Search

Modeling target detection problem given inaccurate annotations as a multiple instance learning (MIL) problem is an effective way for addressing the ground truth uncertainties of remotely sensed hyperspectral imagery. In this paper, we propose a hyperspectral target detection method based on 1D convolution neural network (1DCNN) feature extraction and long short term memory network (LSTM) under the MIL framework, where the LSTM features for each hyperspectral pixel is further refined by a scoring network as to discriminate the real target instance from the inaccurately labeled hyperspectral regions. The proposed method has achieved superior results on both simulated data and real hyperspectral data over the state-of-the-art methods, showing the prospects for further investigation.

Radar systems have been widely used in many fields, where real-time processing and re-configurability of the hardware system are in strong demand. This paper presents the prototype of a System-on-Chip (SoC) architecture of a reconfigurable, high performance, low cost and real-time Radar Signal Processing system. It integrates both the microprocessor and the Field-Programmable Gate Array (FPGA) into a single chip, which makes the system capable of allowing complex system configuration and performing high speed data processing. The proposed SoC system is highly reconfigurable so that the software and hardware configuration can be reconfigured as microprocessor or FPGA.

Template meta-programming is a hot topic in programming languages, and functional programming is often employed in template meta-programming. ATS (Applied Type System) is a functional programming language includes formal specification and implementation, and it is good to be used for generating templates. Since ATS templates are difficult to be designed and lack generality, a method for generating ATS template based on model transformation is proposed in this paper. Firstly, an instance for generating Function from Datatype is given based on ATS Template Meta-programming; Secondly, an ATS meta-model includes the definitions of Datatype and Function is constructed by meta-modeling; Thirdly, the transformation from Datatype model to Function model is described in detail; Finally, an example of ATS Template Meta-programming based on model transformation is given to verify the effectiveness of this method. The experiments show that this method can implement ATS templates and increase the generality of ATS in template meta-programming.

Multiple instance learning (MIL) can be used for solving the imprecisely labeled hyperspectral target detection problems, which only needs the label of an area containing some targets. Furthermore, existing methods decompose this task into a target signature learning task and a follow-on similarity measurement between the estimated signature and the test points. In this paper, we propose a deep multiple instance learning method based on self-attention mechanism, in which the max operation and 1D convolution neural network (1D CNN) are adopted to realize an end-to-end hyperspectral target detection structure without learning the target signature. In the proposed deep MIL target detection method, self-attention mechanism with max operation has advantage in estimating the labels of instances from the positive bag via calculating the contribution of each instance to the bag-level classification. The simulated and real hyperspectral target detection experiments are shown to illustrate the performance of the method.

TA126 solve the job shop dynamic scheduling problem with disruptions, an integrated dynamic scheduling method based on shared mental models is proposed in this paper. Each manufacturing unit in the workshop is mapped into an intelligent agent with the individual mental model for the specific scheduling scenarios. After disturbance occurs, the Q-learning reinforcement learning method is adopted to achieve conflict resolution and consensus consultation among agents, and produce the globally optimized scheduling scheme.

With the development of on-line social networks, more and more users and sellers are interested in group analytics which provides insights into common interests with various relationships. This paper addresses this problem from temporal aspect, i.e., temporal group query (TGQ) which gives people the historical view for group forming and changing. To efficiently achieve our goal, we propose two structures to index temporal social network (TSN), and then a simple naive searching method is designed to process the TGQ, after that we argue a more efficient approach can be implemented by update operations instead of iterative graph generations, which we call optimized method. We conduct experiments on real-synthetic dataset, and the results show that our indexes and searching algorithms are capable, and optimized method is much more efficient than the naive one.

With the development of various Cloud platforms, providing spatio-temporal database services is an essential requirement for many applications, e.g., location-based services in Cloud. However, many previous works on processing queries in distributed environment fail to apply to spatio-temporal queries which is a significant role in spatio-temporal database. In this paper, we propose an efficient and scalable index for answering spatio-temporal queries in the Cloud. The index is a peer-to-peer-based overlay network, which is composed of two ring-structured overlays: spatial ring globally indexing spatial dimension and temporal ring for the temporal one. And it is featured by cost-aware function, i.e., a query is always able to be accomplished at a low cost, utilized by histograms keeping the distributions of Cloud nodes and data, maintained by each Cloud node. Both range query and kNN query can benefit from the mechanism, additionally, an elaborate algorithm for kNN processing is proposed, with which Cloud nodes can deliberatively send messages only to the result-related destinations. Furthermore, optimizations are also proposed for achieving low cost index maintenance and scalable kNN query processing. Experiments on both synthetic and real dataset show that our index is capable to support efficient and scalable range and kNN query, even for a skewed distribution.

Geoscience gives insights into our surroundings and benefits many aspects of our life. Nowadays, with massive sensors deployed to sense all kinds of parameters for environments, tens of billions, even trillions of sensed data are collected and need to be analyzed for surveillance or other purposes. From many perspectives, users always issue queries according to specific spatial and temporal predicates. For these applications, relational databases are overwhelmed by the large scale and high rate insertions, and NoSQL database could be considered a feasible solution. HBase, a popular key-value store system, is capable to solve the storage problem, but fails to provide in-built spatio-temporal querying capability. Many previous works tackle the problem by designing schema, i.e., designing row key and column key formation for HBase, which we don't believe is an effective solution. In this paper, we address this problem from nature level of HBase, and propose an index structure as a built-in component for HBase. STEHIX (Spatio-TEmporal Hbase IndeX) is adapted to two-level architecture of HBase and suitable for HBase to process spatio-temporal queries. It is composed of index in the meta table (the first level) and region index (the second level) for indexing inner structure of HBase regions. Base on this structure, two common queries, range query and kNN query are solved by proposing algorithms, respectively. For achieving load balancing and scalable kNN query, two optimizations are also presented. We implement STEHIX and conduct experiments on real dataset, and the results show our design outperforms a previous work in many aspects.

The acute toxicity of triazoles to earthworms was studied in using a simple paper contact method proposed by OECD testing guideline no. 207. It was found the median lethal concentration (LC50) had no correlation with log Pow (1-octanol/water partition coefficient) of the chemicals. However, the toxicity of triazoles was different for E. fetida and for fresh water organisms by comparing with other research. These results suggest the mechanism of toxicity was different between the effects of pollutants on living things in different environments, fresh water and soil. Therefore, the toxicity could not be explained simply by log Pow alone, even though it is one of the most important parameters.

Long and short vowels in Lhasa dialect is an important language phenomenon. In early studies, the long and short vowels are studied with long and short tones together as one kind of phenomena. Taking Lhasa dialect as an example, this paper summed up the similarities and differences between long and short vowels through statistical analysis and comparison of the two kinds of vowels from the aspects of articulation manners, tone quality, fundamental frequency and duration. On this basis, the paper discussed the role that the length of vowels played on the phonemes system.

The rapid increase of raw data improves a large number of data mining algorithms. But this trend makes it difficult to choose suited data mining algorithms to discover valid knowledge in the data repositories for users with few prior experience of data mining. In this paper, we propose a framework named Intelligent Knowledge Discovery System (IKDS) which help users to select appropriate data mining algorithms to discover useful knowledge. A novel knowledge acquisition methodology called Simplified- EMCUD (SIMCAP) is proposes in IKDS for its Expert system to elicit explicit knowledge as well as implicit knowledge. Then we represent and store the knowledge in IKDS by XML. At last we give the prototypes of SIMCAP and IKDS.