Xintao Ding

Probability method for Cerenkov luminescence tomography based on conformance error minimization

Cerenkov luminescence tomography (CLT) was developed to reconstruct a three-dimensional (3D) distribution of radioactive probes inside a living animal. Reconstruction methods are generally performed within a unique framework by searching for the optimum solution. However, the ill-posed aspect of the inverse problem usually results in the reconstruction being non-robust. In addition, the reconstructed result may not match reality since the difference between the highest and lowest uptakes of the resulting radiotracers may be considerably large, therefore the biological significance is lost. In this paper, based on the minimization of a conformance error, a probability method is proposed that consists of qualitative and quantitative modules. The proposed method first pinpoints the organ that contains the light source. Next, we developed a 0-1 linear optimization subject to a space constraint to model the CLT inverse problem, which was transformed into a forward problem by employing a region growing method to solve the optimization. After running through all of the elements used to grow the sources, a source sequence was obtained. Finally, the probability of each discrete node being the light source inside the organ was reconstructed. One numerical study and two in vivo experiments were conducted to verify the performance of the proposed algorithm, and comparisons were carried out using the hp-finite element method (hp-FEM). The results suggested that our proposed probability method was more robust and reasonable than hp-FEM.

Indoor object recognition using pre-trained convolutional neural network

Indoor object recognition is a key task for mobile robot indoor navigation. In this paper, we proposed a pipeline for indoor object detection based on convolutional neural network (CNN). With the proposed method, we first pre-train an off-line CNN model by using both public Indoor Dataset and private frames of videos (FoV) dataset. This is then followed by a selective search process to extract a region of interest (RoI) after the input video was parsed into frame images. The extracted RoIs are then classified into candidates using the pre-trained deep model and the candidates between the nearest frame images are refined using detection fusion. Finally, the annotated frames are merged to create video as the output. The experiments show that our design is very efficient against indoor object detection.

Outlier-eliminated k-means clustering algorithm based on differential privacy preservation

Individual privacy may be compromised during the process of mining for valuable information, and the potential for data mining is hindered by the need to preserve privacy. It is well known that k-means clustering algorithms based on differential privacy require preserving privacy while maintaining the availability of clustering. However, it is difficult to balance both aspects in traditional algorithms. In this paper, an outlier-eliminated differential privacy (OEDP) k-means algorithm is proposed that both preserves privacy and improves clustering efficiency. The proposed approach selects the initial centre points in accordance with the distribution density of data points, and adds Laplacian noise to the original data for privacy preservation. Both a theoretical analysis and comparative experiments were conducted. The theoretical analysis shows that the proposed algorithm satisfies ε-differential privacy. Furthermore, the experimental results show that, compared to other methods, the proposed algorithm effectively preserves data privacy and improves the clustering results in terms of accuracy, stability, and availability.

Path Planning and Obstacle Avoidance for Mobile Robots in a Dynamic Environment

Because traditional obstacle avoidance path planning methods have a lot of problems, such as large amount of calculation, low efficiency, poor optimization capability, and lack of dealing with dynamic obstacles, a new method which implements real-time path planning of mobile robot is presented. The method builds a neural network model for the robot workspace, and then it uses the model to obtain the relationship between the dynamic obstacles and the network output. It can choose the local optimal collision-free path by the path planning in a dynamic environment (PPIDE) algorithm to find the path between two points for dealing with obstacles. The proposed method is suitable for dynamic environment where both linear and planar obstacles exist. Simulation results prove its effectiveness.

A novel artificial immune algorithm for spatial clustering with obstacle constraint and its applications

An important component of a spatial clustering algorithm is the distance measure between sample points in object space. In this paper, the traditional Euclidean distance measure is replaced with innovative obstacle distance measure for spatial clustering under obstacle constraints. Firstly, we present a path searching algorithm to approximate the obstacle distance between two points for dealing with obstacles and facilitators. Taking obstacle distance as similarity metric, we subsequently propose the artificial immune clustering with obstacle entity (AICOE) algorithm for clustering spatial point data in the presence of obstacles and facilitators. Finally, the paper presents a comparative analysis of AICOE algorithm and the classical clustering algorithms. Our clustering model based on artificial immune system is also applied to the case of public facility location problem in order to establish the practical applicability of our approach. By using the clone selection principle and updating the cluster centers based on the elite antibodies, the AICOE algorithm is able to achieve the global optimum and better clustering effect.

A novel social network hybrid recommender system based on hypergraph topologic structure

With the advent and popularity of social network, more and more people like to share their experience in social network. However, network information is growing exponentially which leads to information overload. Recommender system is an effective way to solve this problem. The current research on recommender systems is mainly focused on research models and algorithms in social networks, and the social networks structure of recommender systems has not been analyzed thoroughly and the so-called cold start problem has not been resolved effectively. We in this paper propose a novel hybrid recommender system called Hybrid Matrix Factorization(HMF) model which uses hypergraph topology to describe and analyze the interior relation of social network in the system. More factors including contextual information, user feature, item feature and similarity of users ratings are all taken into account based on matrix factorization method. Extensive experimental evaluation on publicly available datasets demonstrate that the proposed hybrid recommender system outperforms the existing recommender systems in tackling cold start problem and dealing with sparse rating datasets. Our system also enjoys improved recommendation accuracy compared with several major existing recommendation approaches.

Prior knowledge-based deep learning method for indoor object recognition and application

ABSTRACT Indoor object recognition is a key task for indoor navigation by mobile robots. Although previous work has produced impressive results in recognizing known and familiar objects, the research of indoor object recognition for robot is still insufficient. In order to improve the detection precision, our study proposed a prior knowledge-based deep learning method aimed to enable the robot to recognize indoor objects on sight. First, we integrate the public Indoor dataset and the private frames of videos (FoVs) dataset to train a convolutional neural network (CNN). Second, mean images, which are used as a type of colour knowledge, are generated for all the classes in the Indoor dataset. The distance between every mean image and the input image produces the class weight vector. Scene knowledge, which consists of frequencies of occurrence of objects in the scene, is then employed as another prior knowledge to determine the scene weight. Finally, when a detection request is launched, the two vectors together with a vector of classification probability instigated by the deep model are multiplied to produce a decision vector for classification. Experiments show that detection precision can be improved by employing the prior colour and scene knowledge. In addition, we applied the method to object recognition in a video. The results showed potential application of the method for robot vision.

An Efficient and Secure Itinerary-Based Data Aggregation Algorithm for WSNs

The existing privacy-preserving data aggregation methods in wireless sensor networks (WSNs) generally rely on a network infrastructure, and data privacy is achieved by encryption techniques. However, such an infrastructure is very susceptible to the dynamic network topologies, and excessive encryption process causes a high energy consumption and re-duces the accuracy of the aggregation results. In this paper, we propose a secure and concentric-circle itinerary-based data aggregation algorithm (called SCIDA for short). With the help of a well-designed itinerary for aggregation propagation and data aggregation, SCIDA is not susceptible to network topology structure and thus suitable for wireless sensor net-works with dynamic network topologies and can save energy for network infrastructure maintenance. In addition, SCIDA uses a secure channel to ensure data privacy and avoids dramatic energy consumption caused by heavy encryption operations. SCIDA does not need to carry out encryption during data aggregation, which significantly reduces energy consumption, and prolongs the lifetime of the network. Theoretical analysis and experimental results show that SCIDA enjoys low communication overhead and energy con-sumption, yet high safety and accuracy.

Orthogonal design for scale invariant feature transform optimization

Abstract. To improve object recognition capabilities in applications, we used orthogonal design (OD) to choose a group of optimal parameters in the parameter space of scale invariant feature transform (SIFT). In the case of global optimization (GOP) and local optimization (LOP) objectives, our aim is to show the operation of OD on the SIFT method. The GOP aims to increase the number of correctly detected true matches (NoCDTM) and the ratio of NoCDTM to all matches. In contrast, the LOP mainly aims to increase the performance of recall–precision. In detail, we first abstracted the SIFT method to a 9-way fixed-effect model with an interaction. Second, we designed a mixed orthogonal array, MA(64,23420,2), and its header table to optimize the SIFT parameters. Finally, two groups of parameters were obtained for GOP and LOP after orthogonal experiments and statistical analyses were implemented. Our experiments on four groups of data demonstrate that compared with the state-of-the-art methods, GOP can access more correct matches and is more effective against object recognition. In addition, LOP is favorable in terms of the recall–precision.