Abstract. Fractal geometry is becoming increasingly important in the study of image characteristics. For recognition of regions and objects in natural scenes, there is always a need for features that are invariant and they provide a good set of descriptive values for the region. There are many fractal features that can be generated from an image. In this paper, fractal signatures of nearby galaxies are studied with the aim of classifying them. The fractal signature over a range of scales proved to be an efficient feature set with good discriminating power. Classifiers were designed using nearest neighbour method and neural network technique. Using the nearest distance approach, classification rate was found to be 92%. By the neural network method it has been found to increase to 95%.

Fractal concepts are used to describe the irregular structures and regions of interest of solar images. The most common and easiest way to extract regions of interest from an image is through segmentation. Segmentation techniques vary from conventional edge-detection mechanism to fuzzy c-means clustering. In this study, the pixelwise local fractal dimension of solar images is computed by different techniques. This is followed by different segmentation procedures including the fuzzy-based approach, for extracting the active regions from chromospheric images and assessing their performance. These techniques have also been applied on solar images to extract active regions from Solar Heliospheric Observatory (SOHO) Extreme Ultraviolet Telescope (EIT) images.

The Brain-Computer Interface (BCI) is the technology that enables direct communication between the human brain and the external devices. Electroencephalography (EEG) proves to be the most studied measure of recording brain activity in BCI design. The paper is intended to analyze and extract the features of EEG signal and to classify the signal so that human emotions can be discriminated and serve as the control signal for BCI. The proposed method involves EEG data acquisition and processing which is done by feature extraction and classification of features at different frequency levels for Beta, Alpha, Theta and Delta waves. The Principal Component Analysis(PCA ),and the Wavelet Transform(WT) can be used for dimensionality reduction and feature extraction . The Artificial Neural Network (ANN) which is a computationally powerful model, is used as the classifier. The paper presents the comparison between the two approaches PCA and WT applied on the ANN Classifier.

No abstract

Over the Indian region, the pre-monsoon (i.e., April-May) is a dry summer season. The heatwaves, as well as local temperature variations during this season, are not associated with significant large-scale convective heating like the monsoonal modes, and several studies identified several drivers of heatwaves. Heatwaves are extreme events. Are these extremes arising from low-frequency intraseasonal modes, in the same way, extreme rainfall occurs on a synoptic or intraseasonal mode during monsoon? Studies do not explicitly point out the existence of temperature intraseasonal modes during April-May over the Indian region and it is not clear if some of the drivers of heatwaves can also explain the April-May temperature variations as derivative of some modes. This study identifies the dominant pair of the intrinsic mode of temperature intraseasonal oscillations which can also explain the heatwave spikes. The empirical orthogonal function (EOF) based modes are isolated in the detrended surface temperature data to remove the global warming mode. It was found that the subtropical jet acting as a Rossby wave guide drives the first mode with pan India spatial modal signature, while the second mode is driven by the extratropical Rossby wave modes originating from the latitudes of the eddy-driven jet. Another important result is that the first (second) mode principal component shows a significant decreasing (increasing) trend from 1981-2020 period. The observed spatial heterogeneity in warming and the trend in the spatial distribution of extreme temperature events in India could also be explained by the trend in the two modes of oscillation.

The seamless forecast approach of subseasonal to seasonal scale variability has been succeeding in the forecast of multiple meteorological scales in a uniform framework. In this paradigm, it is hypothesized that reduction in initial error in dynamical forecast would help to reduce forecast error in extended lead-time up to 2-3 weeks. This is tested in a version of operational extended range forecasts based on Climate Forecast System version 2 (CFSv2) developed at Indian Institute of Tropical Meteorology (IITM), Pune. Forecast skills are assessed to understand the role of initial errors on the prediction skill for MJO. A set of lowest and highest initial day error (LIDE & HIDE) cases are defined and the error-growth for these categories are analysed for the strong MJO events during May to September (MJJAS). The MJO forecast initial errors are categorized and defined using the well-known multivariate MJO index introduced by Wheeler & Hendon (2004). The probability distribution of bivariate RMSE and error growth evolution (first order difference of index error for each successive lead days) with respect to extended range lead-time are used as metrics in this analysis. The result showed that initial error is not showing any influence in the skill of model after a lead time of 7-10 days and the error growth remains the same for both set of errors. A rapid error growth evolution of same order is seen for both the classified cases. Further the physical attribution of these errors are studied and found that the errors originate from the events with initial phase in Western Pacific and Indian Ocean. The spatial distribution of OLR and the zonal winds also confirms the same. The study emphasise the importance of better representation of MJO phases especially over Indian ocean in the model to improve the MJO prediction rather than focusing primarily on the initial conditions.

No abstract

Segmentation is performed in recognition applications as a primary step towards extraction of interesting regions of an image. In this paper, the characteristic effects of Weibull and Fractal parameters in the segmentation of Synthetic Aperture Radar (SAR) and Optical images acquired from satellite platform is studied. The algorithms are tested for different window sizes and different number of classes to bring out the effect of these parameters in the segmentation process.