Classification of Buried Objects from Ground Penetrating Radar Images by using Second Order Deep Learning Models

TL;DR

A new classification model based on covariance matrices is built in order to classify buried objects and it is shown that this approach outperform shallow networks designed for GPR data and conventional CNNs typically used in computer vision applications, particularly when the number of training data decreases and in the presence of mislabeled data.

Abstract

In this paper, a new classification model based on covariance matrices is built in order to classify buried objects. The inputs of the proposed models are the hyperbola thumbnails obtained with a classical Ground Penetrating Radar (GPR) system. These thumbnails are then inputs to the first layers of a classical CNN, which then produces a covariance matrix using the outputs of the convolutional filters. Next, the covariance matrix is given to a network composed of specific layers to classify Symmetric Positive Definite (SPD) matrices. We show in a large database that our approach outperform shallow networks designed for GPR data and conventional CNNs typically used in computer vision applications, particularly when the number of training data decreases and in the presence of mislabeled data. We also illustrate the interest of our models when training data and test sets are obtained from different weather modes or considerations.

Figures (15)

• Figure 1: Waveform emitted by classical GPR, called Ricker.
• Figure 2: GPR Principle with acquisition (left), creation of the A-scan (middle) and the B-scan (right).
• Figure 3: Examples of preprocessed GPR images with the 200 MHz antenna in wet sand (after direct wave suppression and histogram correction thanks to GPRpy which is an open-source Ground Penetrating Radar processing and visualization software available in https://github.com/NSGeophysics/GPRPy) for different buried objects: Wooden shelter, dummy shell (Metal) and wooden board coated with rubber (Non-Metal). Noticed that the 3 radargrams have different scales.
• Figure 4: Architecture of model CNN1 almaimani18
• Figure 5: Illustration of the two architectures used in this paper. In RCNet (Residual Covariance Network), we take only the last output of the ResNet blocks while in SRCNet (Stacked Residual Covariance Network), we stack the first 32 outputs (to save memory space) of the outputs features by interpolating them to a common size of $38\times 20$. With RCNet (Residual Covariance Network), we have $h_i=54$ and $w_i=30$ for the first 3 layers and $h_i=28$ and $w_i=15$ for the others.