Self-Training Algorithm Based on Density Peaks Combining Globally Adaptive Multi-Local Noise Filter

Self-training algorithm can quickly train an supervised classifier through a few labeled samples and lots of unlabeled samples. However, self-training algorithm is often affected by mislabeled samples, and local noise filter is proposed to detect the mislabeled samples. Nevertheless, current local noise filters have the problems: (a) Current local noise filters ignore the spatial distribution of the nearest neighbors in different classes. (b) They can’t perform well when mislabeled samples are located in the overlapping areas of different classes. To solve the above challenges, a new self-training algorithm based on density peaks combining globally adaptive multi-local noise filter (STDP-GAMNF) is proposed. Firstly, the spatial structure of data set is revealed by density peak clustering, and it is used for helping self-training to label unlabeled samples. In the meantime, after each epoch of labeling, GAMLNF can comprehensively judge whether a sample is a mislabeled sample from multiple classes or not, and will reduce the influence of edge samples effectively. The corresponding experimental results conducted on eighteen real-world data sets demonstrate that GAMLNF is not sensitive to the value of the neighbor parameter k, and it can be adaptive to find the appropriate number of neighbors of each class.

Reduction of Multiplicative Noise in Radar Images

Introduction. A radar image is an image obtained by remote sensing the earth's surface with a radar device. Radar images are characterized by background graininess caused by speckle noise, which should be filtered to improve the quality of radar images. The structure of speckle noise reduction filters often comprise one or more parameters to control the level of noise smoothing. The values of these parameters have to be selected experimentally. In works devoted to speckle noise filtering, the methods used for selecting filter paraments are rarely clarified.Aim. To present a methodology for selecting the parameters of multiplicative speckle noise filters on a radar image that are optimal in terms of the quality of the resulting image.Materials and methods. The article presents a method for determining the optimal parameters of speckle noise reduction filters. This method was applied to the most conventionally used filters. The search for optimal parameters and testing of the filters were carried out using a specially designed image, which contained the objects most frequently found on radar images. The structural similarity index (SSIM) metric was chosen as a metric that assesses the quality of filtration.Results. After determining the optimal (in terms of SSIM) parameters of speckle noise reduction filters, the filters were compared to select the best filters in terms of the quality of radar image processing. In addition, the operation of the filters under study was tested on images containing various types of objects, namely: large objects, small objects and sharp borders. Knowing which filter copes best with smoothing speckle noise in a particular area and what values of the variable parameters this requires, an optimal quality of radar images can be achieved. Filtering not only improves human perception of radar images, but also reduces the influence of speckle noise during their further processing (object detection, segmentation of areas, etc.).Conclusion. The proposed algorithm allowed optimal parameters for several speckle noise filters to be determined. The quality of filtration was assessed using an expert method (visually) by comparing images before and after filtration, differential images and one-dimensional image slices. The Frost filter and the anisotropic diffusion filter with optimal parameters showed the best processing quality according to the SSIM metric.

Instance Reduction for Avoiding Overfitting in Decision Trees

Decision trees learning is one of the most practical classification methods in machine learning, which is used for approximating discrete-valued target functions. However, they may overfit the training data, which limits their ability to generalize to unseen instances. In this study, we investigated the use of instance reduction techniques to smooth the decision boundaries before training the decision trees. Noise filters such as ENN, RENN, and ALLKNN remove noisy instances while DROP3 and DROP5 may remove genuine instances. Extensive empirical experiments were conducted on 13 benchmark datasets from UCI machine learning repository with and without intentionally introduced noise. Empirical results show that eliminating border instances improves the classification accuracy of decision trees and reduces the tree size, which reduces the training and classification times. In datasets without intentionally added noise, applying noise filters without the use of the built-in Reduced Error Pruning gave the best classification accuracy. ENN, RENN, and ALLKNN outperformed decision trees learning without pruning in 9, 9, and 8 out of 13 datasets, respectively. The datasets reduced using ENN and RENN without built-in pruning were more effective when noise was intentionally introduced in different ratios.

Simulation-Based System-Level Conducted Susceptibility Testing Method and Application to the Evaluation of Conducted-Noise Filters

In this paper, a simulation-based system-level conducted susceptibility (CS) testing method for a wireless power transfer (WPT) system is proposed. The proposed method employs 3-dimensional electromagnetic (3D EM) models as well as equivalent circuit models to replace the measurement-based CS testing method based on the International Electrotechnical Commission 61000-4-6 standard. The conducted-noise source and equipment under test (EUT) are modeled in a circuit simulator. The conduction path, bulk current injection probe, and calibration jig are modeled using the 3D field simulator. A simple WPT system is designed and fabricated as the EUT for the CS test. The proposed method is successfully verified by comparing the voltage waveforms with measurement-based CS testing method. Additionally, as an application of the proposed method, a simulation-based evaluation of the conducted-noise filters is conducted. By using the proposed method, it is expected that the time and cost expense of setting up the test setup, as well as the testing procedure for the conventional measurement-based CS testing, will be drastically reduced. In addition, the proposed method can be used to estimate the conducted immunity of a system in the early stage of the design cycle prior to production.

Speckle Noise Filtering in Side-Scan Sonar Images Based on the Tucker Tensor Decomposition

Real signals are usually contaminated with various types of noise. This phenomenon has a negative impact on the operation of systems that rely on signals processing. In this paper, we propose a tensor-based method for speckle noise reduction in the side-scan sonar images. The method is based on the Tucker decomposition with automatically determined ranks of factoring tensors. As verified experimentally, the proposed method shows very good results, outperforming other types of speckle-noise filters.

M7 subtype leukemic cell edge detection techniques with threshold value comparison and noise filters

The aim of this paper is to study and identify various threshold values for two prevalently used edge detection techniques, which are Sobel and Canny. The purpose is to determine which value gives an accurate result for identifying a leukemic cell. Moreover, evaluating suitability of edge detectors is also essential as feature extraction of cell depends greatly on image segmentation (edge detection). Firstly, an image of M7 subtype of Acute Myelocytic Leukemia (AML) is selected due to its diagnosing which were found lacking. Next, apply noise filters for the best of image quality. Thus by comparing image with no filter, median and average filters, useful information can be acquired. Each edge detectors is fixed with threshold value of 0-0.5 but for Cann edge detection the value can increase until 0.9. From the research, it is found that Canny edge with no filter and a threshold value of 0.7 gives a clearer image with less noise reduction.