Classifying transformation-variant attributed point patterns

Commercial vehicle classification from spectrum parted linked image test‐attributed synthetic aperture radar imagery

“…The models have been advanced to include polarimetric sensor and six degree‐of‐freedom target parameters for monostatic and bistatic conditions [18]. Hence, ASCs can characterise a myriad of structures such as buildings [19], commercial vehicles [15] and military vehicles [20]. The authors in [20] give an excellent and concise summary of what they call global scattering centres for SAR‐based target classification, although the attributions are limited to scattering centre position and amplitude.…”

“…The approach in [20], which follows [24] demonstrates the utility of SAR data synthesis from simple sums of arbitrary scattering centres amplitudes. Adding polarisation and shape attribution to the position and amplitude information as in [15], increases confidence in the ATR classifier, but relies on non‐linear optimisation methods to estimate model parameters. In fact, the works in [15, 21] used high‐performance computing resources and high‐dimensional data to generate training sets.…”

“…In (8), the least trimmed square (LTS) Hausdorf distance [15] compares two sets of attributed points by comparing each point h in set H to each point q in set Q using the Mahalanobis distance In (9), K ≤ N is the trimming factor such that the smallest, sorted Mahalanobis distances are averaged and d k is k th sorted Mahalanobis distance. Moreover, in (10) Σ is the covariance matrix representing the accuracies of the attribute estimates.…”

“…Here, we consider recently reported attributed scattering centres (ASC) models with nearest neighbour classification and propose a classification method using the type of reflection attributed to each scattering centre. We build upon the approaches in [14], where partially attributed scattering centres were proposed as a means for classification and [15], where the least squares trimmed Hausdorf distance measures the similarity between a query and a reference class. However, in this work, the classification is performed with varying signal to noise ratio and with only a sub‐aperture image rather than the complete 360° collection (e.g.…”

Object recognition based on feature matching of scattering centers in SAR imagery

“…One type turns object electromagnetic characterization research to feature extraction from scattering centers [1][2][3][4][5][6][7][8][9] . The other type makes use of matching of scattering center sets [9][10][11][12][13][14][15] .…”