State mixture modelling applied to speech recognition

In state mixture modelling (SMM), the temporal structure of the observation sequences is represented by the state joint probability distribution where mixtures of states are considered. This technique is considered in an iterative scheme via maximum likelihood estimation. A fuzzy estimation approach is also introduced to cooperate with the SMM model. This new approach not only saves calculations from 2N^TT (HMM direct calculation) and N²T (Forward–backward algorithm) to just only 2NT calculations, but also achieves a better recognition result.     

Global continuous fields of vegetation characteristics: A linear mixture model applied to multi-year 8 km AVHRR data

As an alternative to the traditional approach of using predefined classification schemes with discrete numbers of cover types to describe the geographic distribution of vegetation over the Earth's land surface, we apply a linear mixture model to derive global continuous fields of percentage woody vegetation, herbaceous vegetation and bare ground from 8 km Advanced Very High Resolution Radiometer (AVHRR) Pathfinder Land data. Linear discriminants for input into the mixture model are derived from 30 metrics representing the annual phenological cycle, using training data derived from a global network of scenes acquired by Landsat. We test the stability and robustness of the method by assessing the consistency of results derived independently for each year in the 1982 to 1994 AVHRR data set. For those forested locations where land cover variability would not be expected, the percentage woody estimates displayed standard deviations over the 12 years of less than 10%. Problems with the method occur in high latitudes where snow cover in some years and not others produces inconsistencies in the continuous fields. Overall, the results suggest that the method produces fairly consistent results despite apparent problems with artifacts in the multi-year AVHRR data set due to calibration problems, aerosols and other atmospheric effects, bidirectional effects, changes in equatorial crossing time, and other factors. Comparison of continuous fields with other land cover data sets derived from remote sensing suggests 69% to 84% agreement in the per cent woody field, with the highest agreement when per cent woody is averaged over the 12 years. In comparison with regional data sets for the US and Bolivia, the method overestimates per cent woody vegetation for grassland and sparsely wooded locations. We conclude that the method, with possible refinements and more sophisticated methods to include multiple endmembers, improved estimates of endmember values and nonlinear responses of vegetation to proportional cover, can potentially be used to indicate changes in land cover characteristics over time using multi-year data sets as inputs when perfect calibration and consistency between years cannot be assumed.     

Linear mixture model applied to Amazonian vegetation classification

Many research projects require accurate delineation of different secondary succession (SS) stages over large regions/subregions of the Amazon basin. However, the complexity of vegetation stand structure, abundant vegetation species, and the smooth transition between different SS stages make vegetation classification difficult when using traditional approaches such as the maximum likelihood classifier (MLC). Most of the time, classification distinguishes only between forest/non-forest. It has been difficult to accurately distinguish stages of SS. In this paper, a linear mixture model (LMM) approach is applied to classify successional and mature forests using Thematic Mapper (TM) imagery in the Rondônia region of the Brazilian Amazon. Three endmembers (i.e., shade, soil, and green vegetation or GV) were identified based on the image itself and a constrained least-squares solution was used to unmix the image. This study indicates that the LMM approach is a promising method for distinguishing successional and mature forests in the Amazon basin using TM data. It improved vegetation classification accuracy over that of the MLC. Initial, intermediate, and advanced successional and mature forests were classified with overall accuracy of 78.2% using a threshold method on the ratio of shade to GV fractions, a 7.4% increase over the MLC. The GV and shade fractions are sensitive to the change of vegetation stand structures and better capture biophysical structure information.     

Linear search applied to global motion estimation

Gradient-based algorithms for global motion estimation are effective in many image-processing tasks. However, when analytical estimation of derivatives of objective function is not possible, linear search based algorithms such as Powell perform better than the gradient-based ones. In this paper we propose global motion estimation algorithm that exploits linear search based algorithm, particularly Powell, instead of commonly used gradient-based one. We also introduce a new approach for extracting global motion parameters called Two Step Powell-based GME. Using this approach we further improve the Powell-based GME. The proposed Powell-based GME outperforms Gauss–Newton algorithm (gradient-based) in terms of PSNR. The proposed Two Step Powell GME algorithm outperforms Powell-based GME in terms of PSNR and computational time.     

Growth profile based crop yield models: A case study of large area wheat yield modelling and its extendibility using atmospheric corrected NOAA AVHRR data

Wheat growth profile based yield models for 12 districts of Punjab State and 16 districts of Haryana State have been developed using the normalised difference vegetation index (NDVI) derived from NOAA-11 AVHRR data of the 1993-94 cropping season. Atmospheric normalisation of AVHRR data was performed prior to deriving district-level area weighted average NDVI (AWANDVI). The invariant growth profile model suggested by Badhwar was fitted and spectral emergence date, maximum vegetative vigour, peak day value of profile, growth rate and senescence rate, area under the curve, etc. were derived. These parameters were related to the reported district-level wheat yields using multiple regression analysis. A field study was also conducted using a handheld spectro-radiometer at the research station of Punjab Agricultural University (PAU), Ludhiana. From this field experimental data, wheat growth profile parameters were derived which were compared with satellite based parameters. Inversion of the models was carried out to evaluate the results by comparing the reported and predicted wheat yields. The results indicate highly significant fitting of the NDVI profile to the Badhwar model as indicated by multiple linear correlation coefficients and Fisher test. A significant relationship between district-level wheat yields and fractional area under the curve was also observed. The overall correlation of 0.82 for Punjab and Haryana states was obtained between reported yield and growth profile derived parameters. Atmospheric normalisation resulted in improvement of prediction model statistics ( R increased from 0.42 to 0.86). Evaluation of the models indicated that 10 out of 16 districts of Haryana State and 9 out of 12 districts of Punjab State showed relative deviations within 10% between reported and model predicted wheat yields.     

estimation of mixture complexity for count data

For count data, robust estimation of the number of mixture components in finite mixtures is revisited using L₂ distance. An information criterion based on L₂ distance is shown to yield an estimator, which is also shown to be strongly consistent. Monte Carlo simulations show that our estimator is competitive with other procedures in correctly determining the number of components when the data comes from Poisson mixtures. When the data comes from a negative binomial mixture but the postulated model is a Poisson mixture, simulations show that our estimator is highly competitive with the minimum Hellinger distance (MHD) estimator in terms of robustness against model misspecification. Furthermore, we illustrate the performance of our estimator for a real dataset with overdispersion and zero-inflation. Computational simplicity combined with robustness property makes the L₂E approach an attractive alternative to other procedures in the literature.     

Fourier analysis of multi-temporal AVHRR data applied to a land cover classification

A signal processing technique is presented and applied to annual patterns of the Global Vegetation Index (GVI) derived from the Advanced Very High Resolution Radiometer (AVHRR) to examine the frequency distribution of the multi-temporal signal. It is shown that frequencies of the signal are linked to integrated GVI, seasonal variability and subseasonal variability of the land cover type. These characteristics are used to derive a land cover classification.     

Linear modelling and optimal control applied to an indexing process

Indexing (monetary correction) has great consequences in a national economy under inflation; it has been the object of much theoretical debate, possibly because it sets a positive feedback branch around most of the economic activities. One of its few known examples is studied here, from 108 monthly recorded data. The fundamental issue is the least-squares identification of a SISO linear model for the economy; it permits proof of loop stability, and interpretation of the indexing formula under linear minimum variance control theory. This first result shows, however, unsatisfactory behaviour under a step disturbance; this calls for the adoption of an integrator, prior to the design of the optimal stochastic controller. The superiority of the indexing formula that incorporates the integrator and this controller, over the first one and over one of the legal formulae, is shown by simulation.     

Mesoscale grid rainfall estimation from AVHRR and GMS data

Areal rainfall averages derived from rain-gauge observations suffer from limitations not only due to sampling but also because gauges are usually distributed with a spatial bias towards populated areas and against areas with high elevation and slope. For a large river basin, however, heavy rainfall in the mountain upstream can result in severe flooding downstream. In this study, cloud-indexing and cloud model-based techniques were applied to Advanced Very High Resolution Radiometer (AVHRR) and Geostationary Meteorological Satellite (GMS) imager data based on the cloud-top brightness temperature (T _B) and processed for estimating mesoscale grid rainfall. This study aims to improve and refine rainfall estimation in Malaysian monsoons based on cloud model techniques for operational pre-flood forecasting using readily available near-real-time satellite data such as the National Oceanic and Atmospheric Administration (NOAA)-AVHRR and GMS imager. Rain rates between 3 and 12 mm h^?1 were assigned to cloud pixels of hourly coverage AVHRR or GMS data over the Langat Basin area for the duration of the monsoon rainfall event of 27 September to 8 October 2000 in Malaysia. The observed rainfall and quantitative precipitation forecast (QPF) showed an R ² value of 0.9028, while the observed rainfall run-off (RR; recorded) and its simulated data had an R ² value of 0.9263 and the QPF run-off and its simulated data had an R ² value of 0.815. The rainfall estimate was used to simulate the flood event of the catchment. The estimated rainfall over the catchment showed similar flood area coverage to the observed flood event.     

Neural network estimation of air temperatures from AVHRR data

Multilayer feed-forward (MLF) neural networks were employed to estimate air temperatures in Southern Québec (Canada) using Advanced Very High Resolution Radiometer (AVHRR) images. The input variables for the networks were the five bands of the AVHRR image, surface altitude, solar zenith angle, and Julian day. The estimation was carried out using a dataset collected during the growing season from June to September 2000. Levenberg--Marquardt back-propagation (LM-BP) was used to train the networks. The early stopping method was applied to improve the LM-BP and to generalize the networks. Bands 4 and 5, which are used for retrieval of surface temperature, were the most critical components for the estimation. The contribution of Julian day to the precision of estimated air temperature was much superior to that of altitude and solar zenith angle for the dataset of inter-seasonal air temperatures. The network using all five bands, Julian day, altitude, and solar zenith angle provided the best results, with 22 nodes in the hidden layer. In the time series of estimated and station air temperatures, the difference between the temperatures was generally maintained within 2°C on various canopies, even during steep variations in August and September.     

Robust estimation of mixture complexity for count data

For finite mixtures, consistent estimation of unknown number of components, called mixture complexity, is considered based on a random sample of counts, when the exact form of component probability mass functions are unknown but are postulated to belong to some parametric family. Following a recent approach of Woo and Sriram [2006. Robust estimation of mixture complexity. J. Amer. Statist. Assoc., to appear.], we develop an estimator of mixture complexity as a by-product of minimizing a Hellinger information criterion, when all the parameters associated with the mixture model are unknown. The estimator is shown to be consistent. Monte Carlo simulations illustrate the ability of our estimator to correctly determine the mixture complexity when the postulated Poisson mixture model is correct. When the postulated model is a Poisson mixture but the data comes from a negative binomial mixture with moderate to more extreme overdispersion in one of its components, simulation results show that our estimator continues to perform well. These confirm the efficiency of the estimator when the model is correctly specified and the robustness when the model is incorrectly specified. A count dataset with overdispersion and possible zero inflation is analyzed to further illustrate the ability of our estimator to determine the number of components.     

Rice crop identification and area estimation using remotely-sensed data from Indian cropping patterns

Abstract

This study attempts to give an appraisal of the use of remotely-sensed data for different rice cropping patterns specific to India. Using even 80 m spatial resolution data of LANDSAT MSS in the form of false colour composites (FCCs), the rice crop was identified with an accuracy of 90 to 94 per cent from cropping patterns having more than 50 per cent of cropped area under rice. Cropping patterns having less than 50 percent of cropped area under rice and the rest of the area under multiple crops were identified with an accuracy of 75 percent. From the Indian experiences of higher spatial resolution multispcctral scanner data (3m resolution) and in the present study, it was observed that besides higher spatial resolution, acquisition of data at the critical crop windows (the periods of minimum overlap of greenness of rice with other crops) is necessary to reach 90 per cent accuracy for multiple cropping patterns.     

The estimation of atmospheric effects for SPOT using AVHRR channel-1 data

The aim of this paper is to estimate the effect of atmospheric contamination on SPOT satellite data for stereoscopic modelling by using data from the AVHRR instrument carried on the NOAA satellite series. This paper includes the development of an atmospheric correction algorithm for the visible spectral channel data from the AVHRR instrument, and an analysis of the atmospherically corrected AVHRR data from many successive days, bearing in mind that the SPOT data for stereoscopic modelling will be from two orbits which will be separated by several days.     

Earthnet's coordination scheme for AVHRR data

Abstract

Earthnet's plan for a coordinated Advanced Very High Resolution Radiometer (AVHRR) data service in Europe, as outlined by Fusco and Muirhead (1987), is beginning to take'shape. Three High Resolution Picture Transmission (HRPT) stations are contributing to the data acquisition (providing coverage from West Africa to the Middle East and over the North Pole), the AVHRR catalogue is on-line and an autonomous workstation for AVHRR acquisition, processing and archiving to optical disc is in an advanced state of development.     

A multiple-frame approach to crop yield estimation from satellite- remotely sensed data

Many studies have recently explored information from satellite-remotely sensed data (SRSD) for estimating crop production statistics. The value of this information depends on the aerial and spatial resolutions of SRSD. SRSD with fine spatial resolution is costly and aerial coverage is less. Use of multiple frames of SRSD in the estimation process of crop production can increase the precision. We propose an estimator for the average yield of wheat for the state of Haryana, India. This estimator uses information from Wide Field Sensor (WiFS) and Linear Imaging Self Scanner (LISS-III) data from the Indian Remote Sensing (IRS-1D) satellite and crop-cutting experiment data collected by probability sampling design from a list frame of villages. We find that relative efficiencies of multiple-frame estimators are high in comparison to single-frame estimators.     

A hybrid data assimilation scheme for model parameter estimation: Application to morphodynamic modelling

We present a novel algorithm for joint state-parameter estimation using sequential three dimensional variational data assimilation (3D Var) and demonstrate its application in the context of morphodynamic modelling using an idealised two parameter 1D sediment transport model. The new scheme combines a static representation of the state background error covariances with a flow dependent approximation of the state-parameter cross-covariances. For the case presented here, this involves calculating a local finite difference approximation of the gradient of the model with respect to the parameters. The new method is easy to implement and computationally inexpensive to run. Experimental results are positive with the scheme able to recover the model parameters to a high level of accuracy. We expect that there is potential for successful application of this new methodology to larger, more realistic models with more complex parameterisations.     

Mapping sub-pixel burnt percentage using AVHRR data. Application to the Alcalaten area in Spain

The purpose of this work is to estimate at sub-pixel scale the percentage of burnt land using the Advanced Very High Resolution Radiometer (AVHRR) through a simple approach. This methodology is based on multi-temporal spectral mixture analysis (MSMA), which uses a normalized difference vegetation index (NDVI) and a land-surface temperature (LST) image as input bands. The area of study is located in the Alcalaten region in Castellon (Spain), a typical semi-arid Mediterranean region. The results have shown an extension of approximately 55 km² affected by fire, which is only 5% lower than the statistic reports provided by the Environmental Ministry of Spain. Finally, we include a map of the area showing the percentage of estimated burnt area per pixel and its associated uncertainties. The map was validated through supervised classification of an Airborne Hyperspectral Sensor (AHS) image taken on 27 September 2007. Results have a high accuracy, with a mean error of 6.5%.     

Linear regression modelling for the estimation of oil palm age from Landsat TM

This paper investigates the accuracy with which the age since field planting of oil palm (Elaeis guineensis Jacq.) can be estimated from Landsat Thematic Mapper (TM) radiance at pixel and stand scales. The study site, a commercial plantation 30 km south-east of Kuala Lumpur in Selangor, Malaysia, consisted of even-aged blocks from 4 to 21 years old. Spectral data were the six reflective TM bands and three spectral indices. Nonlinear negative relationships between spectral variables and age are compared to published trends in leaf area, stem height and per cent canopy cover for oil palm and other tree plantations. Correlation coefficients between log age and log radiance are moderate and highly significant (p<0.01) for bands 2-5 and 7 (-0.214 to-0.776) at the pixel scale, and increase at the stand scale (r ²=0.985 for log band 5, p<0.01). Relationships are strongest for the mid-infrared bands, especially band 5 (r ²=0.585, p <0.01) and the infrared index (IRI), a normalized difference index of bands 4 and 5 (r ²= 0.48, p<0.01). Direct and inverse linear regression models for log age with log band and log age with IRI squared (IRIsq) were constructed at both scales. Equivalent age was estimated from the models using independent test sets for differing scales and degrees of aggregation of the age classes. Single age classes cannot be estimated accurately at the pixel or stand scales; the lowest RMS error was obtained from the direct model using all bands (RMS error=3.9 years at pixel scale, 2.7 at stand scale). A posteriori aggregation into generalized age classes (<5, 6-10, 11-15, 16-21 years) improved the RMS error but the results were still unacceptably high (2.2, 2.3, 2.7, 6.0 years respectively for direct model 3 using all bands). Acceptable RMS errors down to 0.58 years were obtained for models using IRIsq with generalized age classes developed and applied at the stand scale when variations in ground cover and other variables were averaged out. The spatial pattern of error in equivalent age deserves investigation for precision crop management.     

A new methodology for the estimation of biomass of coniferdominated boreal forest using NOAA AVHRR data

In this paper, a new methodology to estimate the biomass (organic matter) of conifer-dominated boreal forests is developed. It aims to estimate biomass of extensive areas where ground data are limited. First, the principal models are computed using ground measurements and high resolution satellite images. Spectral models are then applied directly to a calibrated AVHRR image mosaic covering the entire area of interest. This methodology was tested quantitatively in Finland, where detailed forest measurement data are available, on an area reaching from the west coast of Norway to the Ural mountains. The methodology appeared to perform beyond pre-test expectation.     

Land-cover classification methods for multi-year AVHRR data

Advanced Very High Resolution Radiometer (AVHRR) data have been extensively used for global land-cover classification, but few studies have taken direct and full advantage of the multi-year properties of AVHRR data. This study focused on generating effective classification features from multi-year AVHRR data to improve classification accuracy. Three types of features were derived from 12-year monthly composite normalized difference vegetation index (NDVI) and channel 4 brightness temperature from the NOAA/NASA Pathfinder AVHRR Land data for land-cover classification. The first is based on the shape of the annual average NDVI or brightness-temperature profile, which was then approximated by a Fourier series. The coefficients estimated by the weighted least-squares method were used for classification. The second and third features were based on the raw periodogram of the time series and the auto-regressive modelling. A global land-cover training database created from Landsat Thematic Mapper and Multi-spectral Scanner imagery was used for training and validation. Both quadrature discriminate analysis (QDA) and linear discriminate analysis (LDA) were explored for classification, and results indicate that QDA performs much better than LDA. The first feature, based on the mean annual shape, produced much better results than the other two features. It was also found that NDVI signals worked better than brightness-temperature signals. That is probably because topof-atmosphere signals were used, and atmospheric contaminations significantly disturb the thermal signals and correlation structures of different cover types. Further validations are needed.