SEARCH · Search NASA
Results for “image classification”
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.
SIDDA: SInkhorn Dynamic Domain Adaptation for image classification with equivariant neural networks
Modern neural networks (NNs) often do not generalize well in the presence of a ‘covariate shift’; that is, in situations where the training and test data distributions differ, but the conditional distribution of classification labels given the data remains unchanged. In such cases, NN generalization can be reduced to a problem of learning more robust, domain-invariant features. Domain adaptation (DA) methods include a broad range of techniques aimed at achieving this; however, these methods have struggled with the need for extensive hyperparameter tuning, which then incurs significant computational costs. In this work, we introduce SInkhorn Dynamic Domain Adaptation (SIDDA), an out-of-the-box DA training algorithm built upon the Sinkhorn divergence, that can achieve effective domain alignment with minimal hyperparameter tuning and computational overhead. We demonstrate the efficacy of our method on multiple simulated and real datasets of varying complexity, including simple shapes, handwritten digits, real astronomical observations, and remote sensing data. These datasets exhibit covariate shifts due to noise, blurring, differences between telescopes, and variations in imaging wavelengths. SIDDA is compatible with a variety of NN architectures, and it works particularly well in improving classification accuracy and model calibration when paired with symmetry-aware equivariant NNs (ENNs). We find that SIDDA consistently enhances the generalization capabilities of NNs, achieving up to a ${\approx}40\%$ improvement in classification accuracy on unlabeled target data, while also providing a more modest performance gain of $\lesssim 1\%$ on labeled source data. We also study the efficacy of DA on ENNs with respect to the varying group orders of the dihedral group DN, and find that the model performance improves as the degree of equivariance increases. Finally, if SIDDA achieves proper domain alignment, it also enhances model calibration on both source and target data, with the most significant gains in the unlabeled target domain—achieving over an order of magnitude improvement in the expected calibration error and Brier score. SIDDA’s versatility across various NN models and datasets, combined with its automated approach to domain alignment, has the potential to significantly advance multi-dataset studies by enabling the development of highly generalizable models.
Binary image classification
Motivated by the LANDSAT problem of estimating the probability of crop or geological types based on multi-channel satellite imagery data, Morris and Kostal (1983), Hill, Hinkley, Kostal, and Morris (1984), and Morris, Hinkley, and Johnston (1985) developed an empirical Bayes approach to this problem. Here, researchers return to those developments, making certain improvements and extensions, but restricting attention to the binary case of only two attributes.
Response Classification Images in Vernier Acuity
Orientation selective and local sign mechanisms have been proposed as the basis for vernier acuity judgments. Linear image features contributing to discrimination can be determined for a two choice task by adding external noise to the images and then averaging the noises separately for the four types of stimulus/response trials. This method is applied to a vernier acuity task with different spatial separations to compare the predictions of the two theories. Three well-practiced observers were presented around 5000 trials of a vernier stimulus consisting of two dark horizontal lines (5 min by 0.3 min) within additive low-contrast white noise. Two spatial separations were tested, abutting and a 10 min horizontal separation. The task was to determine whether the target lines were aligned or vertically offset. The noises were averaged separately for the four stimulus/response trial types (e.g., stimulus = offset, response = aligned). The sum of the two 'not aligned' images was then subtracted from the sum of the 'aligned' images to obtain an overall image. Spatially smoothed images were quantized according to expected variability in the smoothed images to allow estimation of the statistical significance of image features. The response images from the 10 min separation condition are consistent with the local sign theory, having the appearance of two linear operators measuring vertical position with opposite sign. The images from the abutting stimulus have the same appearance with the two operators closer together. The image predicted by an oriented filter model is similar, but has its greatest weight in the abutting region, while the response images fall to nonsignificance there. The response correlation image method, previously demonstrated for letter discrimination, clarifies the features used in vernier acuity.
Assessing Hurricane Katrina Vegetation Damage at Stennis Space Center using IKONOS Image Classification Techniques
Hurricane Katrina inflicted widespread damage to vegetation in southwestern coastal Mississippi upon landfall on August 29, 2005. Storm damage to surface vegetation types at the NASA John C. Stennis Space Center (SSC) was mapped and quantified using IKONOS data originally acquired on September 2, 2005, and later obtained via a Department of Defense ClearView contract. NASA SSC management required an assessment of the hurricane s impact to the 125,000-acre buffer zone used to mitigate rocket engine testing noise and vibration impacts and to manage forestry and fire risk. This study employed ERDAS IMAGINE software to apply traditional classification techniques to the IKONOS data. Spectral signatures were collected from multiple ISODATA classifications of subset areas across the entire region and then appended to a master file representative of major targeted cover type conditions. The master file was subsequently used with the IKONOS data and with a maximum likelihood algorithm to produce a supervised classification later refined using GIS-based editing. The final results enabled mapped, quantitative areal estimates of hurricane-induced damage according to general surface cover type. The IKONOS classification accuracy was assessed using higher resolution aerial imagery and field survey data. In-situ data and GIS analysis indicate that the results compare well to FEMA maps of flooding extent. The IKONOS classification also mapped open areas with woody storm debris. The detection of such storm damage categories is potentially useful for government officials responsible for hurricane disaster mitigation.
A High Speed VMEbus Based Analog Neurocomputer for Image Classification
Explore the source record for details and available documents.
Assessing Hurricane Katrina Vegetation Damage at Stennis Space Center using IKONOS Image Classification Techniques
Hurricane Katrina hit southwestern Mississippi on August 29, 2005, at 9:45 a.m. CDT as a category 3 storm with surges up to approx. 9 m and sustained winds of approx. 120 mph. The hurricane's wind, rain, and flooding devastated several coastal towns, from New Orleans through Mobile. The storm also caused significant damage to infrastructure and vegetation of NASA's SSC (Stennis Space Center). Storm recovery at SSC involved not only repairs of critical infrastructure but also forest damage mitigation (via timber harvests and control burns to reduce fire risk). This presentation discusses an effort to use commercially available high spatial resolution multispectral IKONOS data for vegetation damage assessment, based on data collected over SSC on September 2, 2005.
Machine-Learning Image Classification and Hazard Identification of Dust and Fog in Next-Generation Geostationary Satellite Imagery
No abstract available
Spectral–Spatial Preprocessing Using Multihypothesis Prediction for Noise-Robust Hyperspectral Image Classification
Explore the source record for details and available documents.
Wavelet-Domain Multiview Active Learning for Spatial-Spectral Hyperspectral Image Classification
Explore the source record for details and available documents.
An Active Learning Framework for Hyperspectral Image Classification Using Hierarchical Segmentation
Explore the source record for details and available documents.
DeepAndes: A Self-Supervised Vision Foundation Model for Multispectral Remote Sensing Imagery of the Andes
By mapping sites at large scales usingremotely sensed data, archaeologists can generate unique insights into long-term demographic trends, interregional social networks, and human adaptations in the past. Remote sensing surveys complement field-based approaches, and their reach can be especially great when combined with deep learning and computer vision techniques. However, conventional supervised deep learning methods face challenges in annotating fine-grained archaeological features at scale. In addition, while recent vision foundation models have shown remarkable success in learning large-scale remote sensing data with minimal annotations, most off-the-shelf solutions are designed for RGB images rather than multispectral satellite imagery, such as the eight-band data used in our study. In this article, we introduce DeepAndes, a transformer-based vision foundation model trained on three million multispectral satellite images, specifically tailored for Andean archaeology. DeepAndes incorporates a customized DINOv2 self-supervised learning algorithm optimized for eight-band multispectral imagery, marking the first foundation model designed explicitly for the Andes region. We evaluate its image understanding performance through imbalanced image classification, image instance retrieval, and pixel-level semantic segmentation tasks. Our experiments show that DeepAndes achieves superior F1 scores, mean average precision, and Dice scores in few-shot learning scenarios, significantly outperforming models trained from scratch or pretrained on smaller datasets. This underscores the effectiveness of large-scale self-supervised pretraining in archaeological remote sensing.
FPGA Coprocessor for Accelerated Classification of Images
An effort related to that described in the preceding article focuses on developing a spaceborne processing platform for fast and accurate onboard classification of image data, a critical part of modern satellite image processing. The approach again has been to exploit the versatility of recently developed hybrid Virtex-4FX field-programmable gate array (FPGA) to run diverse science applications on embedded processors while taking advantage of the reconfigurable hardware resources of the FPGAs. In this case, the FPGA serves as a coprocessor that implements legacy C-language support-vector-machine (SVM) image-classification algorithms to detect and identify natural phenomena such as flooding, volcanic eruptions, and sea-ice break-up. The FPGA provides hardware acceleration for increased onboard processing capability than previously demonstrated in software. The original C-language program demonstrated on an imaging instrument aboard the Earth Observing-1 (EO-1) satellite implements a linear-kernel SVM algorithm for classifying parts of the images as snow, water, ice, land, or cloud or unclassified. Current onboard processors, such as on EO-1, have limited computing power, extremely limited active storage capability and are no longer considered state-of-the-art. Using commercially available software that translates C-language programs into hardware description language (HDL) files, the legacy C-language program, and two newly formulated programs for a more capable expanded-linear-kernel and a more accurate polynomial-kernel SVM algorithm, have been implemented in the Virtex-4FX FPGA. In tests, the FPGA implementations have exhibited significant speedups over conventional software implementations running on general-purpose hardware.
The development of an MSS satellite imagery classification expert system
Unsupervised image classification of Landsat MSS imagery entails a significant part of the remote sensing, image analysis effort. Expert systems, a technology developed in the field of artificial intelligence, offers the potential to automate this process, thus greatly increasing the efficiency with which an analyst can perform unsupervised image classification and making the knowledge of the image analyst available to a community of nonexperts. Such a system, under development at the NASA/Ames Research Center, is described and planned enhancements are discussed.
A Rorschach Test for Visual Classification Strategies
Contemporary models of pattern, detection and discrimination often employ template matching, but there have been few direct tests of this proposition. Adopting a method developed by Ahumada, we have analyzed how human observers discriminate between two letters of the alphabet ('c' and 'x'). The stimulus consisted of a one degree tall letter plus a four degree field of static white noise, both displayed for 16 frames at a 67 Hz frame rate. Our font and display dimensions approximated those of Solomon and Pelli. The observer identified the letter presented. A QUEST staircase varied letter contrast to maintain a 75% correct rate. For each trial, we preserved the information required to reconstruct the noise field. Possible trial categories based on (signal, response) pairs are: (c,c), (c,x), (x,c), (x,x). Noise fields were averaged separately for each category, and a final classification image was obtained by averaging the four mean images after inverting the sign of categories in which x was the response. If the observer employs a template, it should be revealed in the classification image. The lowpass-filtered classification image derived from 2048 responses of one observer is shown here, along with the corresponding ideal template. An approximation to the ideal template can be seen appropriately located within the classification image. We have also simulated and will discuss the classification images expected from various discrimination models in this experimental context. The construction of classification images appears to be a powerful tool for studying classification strategies used by human observers. Like a Rorschach test, it surreptitiously discovers the inner desires of the visual system.
Automatic interpretation and classification of images by use of the Fourier domain
Automatic image interpretation and classification using two dimensional digital Fourier transforms
Potential of VIIRS Data for Regional Monitoring of Gypsy Moth Defoliation: Implications for Forest Threat Early Warning System
A NASA RPC (Rapid Prototyping Capability) experiment was conducted to assess the potential of VIIRS (Visible/Infrared Imager/Radiometer Suite) data for monitoring non-native gypsy moth (Lymantria dispar) defoliation of forests. This experiment compares defoliation detection products computed from simulated VIIRS and from MODIS (Moderate Resolution Imaging Spectroradiometer) time series products as potential inputs to a forest threat EWS (Early Warning System) being developed for the USFS (USDA Forest Service). Gypsy moth causes extensive defoliation of broadleaved forests in the United States and is specifically identified in the Healthy Forest Restoration Act (HFRA) of 2003. The HFRA mandates development of a national forest threat EWS. This system is being built by the USFS and NASA is aiding integration of needed satellite data products into this system, including MODIS products. This RPC experiment enabled the MODIS follow-on, VIIRS, to be evaluated as a data source for EWS forest monitoring products. The experiment included 1) assessment of MODIS-simulated VIIRS NDVI products, and 2) evaluation of gypsy moth defoliation mapping products from MODIS-simulated VIIRS and from MODIS NDVI time series data. This experiment employed MODIS data collected over the approximately 15 million acre mid-Appalachian Highlands during the annual peak defoliation time frame (approximately June 10 through July 27) during 2000-2006. NASA Stennis Application Research Toolbox software was used to produce MODIS-simulated VIIRS data and NASA Stennis Time Series Product Tool software was employed to process MODIS and MODIS-simulated VIIRS time series data scaled to planetary reflectance. MODIS-simulated VIIRS data was assessed through comparison to Hyperion-simulated VIIRS data using data collected during gypsy moth defoliation. Hyperion-simulated MODIS data showed a high correlation with actual MODIS data (NDVI R2 of 0.877 and RMSE of 0.023). MODIS-simulated VIIRS data for the same date showed moderately high correlation with Hyperion-simulated VIIRS data (NDVI R2 of 0.62 and RMSE of 0.035), even though the datasets were collected about a half an hour apart during changing weather conditions. MODIS products (MOD02, MOD09, and MOD13) and MOD02-simulated VIIRS time series data were used to generate defoliation mapping products based on image classification and image differencing change detection techniques. Accuracy of final defoliation mapping products was assessed by image interpreting over 170 randomly sampled locations found on Landsat and ASTER data in conjunction with defoliation map data from the USFS. The MOD02-simulated VIIRS 400-meter NDVI classification produced a similar overall accuracy (87.28 percent with 0.72 Kappa) to the MOD02 250-meter NDVI classification (86.71 percent with 0.71 Kappa). In addition, the VIIRS 400-meter NDVI, MOD02 250-meter NDVI, and MOD02 500-meter NDVI showed good user and producer accuracies for the defoliated forest class (70 percent) and acceptable Kappa values (0.66). MOD02 and MOD02-simulated VIIRS data both showed promise as data sources for regional monitoring of forest disturbance due to insect defoliation.