Abstract

This thesis aims to investigate different aspects of timing in Sequence Learning and Adaptation to certain emotional responses. Sequencing and processing of emotions are fundamental to human behaviour and hence are important fields to explore. To investigate the influence of temporal factors in Sequence learning, we conducted a behavioural study using the paradigm of Serial Reaction Time (SRT) task and introduced delay in presentation of successive stimuli, referred to as Response-to-Stimulus Interval (RSI). By varying the RSI between suc- cessive elements of a sequence over a few temporal windows, we attempted to explore what effects this systematic varying of the RSI would have on the sequence learning process. We have also con- ducted post-experimental debriefing sessions to see if such learning is implicit or explicit. The SRT task requires the participants to simply respond to a stimulus that appears on the screen by clicking a key corresponding to the location at which it appeared. Reaction times were recorded for all the stimuli and reduction in this would imply learning. Statistical analysis of the reaction time data showed that learning did not get impaired in the various temporal groups despite delays experienced due to vari- able RSI. Post-experimental results showed that the knowledge acquired in smaller RSIs was implicit while for larger RSIs it was explicit. The results are in general agreement with the existing literature with fixed RSIs and hence we can conclude from the empirical study that varying RSIs do not impair sequence learning and that the sequential knowledge might have become explicit with larger RSIs. Ad- ditionally, we proposed a biologically realistic computational model in the form of a simple recurrent neural network (SRN) to obtain a functional account of the empirical findings. The proposed model has spatial representation of time for incorporating RSI in the neural network. The model illustrates how explicit learning could emerge due to a longer temporal window between stimuli that could potentially give insights into the mechanisms of sequence learning in variable RSI conditions. In the second part of the thesis, we investigated the duration aspect of timing. Specifically, how different kinds of emotional reactions last for different duration. For this, we conducted an empirical study with a sad and self-relevant stimulus and compared the results to a sad but not-so-self-relevant stimulus. We found that the emotional reaction in the current study lasted longer because it is perhaps difficult for human beings to adapt to an emotional response when they can relate to the stimulus that elicited the emotion. The emotional intensity and its duration are thus determined by two factors, the self-relevance of the affective stimulus and how well or poorly understood the stimulus is to the person experiencing it. The greater the self-relevance and poorer the understanding, the stronger the emotion and longer will be its duration. We also describe a preliminary study conducted to collect Physiological data as an attempt to strengthen this observation. The thesis ends with a summary of the work and a general discussion of the findings. Limitations and future work have also been discussed.

Abstract

With the evolution of wireless systems, more communication standards are being proposed while maintaining backward compatibility, therefore, there is an ever growing need for wideband multistandard receivers similar to software defined radio (SDR) receivers. For many years now, SDR or digitally reconfigurable radio research has been quite challenging with only a few reported practical prototypes with limited success. The basic demand of the SDR has been the need for a giant analog-to-digital converter (ADC) and a powerful enough digital signal processing (DSP) so that it serves as a universal radio platform receiving almost all radio standards and services. These demands are difficult to meet even in modern CMOS technologies leading to the necessity of highly flexible architectures compared to traditional receivers. To achieve this objective, digital intensive receiver architectures with passive structures are proposed. Radio frequency (RF) sampling receivers using time-interleaved (TI), RF sampling and discrete-time mixing also provide additional flexibility by employing discrete-time signal processing in analog domain. Owing to the oversampling ratio of up to 8 times in these receivers, high-performance complex frequency synthesizers are needed with stringent phase noise requirement. Moreover, the clocking schemes employed in TI RF sampling architectures result in additional timing jitter and mismatch offsets. These challenges are alleviated in a class of sampling receivers, called subsampling receivers, where the employed bandpass sampling principle results in less than the Nyquist sampling frequency. This leads to the less-complex and less power hungry frequency synthesizers. In this thesis, an intermediate solution to SDR receiver implementation called “Mini-SDR” architecture is proposed by exploiting the bandpass sampling concept. Based on the proposed Mini-SDR architecture, three CMOS radio frontends are proposed for standalone electronic attack (EA) transceiver system, dynamically reconfigurable multistandard subsampling (DRMS) radio receiver and a dual-band subsampling receiver for IEEE WLAN 802.11ac standard. The first architecture is a single chip integrated transceiver for a standalone EA system based on digital radio frequency memory (DRFM) repeater without the need for a separate instantaneous frequency measurement (IFM) receiver. The proposed transceiver architecture employs sub-Nyquist discrete-time frontend to estimate the carrier frequency and bandwidth of the incoming pulse radar signals. The system is verified for linear frequency modulated, constant carrier frequency and phase coded pulse radar signals. These signals are classified on the basis of Wigner-Ville distribution. The performance of the proposed system is validated from 1 GHz to 20 GHz with instantaneous bandwidth of 300 MHz, dynamic range from −60 dBm to −20 dBm and system gain from 30 dB to 60 dB. The minimum delay achieved between reception and retransmission is 13 pulse repetition intervals (PRIs). To the best of author’s knowledge, this work reports the first sub-Nyquist bandpass sampling based single chip architectural solution for a standalone EA system where RF frontend and digital controller are integrated into the system. The second architecture is the DRMS radio receiver. The proposed receiver has a unique capability to detect the carrier frequency of the incoming signal, estimate it’s bandwidth and standard. The RF frontend is modelled in Verilog-AMS behavioural model and the digital signal processing is implemented in Simulink-Matlab. Also, the system level optimizations of the overall receiver performance are presented with the effect of receiver impairments. The complete receiver architecture has been verified to detect and process three different bands belonging to three different standards (GSM, UMTS and WLAN) with the carrier frequency ranging from 0.9 GHz to 2.5 GHz with a maximum signal bandwidth of 22 MHz and the input dynamic range from −109 dBm to −20 dBm. The last implementation is a first dual-band subsampling receiver with subsampling frequency optimization to meet ultimate receiver error vector magnitude (EVM) of −40 dB over wide input power range of 19 dB. Systematic system level optimization of the receiver chain with major impairments such as sampling frequency, synthesizer phase noise, IQ mismatch and unit capacitor in the SC filter is also proposed. Sampling frequency optimization proposed in this work has multi-fold advantages: a) the noise folding effect is reduced leading to smaller noise figures. b) less-complex and less-power hungry clocking scheme. c) quadrature phase readily available between consecutive samples is utilized to separate samples into in-phase (I) and quadrature-phase (Q) paths. d) the ultimate EVM performance of the receiver is significantly better than the subsampling receivers, where the sampling frequency is not optimized. The dual-band subs

Abstract

Decentralized MDPs (Dec-MDPs) provide a rigorous framework for collaborative multi-agent sequential decision making under uncertainty and partial observability. However, their high computational complexity limits their practical impact. To overcome this computational complexity, we focus on a special class of DEC-MDPs consisting of independent agents which collaborate with each other through a global joint-reward function that depend upon their entire histories of states and actions to accomplish joint tasks. We make the following contributions to address the issue of scalability for this class of problems. 1) Dec-NLP: A nonlinear programming (NLP) formulation for such event-based planning model which scales well for small problems. However, we observed empirically that with the increasing number of agents, NLP solvers were unable to scale, slow and often ran out of memory; 2) Dec-EM: A probabilistic inference based approach that scales much better than NLP solvers for a large number of agents. However, the scalability still suffers when the state-space of each agent is exponential, which is often the case for several patrolling and coverage problems; 3) Dec-MARL: A policy gradient based multi-agent Reinforcement Learning (RL) approach that scales well even for exponential state-spaces. (4) Dec-ESR : A new actor-critic based Reinforcement Learning (RL) approach for event based Dec- MDPs that produces better solution quality than Dec-MARL using successor features (SF) which is a value function representation that decouples the dynamics of the environment from the rewards. We then present how Dec-ESR generalizes learning for event based Dec-MDPs using SF within an end-to-end deep RL framework; (5) We show that the proposed method using SF allows useful transfer of information on related but different tasks, hence bootstraps the learning for new tasks and makes convergence much faster on new tasks. Our inference and RL-based advances enable us to solve a large real-world multi-agent coverage problem modeling schedule coordination of agents in a real urban subway network where other approaches fail to scale.

Abstract

Speech often consists of expressive and nonverbal sounds in addition to the message to be conveyed. The objective of this thesis is to analyze expressive voices to determine the features needed for representing, describing and discriminating different types of expressions. Here expression refers to the information in the signal produced by the speech production mechanism, that characterizes different phonations, emotions, singing (normal and artistic) and nonverbal sounds. Most of these expressive features are as a result of the variations in the excitation component of speech production processes, although it is difficult to isolate the excitation component from the dynamic vocal tract system component. Also, the expressiveness is reflected in the dynamics of the excitation component. Few methods for extracting the excitation source features rely on inverse filtering methods, which are known to have limitation for high pitched voices such as emotional speech and singing voice. In this work, we propose a representation of the excitation source in terms of impulse-like sequence to capture variations in the source. Significance of the excitation source is studied for three types of expressive voices, namely, phonation types (in speech and singing), emotional speech and artistic (Noh) singing voice. The primary mode of excitation is due to vibration of the vocal folds at the glottis. Even though the excitation information is present throughout the glottal cycle, it is considered to be significant only when there is large change in a short-time interval, i.e., when it is impulse-like. The impulse-like characteristic is usually exhibited around the instant of glottal closure within each glottal cycle. Apart from the primary impulse-like excitation at the instant of glottal closure, other major and minor excitation impulses may also occur at glottal opening and in creaky voice. Other excitation impulses become prominent in aperiodic voices, like artistic (Noh) singing voice. An approach for extracting the excitation source information in terms of impulse-like sequence is proposed using single frequency filtering (SFF) method. To exploit the impulse-like discontinuities from the signal, impulse properties in magnitude and phase component are explored. An impulse-like characteristic is derived using the spectral variance computed at each instant of time, and the instantaneous frequency (IF) of the filtered signal at each frequency. The slope of the variance plot shows discontinuities at the locations of the impulse-like excitation. The sum of the IFs computed for each SFF output signal shows discontinuities at the locations of the impulse-like excitations. Extraction of the impulse-like sequence is assessed using synthetic and natural speech signals. The SFF-based method also brings out the minor impulse-like excitations within each glottal cycle. Determination of two important features of excitation, namely, glottal closure instant (GCI) and glottal open region (GOR) are derived by exploiting the changes in the spectral characteristics of the SFF spectra and HNGD spectra of the zero time windowing (ZTW) method. A spectral flatness parameter derived from the SFF spectra highlights the impulse-like characteristics at the GCI. The spectral flatness parameter derived from the HNGD spectra highlights the GOR. A new approach is proposed for detection of GCIs from emotional speech. The approach uses the recently proposed modified zero frequency filtering (mZFF) and the impulse-like sequence representation obtained using the SFF method. Using the features derived from the three signal processing methods namely, ZFF, ZTW and SFF, expressive voices such as phonation types, emotional speech and artistic (Noh) singing voice, are analyzed. Features such as locations of GCIs, strength of the glottal closure, different phases (closed and open phase) of glottal vibration and their spectral correlates, are used for analysis and detection of different phonation types in speech and singing. Cepstral coefficients derived from the ZTW (ZTWCC) and SFF (SFFCC) methods are used for discriminating different phonation types. Excitation features derived around the GCIs using ZFF method and linear prediction (LP) analysis are used to study the role of excitation information in emotional speech. One of the main features of the artistic (Noh) singing voice is the aperiodicity in the vocal fold vibration. The aperiodicity characteristics are studied in detail using the spectral characteristics of the impulse-like sequence, which shows harmonics and subharmonics of the fundamental frequency corresponding to pitch. The significance of the impulse-like sequence representation is also studied through the saliency measure. The saliency measure captures the pitch perception information in the aperiodicity regions. The key contributions of the thesis are: • Analysis of expressive voices using the recently proposed signal processing meth

Abstract

Efficient power conversion is an extremely important area of research in modern day electronics.Because of wide range of supply voltages in which the ICs operate, the power conversion or the power management IC (PMIC) needs to work from one voltage range and deliver power at different voltage range. Over the last 20 years, automotive industry has seen a lot of evolution from being a purely mechanical system to an electro-mechanical system. Unlike consumer electronic goods like smartphones, automotive electronics demand high quality, lifetime and durability. Most automobiles available these days run on a 12V battery. The commonly used lead-acid battery when fully charged can exhibit a voltage as high as 14V. Because of the miniaturization of image sensors, automotive rear view cameras in particular have started to find their way into middle and lower segment range automobiles. An automotive camera module needs to withstand high input voltage range and deliver multiple output voltages and load currents at very high step-down ratio. The choice of DSP, sensor and micro-controller will determine overall load current of the application. Though the camera system with PMIC will be quite low power in an automobile, it is also an extremely small board in a hot environment and hence good power efficiency becomes important. Therefore, the PMIC needs to have good efficiency not only at peak load condition but also at moderate and light load condition. Recent trends on high-voltage high step-down ratio power converters are pointing towards voltage regulator modules with high power requirements (>10W). The power transistors are discrete and only the controller is integrated into the IC. However, with increasing number of moderate power applications like automotive camera modules, there is renewed interest to look into design of low-cost power converters that can deliver moderate load power (1-3W). The power management unit (PMU) designed needs to be robust to the wide supply voltage transients and withstand high ambient temperature of 105 ◦ C. Due to self-heating of the PMU, the circuits need to be designed to be robust for IC junction temperatures close to 150 ◦ C. An affordable, reliable and highly-efficient automotive reverse camera system is found to have all the required challenges for a VLSI designer to make it a very interesting research problem for this thesis work. A combination of DC-DC (switching converters) and low-dropout (LDO) regulator, which is known as hybrid regulator, is found to give the best compromise in terms of system efficiency, output voltage ripple while keeping the overall component count to the minimum. State-of-the-art hybrid regulators have high-efficiency in a very limited load range. The LDO dropout voltage which is decided by the peak load current requirement is the main source of inefficiency at moderate and light load condition. There are also several works that deal with the light-load efficiency improvement of the switching converters but they have either low supply voltage or limited load current range. This work focusses on a generic system efficiency improvement solution based on the adaptive dropout of the LDO has been proposed. Optimal circuit design architecture with adaptive LDO dropout has been implemented to handle single DC-DC and upto three LDOs with independent voltage and current requirements. AMS 0.35μm high-voltage CMOS technology with 50V transistors and isolated 3.3V transistors have been used to design the power management circuits. An optimal two-chip PMIC solution with hybrid regulators has been proposed to meet the power supply requirements of rear-view camera application. The design of the two PMICs have been successfully completed and fabricated. The packaged ICs have been tested for their functionality. The innovation has been successfully validated in the lab and the experimental results prove the improved efficiency with respect to the conventional hybrid regulator across the 95% of the load current range and IC junction temperature upto 150 ◦ C. The design has been benchmarked also with state-of-the-art commercially available PMICs. Though a specific application like automotive reverse camera has been chosen to define the thesis goals, the novel contributions of the thesis work is very generic and can be applied to any hybrid voltage regulators with DC-DC and LDO combination.

Abstract

Short Term Load Forecasting (STLF) has always been one of the most critical, sensitive and accuracy demanding factors of the Power Systems. An accurate STLF improves not only the system’s economic viability but also its safety, stability and reliability in operation to realize futuristic Smart Power System. The research presented in this work supports the argument of hybrid approach whereby the complementary strengths of different intelligent techniques are combined to offer better solution to the problem of STLF. This work presents Artificial Intelligence (AI) and Data Mining based formulation for STLF combined with Statistical Techniques. The Temperature, Humidity and Day Type are considered as they are significant factors impacting the effectiveness of an accurate STLF. A Euclidean Norm based Similar Day Approach using the Correction Factors generated by the Fuzzy Inference System has been developed initially for the STLF and novelty is introduced in this methodology by assigning weights to various variables used in the Euclidean Norm to increase the effectiveness of various variables in deciding the similarity. The Fuzzy Inference System (FIS) is designed with three input membership functions, Load Difference (ΔE L ), Temperature Difference (ΔE T ) and Humidity Difference (ΔE H ) taking Trapezoidal Memberships and one output membership function which is the Correction Factor taking a Triangular Membership. Mamdani inference system is used for the FIS. Twenty seven rules are predefined based on the data history and these rules govern the functioning of the FIS. This Euclidean Norm based FIS is used to perform the STLF. The methodology basically uses the Euclidean Norm to select the few similar days of the forecast day from the historic dataset. These similar days of the forecast day are then corrected using the correction factors generated by the FIS that uses the forecast previous day and its similar days’ hourly load, hourly temperature and hourly humidity differences as the input to the FIS. The average of the corrected hourly loads of the similar days of the forecast day is then considered as the hourly load of the forecast day. The Euclidean Norm based FIS has been tested on a dataset of 7 months data. Further to this the input parameter limits of the FIS have been optimized using three Swarm Intelligence Techniques namely Particle Swarm Optimization (PSO), New Particle Swarm Optimization (NPSO) and Evolutionary Particle Swarm Optimization (EPSO). The optimized PSO-FIS, NPSO-FIS and EPSO-FIS were used to perform the STLF on a 7 months dataset as well as data of 3 years as historic dataset. The results of all the three techniques were found to be good. Further another novel methodology, which amalgamates the Clustering Technique of Data Mining with the Regression Technique, has been developed in this thesis to give a more accurate STLF. The framework of the Clustering based Regression methodology is detailed as follows: to forecast the load of a given day, the forecast day and the similar days (all Mondays for Monday and likewise) of the forecast day selected from the previous two years history data are clustered using the k- medoids clustering algorithm, based on their similarity of weather variables i.e. Temperature and Humidity. Further, the proposed method finds the Regression constants from the cluster matrix containing the forecast day. The load forecasting for the forecast day is then carried out using the Regression Equation. The zest of this technique is that clustering brings together the very similar days of the forecast day in one cluster and regression technique further encapsulates the total correlation of load and weather variables of the similar days in the cluster, hence enhancing the forecast efficiency. Historical data set, which captures the impact of weather variables on the load, is used for simulation studies and the proposed technique has been found very successful for load forecast of all days and all seasons. The research presented in this thesis also deals with the very important issue of STLF for special days. This research presents a Hybrid Data Mining based formulation of STLF with emphasis on special days and anomalous days, such as public and national holidays, which are often ignored during the general modeling process. The methodology proposed in present research generates STLF for special days using an Artificial Intelligence based formulation combined with Statistical Techniques. The methodology basically uses the Clustering technique of data mining and combines it with the Regression technique with a variation in the selection of days in historic data set. The proposed framework is detailed as follows:

Abstract

This thesis proposes new signal processing methods to highlight some robust speech-specific features present in the degraded speech. It considers different types of degradations that occur in practice. The signal processing methods are based on single frequency filtering (SFF) of speech signal. The SFF output gives magnitude or envelope and the phase of the speech signal at any desired frequency with high frequency resolution. The SFF output at each frequency gives some segments with high signal-to-noise ratio (SNR), as the noise power in the near zero bandwidth resonator of the single frequency will be very small, whereas the signal component, if it exists, will have high power. Thus the high SNR regions will be at different times for different frequencies. This property of the SFF analysis of speech is exploited for extracting a few robust features from the degraded speech, irrespective of the type and extent of degradation. In particular, the following studies are carried out in this thesis: - Discrimination of speech/nonspeech regions in degraded speech - Determination of speech regions in speech degraded by transient noise - Extraction of the fundamental frequency from degraded speech - Detection of glottal closure instants (GCIs) in degraded speech - Enhancement of degraded speech The major contributions of this work are the following: (a) A new signal processing method called single frequency filtering (SFF) method is proposed which gives high signal-to-noise ratio (SNR) regions in both time and frequency domains for speech affected with different types of degradations. (b) A new method for speech/nonspeech detection is proposed exploiting the high SNR features in the SFF outputs of degraded speech. The procedure works for all types of degradations, without specifically tuning for any specific type of degradation. (c) The high SNR characteristic of the SFF output is also exploited for estimating the fundamental frequency ( f o ) by exploiting information at the frequency that gives the highest SNR for that segment. (d) The noise compensation technique proposed for voice activity detection (VAD) is applied for extracting the location of the significant impulse-like excitation within a glottal cycle. This is because the noise compensated envelopes show distinct changes in the slope of the spectral variance computed as a function of time. (e) The noise compensated SFF envelopes derived at different frequency resolutions are used to derive gross and fine weight functions, as a function of time. The combined weight functions when applied to the degraded speech signal produces enhanced speech for speech affected by different types of degradations, thus improving the comfort level of listening.

Abstract

Sentence Classification is one of the most fundamental tasks in Natural Language Processing (NLP), wherein the aim is to classify any given sentence into one of a given set of classes. The set of classes usually depends upon the specific task under consideration, such as determining the type of answer in Question Classification or finding the sentiment of a sentence in Sentiment Analysis or finding whether a sentence is subjective or objective in Subjectivity Analysis. These are some of the tasks that can be included in Sentence Classification. Some examples of other tasks being sarcasm or humour detection, detection of abusive text etc. In sentence classification, the aim is to have a common network architecture for a range of classification tasks, rather than building an independent one for each task. In other words, the aim is to build a task-independent classification system. Many methods have been developed for various sentence classification tasks for English, which usually exploit linguistic resources like parsers making it difficult to adapt them to other languages. There are a huge number of languages spoken around the world, and it is not possible to build an independent classifier for each task in every language. Thus, in this thesis, we explore language-independent task-independent sentence classification focusing on improvement in classification of morphologically rich languages. We perform sentence classification for five datasets in two tasks - Sentiment Analysis and Question Classification and three languages - English, Hindi and Telugu. We present an evaluation of popular deep learning methods for sentence classification on the morphologically rich Indian languages, specifically Hindi and Telugu. For this purpose, we also created a question classification dataset for Hindi, by translating the TREC-UIUC English dataset. We show that character based input can enhance the performance of current classification systems for morphologically rich languages. Finally, we show that our proposed multiInput-CNN variant is able to perform better than our baselines in two out of three tasks in Hindi and Telugu, while giving comparable results for others. Since a huge proportion of the work in sentence classification is specifically for English, the methods are often designed to best suit that language. A lot of the works represent the input sentences as asequence of words in their models. Only a few of them rely on character level representation. Through this work, we introduce a new method for representing a sentence - as a sequence of syllables. It is essential to capture sub-word level information while classifying in a morphologically rich language. Syllables are an ideal choice for this, as in many languages most of the morphemes can be seen as n-grams of syllables. Thus, making them a more intuitive choice when compared with character n-grams to capture morphological information. Also, syllables can lead to a reduction in noise caused due to the usage of character n-grams. Through extensive evaluation, we show that syllables are the best performing input type when compared to words or characters for the morphologically rich languages - Hindi and Telugu. Finally, we make use of attention mechanism to automatically find the best combination of inputs for every dataset and language. We experiment with a number of different attention based CNN models in this part and find that attention can be used to obtain a model that performs decently well across all datasets.

Abstract

Research in Natural Language Processing is expanding in multiple domains and is seeping into all aspects of life with time. With every advancement, the variety of text that can be processed is growing. One such domain is lyrics processing. Songs are vital to the music and film industry and can be analysed to obtain important information such as genre, theme, mood, etc. of the song and supplement the information gathered by the study of its audio features. Bollywood, the Indian film industry makes a lot of revenue making use of songs. The number of songs churned out by this industry is massive and is a rich source of audio and textual data for Natural Language Processing tasks. It also gives us an opportunity to work on data in Hindi which is a relatively less explored field. The focus of this thesis is on the textual part of the data. In an attempt to create a data resource for this domain, this work presents a corpus of Bollywood song lyrics and its metadata, annotated with sentiment polarity. We call this BolLy. It contains lyrics of 1055 songs ranging from those composed in the year 1970 to the most recent ones. This dataset is of utmost value as all the annotation is done manually by three annotators and this makes it a very rich dataset for training purposes. In this work, we describe the creation and annotation process, content, and the possible uses of the dataset. As an experiment, we have built a basic classification system to identify the emotion polarity of the song based solely on the lyrics and this can be used as a baseline algorithm for the same. The lyrics of Hindi songs have been used for their classification as having positive or negative sentiment by extraction of opinions. Some experiments employing subjectivity lexicons and probabilistic approaches were conducted for this on a dataset which was just a subset of ‘BolLy’. This motivated us to work with a bigger dataset and focus our efforts towards expanding it. With the complete ‘BolLy’ dataset, the experiments have been created using different variations of the Naive Bayes Classifier. There can be a multitude of Natural Language Processing applications on the presented dataset. This thesis work contains one of them explored in detail. Keywords of a document are a representative of its content, and it helps to have meaningful words to facilitate search and organization of documents. Hence, finding methods that can automatically identify keywords in a document is very important as manual processes for this is very cumbersome and error-prone.If this task is accomplished for song lyrics, it has varied applications such as recommendation systems and digital music library management. This work proposes and compares methods to identify keywords from lyrics of Bollywood songs. We use a collection of lyrics of 1055 Bollywood songs, all written in the Devanagari script. Experiments include looking at the spatial distribution of the terms, their occurrence in a certain context or position, and using WordNet to generate keywords not present in the document. Validation was done by human annotators by providing a score to each method based on the results obtained on a subset of the data. We also used Latent Dirichlet Allocation and Latent Semantic Indexing to validate the results, as further explained in the paper.

Abstract

Many socio-environmental aspects manifest themselves over space and time, interacting at varying scales of these dimensions. Satellite imagery, available repetitively over a region, provide important clues of these observations across these dimensions. But, also pose enormous challenges in terms of data processing, extracting significant patterns (indicating the underlying processes) and be able to fur- ther model them as scientific knowledge of the environmental process. Study of cropping practices and changes occurring in it over time for a region is critical to the field of agronomics. The increase of temporal resolution of Earth Observation Satellite platforms has re- sulted in easier monitoring of the Earth’s surfaces. Satellite imagery is an exemplary data source that captures the vegetation responses of the field and such dataset can be used to effectively monitor and track vegetation health over time. The need for large-scale and structured analysis of satellite data for agricultural monitoring is growing. One of the current challenges in understanding agricultural prac- tices is to build a history of land-use outcomes and extracting the local phenological responses that help in building field-based inventory and understanding gross estimate of various cropping systems. Also, monitoring of vegetation health and identification of the major changes in crop practice described here as change-event can provide valuable insights into the possible factors that were responsible for such an event be it biological, physical, hydrological or climatic. In this study, an effort has been made to propose a time-variant analysis method based on the shape characteristics of the vegetation response over time to help identify regions of significant changes. This study attempts to build a spatio-temporal library of vegetation practices and presents an approach, supervised and unsupervised, to process the raw data obtained from the satellite images into representative phenological growth curves. The statis- tics contributed by the current work is critical to understanding and assessing the changes in cropping practices over time and detecting possible drivers that triggered to cause such change-events. The study covers four agricultural-year periods between 2008 and 2012 over the district of West Godavari, in the south of India. An interesting result from this study was to observe the impact of drought on cropping cycles during and after the drought year for the given region. The work finds that the immediate effects of drought in a given year are limited if the region is well endowed with other resources such as irrigation facility etc. Whereas its impacts on cropping patterns can also be seen in the season following the drought year. And such changes vary across the district depending on where vi vii cropping is spatially located. The approach uncovers that the effect of the 2009 drought year on the agricultural practices vary spatially depending on the access to resources and the time-lag that manifests itself in such processes. In this study, we also find that nearly 80% of the region is well endowed and hence resilient to the climatic vagaries. The major contribution of this work is the analysis of spatio-temporal data to detect changes that may help in capturing episodic and periodic change-events occurring in crop cycles of any given region. With vast data spanning across multiple regions in hand, the proposed shape-based approach that characterize the phenological nature of these time-dependent sequences help to discover regions that are spatially correlated and find crop practices that are prevalent across similarly endowed regions.