Robustness In Noise Environments Research Articles

Electric Signature Detection and Analysis for Power Equipment Failure Monitoring in Smart Grid

Power equipment is one kind of basic element in smart grid, and how to design an efficient detection and analysis scheme of electric signature (ES) for power equipment failure (PEF) monitoring is a key and challenging issue. This article proposes an ES detection and analysis method which can monitor multiple kinds of PEF in smart substation. The bottleneck of ES analysis is explored in the view of Heisenberg uncertainty, and an optimal time–frequency analysis method is designed to solve the problems. The proposed method (PM) is based on union of time and frequency bases whose decomposition is realized by Bayesian compressive sensing using Laplace prior. Simulated and field ESs are employed to test PM with comparisons of existing methods. Also, PM is applied in a smart substation of China. Several typical PEFs and measurement soft failures caused by electromagnetic interference are discussed. The results indicate that the PM can accurately monitor PEFs whose mechanism can be revealed by time–frequency features of ESs, if the required sampling rate and sampling time are satisfied because of its immunity of the uncertainty principle restriction. The robustness in noise environment and optimal time–frequency representation of ESs make the PM an efficient general-purpose PEF monitoring in smart grid by time–frequency analysis.

Plant recognition based on Jaccard distance and BOW

Plant recognition is a meaningful research that has attracted many researchers. Due to the variety of plants, it is difficult for the existing identification methods to identify their species efficiently. We proposes a plant recognition method based on Jaccard distance and Bag of words (BOW). Firstly, Jaccard distance is employed to calculate the similarity between the test sample and part of the training samples of all species, $$C_{1}$$ species with the highest similarity are selected as candidate species of the test image, which not only reduce the amount of computation but also shorten the time consumption. Secondly, BOW is employed to extract features from texture image and contour image, and support vector machine is used for training and classification. In our method, the texture and contour features of leaf images are extracted by Laws texture measure and Sobel operators respectively. The local and global features of the leaf can be described well. Some representative datasets are used to evaluate the proposed method and obtain high accuracy. Comparison with existing methods proves that the proposed method not only has a high accuracy, but also has robustness in noise environment.

Time reversal imaging based on joint space–frequency and frequency–frequency data

AbstractA new time reversal (TR) method for target imaging is proposed in this paper. Through single measurement by the antenna array, the received signals are utilized to form the space–frequency–frequency multistatic data matrix (MDM). Singular value decomposition is applied to the matrix to obtain the left singular vectors which span the signal subspace. The obtained vectors are divided into multiple subvectors by two different schemes and used to provide target signatures in the form of coarse frequency dependence and relative phase shifts that can be exploited to construct the imaging function. The performance of the proposed method is investigated through numerical simulations for both single and multiple targets, and the results are compared with the traditional TR method using the frequency–frequency MDM. It turned out that the proposed method is able to achieve high resolution with limited array aperture and shows satisfactory robustness in noise environment. Furthermore, experimental results are provided to show the availability of the method in practical applications.

A Note on Lewicki-Sejnowski Gradient for Learning Overcomplete Representations

Overcomplete representations have greater robustness in noise environment and also have greater flexibility in matching structure in the data. Lewicki and Sejnowski (2000) proposed an efficient extended natural gradient for learning the overcomplete basis and developed an overcomplete representation approach. However, they derived their gradient by many approximations, and their proof is very complicated. To give a stronger theoretical basis, we provide a brief and more rigorous mathematical proof for this gradient in this note. In addition, we propose a more robust constrained Lewicki-Sejnowski gradient.

An autoregressive moving average tone detector

A new method for estimating tones in an arbitrary spectrum is presented. An autoregressive-moving average estimator is formulated and transformed into a linear regression problem. Many of the shortcomings of an "all pole" model are overcome and simulated test results indicate that the estimates are not particularly sensitive to additive noise. The main advantages of this new method are computational simplicity and robustness in noise environments. The algorithm can be useful in all areas where spectral information must be extracted in a computationally efficient fashion.