Split-Attention Networks with Self-Calibrated Convolution for Moon Impact Crater Detection from Multi-Source Data

Impact craters are the most prominent features on the surface of the Moon, Mars, and Mercury. They play an essential role in constructing lunar bases, the dating of Mars and Mercury, and the surface exploration of other celestial bodies. The traditional crater detection algorithms (CDA) are mainly based on manual interpretation which is combined with classical image processing techniques. The traditional CDAs are, however, inefficient for detecting smaller or overlapped impact craters. In this paper, we propose a Split-Attention Networks with Self-Calibrated Convolution (SCNeSt) architecture, in which the channel-wise attention with multi-path representation and self-calibrated convolutions can generate more prosperous and more discriminative feature representations. The algorithm first extracts the crater feature model under the well-known target detection R-FCN network framework. The trained models are then applied to detecting the impact craters on Mercury and Mars using the transfer learning method. In the lunar impact crater detection experiment, we managed to extract a total of 157,389 impact craters with diameters between 0.6 and 860 km. Our proposed model outperforms the ResNet, ResNeXt, ScNet, and ResNeSt models in terms of recall rate and accuracy is more efficient than that other residual network models. Without training for Mars and Mercury remote sensing data, our model can also identify craters of different scales and demonstrates outstanding robustness and transferability.

Unsupervised Path Representation Learning with Curriculum Negative Sampling

Path representations are critical in a variety of transportation applications, such as estimating path ranking in path recommendation systems and estimating path travel time in navigation systems. Existing studies often learn task-specific path representations in a supervised manner, which require a large amount of labeled training data and generalize poorly to other tasks. We propose an unsupervised learning framework Path InfoMax (PIM) to learn generic path representations that work for different downstream tasks. We first propose a curriculum negative sampling method, for each input path, to generate a small amount of negative paths, by following the principles of curriculum learning. Next, PIM employs mutual information maximization to learn path representations from both a global and a local view. In the global view, PIM distinguishes the representations of the input paths from those of the negative paths. In the local view, PIM distinguishes the input path representations from the representations of the nodes that appear only in the negative paths. This enables the learned path representations encode both global and local information at different scales. Extensive experiments on two downstream tasks, ranking score estimation and travel time estimation, using two road network datasets suggest that PIM significantly outperforms other unsupervised methods and is also able to be used as a pre-training method to enhance supervised path representation learning.

Strategies for smarter catchment hydrology models: incorporating scaling and better process representation

Hydrological models have proliferated in the past several decades prompting debates on the virtues and shortcomings of various modelling approaches. Rather than critiquing individual models or modelling approaches, the objective here is to address the critical issues of scaling and hydrological process representation in various types of models with suggestions for improving these attributes in a parsimonious manner that captures and explains their functionality as simply as possible. This discussion focuses mostly on conceptual and physical/process-based models where understanding the internal catchment processes and hydrologic pathways is important. Such hydrological models can be improved by using data from advanced remote sensing (both spatial and temporal) and derivatives, applications of machine learning, flexible structures, and informing models through nested catchment studies in which internal catchment processes are elucidated. Incorporating concepts of hydrological connectivity into flexible model structures is a promising approach for improving flow path representation. Also important is consideration of the scale dependency of hydrological parameters to avoid scale mismatch between measured and modelled parameters. Examples are presented from remote high-elevation regions where water sources and pathways differ from temperate and tropical environments where more attention has been focused. The challenge of incorporating spatially and temporally variable water inputs, hydrologically pathways, climate, and land use into hydrological models requires modellers to collaborate with catchment hydrologists to include important processes at relevant scales—i.e. develop smarter hydrological models.

On the Use of an Inflatable Rubber Lip to Improve the Reverse Thrust Flow Field in a Variable Pitch Fan

The installed Variable Pitch Fan (VPF) reverse thrust flow field is obtained from the flow solution of an integrated airframe-engine research model for the complete reverser engagement regime during the aircraft landing run from 140 knots to 40 knots. The model includes a twin-engine airframe, complete flow path representation of a future 40000 lbf high bypass ratio geared turbofan engine, and a bespoke reverse flow-capable VPF design. The reverse thrust flow field, at all speeds, indicates that the reverse flow out of the nacelle inlet is washed downstream by the freestream towards the engine exit regions. Consequently, reverse flow enters the engine through the bypass nozzle from a 180° turn of the washed-down stream. This results in a region of circumferentially varying separated flow at the nozzle lip that acts as a blockage to the reverse flow entry into the engine. To mitigate the blockage issue, a smooth guidance of the reverse flow into the engine to avoid separation can be achieved by using an inflatable rubber lip that would define a bell-mouth like geometric feature with a round radius at the nacelle exit region. In nominal engine operation, the rubber lip would be stowed flush within the contours of the optimized nacelle surface. The design space of the rubber lip is studied by considering different rounding radii and locations of the turn radius with respect to the nacelle trailing edge. The choices of the design parameters are chosen by considering the nacelle edge thickness, inflation air volume requirement, weight, and thickness of support structures. The effect of these designs on the reverse thrust flow field is studied by incorporating the designs into the integrated model, with realistic installation related restrictions. It is observed that a rounding radius of 0.1x nacelle length is sufficient to reduce the blockage and increase the ingested reverse flow by 47% to 18% in the 140 to 40 knots landing speed range. The inflatable rubber lip represents a design modification that can aid in the improvement of VPF reverse thrust operation, in cases where an augmentation of reverse thrust capability over the baseline is desired.

A Generic Interface Enabling Combinations of State-of-the-Art Path Planning and Tracking Algorithms

In the development of Level 4 automated driving functions, very specific, but diverse, requirements with respect to the operational design domain have to be considered. In order to accelerate this development, it is advantageous to combine dedicated state-of-the-art software components, as building blocks in modular automated driving function architectures, instead of developing special solutions from scratch. However, e.g., in local motion planning and control, the combination of components is still limited in practice, due to necessary interface alignments, which might yield sub-optimal solutions and additional development overhead. The application of generic interfaces, which manage the data transfer between the software components, has the potential to avoid these drawbacks and hence, to further boost this development approach. This publication contributes such a generic interface concept between the local path planning and path tracking systems. The crucial point is a generalization of the lateral tracking error computation, based on an introduced error classification. It substantiates the integration of an internal reference path representation into the interface, to resolve the component interdependencies. The resulting, proposed interface enables arbitrary combinations of components from a comprehensive set of state-of-the-art path planning and tracking algorithms. Two interface implementations are finally applied in an exemplary automated driving function assembly task.

The image edge model

Noise from traffic, industry and neighborhood is a prominent feature in urban environments. In these environments, sound reaches receiver points through reflections and diffractions. Real-time auralization of outdoor scenarios is a common goal for presenting sound characteristics in a realistic and intuitive fashion. Challenges in this attempt can be identified on many levels, however the most prominent part is sound propagation simulation. Geometrical acoustics has become the de-facto standard for the prediction of acoustic propagation in a virtual scenario. A considerable difficulty is the determination of the diffracted sound field component, because it is a wave effect that must be be explicitly integrated into the search algorithm of valid propagation paths. A deterministic solution to this problem is implemented that establishes propagation paths with an arbitrary constellation of far-field interactions at geometrical boundaries, i.e. reflecting surfaces and diffracting edges in large distance to each other. The result is an open-source code algorithm for propagation paths that follows the wave front normal and assembles metadata required for further acoustic modelling, such as incoming and outgoing angles, reflection material and geometrical details for the construction of the diffracting wedge. Calculation times are outlined and a proof of concept is presented that describes the employment of the propagation algorithm as well as the determination of an acoustic transfer function based on the input of the intermediate path representation. Future research will focus on prioritization of path contributions according to physical and psychoacoustical culling schemes.

Inferring Drug-Related Diseases Based on Convolutional Neural Network and Gated Recurrent Unit

Predicting novel uses for drugs using their chemical, pharmacological, and indication information contributes to minimizing costs and development periods. Most previous prediction methods focused on integrating the similarity and association information of drugs and diseases. However, they tended to construct shallow prediction models to predict drug-associated diseases, which make deeply integrating the information difficult. Further, path information between drugs and diseases is important auxiliary information for association prediction, while it is not deeply integrated. We present a deep learning-based method, CGARDP, for predicting drug-related candidate disease indications. CGARDP establishes a feature matrix by exploiting a variety of biological premises related to drugs and diseases. A novel model based on convolutional neural network (CNN) and gated recurrent unit (GRU) is constructed to learn the local and path representations for a drug-disease pair. The CNN-based framework on the left of the model learns the local representation of the drug-disease pair from their feature matrix. As the different paths have discriminative contributions to the drug-disease association prediction, we construct an attention mechanism at the path level to learn the informative paths. In the right part, a GRU-based framework learns the path representation based on path information between the drug and the disease. Cross-validation results indicate that CGARDP performs better than several state-of-the-art methods. Further, CGARDP retrieves more real drug-disease associations in the top part of the prediction result that are of concern to biologists. Case studies on five drugs demonstrate that CGARDP can discover potential drug-related disease indications.

Convolutional Neural Network and Bidirectional Long Short-Term Memory-Based Method for Predicting Drug–Disease Associations

Identifying novel indications for approved drugs can accelerate drug development and reduce research costs. Most previous studies used shallow models for prioritizing the potential drug-related diseases and failed to deeply integrate the paths between drugs and diseases which may contain additional association information. A deep-learning-based method for predicting drug–disease associations by integrating useful information is needed. We proposed a novel method based on a convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM)—CBPred—for predicting drug-related diseases. Our method deeply integrates similarities and associations between drugs and diseases, and paths among drug-disease pairs. The CNN-based framework focuses on learning the original representation of a drug-disease pair from their similarities and associations. As the drug-disease association possibility also depends on the multiple paths between them, the BiLSTM-based framework mainly learns the path representation of the drug-disease pair. In addition, considering that different paths have discriminate contributions to the association prediction, an attention mechanism at path level is constructed. Our method, CBPred, showed better performance and retrieved more real associations in the front of the results, which is more important for biologists. Case studies further confirmed that CBPred can discover potential drug-disease associations.

Optimal scheduling for aerial recovery of multiple unmanned aerial vehicles using genetic algorithm

The ability to deploy multiple unmanned aerial vehicles expands their application range, but aerial recovery of unmanned aerial vehicles presents many unique challenges owing to the number of unmanned aerial vehicles and the limited recovery time. In this paper, scheduling the aerial recovery of multiple unmanned aerial vehicles by one mothership is posed as a combinatorial optimization problem. A mathematical model with recovery time windows of the unmanned aerial vehicles is developed to formulate this problem. Furthermore, a genetic algorithm is proposed for finding the optimal recovery sequence. The algorithm adopts the path representation of chromosomes to simplify the encoding process and the genetic operations. It also resolves decoding difficulties by iteration, and thus can efficiently generate a recovery timetable for the unmanned aerial vehicles. Simulation results in stochastic scenarios validate the performance of the proposed algorithm compared with the random search algorithm and the greedy algorithm.