Синтез нейросетевых алгоритмов классификации морских объектов для низкочастотных стационарных гидроакустических систем

The fusion of multiple Computer Aided Detection/Computer Aided Classification (CAD/CAC) algorithms has been shown to be effective in reducing the false alarm rate associated with the automated classification of bottom mine-like objects when applied to side-scan sonar images taken in Very Shallow Water (VSW) environments [C.M. Ciany et al., 2001],[C.M. Ciany et al., 2000]. The fusion of CAD/CAC algorithms from Raytheon and NSWC Coastal Systems Station (CSS) also has been demonstrated in the shallow water environment on a single sonar data set [C.M. Ciany et al., 2002]. This paper extends the shallow water CAD/CAC/Fusion performance analysis to an additional set of sonar data taken in the Gulf of Mexico during June 1998, and adds the outputs of a third CAD/CAC algorithm from Lockheed Martin to the fusion processing. The fusion algorithm accepts the classification confidence levels and associated contact locations from the three different CAD/CAC algorithms, clusters the contacts based on the distance between their locations, and then declares a valid target when a clustered contact passes a prescribed fusion criterion. Four different fusion criteria are evaluated: the first based on the Fisher Discriminant, the second and third based on simple and constrained binary combinations of the multiple CAD/CAC processor outputs, and the fourth based on a constrained optimization approach that minimizes the total number of false alarms over the clustering distance and cluster confidence factor thresholds for a given probability of correct classification. The resulting performance of the four fusion algorithms is compared, and the overall performance benefit of a significant reduction of false alarms at high correct classification probabilities is quantified.

Land cover change detection plays an important role in natural disaster monitoring, tracking urban expansion, and many social benefit areas. The spectral-based direct comparison (SDC) methods are commonly used for change detection, but such methods are vulnerable to the influence of external factors. In general, land changes among different land cover types have different characters of change magnitude. The class probability-based direct comparison (CPDC) methods consider land type information and reduce the influence of external factors, but these methods are strongly dependent on the training samples. To address the above problems, we proposed a novel change detection method that integrates spectral values and class probabilities (SVCP). First, a new change magnitude map based on spectral values and class probabilities is constructed by using the maximum interclass variance and Gaussian mixture model (GMM), which greatly enhances the differentiation between changed and unchanged areas. Second, the Kapur threshold selection method is improved by using the variance of the changed and unchanged areas as well as the class probabilities for adaptive thresholding. The SVCP approach was assessed by two case studies from Landsat 8 Operational Land Imager (OLI) images. The “change/no-change” detection and “from-to” change types were evaluated. The experimental results indicated that the SVCP method is more accurate in the change detection, with lower false and missed detection rates than the traditional methods.

Human object classification in video is an important problem in computer vision. The main challenge is to correctly classify any human object in presence of several objects, occlusion of object, varying background and lighting conditions, etc. Object classification is much desired in surveillance like applications. Several classification algorithms have been proposed in spatial and wavelet domain. In this paper we present an object classification application of machine learning which is based on the single feature of object. The feature chosen for classification is energy of biorthogonal wavelet transform (BWT) coefficients of the object and is used to classify the object in a video into two categories: human object and non-human object. Two important properties of biorthogonal wavelet transform - shift invariance and symmetry, are useful for object classification. Translation in object is well handled by the shift invariance property while symmetry property is used to maintain the object boundaries. Classification has been performed using Adaboost classifier. Quantitative analysis of the results demonstrates better performance of the proposed method over other state-of-the-art methods.

Deep learning breakthrough stimulates the new research trend in infrastructure defects inspection. The lack of quality-controlled, human-annotated, free of charge, and publicly available defect datasets with sufficient amounts of data hinders the progress of deep learning in defects inspection. To boost research in deep learning-based defects inspection, we first summarize 37 publicly available defect datasets in this two-part survey. These defect datasets cover common defects in various types of infrastructure, and the taxonomy of the datasets is based on specific deep learning objectives. Besides, taking crack as the research target, we have combined the existing datasets with self-labeled crack images to establish a benchmark dataset for crack classification and segmentation. Moreover, based on the established crack dataset, we make a comprehensive comparison between state-of-the-art algorithms for classification, segmentation, and detection. In this paper, we concentrate on datasets and algorithms for defect classification. Altogether 11 classification-oriented defect datasets are summarized and demonstrated in detail. Based on the established crack dataset, we comprehensively compare existing state-of-the-art algorithms for object classification, which provides a baseline for future research in defects inspection. The companion paper of this work surveys datasets and algorithms for defect segmentation and detection, where 26 defect datasets are elaborated, and systematic comparison between state-of-the-art algorithms has been conducted. The classification algorithms illustrated in this paper are conducive to the coarse localization of the defects' position in images.

This paper presents an improved method for highly accurate identification of the speaker who speaks sentences in the Uzbek language from a limited set of sentences. It is proposed the use of an advanced method for identifying the speaker speaking in the Uzbek language through the use of classification method - Gaussian mixture model (GMM) for decision making. The advanced method is a combination of Gaussian mixture model and MFCC. Current scientific research is aimed at identifying opportunities to improve performance. One of the latest implementations of MFCC are Delta MFCC (DMFCC) and Delta Delta Delta MFCC (DDMFCC), which improve speaker identification. Identification accuracy was obtained from the results of the research. According to the experimental results, speaker identification by conventional MFCC combined with GMM classification algorithm (94.5%) lower than the second (DMFCC+GMM - 96.8%) and third (DDMFCC+GMM-98.1%) feature extraction methods. It is easy to see the performance improvement observed when using DDMFCC in combination with GMM and is 98.1%. first, second, and third level headings (first level heading).

In recent times, there have been increased rates at which researchers are searching for advanced ways of carrying out land-use land-cover (LULC) mapping, especially in developing countries. Four machine-learning algorithms, namely Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbour (K-NN), and Gaussian Mixture Models (GMM) were examined. This study also attempted to validate the various models using the index-based validation method. Accuracy assessment was performed by using the Kappa coefficient. The results of the LULC showed that RF classified 23% of the study area as bare land, SVM has 24% of the study area classified as bare land, K-NN also allotted 24% to bare land, while that of GMM classifier was 30%. The overall accuracy of RF, SVM, K-NN and GMM were 0.9840, 0.9780, 0.9641 and 0.9421 respectively. The Kappa Coefficient of the various classifiers were RF (0.9695), SVM (0.9580), K-NN (0.9319) and GMM (0.8916). This study showed that though all the algorithms performed relatively very well, RF performed better than the other classifiers. It suffices to state that, there is a need for further studies since other extraneous environmental variables may be underpinning these conclusions.

The acoustic recognition module of the unattended ground sensor (UGS) system applied in wild environments is faced with the challenge of complicated noise interference. In this paper, a small-aperture microphone array (MA)-based acoustic target classification system, including the system hardware architecture and classification algorithm scheme, is designed as a node-level sensor for the application of UGS in noisy situation. Starting from the analysis of signature of the acoustic signal in wild environments and the merits of small-aperture array in noise reduction, a closely arranged microelectromechanical systems MA is designed to improve the signal quality. Considering the similarities between speaker discrimination and acoustic target recognition, a classification algorithm scheme, consisting of a simplified Mel-frequency cepstrum coefficients and the Gaussian mixture model, is developed to distinguish acoustic targets’ patterns. The proposed classification algorithm has been implemented on embedded system after being tested on training datasets. By combining the small-aperture array and low-complexity classification algorithm, the presented acoustic classification prototype system is portable and efficient. To demonstrate the efficiency of the design, the prototype system is verified in a practical situation with the employment of wheeled and tracked vehicles. Evaluation of the system performances in comparison with other state-of-the-art methods indicates that the proposed design is practical for the acoustic target classification and may be widely adopted by UGS.

Object classification from images is an important task for machine vision and it is a crucial ingredient for many computer vision applications, ranging from security and surveillance to marketing. Image based object classification techniques properly integrate image processing and machine learning (i.e., classification) procedures. In this paper we present a system for automatic simultaneous optimization of algorithms and parameters for object classification from images. More specifically, the proposed system is able to process a dataset of labelled images and to return a best configuration of image processing and classification algorithms and of their parameters with respect to the accuracy of classification. Experiments with real public datasets are used to demonstrate the effectiveness of the developed system.

In this paper, we propose a new feature selection algorithm for multi-class classification. The proposed algorithm is based on Gaussian mixture models (GMMs) of the features, and it uses the distance between the two least separable classes as a metric for feature selection. The proposed system was tested with a support vector machine (SVM) for multi-class classification of music. Results show that the proposed feature selection scheme is superior to conventional schemes.

A fusion of dolphin swarm optimization and improved sine cosine algorithm for automatic detection and classification of objects from surveillance videos

Highly automated driving in real world situations requires perception and planning to account for other forms of human locomotion, principally bicycles, and pedestrians, in addition to vehicles. Here we present a complete perception-planning-control system with high quality perception including object classification and tracking, and a path planning algorithm that makes use of probabilistic classification results. Our classification algorithm is fast enough to provide results for the vast majority of objects in on-line operation, but to ensure output for real-time decision-making, the computation and integration of classification results is done using an anytime algorithm. This work builds on recent research in object classification using LIDAR data (Teichman et. al., 2011). Our focus is on safe driving behavior in the presence of dynamic obstacles on the side of the road, which are typically bicycles, using LIDAR data. Our resulting system is extensively tested in real-world situations, and shows good performance on a large test set of interactions with bicycles.

Purpose: To develop an automatic algorithm for segmentation and classification of Ultrashort TE (UTE) MRI into bone, air and water and to use these images for .dose calculation. Method and Materials: In this work a method for segmentation of MR images of the head for use in RT planning is proposed. Three tissue class need to be segmented: soft tissue, cortical bone and air‐filled cavities. A special MRI technique, ultra‐short TE (UTE) imaging, is used to obtain a signal from bones, which remains dark in conventional MR sequences. The segmentation is done by fitting a Gaussian mixture model (GMM) to the data. Results: The separation of fore‐ and background works reliable. Due to partial volume effects and image noise, parts of trachea are sometime falsely classified.. In a few images bone classification failed partially. Two determine the optimum number of classes different GMM models are estimated for each patient dataset and the Bayesian information criterion (BIC) is used to determine the model with the optimum number of clusters. The optimum BIC value varied. Without normalization GMMs with three clusters are optimal, with normalization more clusters are optimal. Experimental dose calculation was performed on the MRIimages and resulting distributions were compared to CT‐based dose calculation results. Dose difference was below or in the range of 3–4%. Conclusion: The presented method is fully automatic and could improve accuracy in RT planning for certain indications, e.g. existence of high‐Z implants or for planning in low soft‐tissue contrast regions where MR data often can only be used when rigid fusion allows to overlay MRI and CT, which is not often the case for anatomical regions outside of the head. Our method could allow for directly planning the treatment on the MR making acquisition of the CT obsolete.

Intelligent vehicles need reliable information about the environment in order to operate with total safety. In this paper we propose a flexible multi-module architecture for a multi-target detection and tracking system (MTDTS) complemented with a Bayesian object classification layer based on finite Gaussian mixture models (GMM). The GMM parameters are estimated by an expectation maximization (EM) algorithm, hence finite-component models were generated based on feature-vectors extracted from object's classes during the training stage. Using the joint mixture Gaussian pdf modelled for each class, a Bayesian approach is used to distinct the object's categories (persons, tree-trunks/posts, and cars) in a semi-structured outdoor environment based on data from a laser range finder (LRF). Experiments using real data scan confirm the robustness of the proposed architecture. This paper investigates a particular problem: detection, tracking and classification of objects in cybercars-like outdoor environments

For a variety of applications, machine learning algorithms are required to construct models that minimize the total loss associated with the decisions, rather than the number of errors. One of the most efficient approaches to building models that are sensitive to non-uniform costs of errors is to first estimate the class probabilities of the unseen instances and then to make the decision based on both the computed probabilities and the loss function. Although all classification algorithms can be converted into algorithms for learning models that compute class probabilities, in many cases the computed estimates have proven to be inaccurate. As a result, there is a large research effort to improve the accuracy of the estimates computed by different algorithms. This paper presents a novel approach to cost-sensitive learning that addresses the problem of minimizing the actual cost of the decisions rather than improving the overall quality of the probability estimates. The decision making step for our methods is based on the distribution of the individual scores computed by classifiers that are built by different types of ensembles of decision trees. The new approach relies on statistics that measure the probability that the computed estimates are on one side or the other of the decision boundary, rather than trying to improve the quality of the estimates. The experimental analysis of the new algorithms that were developed based on our approach gives new insight into cost-sensitive decision making and shows that for some tasks, the new algorithms outperform some of the best probability-based algorithms for cost-sensitive learning.

IMPROVEMENTS OF CONDENSATION HEAT TRANSFER MODELS IN MARS CODE FOR LAMINAR FLOW IN PRESENCE OF NON-CONDENSABLE GAS