Statistical Fusion Research Articles

Omnibus proteome-wide association study identifies 43 risk genes for Alzheimer disease dementia

Transcriptome-wide association study (TWAS) tools have been applied to conduct proteome-wide association studies (PWASs) by integrating proteomics data with genome-wide association study (GWAS) summary data. The genetic effects of PWAS-identified significant genes are potentially mediated through genetically regulated protein abundance, thus informing the underlying disease mechanisms better than GWAS loci. However, existing TWAS/PWAS tools are limited by considering only one statistical model. We propose an omnibus PWAS pipeline to account for multiple statistical models and demonstrate improved performance by simulation and application studies of Alzheimer disease (AD) dementia. We employ the Aggregated Cauchy Association Test to derive omnibus PWAS (PWAS-O) p values from PWAS p values obtained by three existing tools assuming complementary statistical models-TIGAR, PrediXcan, and FUSION. Our simulation studies demonstrated improved power, with well-calibrated type I error, for PWAS-O over all three individual tools. We applied PWAS-O to studying AD dementia with reference proteomic data profiled from dorsolateral prefrontal cortex of postmortem brains from individuals of European ancestry. We identified 43 risk genes, including 5 not identified by previous studies, which are interconnected through a protein-protein interaction network that includes the well-known AD risk genes TOMM40, APOC1, and APOC2. We also validated causal genetic effects mediated through the proteome for 27 (63%) PWAS-O risk genes, providing insights into the underlying biological mechanisms of AD dementia and highlighting promising targets for therapeutic development. PWAS-O can be easily applied to studying other complex diseases.

An Integrated Deep Learning-Based Data Fusion and Degradation Modeling Method for Improving Prognostics

Accurate prognostics are crucially important to prevent unexpected failures in industrial and service systems. This process aims to monitor the degradation status of units and predict their remaining useful lifetime (RUL) by analyzing the data collected from multiple sensors. Existing studies for prognostics either focus on health index (HI)-based statistical fusion methods that are limited by restrictive assumptions or machine learning methods that model the HI and degradation status in two separate steps. However, the restrictive assumptions are often invalid in practice, and the intrinsic connection between the HI and degradation status is missing if the two parts are modeled separately, leading to poor prognostic results. This paper proposes an integrated deep learning-based data fusion and degradation modeling method by integrating a deep neural network (DNN) and a long short-term memory (LSTM) to characterize the nonlinear relationship between the HI and multiple sensor signals and to describe the underlying degradation status of units. In particular, our innovative idea is to develop an integrated backpropagation parameter estimation algorithm to solve the fusion procedure and the degradation modeling in an integrated manner by considering the properties of HI construction in the loss functions. Thus, the constructed HI is expected to better characterize the underlying degradation process and lead to a superior prognostic result. In the case study on the degradation of aircraft gas turbine engines, the proposed method achieves promising performance compared with the existing benchmarks of statistical models and other deep learning models. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —This paper develops an integrated deep learning-based data fusion and degradation modeling method for improving prognostics when multiple sensors are available to monitor the degradation status of a unit. There are four steps for implementing this method in practice: 1) collecting multiple sensor signals of historical units; 2) constructing the HI and modeling the degradation status of units by combining a DNN model and an LSTM model; 3) solving the fusion procedure and the degradation modeling in an integrated manner by developing an integrated backpropagation parameter estimation algorithm; and 4) making prognostics for in-service units. The novelty of the proposed method is that it conducts prognostics by combining a DNN fusion model and an LSTM degradation model, and seamlessly integrates the fusion procedure with the degradation modeling to construct the HI for better characterizing the status of a unit. As a result, the proposed method has two main advantages: (i) capable of characterizing various degradation processes of different engineering systems; and (ii) superior prognostic results by constructing more suitable HI for the degradation process.

Power transformer fault diagnosis using dynamic multiscale graph modeling and M2SGCN network based on statistical fusion

Power equipment fault diagnostics hold significant importance for the stability of power grid systems. In pursuit of this objective, this paper proposes a fault diagnosis method that utilizes dynamic multiscale graph (DMG) modeling and the multiscale multi-stream GCN(M2SGCN) network, incorporating statistical fusion. Specifically, a novel DMG modeling method is proposed to derive visibility graph data and horizontal visibility graph data from vibration signals across multiple scales. Next, a comprehensive neural network architecture named M2SGCN is established to learn global and local features simultaneously, providing a more precise representation. Subsequently, a Dempster Shafer evidence theory statistical fusion technique combined with an adaptive threshold model (DSTFusion) is utilized to integrate primary decision results for enhanced fault diagnosis accuracy. In addition, two datasets obtained from single-phase and three-phase power transformers are analyzed to demonstrate the evolution process. When compared to state-of-the-art indicators such as accuracy, precision, recall, and F1 scores, the method proposed excels in multiple aspects, successfully detecting fault states before their occurrence and achieving outstanding performance.

Diverse types of expertise in facial recognition

Facial recognition errors can jeopardize national security, criminal justice, public safety and civil rights. Here, we compare the most accurate humans and facial recognition technology in a detailed lab-based evaluation and international proficiency test for forensic scientists involving 27 forensic departments from 14 countries. We find striking cognitive and perceptual diversity between naturally skilled super-recognizers, trained forensic examiners and deep neural networks, despite them achieving equivalent accuracy. Clear differences emerged in super-recognizers’ and forensic examiners’ perceptual processing, errors, and response patterns: super-recognizers were fast, biased to respond ‘same person’ and misidentified people with extreme confidence, whereas forensic examiners were slow, unbiased and strategically avoided misidentification errors. Further, these human experts and deep neural networks disagreed on the similarity of faces, pointing to differences in their representations of faces. Our findings therefore reveal multiple types of facial recognition expertise, with each type lending itself to particular facial recognition roles in operational settings. Finally, we show that harnessing the diversity between individual experts provides a robust method of maximizing facial recognition accuracy. This can be achieved either via collaboration between experts in forensic laboratories, or most promisingly, by statistical fusion of match scores provided by different types of expert.

Assessment of the impact of discontinuity in satellite instruments and retrievals on global PM2.5 estimates

Exposure to fine particulate matter (PM2.5) is the leading environmental risk factor for mortality globally. Satellite-derived estimates of surface PM2.5 developed from a combination of satellites, simulations, and ground monitor data are relied upon for health impact studies. The ability to develop satellite-derived PM2.5 estimates requires the continued availability of aerosol optical depth (AOD) sources. This work examines the impact of the addition or loss of satellite AOD data sources on global PM2.5 estimation and the impact of continuing the long-term record with AOD from the Visible Infrared Imaging Radiometer Suite (VIIRS) on the Suomi-National Polar orbiting Partnership (S-NPP) satellite after the loss of the MODIS (MODerate resolution Imaging Spectroradiometer) and MISR (Multi-angle Imaging Spectroradiometer) instruments on board the Terra and Aqua satellites. We find that the addition of VIIRS S-NPP AOD products to geophysical PM2.5 estimates from satellites and simulations causes regional differences that correspond to differences in the VIIRS and MODIS Deep Blue AOD algorithms and sampling. Changes in long-term trends and timeseries due to the addition or loss of AOD data sources are generally within their uncertainties. Statistical fusion with ground monitor data partially corrects for changes due to sampling differences when introducing the VIIRS AOD products, but uncertainty remains over desert regions where ground monitor coverage is sparse. This work provides promise for the sustained development of global satellite-derived PM2.5 estimates, despite discontinuities in instruments and retrieval methods.

Spatial correction of low‐cost sensors observations for fusion of air quality measurements

AbstractThe context for this article is the statistical fusion of several pollutant measurement networks: a reference one of fixed sensors of high quality and others of fixed or mobile micro‐sensors of heterogeneous quality. The challenge is to use together the measurements of such different networks to obtain a better air quality map. Since pollution maps are often obtained from the correction of numerical model outputs by the measurements provided by the monitoring stations of air quality networks, the quality of the reconstructed map may be improved by increasing the density of sensors by adding low‐cost micro‐sensors. A geostatistical approach is very often used for the fusion of measurements. But the first step is to correct micro‐sensors measures using those given by the reference sensors. Usually, this preprocessing is performed during an offline preliminary study for which reference and micro‐sensor are located at the same position, which does not allow to adapt quickly to changes and to cope with time‐related nonstationarities. We propose in this article to complement these approaches by a simple online spatial correction of micro‐sensors. The principle is to use the reference measurements to correct the network of micro‐sensors. More precisely, by kriging only the measurements from micro‐sensors, the reference measurements are estimated; allowing to calculate a correction by kriging the differences, finally applied to the micro‐sensors. Then one can iterate this fundamental sequence of steps. Numerical experiments exploring the proposed algorithm by simulation and an application to a real‐world dataset are provided.

Accident risk prediction model based on attention-mechanism LSTM using modality convergence in multimodal

Anytime, anywhere, anyone can collect not only personal information but also surrounding situation information from home or a vehicle on the move without being restricted by time and space. To this end, the need for various devices such as biosensors, wearable devices, and temperature sensors is increasing. In this circumstance, there are many problems with the interaction of multiple devices and data processes, and there is a lack of structured or unstructured emerging contextual data. Accordingly, it is necessary to come up with a data collection method to obtain a variety of information efficiently. Such a method can obtain a diversity of information consistently with the use of data convergence. Since the method makes possible decision making, entity identification, and context prediction, it is applicable to diverse areas, including object detection and context awareness. The data generated through convergence makes it possible to the improve performance of data analysis through machine learning and deep learning. For this reason, a variety of modal types are applied to data analysis. Modal convergence for analysis can improve analysis performance more than unimodal. In this study, propose the accident risk prediction model based on attention-mechanism LSTM using modality convergence in multi-modal. In order to predict the risk of an accident, the proposed method makes the convergence of structured and unstructured modality data on the basis of judgmental fusion and statistical fusion and generates data sets. It is capable of solving the problem of data shortage and obtaining a variety of information consistently. In the proposed method, firstly, preprocessing is performed, and data sets of accident risk information are generated with the uses of the road risk equation and accident risk equation. Next, the correlation coefficient and regression coefficient for each variable is calculated through regression analysis modal and correlation coefficient analysis modal. A correlation coefficient is used to judge which variable needs to be learned intensively in order for the attention-mechanism of the accident risk prediction model. A regression coefficient is used as attention-weight to adjust a data weighting factor in attention-mechanism LSTM. Accordingly, whether there is the risk of an accident is judged. Performance evaluation is conducted in two ways. Firstly, epoch-based loss is evaluated in the comparison between the attention-mechanism LSTM-based accident risk prediction model with convergence modal and the model without it. In short, their RMSE is evaluated and compared. Secondly, the proposed model is compared with conventional models in terms of RMSE.

A terahertz multi-band metamaterial absorber and its synthetic evaluation method based on multivariate resonant response fusion for trace pesticide detection

A synthetic multivariate response evaluation of a multi-band terahertz (THz) metamaterial absorber (MMA) for highly sensitive and accurate detection of trace pesticides including indole-3-acetic acid, thidiazuron, benomyl, and chlorpyrifos in solution is presented. The THz spectral response of the prepared MMA sensor illustrated that with an increase in pesticide concentration (10∼520 ng/L), all absorption bands (0.91, 1.59, 1.99, and 2.42 THz) exhibited a red-shift in frequency, part of which showed a decrease in absorption intensity. The modeling results between the pesticide concentrations and spectral responses demonstrated that the proposed multivariate resonant response fusion (MRRF) framework outperformed all the univariate regressions in improving accuracy and simplifying evaluation through the effective multivariate selection and statistical fusion. The highest detection accuracy was achieved for chlorpyrifos (CPF). Consequently, trace CPF in four vegetable extracts was detected to further verify the stability and reliability of the multi-band MMA combined with MRRF. The results indicated that the predicted recoveries using our approach were accurate, and the detection limits of all pesticide solutions reached 10 ng/L. This work presents a new prospect for highly sensitive detection of trace hazardous materials and the potential application of THz metamaterial technology in food safety inspection.

A central multimodal fusion framework for outdoor scene image segmentation

Robust multimodal fusion is one of the challenging research problems in semantic scene understanding. In real-world applications, the fusion system can overcome the drawbacks of individual sensors by taking different feature representations and statistical properties of multiple modalities (e.g., RGB-depth cameras, multispectral cameras). In this paper, we propose a novel central multimodal fusion framework for semantic image segmentation of road scenes, aiming to effectively learn joint feature representations and optimally combine deep neural networks with statistical priors. More specifically, the proposed fusion framework can automatically generate a central branch by sequentially mapping multimodal features into a common space, including both low-level and high-level features. Besides, in order to reduce the model uncertainty, we employ statistical fusion to compute the final prediction, which leads to significant performance improvement. We conduct extensive experiments on various outdoor scene datasets. Both qualitative and quantitative experiments demonstrate that our central fusion framework achieves competitive performance against existing multimodal fusion methods.

Sports Event Level Measurement Indicator System Using Multisensor Information Fusion

In view of the imperfection of the measurement index of the level of sports events, the coverage rate of the measurement index is low and the stability is poor. Therefore, this paper puts forward a sports event level measurement index system based on multisensor information fusion. First, the simulated annealing algorithm is used to cluster the grouped sensors and fuse the optimal Bayesian estimation of compatible sensors. Second, the relative entropy measure method is used to expand the compatibility measure of the sensor information in the group, and the optimal Bayesian estimation value of the consistent measure test is obtained. The outliers are eliminated, the optimal fusion value is obtained by the overall weighted statistical fusion, and the alternative measure index system is constructed. Finally, analytic hierarchy process (AHP) is used to calculate the weight of each alternative index, so as to achieve the final measurement index. The results show that the standard deviation of clustering average energy consumption is low, and the energy loss is small. The system can effectively construct the measurement index, and the index coverage rate is as high as 95%.

High-resolution 3D abdominal segmentation with random patch network fusion.

Deep learning for three dimensional (3D) abdominal organ segmentation on high-resolution computed tomography (CT) is a challenging topic, in part due to the limited memory provide by graphics processing units (GPU) and large number of parameters and in 3D fully convolutional networks (FCN). Two prevalent strategies, lower resolution with wider field of view and higher resolution with limited field of view, have been explored but have been presented with varying degrees of success. In this paper, we propose a novel patch-based network with random spatial initialization and statistical fusion on overlapping regions of interest (ROIs). We evaluate the proposed approach using three datasets consisting of 260 subjects with varying numbers of manual labels. Compared with the canonical "coarse-to-fine" baseline methods, the proposed method increases the performance on multi-organ segmentation from 0.799 to 0.856 in terms of mean DSC score (p-value < 0.01 with paired t-test). The effect of different numbers of patches is evaluated by increasing the depth of coverage (expected number of patches evaluated per voxel). In addition, our method outperforms other state-of-the-art methods in abdominal organ segmentation. In conclusion, the approach provides a memory-conservative framework to enable 3D segmentation on high-resolution CT. The approach is compatible with many base network structures, without substantially increasing the complexity during inference. Given a CT scan with at high resolution, a low-res section (left panel) is trained with multi-channel segmentation. The low-res part contains down-sampling and normalization in order to preserve the complete spatial information. Interpolation and random patch sampling (mid panel) is employed to collect patches. The high-dimensional probability maps are acquired (right panel) from integration of all patches on field of views.

A lifecycle operating performance assessment framework for hot strip mill process based on robust kernel canonical variable analysis

In the modern hot strip mill process (HSMP), the operating performance may deteriorate because of wear of equipment, mode transitions, and random disturbances. If the process is not adjusted and maintained, faults may occur, resulting in greater economic losses and potential safety hazards. Therefore, it is of great practical significance to carry out comprehensive operating performance assessment. In this paper, a lifecycle operating performance assessment framework based on robust kernel canonical variable analysis (RKCVA) is proposed to deal with automation hierarchy, nonlinearity, and outliers in the plant-wide HSMP. First, the HSMP is divided into upstream, midstream, and downstream in real-time control level (L1). Then, based on kernel canonical variable analysis (KCVA) and partial robust M-regression (PRM), the RKCVA models are developed for each stream and process control level (L2). Based on the Bayesian inference, statistical fusion is implemented to judge whether the process is in normal or faulty operating condition. After that, according to different evaluation rules of different operating conditions, the lifecycle operating performance assessment is realized. Finally, the framework is illustrated with a case study on a real HSMP. The assessment results show that the accuracy of the RKCVA is more than 10% higher than that of the KCVA.

Heterogeneous iris certification method with universal sensors and safety output based on multisource fusion features and entropy labels of lightweight samples

Current iris recognition technology faces practical difficulties. For example, the unsteady morphology of a heterogeneous iris is generated by a variety of different devices and environments. The existing iris data sets have difficulty meeting the requirements of learning methods under a single deep learning framework. Research on iris recognition security has limited response to results stealing. Therefore, we propose a one-to-one heterogeneous iris certification method with universal sensors and safety output based on multisource fusion features and entropy labels of lightweight samples. This method is based on the classic convolutional neural network structure, and a lightweight neural network certification structure is designed. At first, convert the constrained multistate iris image into the digital features based on statistical learning and multisource feature fusion mechanism. The information entropy of the iris feature label is used to set the iris entropy feature category label and design a certification function that meets the requirements of different acquisition sensors according to the category label to obtain the certification result. Through the result encryption output module, the security output is achieved between the user and the certification result, and measures can be taken in time to confirm the stealing attack to improve the security of the output of certification result. As the requirement for the number and quality of irises changes, the category labels in the certification function are dynamically adjusted using a feedback learning mechanism. Three different acquisition sensors in the JLU iris library are used to do the experiments. The results prove that, for lightweight constrained multistate irises, the above mentioned problems are ameliorated to a certain extent by this method.

A new patch-based two-scale decomposition for infrared and visible image fusion

Typically, in existing two-scale fusion method, source images are first processed by a filter to obtain two layers, including a base layer and a detail layer. Second, a specific fusion strategy is applied to fuse these layers. Herein, the fusion of the base layer influences the contrast and clearness of final fusion result because the base layer may contain large scale variations in intensity. However, due to the inherent limitation of the base layer based on the filtering decomposition, the contrast of fusion result is easily reduced, or even the detail of the source images, in the fusion result, may be blurred. In this paper, a new two-scale fusion approach is proposed. In the process of two-scale decomposition, a new patch-based low-rank representation is proposed to obtain the two layers of the source images. In the proposed decomposition approach, the expression of the base layer is converted from a single filtering decomposition to a transform domain. It consists of a dictionary and corresponding coefficients. In the process of fusion, we design a global statistical fusion rule based on the coefficient value to achieve the base layer fusion, which improves the contrast of the final fusion result. For detail layer fusion, the L1-norm strategy is chosen to integrate small scale features contained in the detail layer into the final fusion result, which can contain clear detail. Experimental results illustrate that the proposed framework outperforms traditional approaches on the public database.

The statistical fusion identification of dairy products based on extracted Raman spectroscopy.

At present, practical and rapid identification techniques for dairy products are still scarce. Taking different brands of pasteurized milk as an example, they are all milky white in appearance, and their Raman spectra are very similar, so it is not feasible to identify them directly using the naked eye. In the current work, a clear feature extraction and fusion strategy based on a combination of Raman spectroscopy and a support vector machine (SVM) algorithm was demonstrated. The results showed a 58% average recognition accuracy rate for dairy products as based on the original Raman full spectral data and up to nearly 70% based on a single spectral interval. Data normalization processing effectively improved the recognition accuracy rate. The average recognition accuracy rate of dairy products reached 91% based on the normalized Raman full spectral data or nearly 85% based on a normalized single spectral interval. The fusion of multispectral feature regions yielded high accuracy and operation efficiency. After screening and optimizing based on SVM algorithm, the best spectral feature intervals were determined to be 335–354 cm−1, 435–454 cm−1, 485–540 cm−1, 820–915 cm−1, 1155–1185 cm−1, 1300–1414 cm−1, and 1415–1520 cm−1 under the experimental conditions, and the average identification accuracy rate here reached 93%. The developed scheme has the advantages of clear feature extraction and fusion, and short identification time, and it provides a technical reference for food quality control.

Cooperative sensor fusion in centralized sensor networks using Cauchy–Schwarz divergence

This paper presents a new solution for statistical fusion of multi-sensor information acquired from different fields of view, in a centralized sensor network. The focus is on applications that involve tracking unknown number of objects with time-varying states. Our solution is a track-to-track fusion method in which the information contents of posteriors are combined. Existing information-theoretic solutions for track-to-track fusion in sensor networks are commonly devised based on minimizing the average information divergence from the local posteriors to the fused one. A common approach is to use Generalized Covariance Intersection rule for sensor fusion. This approach works best when all the sensors detect the same object(s), and performs poorly when fields-of-view are different. We suggest Cauchy–Schwarz divergence to be used for measuring information divergence. We demonstrate that employing Cauchy–Schwarz divergence leads to fusion rules that are generally more tolerant to imperfect consensus. We show that the proposed fusion rule for multiple Poisson posteriors is the weighted arithmetic mean of the Poisson densities. Furthermore, we derive the fusion rule for labeled multi Bernoulli filter by approximating the labeled multi Bernoulli density to its first order moment. Numerical experiments show the superior performance of our solution compared to Kullback–Leibler averaging method.

Improved detection of gene fusions by applying statistical methods reveals oncogenic RNA cancer drivers

The extent to which gene fusions function as drivers of cancer remains a critical open question. Current algorithms do not sufficiently identify false-positive fusions arising during library preparation, sequencing, and alignment. Here, we introduce Data-Enriched Efficient PrEcise STatistical fusion detection (DEEPEST), an algorithm that uses statistical modeling to minimize false-positives while increasing the sensitivity of fusion detection. In 9,946 tumor RNA-sequencing datasets from The Cancer Genome Atlas (TCGA) across 33 tumor types, DEEPEST identifies 31,007 fusions, 30% more than identified by other methods, while calling 10-fold fewer false-positive fusions in nontransformed human tissues. We leverage the increased precision of DEEPEST to discover fundamental cancer biology. Namely, 888 candidate oncogenes are identified based on overrepresentation in DEEPEST calls, and 1,078 previously unreported fusions involving long intergenic noncoding RNAs, demonstrating a previously unappreciated prevalence and potential for function. DEEPEST also reveals a high enrichment for fusions involving oncogenes in cancers, including ovarian cancer, which has had minimal treatment advances in recent decades, finding that more than 50% of tumors harbor gene fusions predicted to be oncogenic. Specific protein domains are enriched in DEEPEST calls, indicating a global selection for fusion functionality: kinase domains are nearly 2-fold more enriched in DEEPEST calls than expected by chance, as are domains involved in (anaerobic) metabolism and DNA binding. The statistical algorithms, population-level analytic framework, and the biological conclusions of DEEPEST call for increased attention to gene fusions as drivers of cancer and for future research into using fusions for targeted therapy.

Abdominal multi-organ segmentation with organ-attention networks and statistical fusion.

Accurate and robust segmentation of abdominal organs on CT is essential for many clinical applications such as computer-aided diagnosis and computer-aided surgery. But this task is challenging due to the weak boundaries of organs, the complexity of the background, and the variable sizes of different organs. To address these challenges, we introduce a novel framework for multi-organ segmentation of abdominal regions by using organ-attention networks with reverse connections (OAN-RCs) which are applied to 2D views, of the 3D CT volume, and output estimates which are combined by statistical fusion exploiting structural similarity. More specifically, OAN is a two-stage deep convolutional network, where deep network features from the first stage are combined with the original image, in a second stage, to reduce the complex background and enhance the discriminative information for the target organs. Intuitively, OAN reduces the effect of the complex background by focusing attention so that each organ only needs to be discriminated from its local background. RCs are added to the first stage to give the lower layers more semantic information thereby enabling them to adapt to the sizes of different organs. Our networks are trained on 2D views (slices) enabling us to use holistic information and allowing efficient computation (compared to using 3D patches). To compensate for the limited cross-sectional information of the original 3D volumetric CT, e.g., the connectivity between neighbor slices, multi-sectional images are reconstructed from the three different 2D view directions. Then we combine the segmentation results from the different views using statistical fusion, with a novel term relating the structural similarity of the 2D views to the original 3D structure. To train the network and evaluate results, 13 structures were manually annotated by four human raters and confirmed by a senior expert on 236 normal cases. We tested our algorithm by 4-fold cross-validation and computed Dice-Sørensen similarity coefficients (DSC) and surface distances for evaluating our estimates of the 13 structures. Our experiments show that the proposed approach gives strong results and outperforms 2D- and 3D-patch based state-of-the-art methods in terms of DSC and mean surface distances.

Building extraction from very high-resolution synthetic aperture radar images based on statistical and structural information fusion

ABSTRACT To investigate the limits of building detection from very high-resolution (VHR) synthetic aperture radar (SAR) images, a new method, based on statistical and structural information fusion, is proposed in this paper. The proposed method contains two stages: First, using order statistics constant false alarm rate (OS-CFAR) and power ratio (PR) detectors, a set of detections are made. These detections have different statistical properties, compared to the other objects, and these properties are selected for discriminating buildings from clutters. Second, the morphological analysis is used for increasing the precision of the detection. In this stage, segments, which have the most similarities to buildings in terms of shape and size, are extracted via various structural elements (SEs). The final result is obtained by fusing the two sets of detections. The experimental results on the four real VHR SAR images show that the proposed method has a high detection rate (DR) and low false alarm rate (FAR).

The Robustness of Echoic Log-Surprise Auditory Saliency Detection

The concept of saliency describes how relevant a stimulus is for humans. This phenomenon has been studied under different perspectives and modalities, such as audio, visual, or both. It has been employed in intelligent systems to interact with their environment in an attempt to emulate or even outperform human behavior in tasks, such as surveillance and alarm systems or even robotics. In this paper, we focus on the aural modality and our goal consists in measuring the robustness of Echoic log-surprise in comparison with a set of auditory saliency techniques when tested on noisy environments for the task of saliency detection. The acoustic saliency methods that we have analyzed include Kalinli’s saliency model, Bayesian log-surprise, and our proposed algorithm, Echoic log-surprise. This last method combines an unsupervised approach based on the Bayesian log-surprise and the biological concept of echoic or auditory sensory memory by means of a statistical fusion scheme, where the use of different distance metrics or statistical divergences, such as Renyi’s or Jensen-Shannon’s among others, are considered. Additionally, for comparison purposes, we have also compared some classical onset detection techniques, such as those based on voice activity detection or energy thresholding. Results show that Echoic log-surprise outperforms the detection capabilities of the rest of the techniques analyzed in this paper under a great variety of noises and signal-to-noise ratios, corroborating its robustness in noisy environments. In particular, our algorithm with the Jensen-Shannon fusion scheme produces the best F-scores. With the aim of better understanding the behavior of Echoic log-surprise, we have also studied the influence of its control parameters, depth and memory, and their influence at different noise levels.