A fisher vector representation of GPR data for detecting buried objects

Andrew Karem; Amine B. Khalifa; Hichem Frigui

doi:10.1117/12.2223731

3 May 2016 A fisher vector representation of GPR data for detecting buried objects

Andrew Karem, Amine B. Khalifa, Hichem Frigui

Proceedings Volume 9823, Detection and Sensing of Mines, Explosive Objects, and Obscured Targets XXI; 98231C (2016) https://doi.org/10.1117/12.2223731
Event: SPIE Defense + Security, 2016, Baltimore, MD, United States

Abstract

We present a new method, based on the Fisher Vector (FV), for detecting buried explosive objects using ground- penetrating radar (GPR) data. First, low-level dense SIFT features are extracted from a grid covering a region of interest (ROIs). ROIs are identified as regions with high-energy along the (down-track, depth) dimensions of the 3-D GPR cube, or with high-energy along the (cross-track, depth) dimensions. Next, we model the training data (in the SIFT feature space) by a mixture of Gaussian components. Then, we construct FV descriptors based on the Fisher Kernel. The Fisher Kernel characterizes low-level features from an ROI by their deviation from a generative model. The deviation is the gradient of the ROI log-likelihood with respect to the generative model parameters. The vectorial representation of all the deviations is called the Fisher Vector. FV is a generalization of the standard Bag of Words (BoW) method, which provides a framework to map a set of local descriptors to a global feature vector. It is more efficient to compute than the BoW since it relies on a significantly smaller codebook. In addition, mapping a GPR signature into one global feature vector using this technique makes it more efficient to classify using simple and fast linear classifiers such as Support Vector Machines. The proposed approach is applied to detect buried explosive objects using GPR data. The selected data were accumulated across multiple dates and multiple test sites by a vehicle mounted mine detector (VMMD) using GPR sensor. This data consist of a diverse set of conventional landmines and other buried explosive objects consisting of varying shapes, metal content, and burial depths. The performance of the proposed approach is analyzed using receiver operating characteristics (ROC) and is compared to other state-of-the-art feature representation methods.

Citation Download Citation

Andrew Karem, Amine B. Khalifa, and Hichem Frigui "A fisher vector representation of GPR data for detecting buried objects", Proc. SPIE 9823, Detection and Sensing of Mines, Explosive Objects, and Obscured Targets XXI, 98231C (3 May 2016); https://doi.org/10.1117/12.2223731

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available