Deep Neural Network-Based Landmark Selection Method for Optical Navigation on Lunar Highlands

Lee, Hoonhee; Choi, Han‐Lim; Jung, Dawoon; Choi, Su-Jin

doi:10.1109/access.2020.2996403

Cited by 12 publications

(12 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Many researchers have also proposed a series of lunar surface impact craters extraction algorithms. According to the different design ideas of these algorithms, their development can be roughly divided into two directions: the traditional algorithm that uses image processing technology to identify impact craters [5,6] and the intelligent algorithm that introduces a deep learning model to extract impact craters [7][8][9][10].…”

Section: Introductionmentioning

confidence: 99%

“…Silburt [7] based on the U-Net model of image semantic segmentation in deep learning, proposed a lunar surface impact crater recognition model, and transferred the model to the surface of Mercury impact crater recognition, and achieved good results. Lee [8] extended this model structure to impact craters on the surface of Mars and then proposed the DeepMars model to realize the rapid identification of impact craters on the surface of Mars. Wang [20] optimized the model, combined with a residual network, and proposed an effective residual U-Net (ERU-Net), which achieved good results in the test.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Moon Impact Crater Detection Using Nested Attention Mechanism Based UNet++

2021

View full text Add to dashboard Cite

Impact craters are the most prominent topographic feature on the lunar surface, which will play a significant role in constructing lunar bases and lunar surface activities in the future. Traditional meteorite crater recognition methods are mainly based on artificial interpretation, usually combined with classical image processing methods. However, due to the different diameters and shapes of impact craters, the traditional crater identification methods have significant errors and low efficiency for small or overlapping impact craters. This paper proposes an automated algorithm termed nested attention-aware U-Net (NAU-Net) for crater detection using a lunar's digital elevation model (DEM). It then uses template matching to effectively calculate the longitude, latitude, and radius of the crater. It is worth mentioning that NAU-Net is primarily based on UNet++ and attention networks and applies a sequence of nested dense convolution blocks, preferably of classical convolution, which combines U-Net++ and Attention Gates advantages. Since our network uses nested intensive attentional aware connections and in-depth supervision, the training process of the NAU-Net is simple, which can achieve more accurate detection and recognition. In fact, the network can recognize smaller or overlapping impact craters and larger and more complex ones. In honor of lunar impact craters, compared with U-Net, UNet++, Dense-Unet, Attention-Unet, R2-Unet, our model achieved recall rates and accuracy of 0.791 and 0.856, better than other improved U-Net models. The experimental results show that the NAU-Net model can be used to extract impact craters.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Moon Impact Crater Detection Using Nested Attention Mechanism Based UNet++

2021

View full text Add to dashboard Cite

show abstract

“…Thus, predicting the landing stability at a given landing site is essential for lunar exploration missions [7][8][9][10][11][12][13][14]. As a method of predicting landing stability, many studies have been conducted on selecting an optimal landing site by analyzing the slope angle of the Moon's surface.…”

Section: Introductionmentioning

confidence: 99%

Prediction and Validation of Landing Stability of a Lunar Lander by a Classification Map Based on Touchdown Landing Dynamics’ Simulation Considering Soft Ground

et al. 2021

View full text Add to dashboard Cite

In this paper, a method for predicting the landing stability of a lunar lander by a classification map of the landing stability is proposed, considering the soft soil characteristics and the slope angle of the lunar surface. First, the landing stability condition in terms of the safe (=stable), sliding (=unstable), and tip-over (=statically unstable) possibilities was checked by dropping a lunar lander onto flat lunar surfaces through finite-element (FE) simulation according to the slope angle, friction coefficient, and soft/rigid ground, while the vertical touchdown velocity was maintained at 3 m/s. All of the simulation results were classified by a classification map with the aid of logistic regression, a machine-learning classification algorithm. Finally, the landing stability status was efficiently predicted by Monte Carlo (MC) simulation by just referring to the classification map for 10,000 input datasets, consisting of the friction coefficient, slope angles, and rigid/soft ground. To demonstrate the performance, two virtual lunar surfaces were employed based on a 3D terrain map of the LRO mission. Then, the landing stability was validated through landing simulation of an FE model of a lunar lander requiring high computation cost. The prediction results showed excellent agreement with those of landing simulations with a negligible computational cost of around a few seconds.

show abstract

“…Frommberger [44] in 2008 investigated a qualitative representation of landmarks for the selection process, while Beinhofer et al [47] linearized the whole navigation cycle representing the landmark locations by a discrete set and then used a user-defined bound for conservative approximation of landmark visibility and selection. Lee et al [48] presented a deep neural network-based landmark selection method for optical navigation on lunar highlands.…”

Section: Introductionmentioning

confidence: 99%

Stabilizing Controllers for Landmark Navigation of Planar Robots in an Obstacle-Ridden Workspace

Prasad

Sharma

Vanualailai

et al. 2020

Journal of Advanced Transportation

View full text Add to dashboard Cite

This paper essays a new solution to the landmark navigation problem of planar robots in the presence of randomly fixed obstacles through a new dynamic updating rule involving the orientation and steering angle parameters of a robot. The dynamic updating rule utilizes a first-order nonlinear ordinary differential equation for the changing of landmarks so that whenever a landmark is updated, the path followed by the robot remains continuous and smooth. This waypoints guidance is via specific landmarks selected from a new set of rules governing the robot’s field of view. The governing control laws guarantee asymptotic stability of the 2D point robot system. As an application, the landmark motion planning and control of a car-like mobile robot navigating in the presence of fixed elliptic-shaped obstacles are considered. The proposed control laws take into account the geometrical constraints imposed on steering angle and guarantee eventual uniform stability of the car-like system. Computer simulations, using Matlab software, are presented to illustrate the effectiveness of the proposed technique and its stabilizing algorithm.

show abstract

Deep Neural Network-Based Landmark Selection Method for Optical Navigation on Lunar Highlands

Cited by 12 publications

References 23 publications

Moon Impact Crater Detection Using Nested Attention Mechanism Based UNet++

Moon Impact Crater Detection Using Nested Attention Mechanism Based UNet++

Prediction and Validation of Landing Stability of a Lunar Lander by a Classification Map Based on Touchdown Landing Dynamics’ Simulation Considering Soft Ground

Stabilizing Controllers for Landmark Navigation of Planar Robots in an Obstacle-Ridden Workspace

Contact Info

Product

Resources

About