Radar Object Identification Using Transparent Spectral Feature Tests

The patent segments the classification process into distinct modules: a visualization module that extracts and displays specific spectral features (such as spectral peaks, bandwidth, frequency distribution), and a classification module that applies predefined rules to these visualized features. This segmentation makes the decision-making process transparent by showing exactly which features are examined and how classification decisions are reached, eliminating the black box nature of conventional neural networks.

The patent introduces an intermediary visualization module between the raw radar signal and the classification decision. This module acts as a mediator that transforms the raw signal into interpretable spectral features (amplitude spectrum, frequency distribution, spectral centroids) that can be visually inspected and logically analyzed, thereby bridging the gap between complex signal processing and transparent decision-making.

The patent applies local quality by focusing computational resources on extracting specific, locally-relevant spectral features that are characteristic of different target types (such as spectral peak positions, bandwidth measurements, frequency distribution patterns). Rather than analyzing the entire spectrum uniformly, the algorithm identifies and examines only the locally-significant features that provide discriminative power for classification, thereby improving accuracy without proportionally increasing complexity.

The patent employs parameter changes by transforming the raw radar signal into multiple spectral domain representations (amplitude spectrum, phase spectrum, power spectral density) and extracting specific parameters from each (frequency centroids, bandwidth, peak amplitudes). These transformed parameters provide object-specific characteristics that enhance classification accuracy while maintaining manageable algorithmic complexity through systematic feature extraction.