Local Competition-Based Superpixel Segmentation Algorithm in Remote Sensing.

Remote sensing technologies have been widely applied in urban environments' monitoring, synthesis and modeling. Incorporating spatial information in perceptually coherent regions, superpixel-based approaches can effectively eliminate the "salt and pepper" phenomenon which is common in pixel-wise approaches. Compared with fixed-size windows, superpixels have adaptive sizes and shapes for different spatial structures. Moreover, superpixel-based algorithms can significantly improve computational efficiency owing to the greatly reduced number of image primitives. Hence, the superpixel algorithm, as a preprocessing technique, is more and more popularly used in remote sensing and many other fields. In this paper, we propose a superpixel segmentation algorithm called Superpixel Segmentation with Local Competition (SSLC), which utilizes a local competition mechanism to construct energy terms and label pixels. The local competition mechanism leads to energy terms locality and relativity, and thus, the proposed algorithm is less sensitive to the diversity of image content and scene layout. Consequently, SSLC could achieve consistent performance in different image regions. In addition, the Probability Density Function (PDF), which is estimated by Kernel Density Estimation (KDE) with the Gaussian kernel, is introduced to describe the color distribution of superpixels as a more sophisticated and accurate measure. To reduce computational complexity, a boundary optimization framework is introduced to only handle boundary pixels instead of the whole image. We conduct experiments to benchmark the proposed algorithm with the other state-of-the-art ones on the Berkeley Segmentation Dataset (BSD) and remote sensing images. Results demonstrate that the SSLC algorithm yields the best overall performance, while the computation time-efficiency is still competitive.

A hierarchical framework approach for voice activity detection and speech enhancement.

Accurate and effective voice activity detection (VAD) is a fundamental step for robust speech or speaker recognition. In this study, we proposed a hierarchical framework approach for VAD and speech enhancement. The modified Wiener filter (MWF) approach is utilized for noise reduction in the speech enhancement block. For the feature selection and voting block, several discriminating features were employed in a voting paradigm for the consideration of reliability and discriminative power. Effectiveness of the proposed approach is compared and evaluated to other VAD techniques by using two well-known databases, namely, TIMIT database and NOISEX-92 database. Experimental results show that the proposed method performs well under a variety of noisy conditions.

Cooperative search and rescue with artificial fishes based on fish-swarm algorithm for underwater wireless sensor networks.

This paper presents a searching control approach for cooperating mobile sensor networks. We use a density function to represent the frequency of distress signals issued by victims. The mobile nodes' moving in mission space is similar to the behaviors of fish-swarm in water. So, we take the mobile node as artificial fish node and define its operations by a probabilistic model over a limited range. A fish-swarm based algorithm is designed requiring local information at each fish node and maximizing the joint detection probabilities of distress signals. Optimization of formation is also considered for the searching control approach and is optimized by fish-swarm algorithm. Simulation results include two schemes: preset route and random walks, and it is showed that the control scheme has adaptive and effective properties.

TargetATPsite: a template-free method for ATP-binding sites prediction with residue evolution image sparse representation and classifier ensemble.

Understanding the interactions between proteins and ligands is critical for protein function annotations and drug discovery. We report a new sequence-based template-free predictor (TargetATPsite) to identify the Adenosine-5'-triphosphate (ATP) binding sites with machine-learning approaches. Two steps are implemented in TargetATPsite: binding residues and pockets predictions, respectively. To predict the binding residues, a novel image sparse representation technique is proposed to encode residue evolution information treated as the input features. An ensemble classifier constructed based on support vector machines (SVM) from multiple random under-samplings is used as the prediction model, which is effective for dealing with imbalance phenomenon between the positive and negative training samples. Compared with the existing ATP-specific sequence-based predictors, TargetATPsite is featured by the second step of possessing the capability of further identifying the binding pockets from the predicted binding residues through a spatial clustering algorithm. Experimental results on three benchmark datasets demonstrate the efficacy of TargetATPsite.

Learning protein multi-view features in complex space.

Protein attribute prediction from primary sequences is an important task and how to extract discriminative features is one of the most crucial aspects. Because single-view feature cannot reflect all the information of a protein, fusing multi-view features is considered as a promising route to improve prediction accuracy. In this paper, we propose a novel framework for protein multi-view feature fusion: first, features from different views are parallely combined to form complex feature vectors; Then, we extend the classic principal component analysis to the generalized principle component analysis for further feature extraction from the parallely combined complex features, which lie in a complex space. Finally, the extracted features are used for prediction. Experimental results on different benchmark datasets and machine learning algorithms demonstrate that parallel strategy outperforms the traditional serial approach and is particularly helpful for extracting the core information buried among multi-view feature sets. A web server for protein structural class prediction based on the proposed method (COMSPA) is freely available for academic use at: http://www.csbio.sjtu.edu.cn/bioinf/COMSPA/ .

Depth and Video Segmentation Based Visual Attention for Embodied Question Answering.

Embodied Question Answering (EQA) is a newly defined research area where an agent is required to answer the user's questions by exploring the real-world environment. It has attracted increasing research interests due to its broad applications in personal assistants and in-home robots. Most of the existing methods perform poorly in terms of answering and navigation accuracy due to the absence of fine-level semantic information, stability to the ambiguity, and 3D spatial information of the virtual environment. To tackle these problems, we propose a depth and segmentation based visual attention mechanism for Embodied Question Answering. First, we extract local semantic features by introducing a novel high-speed video segmentation framework. Then guided by the extracted semantic features, a depth and segmentation based visual attention mechanism is proposed for the Visual Question Answering (VQA) sub-task. Further, a feature fusion strategy is designed to guide the navigator's training process without much additional computational cost. The ablation experiments show that our method effectively boosts the performance of the VQA module and navigation module, leading to 4.9 % and 5.6 % overall improvement in EQA accuracy on House3D and Matterport3D datasets respectively.

Infrared human tracking with improved mean shift algorithm based on multicue fusion.

Mean shift has been presented as a well known and efficient algorithm for tracking infrared targets. However, under complex backgrounds, such as clutter, varying illumination, and occlusion, the traditional mean tracking method often converges to a local maximum and loses the real infrared target. To cope with these problems, an improved mean shift tracking algorithm based on multicue fusion is proposed. According to the characteristics of the human in infrared images, the algorithm first extracts the gray and edge cues, and then uses the motion information to guide the two cues to obtain improved motion-guided gray and edge cues that are fused adaptively into the mean shift framework. Finally an automatic model update is used to improve the tracking performance further. The experimental results show that, compared with the traditional mean shift algorithm, the presented method greatly improves the accuracy and effectiveness of infrared human tracking under complex scenes, and the tracking results are satisfactory.

Extracting Privileged Information for Enhancing Classifier Learning.

The accuracy of data-driven learning approaches is often unsatisfactory when the training data is inadequate either in quantity or quality. Manually labeled privileged information (PI), e.g., attributes, tags or properties, is usually incorporated to improve classifier learning. However, the process of manually labeling is time-consuming and labor-intensive. Moreover, due to the limitations of personal knowledge, manually labeled PI may not be rich enough. To address these issues, we propose to enhance classifier learning by exploring PI from untagged corpora, which can effectively eliminate the dependency on manually labeled data and obtain much richer PI. In detail, we treat each selected PI as a subcategory and learn one classifier for each subcategory independently. The classifiers for all subcategories are integrated together to form a more powerful category classifier. Particularly, we propose a novel instancelevel multi-instance learning (MIL) model to simultaneously select a subset of training images from each subcategory and learn the optimal SVM classifiers based on the selected images. Extensive experiments on four benchmark datasets demonstrate the superiority of our proposed approach.

Sparse representation-based ECG signal enhancement and QRS detection.

Electrocardiogram (ECG) signal enhancement and QRS complex detection is a critical preprocessing step for further heart disease analysis and diagnosis. In this paper, we propose a sparse representation-based ECG signal enhancement and QRS complex detection algorithm. Unlike traditional Fourier or wavelet transform-based methods, which use fixed bases, the proposed algorithm models the ECG signal as the superposition of a few inner structures plus additive random noise, where these structures (referred to here as atoms) can be learned from the input signal or a training set. Using these atoms and their properties, we can accurately approximate the original ECG signal and remove the noise and other artifacts such as baseline wandering. Additionally, some of the atoms with larger kurtosis values can be modified and used as an indication function to detect and locate the QRS complexes in the enhanced ECG signals. To demonstrate the robustness and efficacy of the proposed algorithm, we compare it with several state-of-the-art ECG enhancement and QRS detection algorithms using both simulated and real-life ECG recordings.

Hashing on nonlinear manifolds.

Learning-based hashing methods have attracted considerable attention due to their ability to greatly increase the scale at which existing algorithms may operate. Most of these methods are designed to generate binary codes preserving the Euclidean similarity in the original space. Manifold learning techniques, in contrast, are better able to model the intrinsic structure embedded in the original high-dimensional data. The complexities of these models, and the problems with out-of-sample data, have previously rendered them unsuitable for application to large-scale embedding, however. In this paper, how to learn compact binary embeddings on their intrinsic manifolds is considered. In order to address the above-mentioned difficulties, an efficient, inductive solution to the out-of-sample data problem, and a process by which nonparametric manifold learning may be used as the basis of a hashing method are proposed. The proposed approach thus allows the development of a range of new hashing techniques exploiting the flexibility of the wide variety of manifold learning approaches available. It is particularly shown that hashing on the basis of t-distributed stochastic neighbor embedding outperforms state-of-the-art hashing methods on large-scale benchmark data sets, and is very effective for image classification with very short code lengths. It is shown that the proposed framework can be further improved, for example, by minimizing the quantization error with learned orthogonal rotations without much computation overhead. In addition, a supervised inductive manifold hashing framework is developed by incorporating the label information, which is shown to greatly advance the semantic retrieval performance.

Constructing query-driven dynamic machine learning model with application to protein-ligand binding sites prediction.

We are facing an era with annotated biological data rapidly and continuously generated. How to effectively incorporate new annotated data into the learning step is crucial for enhancing the performance of a bioinformatics prediction model. Although machine-learning-based methods have been extensively used for dealing with various biological problems, existing approaches usually train static prediction models based on fixed training datasets. The static approaches are found having several disadvantages such as low scalability and impractical when training dataset is huge. In view of this, we propose a dynamic learning framework for constructing query-driven prediction models. The key difference between the proposed framework and the existing approaches is that the training set for the machine learning algorithm of the proposed framework is dynamically generated according to the query input, as opposed to training a general model regardless of queries in traditional static methods. Accordingly, a query-driven predictor based on the smaller set of data specifically selected from the entire annotated base dataset will be applied on the query. The new way for constructing the dynamic model enables us capable of updating the annotated base dataset flexibly and using the most relevant core subset as the training set makes the constructed model having better generalization ability on the query, showing "part could be better than all" phenomenon. According to the new framework, we have implemented a dynamic protein-ligand binding sites predictor called OSML (On-site model for ligand binding sites prediction). Computer experiments on 10 different ligand types of three hierarchically organized levels show that OSML outperforms most existing predictors. The results indicate that the current dynamic framework is a promising future direction for bridging the gap between the rapidly accumulated annotated biological data and the effective machine-learning-based predictors. OSML web server and datasets are freely available at: http://www.csbio.sjtu.edu.cn/bioinf/OSML/ for academic use.

Approximate least trimmed sum of squares fitting and applications in image analysis.

The least trimmed sum of squares (LTS) regression estimation criterion is a robust statistical method for model fitting in the presence of outliers. Compared with the classical least squares estimator, which uses the entire data set for regression and is consequently sensitive to outliers, LTS identifies the outliers and fits to the remaining data points for improved accuracy. Exactly solving an LTS problem is NP-hard, but as we show here, LTS can be formulated as a concave minimization problem. Since it is usually tractable to globally solve a convex minimization or concave maximization problem in polynomial time, inspired by , we instead solve LTS' approximate complementary problem, which is convex minimization. We show that this complementary problem can be efficiently solved as a second order cone program. We thus propose an iterative procedure to approximately solve the original LTS problem. Our extensive experiments demonstrate that the proposed method is robust, efficient and scalable in dealing with problems where data are contaminated with outliers. We show several applications of our method in image analysis.

Enhancing membrane protein subcellular localization prediction by parallel fusion of multi-view features.

Membrane proteins are encoded by ~ 30% in the genome and function importantly in the living organisms. Previous studies have revealed that membrane proteins' structures and functions show obvious cell organelle-specific properties. Hence, it is highly desired to predict membrane protein's subcellular location from the primary sequence considering the extreme difficulties of membrane protein wet-lab studies. Although many models have been developed for predicting protein subcellular locations, only a few are specific to membrane proteins. Existing prediction approaches were constructed based on statistical machine learning algorithms with serial combination of multi-view features, i.e., different feature vectors are simply serially combined to form a super feature vector. However, such simple combination of features will simultaneously increase the information redundancy that could, in turn, deteriorate the final prediction accuracy. That's why it was often found that prediction success rates in the serial super space were even lower than those in a single-view space. The purpose of this paper is investigation of a proper method for fusing multiple multi-view protein sequential features for subcellular location predictions. Instead of serial strategy, we propose a novel parallel framework for fusing multiple membrane protein multi-view attributes that will represent protein samples in complex spaces. We also proposed generalized principle component analysis (GPCA) for feature reduction purpose in the complex geometry. All the experimental results through different machine learning algorithms on benchmark membrane protein subcellular localization datasets demonstrate that the newly proposed parallel strategy outperforms the traditional serial approach. We also demonstrate the efficacy of the parallel strategy on a soluble protein subcellular localization dataset indicating the parallel technique is flexible to suite for other computational biology problems. The software and datasets are available at: http://www.csbio.sjtu.edu.cn/bioinf/mpsp.

Thinning, fertilization, and crown position interact to control physiological responses of loblolly pine.

To examine physiological responses to thinning, fertilization, and crown position, we measured net photosynthesis (P(n)), transpiration (E), vapor pressure difference (VPD), stomatal conductance (g(s)), and xylem pressure potential (Psi(1)) between 0930 and 1130 h under ambient conditions in the upper and lower crowns of a 13-year-old loblolly pine (Pinus taeda L.) plantation six years (1994) after the treatments were applied. Photosynthetic photon flux density (PPFD) and air temperature (T(a)) within the canopy were also recorded. Needle P(n) of thinned trees was significantly enhanced by 22-54% in the lower crown, because canopy PPFD increased by 28-52%. Lower crown foliage of thinned plots also had higher E and g(s) than foliage of unthinned plots, but thinning had no effect on needle Psi(1) and predawn xylem pressure potential (0430-0530 h; Psi(pd)). Tree water status did not limit P(n), E and g(s) during the late-morning measurements. Fertilization significantly decreased within-canopy PPFD and T(a). Needle Psi(1) was increased in fertilized stands, whereas P(n), E and g(s) were not significantly altered. Upper crown foliage had significantly greater PPFD, P(n), VPD, g(s), E, and more negative Psi(1) than lower crown foliage. In both crown positions, needle P(n) was closely related to g(s), PPFD and T(a) (R(2) = 0.77 for the upper crown and 0.82 for the lower crown). We conclude that (1) silvicultural manipulation causes microclimate changes within the crowns of large trees, and (2) needle physiology adjusts to the within-crown environmental conditions.