Ning Jing

Pointly Supervised Object Detection (PSOD) has attracted considerable interests due to its lower labeling cost as compared to box-level supervised object detection. However, the complex scenes, densely packed and dynamic-scale objects in Remote Sensing (RS) images hinder the development of PSOD methods in RS field. In this paper, we make the first attempt to achieve RS object detection with single point supervision, and propose a PSOD method tailored for RS images. Specifically, we design a point label upgrader (PLUG) to generate pseudo box labels from single point labels, and then use the pseudo boxes to supervise the optimization of existing detectors. Moreover, to handle the challenge of the densely packed objects in RS images, we propose a sparse feature guided semantic prediction module which can generate high-quality semantic maps by fully exploiting informative cues from sparse objects. Extensive ablation studies on the DOTA dataset have validated the effectiveness of our method. Our method can achieve significantly better performance as compared to state-of-the-art image-level and point-level supervised detection methods, and reduce the performance gap between PSOD and box-level supervised object detection. Code is available at https://github.com/heshitian/PLUG.

Hunting is an important factor in the safe operation of high-speed trains. Most techniques used for monitoring hunting aim at detecting hunting occurrence and some of them can deal with small-amplitude hunting. However, they do not provide information about evolution of small-amplitude hunting. The present work studies the evolution of small-amplitude hunting with the goal of predicting the occurrence of the hunting instability. An improved method contained multiple hidden for predicting the interval of the amplitude of small-amplitude hunting is proposed, which is improved via a two-level model. The method is computationally efficient and converges rapidly. Upon applying the proposed method to high-speed train data, the coverage probability of the resulting prediction interval is 100% and its normalized average width is 0.187, which means a higher coverage and smaller width than the interval predicted by existing methods. The confidence-level of the prediction is also high.

The hunting motion of the vehicle increases the wear of the wheel tracks and affects the lateral stability of the vehicle system and safety, so the hunting motion of the vehicle needs to be recognized. However, the existing recognition methods ignore the small-amplitude hunting motion that precedes the onset of the running motion. In addition, due to the scarcity of real hunting data of high-speed trains, there is a problem of underfitting when training with traditional deep learning methods. In this paper, first, the dynamics model of a high-speed train is established, and the normal, small-amplitude hunting and hunting motions of high-speed trains in the process are simulated. Second, this paper proposes a transfer learning-based method for high-speed train hunting motion recognition. The method uses easily collected normal data samples in the training process, does not use real data samples of small-amplitude hunting and hunting motions, and completes the high-speed train running motions recognition task by transferring from simulation data (source domain) to real data (target domain). Finally, the validation is carried out using the real data, which proves that the method-related approach has some engineering application value in the intelligent monitoring of high-speed trains.

Joint Approximate Diagonalization of Eigen-matrices (JADE) cannot deal with non-stationary data. Therefore, in this paper, a method called Non-stationary Kernel JADE (NKJADE) is proposed, which can extract non-stationary features and fuse multi-sensor features precisely and rapidly. In this method, the non-stationarity of the data is considered and the data from multi-sensor are used to fuse the features efficiently. The method is compared with EEMD-SVD-LTSA and EEMD-JADE using the bearing fault data of CWRU, and the validity of the method is verified. Considering that the vibration signals of high-speed trains are typically non-stationary, it is necessary to utilize a rapid feature fusion method to identify the evolutionary trends of hunting motions quickly before the phenomenon is fully manifested. In this paper, the proposed method is applied to identify the evolutionary trend of hunting motions quickly and accurately. Results verify that the accuracy of this method is much higher than that of the EEMD-JADE and EEMD-SVD-LTSA methods. This method can also be used to fuse multi-sensor features of non-stationary data rapidly.

The relevance of geographic entities has long been the focus of research on geographic information retrieval, geographic knowledge graph, and recommendation systems. Traditional research methods use spatial similarity or semantic similarity to calculate the correlation strength between regions. The calculation process is complex and has certain one-sidedness and limitations, and it is difficult to reveal the essential characteristics of geographic entities. The rapid development of graph models and related algorithms provides researchers with new ideas. By constructing network topology, it focuses on the relationship between entities and is more suitable for mining potential correlations between entities. This study attempts to explore the geographic entity association of integrated semantics and spatial factors by using network mining algorithms. At the same time, unlike traditional semantic information, this paper uses text data as a source of semantically related information. Experiments show that this method can effectively represent geographical entity associations and reflect regional characteristics.

We interpret the geographic world as a combination of geographic entities (such as city, road, river). Using these geographic entities as the basic granularity for description, storage and manipulation, therefore, can provide a bridge between conceptual understanding and practical application. However, the issue arises of how to design a data model that is generic enough for the geographic entities when the data can change spatially and temporally. This difficulty is compounded when there is a need for geographic inference. In this paper, we offer a feasible solution by presenting a Spatio-Temporal Data Model of Geographic Entities (STDMGE), which describes each geographic entity as geographic object part and geographic relation part, both of which are well designed to organize the multi-source data (traditional geospatial data, sensor data, relationship data among geographic entities, etc.). Spatial-temporal changes can be accurately expressed and efficiently implemented using this model. Moreover, the formal description of geographic rules and events are incorporated in the data model for the possible future requirements of geographic reasoning. We further illustrate the data model by applying it to the case study of the river data.

The electromagnetic detection satellite (EDS) is a type of earth observation satellites (EOSs). The Information collected by EDSs plays an important role in some fields, such as industry, science and military. The scheduling of EDSs is a complex combinatorial optimization problem. Current research mainly focuses on the scheduling of imaging satellites and SAR satellites, but little work has been done on the scheduling of EDSs for its specific characteristics. A multi-satellite scheduling model is established, in which the specific constrains of EDSs are considered, then a scheduling algorithm based on the genetic algorithm (GA) is proposed. To deal with the specific constrains of EDSs, a penalty function method is introduced. However, it is hard to determine the appropriate penalty coefficient in the penalty function. Therefore, an adaptive adjustment mechanism of the penalty coefficient is designed to solve the problem, as well as improve the scheduling results. Experimental results are used to demonstrate the correctness and practicability of the proposed scheduling algorithm.