MFN Decomposition and Related Metrics for High-Resolution Range Profiles Generative Models

TL;DR

This work focuses on decomposing HRRP data into three components: the mask, the features, and the noise, and proposes two metrics based on the physical interpretation of those data.

Abstract

High-resolution range profile (HRRP ) data are in vogue in radar automatic target recognition (RATR). With the interest in classifying models using HRRP, filling gaps in datasets using generative models has recently received promising contributions. Evaluating generated data is a challenging topic, even for explicit data like face images. However, the evaluation methods used in the state-ofthe-art of HRRP generation rely on classification models. Such models, called ''black-box'', do not allow either explainability on generated data or multi-level evaluation. This work focuses on decomposing HRRP data into three components: the mask, the features, and the noise. Using this decomposition, we propose two metrics based on the physical interpretation of those data. We take profit from an expensive dataset to evaluate our metrics on a challenging task and demonstrate the discriminative ability of those.

Figures (6)

• Figure 1: Three HRRPs of the same ship captured the same day at very close aspect angles. X-axis: range cells, Y-axis: amplitude.
• Figure 2: Example of TLOP and LRP of multiple data at various aspect angles. Red : TLOP, theoretical projection length of a rectangle on the LOS of the radar. Blue : LRP calculated on the data depending on the aspect angle. Each cross corresponds to a data.
• Figure 3: Examples of decomposition (with $\sigma = 0.5$) with comparison metrics between the standard data (cyan) and transformed data (red). From left to right: raw data, m components, f components, and n components. Top: Adding noise to the data transforms slightly the m and f component, while the n component captures most of the noise. Middle: The transformed data is the standard data at lower amplitude. The MSE metric captures this difference, while the cosine similarity is the same for both data. Bottom: We add low-frequency components to the standard data. Our cosine similarity is way more discriminant than the standard one without decomposition.
• Figure 4: Top metrics comparison. Cyan: top metrics between a data and a set of data from the same ship. Red: top metrics between data from different ships. We note a larger difference with our decomposition.
• Figure 5: Relative evolution for both metrics. Cyan: metrics introduced in \ref{['subsec:proposed_metrics']} using the decomposition. Red: standard metrics, we recall that we normalized the standard MSE by the mean LRP of targets.