SEARCH · Engineering Papers
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Classification, decomposition and modeling of polarimetric SAR data has received a great deal of attention in the recent literature. The objective behind these efforts is to better understand the scattering mechanisms which give rise to the polarimetric signatures seen in SAR image data.
This work provides a synopsis of aerosol phase function (F (sub 11)) and polarized phase function (F (sub 12)) measurements made by the Polarized Imaging Nephelometer (PINeph) during the Studies of Emissions, Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys (SEAC4RS) and the Deep Convection Clouds and Chemistry (DC3) field campaigns. In order to more easily explore this extensive dataset, an aerosol classification scheme is developed that identifies the different aerosol types measured during the deployments. This scheme makes use of ancillary data that include trace gases, chemical composition, aerodynamic particle size and geographic location, all independent of PI-Neph measurements. The PI-Neph measurements are then grouped according to their ancillary data classifications and the resulting scattering patterns are examined in detail. These results represent the first published airborne measurements of F (sub 11) and minus F (sub 12) divided by F (sub 11) for many common aerosol types. We then explore whether PI-Neph light-scattering measurements alone are sufficient to reconstruct the results of this ancillary data classification algorithm. Principal component analysis (PCA) is used to reduce the dimensionality of the multi-angle PI-Neph scattering data and the individual measurements are examined as a function of ancillary data classification. Clear clustering is observed in the PCA score space, corresponding to the ancillary classification results, suggesting that, indeed, a strong link exists between the angular scattering measurements and the aerosol type or composition. Two techniques are used to quantify the degree of clustering and it is found that in most cases the results of the ancillary data classification can be predicted from PI-Neph measurements alone with better than 85 percent recall. This result both emphasizes the validity of the ancillary data classification as well as the PI-Neph's ability to distinguish common aerosol types without additional information.
Classification and statistical sampling techniques for crop type discrimination using Landsat digital data have been developed by the University of California in cooperation with NASA and the California Department of Water Resources. Ratioed bands (MSS 7/5 and 5/4) and a sun-angle corrected Euclidean albedo band were prepared from data for the Sacramento Valley for five different dates. The test area was stratified into general crop groupings based on the particular patterns of irrigation timing for each crop. Data classified within each stratum were used to produce a crop type map. Comparison with ground data indicates that certain crops and crop groups are discernable. Small grains and rice are easily identifiable, as are deciduous fruit varieties as a group. However, it is not feasible to separate various fruit and nut varieties, or separate vegetable crops with these techniques at present.
We investigate how multifrequency and polarimetric synthetic aperture radar (SAR) imagery enhances present capability to discriminate different ice conditions in single-frequency, single-polarization satellite SAR data. Frequencies considered are C- (lambda = 5.6cm), L- (lambda = 24cm) and P- (lambda = 68cm) band. Radar backscatter characteristics of six radiometrically and polarimetrically distinct ice types are selected from a cluster analysis of the multifrequency polarimetric SAR data and used to classify SAR images. Validation of these ice conditions is based on information provided by aerial photos, weather and ice surface measurements acquired at an ice camp, together with airborne passive microwave imagery, and visual analysis of the SAR data. The six identified sea-ice types are: (1) multiyear sea-ice; (2) compressed first year ice; (3) first year rubble and ridges; (4) first year rough ice; (5) first year smooth ice; and (6) thin ice. Open water is absent in all analyzed data. Classification of the SAR imagery into those six ice types is performed using a Bayesian Maximum A Posteriori classifier. Two complete scenes acquired at different dates in different locations are classified. The scenes were chosen such that they are representative of typical ice conditions in the Beaufort Sea in March 1988 and because ancillary information is available for validating the segmentation of various ice surface conditions.
This paper describes a concept of Joint Classification and Data Compression of multidimensional information sources in the context of applications to Earth Resources Technology Satellites. The concept is used to define an Information Extraction System intended to remove data redundancy early in the sensor to user link in both classification and film interpretation applications. The approach uses cascaded partially supervised clustering to extract spectral intensity features from spatially local sources. Data compression is then used to efficiently represent spatial features within the spectral intensity feature map.
A computer-implemented classification has been derived from Landsat-4 Thematic Mapper data acquired over Baldwin County, Alabama on January 15, 1983. One set of spectral signatures was developed from the data by utilizing a 3x3 pixel sliding window approach. An analysis of the classification produced from this technique identified forested areas. Additional information regarding only the forested areas. Additional information regarding only the forested areas was extracted by employing a pixel-by-pixel signature development program which derived spectral statistics only for pixels within the forested land covers. The spectral statistics from both approaches were integrated and the data classified. This classification was evaluated by comparing the spectral classes produced from the data against corresponding ground verification polygons. This iterative data analysis technique resulted in an overall classification accuracy of 88.4 percent correct for slash pine, young pine, loblolly pine, natural pine, and mixed hardwood-pine. An accuracy assessment matrix has been produced for the classification.
Manual labeling for machine learning tasks such as image classification is tedious and labor-intensive; as a result, scientific datasets suitable for deep learning applications are scarce and limited. While data augmentation techniques have shown promise for extending image datasets, very little work has been done to understand the impact of combining multiple augmentation methods sequentially or the limits of their effectiveness when combined. Our work addresses this gap by examining how standard and combinatorial data augmentation affects the performance of machine learning models when trained on small datasets for label classification tasks. For our analysis, we generate single, double and quadruple-augmented datasets for a microscopy image classification task using six standard augmentation methods, and compare the resultant improvements observed in binary classification accuracy with three standard image classification models (DenseNet169, MobileNetV2, ResNet101V2). Our experiments show a non-monotonic relationship between the number of simultaneous augmentation methods and classification accuracy, indicating that there is a trade-off between the degree of augmentation and the model performance. These findings suggest that the optimal number of augmentation methods will vary by domain and use case. We also find that the order in which augmentation methods are applied to a limited dataset matters when combining augmentation schemes, with our use case showing performance differences up to 2.6% when the augmentation order is reversed for double-augmented datasets. Our work offers insights to the limits of data augmentation when working on image classification tasks with limited datasets.
We present maximum-likelihood (ML) coherent and noncoherent classifiers for discriminating between NRZ and Manchester coded (biphase-L) data formats for binary phase-shift-keying (BPSK) modulation. Such classification of the data format is an essential element of so-called autonomous software defined radio (SDR) receivers (similar to so-called cognitive SDR receivers in the military application) where it is desired that the receiver perform each of its functions by extracting the appropriate knowledge from the received signal and, if possible, with as little information of the other signal parameters as possible. Small and large SNR approximations to the ML classifiers are also proposed that lead to simpler implementation with comparable performance in their respective SNR regions. Numerical performance results obtained by a combination of computer simulation and, wherever possible, theoretical analyses, are presented and comparisons are made among the various configurations based on the probability of misclassification as a performance criterion. Extensions to other modulations such as QPSK are readily accomplished using the same methods described in the paper.
The spectral and textural characteristics of polar clouds and surfaces for a 7-day summer series of AVHRR data in two Arctic locations are examined, and the results used in the development of a cloud classification procedure for polar satellite data. Since spatial coherence and texture sensitivity tests indicate that a joint spectral-textural analysis based on the same cell size is inappropriate, cloud detection with AVHRR data and surface identification with passive microwave data are first done on the pixel level as described by Key and Barry (1989). Next, cloud patterns within 250-sq-km regions are described, then the spectral and local textural characteristics of cloud patterns in the image are determined and each cloud pixel is classified by statistical methods. Results indicate that both spectral and textural features can be utilized in the classification of cloudy pixels, although spectral features are most useful for the discrimination between cloud classes.
The author has identified the following significant results. Color-combined ERTS-1 MSS spectral slices were analyzed to determine the maximum (repeatable) level of meaningful forest resource classification data visually attainable by skilled forest photointerpreters for the following purposes: (1) periodic updating of the Minnesota Land Management Information System (MLMIS) statewide computerized land use data bank, and (2) to provide first-stage forest resources survey data for large area forest land management planning. Controlled tests were made of two forest classification schemes by experienced professional foresters with special photointerpretation training and experience. The test results indicate it is possible to discriminate the MLMIS forest class from the MLMIS nonforest classes, but that it is not possible, under average circumstances, to further stratify the forest classification into species components with any degree of reliability with ERTS-1 imagery. An ongoing test of the resulting classification scheme involves the interpretation, and mapping, of the south half of Itasca County, Minnesota, with ERTS-1 imagery. This map is undergoing field checking by on the ground field cooperators, whose evaluation will be completed in the fall of 1973.
A classification scheme based on unsupervised self-organizing maps (SOM) is described. Results from its application to the ASU mineral spectral database are presented. Applications to the Martian Thermal Emission Spectrometer data are discussed. Additional information is contained in the original extended abstract.
The NASA Earth Resources Laboratory has developed a transferrable system for processing Landsat and disparate data with capabilities for digital data classification, georeferencing, overlaying, and data base management. This system is known as the Earth Resources Data Analysis System. The versatility of the system has been demonstrated with applications in several disciplines. A description is given of a low-cost data system concept that is suitable for transfer to one's available in-house minicomputer or to a low-cost computer purchased for this purpose. Software packages are described that process Landsat data to produce surface cover classifications and that geographically reference the data to the UTM projection. Programs are also described that incorporate several sets of Landsat derived information, topographic information, soils information, rainfall information, etc., into a data base. Selected application algorithms are discussed and sample products are presented. The types of computers on which the low-cost data system concept has been implemented are identified, typical implementation costs are given, and the source where the software may be obtained is identified.
A classification algorithm incorporating contextual information in a general, statistical manner is presented. Methods are investigated for obtaining adequate estimates of the context distribution (a statistical characterization of context) upon which the classification algorithm depends. Finally, a method of estimating optimal algorithm parameters prior to performing preliminary classifications is explored.
A computer-implemented method of deep packet inspection (DPI) in a network is provided. The method comprises collecting data packets comprising a number of traffic flows from a number of devices via a number of traffic taps and classifying each traffic flow according to data about network protocol layers of the packets comprising the traffic flow. Application layer metadata is extracted from the packets. Traffic flow classification data and the extracted metadata are ingested into a data cluster and normalized. The normalized classification data and extracted metadata is then correlated to other data sets.
Quantitative evaluation of computer-processed ERTS MSS data classifications has shown that major crop species (corn and soybeans) can be accurately identified. The classifications of satellite data over a 2000 square mile area not only covered more than 100 times the area previously covered using aircraft, but also yielded improved results through the use of temporal and spatial data in addition to the spectral information. Furthermore, training sets could be extended over far larger areas than was ever possible with aircraft scanner data. And, preliminary comparisons of acreage estimates from ERTS data and ground-based systems agreed well. The results demonstrate the potential utility of this technology for obtaining crop production information.
A method of classification of digitized multispectral image data is described. It is designed to exploit a particular type of dependence between adjacent states of nature that is characteristic of the data. The advantages of this, as opposed to the conventional per point approach, are greater accuracy and efficiency, and the results are in a more desirable form for most purposes. Experimental results from both aircraft and satellite data are included.
Computer classifications into seven and ten classes of two areas in central Oregon of interest to geology and forestry demonstrate the extraction of information from ERTS-1 data. The area around Newberry Caldera was classified into basalt, rhyolite obsidian, pumice flats, Newberry pumice, ponderosa pine, lodgepole pine, and water classes. The area around Mt. Washington was classified into two basalts, three forest, two clearcut, burn, snow, and water classes. Both also include an unclassified category. Significant details that cannot be extracted from photographic reconstitutions of the data emerge from these classifications, such as moraine locations and paleowind directions. Spectral signatures for the various rocks are comparable to those published elsewhere.
A classification method for digitized multispectral-image data is described. This method is designed to exploit a particular type of dependence between adjacent states of nature that is characteristic of the data. The advantages of this, as opposed to the conventional 'per point' approach, are greater accuracy and efficiency, and the results are in a more desirable form for most purposes. Experimental results from both aircraft and satellite data are included.