, the existing meteorite crater target detection method cannot adapt to multi-scale changes of targets, so that the target identification error rate is high, and the target detection precision needs to be improved.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art. Therefore, the first purpose of the invention is to provide a lunar landing merle crate obstacle detection method, which can autonomously identify the merle crate obstacle with multi-scale changes, improve the target detection precision and complete the avoidance of the dangerous landing area in the lunar landing rough obstacle avoidance stage.
A second object of the present invention is to provide a computer-readable storage medium.
A third object of the present invention is to provide an electronic apparatus.
In order to achieve the purpose, the invention is realized by the following technical scheme:
a method for detecting a lunar landing merle crate disorder, comprising: step S1, acquiring lunar remote sensing image data with a preset resolution, and making the lunar remote sensing image data into a meteor crater data set; step S2, performing data enhancement processing on the image in the meteor crater data set, and labeling the meteor crater data set after the data enhancement processing; step S3, establishing a dense connection network model based on a YOLO-V4 backbone network structure; step S4, inputting a data set containing meteor crater labels into the dense connection network model in a preset size, and setting preset training parameters to train the dense connection network model; and step S5, inputting the meteor crater data set to be detected into the trained dense connection network model, and detecting the target meteor crater through the trained dense connection network model.
Optionally, the step S1 includes: and (3) performing equal-depth image cutting of different depth value batches on the image in the lunar remote sensing image data to obtain meteor crater data sets in different depth ranges.
Optionally, in step S2, the data enhancement processing on the image in the meteorite crater dataset includes: and carrying out random scaling, random size cutting, random angle rotation, random change of image saturation, contrast, brightness and color and image pyramid multilayer sampling processing on the image.
Optionally, the step S3 includes: and replacing the residual error units with low resolution in the YOLO-V4 backbone network structure by dense connection units to obtain the dense connection network model.
Optionally, the preset training parameters in step S4 include at least one of training step number, training batch number, training momentum, and learning rate.
Optionally, when an image is input to an input end of the densely connected network model, the image is divided into a plurality of grid cells, where each grid cell is composed of a plurality of bounding boxes, and each bounding box has a corresponding confidence score.
Optionally, in step S5, the step of performing target merle crate detection through the densely connected network model after training includes: target merle detection is performed by comparing the confidence scores of several of the bounding boxes.
Optionally, the step of performing target merle detection by comparing the confidence scores of a plurality of bounding boxes specifically includes: step S51: obtaining a bounding box with the highest confidence score; step S52: carrying out IOU calculation on a plurality of bounding boxes and the bounding box with the highest confidence score to obtain an IOU value, and carrying out weight function operation on the IOU value to obtain a corresponding final bounding box score, wherein the IOU value is an overlapped region value between the bounding box and a target true value; step S53: and acquiring the most accurate bounding box according to the final bounding box score value so as to detect the target meteor crater through the most accurate bounding box.
To achieve the above object, a second aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, where the computer program is executed by a processor to implement the method for detecting lunar landing merle hole obstacle.
In order to achieve the above object, a third aspect of the present invention provides an electronic device, including a processor and a memory, where the memory stores a computer program, and the computer program, when executed by the processor, implements the method for detecting lunar landing merle obstacle.
The invention has at least one of the following technical effects:
(1) the method comprises the steps of performing equal-depth image cutting of different depth value batches on an image in lunar remote sensing image data to construct a multi-scale image, and enhancing the diversity of data types by performing data enhancement processing on the image in meteor crater data set, so that the meteor crater obstacle detection method can be suitable for the meteor craters on the lunar surface with different complexity degrees and under different illumination conditions, is particularly suitable for multi-scale change of meteor crater targets in the descending process of a lander, and reduces the target identification error rate;
(2) dense connection units are adopted to replace low-resolution residual error units in a YOLO-V4 backbone network structure, and all layers can be connected to carry out channel combination, so that repeated use of features is realized, and the reverse propagation of network gradients is enhanced;
(3) the meteorite crater obstacle can be identified autonomously through the improved dense connection network model, the target detection precision is effectively improved on the premise of keeping better detection speed, and the robustness is better.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
FIG. 1 is a flowchart of a method for detecting lunar landing merle-crate obstacle according to an embodiment of the present invention;
FIG. 2 is a schematic view of images at different heights from the lunar surface according to an embodiment of the present invention;
FIG. 3 is a merle plot after data enhancement processing according to an embodiment of the present invention;
FIG. 4 is a diagram of a dense connection network architecture according to an embodiment of the present invention;
FIGS. 5-8 are graphs of the meteorite crater test results provided by one embodiment of the present invention.
Detailed Description
Reference will now be made in detail to the present embodiments, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The method for detecting a lunar landing merle-crater obstacle, the storage medium, and the electronic device of the present embodiment are described below with reference to the drawings.
FIG. 1 is a flowchart of a method for detecting lunar landing merle-crater disturbance according to an embodiment of the present invention. As shown in fig. 1, the method includes:
and step S1, acquiring lunar remote sensing image data with preset resolution, and making the lunar remote sensing image data into meteor crater data sets.
Wherein, step S1 includes: and (3) performing equal-depth image cutting of different depth value batches on the image in the lunar remote sensing image data to obtain meteor crater data sets in different depth ranges.
Specifically, the data set is important for training the whole neural network model, and the meteorite crater is used as a detection target of the detection method, the definition and the authenticity of the data are considered firstly, and secondly, as shown in fig. 2, the multi-scale detection of the meteorite crater is used as a multi-scale detection of the simulated lander in the descending process, and the multi-scale property of the data is also an important factor to be considered. Therefore, in order to facilitate the processing and analysis of the image, Lunar remote sensing image data with a preset resolution, such as LRO (Lunar Reconnaissance orbiter) NAC (Narrow Angle Camera) image data, can be made into a meteorite pit data set.
In this embodiment, the preset resolution may be 2m, but the invention is not limited thereto; that is, in this embodiment, the LRO NAC image data image with a resolution of 2m may be divided into a plurality of image blocks with equal depth, that is, the equal-depth images of different depth value batches are cropped to construct a multi-scale image, so that different-scale images of the same meteorite crater in the descent process can be simulated, data overfitting is prevented, and the detection method can be applied to multi-scale changes of the meteorite crater target in the descent process of the lander.
And step S2, performing data enhancement processing on the image in the meteor crater dataset, and labeling the meteor crater dataset after the data enhancement processing.
In step S2, the data enhancement processing on the image in the meteor crater dataset includes: and (3) carrying out random scaling, random size cutting, random angle rotation, random change of image saturation, contrast, brightness and color and image pyramid multi-layer sampling processing on the image.
Specifically, in order to expand the diversity of the training data, the image of the data set is expanded to prevent overfitting, some data enhancement methods may be adopted to expand the meteor crater data set, specifically, data enhancement methods such as scaling the image, clipping with random size, random horizontal and vertical flipping, random angular rotation, and random variation of brightness, contrast, saturation, and color may be adopted, and in order to increase the diversity of the data scale, the image may be subjected to pyramid multi-layer sampling as shown in fig. 3. Therefore, the meteorite crater obstacle detection method can be suitable for the meteorite craters on the lunar surface under different complex terrains and different illumination conditions by combining the step S1, is particularly suitable for the multi-scale change of the meteorite crater target in the descending process of the lander, and reduces the error rate of target identification.
And step S3, establishing a dense connection network model based on the YOLO-V4 backbone network structure.
Wherein, step S3 includes: and replacing the residual error units with low resolution in the YOLO-V4 backbone network structure by the dense connection units to obtain a dense connection network model.
In this embodiment, a dense connection structure of convolutional layers may be adopted to improve the backbone network structure of YOLO-V4, for example, a low-resolution residual error unit in the backbone network of YOLO-V4 is connected by a dense connection unit, that is, a part of low-resolution layers of the feature extraction network is replaced by an improved dense connection network, so as to obtain a dense connection network model. Therefore, all layers can be connected through the dense connection units to carry out channel combination so as to realize the repeated use of the characteristics, thereby enhancing the reverse propagation of the gradient, better utilizing the characteristic information and improving the transmission efficiency of the information and the gradient in the network.
Specifically, YOLO-v4 adopts the CSPDarknet53 network structure as a feature extraction network according to the principles of larger network input resolution, deeper network layer and more parameters, and the network structure is shown in table 1 below:
table 1 network architecture using CSPDarknet53 as a feature extraction network
To extract deeper features and avoid gradient vanishing, the Darknet53 structure adds 5 residual blocks in the network, each block consisting of one or more residual units. Wherein the CBL module (network fabric component) contains convolution, batch normalization and activation functions.
And YOLO-v4 adopts SPP-net network structure, 5 × 5, 9 × 9 and 13 × 13 maximal pooling is performed on the 107 th layer of the network respectively to obtain the 108 th layer, the 110 th layer and the 112 th layer, after the pooling is completed, the 107 th layer, the 108 th layer, the 110 th layer and the 112 th layer are connected into the 114 th layer of the feature map, and dimension reduction is performed to 512 channels, so that the network receptive field is increased, and the global grasping capability of the target image is improved.
Among them, the Darknet53 uses several Residual units, and ResNet (Residual Network) composed of these Residual units contains a large number of parameters, which are responsible for the main calculation of the YOLO v4 Network. In contrast, ResNet is summation, and the enhanced deolo-v 4 is a concatenation where the inputs of each layer network include the outputs of all previous layers. E.g., the input at layer l equals k (l-1) + k0Wherein k is growth rate, representing the number of channels of each layer, and DenseNet connects all layers for channel merging, so that feature reuse can be realized, thereby enhancing gradient back propagation, better utilizing feature information, and improving information and gradient in networkThe transmission efficiency of (1).
The network structure of DenseNet in this embodiment is shown in fig. 4. Wherein x is1,x2,x3And x4A characteristic diagram of the output layer is shown. H1,H2,H3And H4Representing a non-linear transformation. The network of l layers contains l (l +1)/2 layer connections, each layer being connected to the other layers, so that each layer can receive all the signatures of the preceding l-1 layer. The characteristic diagram of each layer is shown as xl=Hl[x0,x1,···,xl-1]. Thus, in order to reduce the dependency of the network on the residual unit, a part of the low resolution layer of the feature extraction network may be replaced with a modified dense connection network, and the structure of the modified feature extraction network is shown in table 2 below.
Table 2 improved feature extraction network architecture
And step S4, inputting the data set containing the meteor crater labels into the dense connection network model in a preset size, and setting preset training parameters to train the dense connection network model.
Wherein the preset training parameters in step S4 include at least one of training step number, training batch number, training momentum, and learning rate.
Specifically, a dataset containing merle crate labels may be input into the dense connection network model in a preset size (e.g., 3 × 416 × 416) for experimental training. Wherein, the training parameters can be set to 4000 training steps, 8 training batches and 0.001 learning rate.
And S5, inputting the meteor crater data set to be detected into the trained dense connection network model, and detecting the target meteor crater through the trained dense connection network model.
Specifically, after model training is completed, the meteor crater data set to be detected is input to the trained dense connection network model for verification, and if a lunar surface image is input, the dense connection network model can automatically detect the target meteor crater.
When an image is input at the input end of the densely connected network model, the image is divided into a plurality of grid units, wherein each grid unit consists of a plurality of boundary boxes, and each boundary box has a corresponding confidence score.
Optionally, in step S5, the step of performing target merle crate detection through the densely connected network model after training includes: target merle detection is performed by comparing the confidence scores of several bounding boxes.
The network model in this scheme is improved based on the YOLO network, and for this reason, the foregoing embodiment is described below by taking the YOLO network as an example. The YOLO network is a CNN model for object detection and classification using a one-step method, which outputs the type and location of an object with bounding box and class probabilities in an image as a regression problem. At the initial end of the network, the input image may be divided into S × S grid cells, each of which consists of several bounding boxes and corresponding confidence scores, each bounding box containing a Class C probability, denoted as P (Class)iI Object), if the bounding box contains an Object, p (Object) is 1, otherwise p (Object) is 0, wherein the confidence score is defined asIndicating the probability of the presence of the target and the accuracy of the prediction of the position of the bounding box, the IOU indicates the value of the overlap area between the bounding box and the true value. And finally, determining a target detection result by comparing the confidence scores through the YOLO, thereby realizing the target merle crate detection. Wherein, the confidence coefficient calculation formula is as follows:
wherein, P (Class)iI Object) represents the probability that each bounding box contains class i, p (Object) represents the probability that a bounding box contains an Object,representing the overlap region between the bounding box and the truth.
In one embodiment of the present invention, the step of performing target merle detection by comparing confidence scores of several bounding boxes specifically comprises:
step S51, obtaining a bounding box with the highest confidence score;
step S52, IOU calculation is carried out on a plurality of bounding boxes and the bounding box with the highest confidence score to obtain an IOU value, and the IOU value is subjected to weight function operation to obtain a corresponding final bounding box score, wherein the IOU value is an overlapped area value between the bounding box and a target true value;
and step S53, acquiring the most accurate bounding box according to the final bounding box score value, and detecting the target meteor crater through the most accurate bounding box.
Specifically, YOLO displays the bounding box b by estimating the position of the object and the category of the inputiIn this case, a single true value may generate a plurality of bounding boxes Blist. YOLO then uses NMS (Non-Maximum Suppression) to remove the repeat boxes. Firstly, find out the bounding box M with the highest confidence score, and then find out BlistFrame b iniIOU calculation with M, then delete BlistIs greater than a set threshold value NtRepeating the steps until the most accurate boundary frame is extracted. While YOLO-v4 uses Soft-NMS (softening non-maximum suppression), which differs from conventional NMS in the handling of confidence score adjustment. If conventional NMS operation, as described above, when biThe IOU value of sum M is greater than threshold NtThen from BlistRemoving the frame; for Soft-NMS, M and b are calculated firstiI of (A)OU, then the IOU outputs the finally obtained bounding box score through a weight function, wherein the larger the overlapping area is, the more serious the confidence attenuation is. And then, acquiring the most accurate bounding box according to the final bounding box score value so as to detect the target meteor crater through the most accurate bounding box.
As a specific example, a lunar image is input, as shown in FIGS. 5-8, the dense connection network model can automatically output the position bounding box of the target merle crate, and the detection of the target merle crate can be realized through the position bounding box of the target merle crate.
In this embodiment, the performance of the trained YOLO-v4 can be represented by the mAP (Mean Average Precision), which is a general deep learning performance index. The average precision is the ratio of positive values in all bounding boxes obtained after NMS on the input and is the average of the ratios obtained for all classes.
Finally, the method is compared with a YOLO-v4 and a YOLO-v4 tiny algorithm on the manufactured meteorite crater data set, and the experimental conditions are as follows: a Darknet framework is adopted, and a computer: NVIDIA Jetson TX2, operating system: ubuntu16.04, CPU: 64-bit Denver 2and A57 CPUs, GPU: NVIDIA Pascal, 256-cores. In the experiment, firstly, a data set containing the meteor crater label is input into a dense connection network model in a size of 3 multiplied by 416, 4000 training steps are set in the experiment training, and the training initial parameters and the experiment comparison result are respectively shown in the following table 3 and the following table 4.
TABLE 3 training initial parameters
TABLE 4 Experimental comparison of modified YOLO-V4 with other algorithms
Compared with YOLO-v4, the detection speed of the improved YOLO-v4 is not obviously reduced. And effectively improves the detection precision. Experimental results show that the improved YOLO-v4 can effectively detect the lunar surface target under a complex background under the condition of real-time detection.
The method for detecting meteorite crater obstacle during lunar landing obtains a two-dimensional plane image of an approximate landing area on the lunar surface continuously through an optical imaging sensor installed on a lander, then constructs a meteorite crater model training data set, cuts images in lunar remote sensing image data of the training data set with equal-depth images of different depth values in batches to construct a multi-scale image, enhances the diversity of data types through data enhancement processing of the images in the meteorite crater data set, replaces low-resolution residual error units in a skeleton network structure of YOLO-V4 with dense connection units, can connect all layers for channel combination, and is suitable for terrains with different complexity degrees on the lunar surface and meteorite craters under different illumination conditions, and is particularly suitable for multi-scale changes of the meteorite crater target during the descent process of the lander, the target identification error rate is reduced, the probability of safe landing of the system is improved, the improved YOLO-V4 realizes the repeated use of features, the backward propagation of network gradient is enhanced, meteorite crater obstacle can be identified independently, the target detection precision is effectively improved on the premise of keeping better detection speed, and the robustness is better.
Further, the present embodiment may also provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method for detecting lunar landing merle obstacle can be implemented.
In particular, the storage medium may be a non-transitory computer readable storage medium, for example, the non-transitory computer readable storage medium may be a read-only memory, a random access memory, a read-only optical disc, a magnetic tape, a floppy disk, an optical data storage device, and the like.
Further, this embodiment may also provide an electronic device, which includes a processor and a memory, where the memory stores a computer program, and when the computer program is executed by the processor, the method for detecting lunar landing merle crate obstacle as described above may be implemented.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
While the present invention has been described in detail with reference to the preferred embodiments, it should be understood that the above description should not be taken as limiting the invention. Various modifications and alterations to this invention will become apparent to those skilled in the art upon reading the foregoing description. Accordingly, the scope of the invention should be determined from the following claims.
Claims (10)
1. A method for detecting a lunar landing merle crate obstacle is characterized by comprising the following steps:
step S1: acquiring lunar remote sensing image data with a preset resolution, and making the lunar remote sensing image data into a meteor crater data set;
step S2: performing data enhancement processing on the image in the meteor crater dataset, and labeling the meteor crater dataset after the data enhancement processing;
step S3: establishing a dense connection network model based on a YOLO-V4 backbone network structure;
step S4: inputting a data set containing a meteor crater label to the dense connection network model in a preset size, and setting a preset training parameter to train the dense connection network model;
step S5: inputting a meteor crater data set to be detected into the trained dense connection network model, and detecting the target meteor crater through the trained dense connection network model.
2. The method for detecting lunar landing merle obstacle as claimed in claim 1, wherein the step S1 includes: and (3) performing equal-depth image cutting of different depth value batches on the image in the lunar remote sensing image data to obtain meteor crater data sets in different depth ranges.
3. The method for detecting lunar landing merle obstacle as claimed in claim 1, wherein in step S2, the data enhancement processing of the image in the merle data set comprises: and scaling the images in random proportion, cutting the images in random size, rotating the images in random angle, and carrying out image saturation, contrast, brightness, color random change and image pyramid multilayer sampling processing.
4. The method for detecting lunar landing merle obstacle as claimed in claim 1, wherein the step S3 includes: and replacing the residual error units with low resolution in the YOLO-V4 backbone network structure by dense connection units to obtain the dense connection network model.
5. The method for detecting lunar landing merle crate disturbance according to claim 1, wherein the preset training parameters in step S4 include at least one of training step number, training batch number, training momentum and learning rate.
6. The method for detecting lunar landing merle crate obstacle as claimed in claim 1, wherein the input end of the dense connection network model, when inputting the image, segments the image into a number of grid cells, wherein each grid cell is composed of a number of bounding boxes, each of the bounding boxes having a corresponding confidence score.
7. The method for detecting lunar landing merle obstacle as claimed in claim 6, wherein in step S5, the step of performing target merle detection through the densely connected network model after training comprises: target merle detection is performed by comparing the confidence scores of several of the bounding boxes.
8. The method of detecting lunar landing merle obstacle as claimed in claim 7, wherein the step of performing target merle detection by comparing confidence scores of several of the bounding boxes specifically comprises:
step S51: obtaining a bounding box with the highest confidence score;
step S52: carrying out IOU calculation on a plurality of bounding boxes and the bounding box with the highest confidence score to obtain an IOU value, and carrying out weight function operation on the IOU value to obtain a corresponding final bounding box score, wherein the IOU value is an overlapped region value between the bounding box and a target true value;
step S53: and acquiring the most accurate bounding box according to the final bounding box score value so as to detect the target meteor crater through the most accurate bounding box.
9. A computer-readable storage medium, having a computer program stored thereon, which, when executed by a processor, implements the method for detecting lunar landing merle obstacle as claimed in any one of claims 1-8.
10. An electronic device comprising a processor and a memory, the memory having stored thereon a computer program that, when executed by the processor, implements the method of detecting lunar landing merle obstacle as claimed in any of claims 1-8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111333774.2A CN114241296A (en) | 2021-11-11 | 2021-11-11 | Method for detecting meteorite crater obstacle during lunar landing, storage medium and electronic device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111333774.2A CN114241296A (en) | 2021-11-11 | 2021-11-11 | Method for detecting meteorite crater obstacle during lunar landing, storage medium and electronic device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114241296A true CN114241296A (en) | 2022-03-25 |
Family
ID=80749175
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111333774.2A Pending CN114241296A (en) | 2021-11-11 | 2021-11-11 | Method for detecting meteorite crater obstacle during lunar landing, storage medium and electronic device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114241296A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114972720A (en) * | 2022-05-30 | 2022-08-30 | 西北工业大学 | High-precision unmanned positioning method based on autonomous image perception |
CN115393730A (en) * | 2022-07-15 | 2022-11-25 | 南京林业大学 | Accurate identification method for Mars meteorite crater, electronic equipment and storage medium |
CN115410096A (en) * | 2022-11-03 | 2022-11-29 | 成都国星宇航科技股份有限公司 | Satellite remote sensing image multi-scale fusion change detection method, medium and electronic device |
CN116051821A (en) * | 2023-03-01 | 2023-05-02 | 吉林大学 | Engineering-oriented single sliding window full-month landing zone selection method and system thereof |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020164282A1 (en) * | 2019-02-14 | 2020-08-20 | 平安科技(深圳)有限公司 | Yolo-based image target recognition method and apparatus, electronic device, and storage medium |
CN112819794A (en) * | 2021-02-04 | 2021-05-18 | 青岛科技大学 | Small celestial body meteorite crater detection method based on deep learning |
CN113139592A (en) * | 2021-04-14 | 2021-07-20 | 中国地质大学(武汉) | Method, device and storage medium for identifying lunar meteorite crater based on depth residual error U-Net |
-
2021
- 2021-11-11 CN CN202111333774.2A patent/CN114241296A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020164282A1 (en) * | 2019-02-14 | 2020-08-20 | 平安科技(深圳)有限公司 | Yolo-based image target recognition method and apparatus, electronic device, and storage medium |
CN112819794A (en) * | 2021-02-04 | 2021-05-18 | 青岛科技大学 | Small celestial body meteorite crater detection method based on deep learning |
CN113139592A (en) * | 2021-04-14 | 2021-07-20 | 中国地质大学(武汉) | Method, device and storage medium for identifying lunar meteorite crater based on depth residual error U-Net |
Non-Patent Citations (4)
Title |
---|
HU, TAO, ET AL.: ""Crater Obstacle Recognition and Detection of Lunar Landing Based on YOLO v4"", 2021 33RD CHINESE CONTROL AND DECISION CONFERENCE (CCDC), 24 May 2021 (2021-05-24), pages 1748 - 1752, XP034031569, DOI: 10.1109/CCDC52312.2021.9601421 * |
丁萌;曹云峰;吴庆宪;: "一种从月面图像检测陨石坑的方法", 宇航学报, no. 03, 15 May 2009 (2009-05-15), pages 1243 - 1248 * |
卢健 等: ""基于深度学习的目标检测综述"", 电光与控制, vol. 27, no. 5, 31 May 2020 (2020-05-31), pages 56 - 63 * |
张为;魏晶晶;: "嵌入DenseNet结构和空洞卷积模块的改进YOLO v3火灾检测算法", 天津大学学报(自然科学与工程技术版), no. 09, 29 June 2020 (2020-06-29), pages 976 - 983 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114972720A (en) * | 2022-05-30 | 2022-08-30 | 西北工业大学 | High-precision unmanned positioning method based on autonomous image perception |
CN114972720B (en) * | 2022-05-30 | 2024-02-20 | 西北工业大学 | Autonomous image perception-based high-precision unmanned positioning method |
CN115393730A (en) * | 2022-07-15 | 2022-11-25 | 南京林业大学 | Accurate identification method for Mars meteorite crater, electronic equipment and storage medium |
CN115393730B (en) * | 2022-07-15 | 2023-05-30 | 南京林业大学 | Mars meteorite crater precise identification method, electronic equipment and storage medium |
CN115410096A (en) * | 2022-11-03 | 2022-11-29 | 成都国星宇航科技股份有限公司 | Satellite remote sensing image multi-scale fusion change detection method, medium and electronic device |
CN116051821A (en) * | 2023-03-01 | 2023-05-02 | 吉林大学 | Engineering-oriented single sliding window full-month landing zone selection method and system thereof |
CN116051821B (en) * | 2023-03-01 | 2023-06-02 | 吉林大学 | Engineering-oriented single sliding window full-month landing zone selection method and system thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yu et al. | A real-time detection approach for bridge cracks based on YOLOv4-FPM | |
CN111429514B (en) | Laser radar 3D real-time target detection method integrating multi-frame time sequence point cloud | |
CN110084292B (en) | Target detection method based on DenseNet and multi-scale feature fusion | |
CN114241296A (en) | Method for detecting meteorite crater obstacle during lunar landing, storage medium and electronic device | |
US20220036562A1 (en) | Vision-based working area boundary detection system and method, and machine equipment | |
CN110569738B (en) | Natural scene text detection method, equipment and medium based on densely connected network | |
CN110599537A (en) | Mask R-CNN-based unmanned aerial vehicle image building area calculation method and system | |
CN114120115B (en) | Point cloud target detection method integrating point features and grid features | |
Li et al. | Toward in situ zooplankton detection with a densely connected YOLOV3 model | |
WO2023273337A1 (en) | Representative feature-based method for detecting dense targets in remote sensing image | |
Xu et al. | Feature-based constraint deep CNN method for mapping rainfall-induced landslides in remote regions with mountainous terrain: An application to Brazil | |
CN109325407B (en) | Optical remote sensing video target detection method based on F-SSD network filtering | |
CN115100741B (en) | Point cloud pedestrian distance risk detection method, system, equipment and medium | |
CN116469020A (en) | Unmanned aerial vehicle image target detection method based on multiscale and Gaussian Wasserstein distance | |
CN113160117A (en) | Three-dimensional point cloud target detection method under automatic driving scene | |
CN116309817A (en) | Tray detection and positioning method based on RGB-D camera | |
Yildirim et al. | Ship detection in optical remote sensing images using YOLOv4 and Tiny YOLOv4 | |
Yang et al. | Lightweight Attention-Guided YOLO With Level Set Layer for Landslide Detection From Optical Satellite Images | |
CN112560907B (en) | Finite pixel infrared unmanned aerial vehicle target detection method based on mixed domain attention | |
Bagwari et al. | A comprehensive review on segmentation techniques for satellite images | |
Wu et al. | Detection algorithm for dense small objects in high altitude image | |
CN117218545A (en) | LBP feature and improved Yolov 5-based radar image detection method | |
Wu et al. | Research on asphalt pavement disease detection based on improved YOLOv5s | |
CN116953702A (en) | Rotary target detection method and device based on deduction paradigm | |
Zhang et al. | Remote sensing image land classification based on deep learning |
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 |