Papers at Engineering Conferences

These papers are typically written from a machine intelligence or information-processing view point and are strong on technique, but only glance upon results, motivations and underlying scientific ideas.

Overview of radar compression (SPIE 2007, invited paper)

In (13), we describe how radar data is transmitted, compressed and archived. We note that although custom compression techniques have been devised for radar data that outperform generic techniques, radar operations groups ultimately use off-the-shelf solutions. We also point out that the underlying ideas behind compressibity are useful beyond just reducing the amount of data for transmission and archival. The compressibility of radar data has been found useful for devising quality control algorithms, especially for the detection and removal of test patterns.

[ read (pdf) online ]. [ presentation slides (ppt) ].

A Technique for creating probabilistic spatio-temporal forecasts (ICAPR 2007, invited paper)

In (14), we describe how to create good probabilistic forecasts when the entity to be forecast can move and morph. The technique involves clustering Doppler radar-derived fields such as low-level shear and reflectivity to form candidate regions. Assuming stationarity, the spatial probability distribution of the regions is estimated, conditioned based on the level of organization within the regions and combined with the probability that a candidate region becomes severe. [ read (pdf) online ].

A Spatiotemporal Approach to Tornado Prediction (IJCNN 2005)

In (15), we formulate the tornado prediction problem differently. Instead of devising a machine intelligence approach to classify detections, we formulate the problem as a spatio-temporal one: of estimating the probability of a tornado event at a particular spatial location within a given time window. We also describe our initial approach to addressing this differently formulated problem. [ read (pdf) online ].

Quality Control of Weather Radar Data Using Texture Features and a Neural Network (Int'l Conf. on Patt. Recog. 2003 (Invited Paper))

Weather radar data is subject to many contaminants, mainly due to non-precipitating targets (such as insects and wind-borne particles) and due to anamalous propagation (AP) or ground clutter. Although weather forecasters can usually identify, and account for, the presence of such contamination, automated weather algorithms are affected drastically. We discuss several local texture features and image processing steps that can be used to discriminate some of these types of contaminants. None of these features by themselves can discriminate between precipitating and non-precipitating areas. A neural network is used for this purpose. We discuss training this neural network using a million-point data set, and accounting for the fact that even this data set is necessarily incomplete (16). [ read (pdf) online ]. [ read (powerpoint) online ].

Nested Partitions Using Texture Segmentation (SSIAI 2002)

A multistep method of partioning the pixels of an image such that the partitions at one step are wholly nested inside the partitions of the next step is described, i.e. we describe an agglomerative, hierarchical segmentation technique that uses texture information to perform the segmentation (17). The image is requantized using K-Means clustering. Then, clusters are expanded using region growing and morphological processing. This provides the most detailed level of segmentation. The next coarser segmentation levels are obtained by steadily relaxing the intercluster distance between the clusters that is allowed by the morphological processing. Results are demonstrated on real-world images and swathes of Brodatz textures. [ read (pdf) online ].

Texture-Based Segmentation of Satellite Weather Imagery (ICIP 2000)

This paper, in the International Conference on Image Processing (ICIP 2000) (18), describes a method of segmenting satellite weather images using a Kolmogorov-Smirnov test on the distribution of local texture within an image. [ read (pdf) online ].

Detecting Rare Signatures (ANNIE 97)

In (19), you will find a description of problems associated with detecting rare signatures. I describe various skill measures and list some of the genetic algorithm and fuzzy classifier modifications that need to be made to handle rare signature detection. [ read (html) online ].

BWER - fuzzy logic classifier (ANNIE 96)

At Annie (20), we discussed the limitations of classification done by thresholding a weighted sum and introduced a fuzzy logic classifier, comparing its performance with that of the weighted-sum thresholder used in (21). [read online]

BWER - fuzzy logic approach (IAPR 96)

At the IAPR, (22) and (21) we described a fuzzy logic approach where the various uncertainties associated with a BWER's radar profile (capping, three-dimensional structure, vertical height, reflectivity ranges, proximity to a mesocyclone, etc.) are taken into account. [ read (html) online ].

BWER - fuzzy aggregator for disproportinate classes (IASTED 96)

The fuzzy aggregration operators in the general literature can not handle cases where the categories are disproportionately distributed. This paper (23) describes a fuzzy aggregation operator that works in aggregating disproportionate classes. [ read (html) online ].