The detection and identification of impact craters on a planetary surface are crucially important for planetary studies and autonomous navigation. Crater detection refers to finding craters in a given image, whereas identification means to actually mapping them to particular reference craters. However, no method is available for simultaneously detecting and identifying craters with sufficient accuracy and robustness. Thus, this study proposes a novel end-to-end fully convolutional neural network (CNN), namely, CraterIDNet, which takes remotely sensed planetary images of any size as input and outputs detected crater positions, apparent diameters, and identification results. CraterIDNet comprises two pipelines, namely, crater detection pipeline (CDP) and crater identification pipeline (CIP). First, we propose a pre-trained model with high generalization performance for transfer learning. Then, anchor scale optimization and anchor density adjustment are proposed for CDP. In addition, multi-scale impact craters are detected simultaneously by using different feature maps with multi-scale receptive fields. These strategies considerably improve the detection performance of small craters. Furthermore, a grid pattern layer is proposed to generate grid patterns with rotation and scale invariance for CIP. The grid pattern integrates the distribution and scale information of nearby craters, which will remarkably improve identification robustness when combined with the CNN framework. We comprehensively evaluate CraterIDNet and present state-of-the-art crater detection and identification performance with a small network architecture (4 MB).

The most accurate star centroiding method for star sensors is the Gaussian fitting (GF) algorithm, because the intensity distribution of a star spot conforms to the Gaussian function, but the computational complexity of GF is too high for real-time applications. In this paper, we develop the fast Gaussian fitting method (FGF), which approximates the solution of the GF in a closed-form, thus significantly speeding up the GF algorithm. Based on the fast Gaussian fitting method, a novel star centroiding algorithm is proposed, which sequentially performs the FGF twice to calculate the star centroid: the first FGF step roughly calculates the Gaussian parameters of a star spot and the noise intensity of each pixel; subsequently the second FGF accurately calculates the star centroid utilizing the noise intensity provided in the first step. In this way, the proposed algorithm achieves both high accuracy and high efficiency. Both simulated star images and star sensor images are used to verify the performance of the algorithm. Experimental results show that the accuracy of the proposed algorithm is almost the same as the GF algorithm, higher than most existing centroiding algorithms, meanwhile, the proposed algorithm is about 15 times faster than the GF algorithm, making it suitable for real-time applications.

Wheel diameter is a significant parameter related to the safety and comfort level of trains. The traditional circle fitting method can be sensitive to noise because of the small distribution of contact points. This paper proposes a structured-light measurement based on the cycloid constraint to obtain a wide distribution of contact points. The wheel is measured twice while running. The cycloid is used to integrate all the contact points to one circle. Simulations are implemented to analyze the possible errors, and the results are compared with the traditional method, which shows the method is more precise and has good stability against 3D reconstruction noise and out-of-roundness. The physical experiment also shows that the result is precise and robust.

Aircraft attitude has always been an important technical parameter during flights, especially for stages such as aircraft take-off and landing. This article presents a monocular off-board vision system for relative attitude measurement of fixed-wing aircraft. In this system, a zoom imaging subsystem is set up to track and shoot the aircraft over a large distance range. Image sequences and corresponding servo data are transmitted to the processing computer for vision algorithms. A model-based attitude tracking algorithm, which requires the computer-aided design model of the aircraft, is performed to measure the relative attitude of the aircraft with respect to the reference platform. A 2-D visual tracking algorithm and a target detection algorithm are also employed to control the vision system. The proposed attitude tracking algorithm achieves competitive results on a publicly available dataset. The full system is validated in a real-flight experiment. The results revealed that the proposed system can run in real time on 1920×1080@40 Hz videos with an accuracy better than 0.5 $^\circ$ .

In this article, we consider the problem of extracting star spots for traditional star sensors under highly dynamic conditions. Under this setting, it is difficult to effectively extract star spots with a low signal-to-noise ratio (SNR), resulting in a failure to estimate the attitude of star sensors. We propose a method named optimal directional connected component (ODCC) for this task. According to the dynamic characteristics of star sensors, we propose an image enhancement method that can adaptively estimate the directions of star spots and integrate the star image so that the SNR of the star spots is increased. This aids in searching for spots with a low SNR. According to the imaging characteristics of a star, we model the imaging region of a star as a rectangle and then creatively cast the problem of extracting spots into a problem of estimating the endpoints of the rectangle. A maximum likelihood estimation method based on the Gaussian mixture model is proposed to determine the region of the star spot. Experiments demonstrate the extraction capability and accuracy of the method. In conclusion, this method has a promising application value for improving the dynamic performance of traditional star sensors.

Images collected by linear scan cameras are stretched and compressed due to the speed regulation of trains. This condition changes the shape of objects and considerably increases the false and missed alarm rates. Moreover, small size and dense distribution of the key components in railway trains increase the difficulty in fault diagnosis. Therefore, a two-module troubleshooting and positioning methodology is proposed in this study for state diagnosis of key small components of high-speed running trains. Firstly, the image with deformation is reshaped to be exactly the same as the standard one via omnidirectional scale correlation normalization (OSCN), which performs well even in low-texture, high-light situations. Secondly, we propose three feature-enhanced models to expand the receptive field of deep feature maps. This novel detector, namely, refine-inception net (RIN), not only reduces the rate of missed and false detection, but also minimizes the influence of target size and occlusion. Augmentation is performed to increase robustness because the detector is data driven. Experimental results show that the optimal strategy combining OSCN and RIN can troubleshoot high-speed trains of different models with an accuracy higher than 99%. Our method can be extended to foreign object recognition on the train roof while maintaining railway safety.