Simultaneous spectral analysis of multiple video sequence data for LWIR gas plumes

Abstract

We consider the challenge of detection of chemical plumes in hyperspectral image data. Segmentation of gas is difficult due to the diffusive nature of the cloud. The use of hyperspectral imagery provides non-visual data for this problem, allowing for the utilization of a richer array of sensing information. We consider several videos of different gases taken with the same background scene. We investigate a technique known as "manifold denoising" to delineate different features in the hyperspectral frames. With manifold denoising, we can bring more pertinent eigenvectors to the forefront. One can also simultaneously analyze frames from multiple videos using efficient algorithms for high dimensional data such as spectral clustering combined with linear algebra methods that leverage either subsampling or sparsity in the data. Analysis of multiple frames by the Nyström extension shows the ability to differentiate between different gasses while being able to group the similar items together, such as gasses or background signatures.

Multi-class Graph Mumford-Shah Model for Plume Detection using the MBO Scheme

“…Unlike the segmentation in Figure 2 (a) where the mountain component (red, the third in the background) has a well defined outline, the spectral clustering results do not provide clear boundaries. Our approach performs better than other unsupervised clustering results on this dataset [12,20].…”

“…The data is provided by the Applied Physics Laboratory at Johns Hopkins University, (see more details in [3]). Prior analysis of this dataset can be found in the works [12,15,20,22]. The authors of [15] implement a semisupervised graph model using a similar MBO scheme.…”

Hyperspectral Image Classification Using Graph Clustering Methods

“…In [71], a binary partition tree method is used to retrieve the real location and the extent of the plume. Moreover, the author of [68] proposes two ways to compute meaningful eigenvectors of the graph Laplacian. Other detection methods for hyperspectral plumes include [40] (MWIR) and [54] …”

Graph MBO method for multiclass segmentation of hyperspectral stand-off detection video

“…This is done by obtaining a small sample set from the data and performing matrix completion that utilizes properties of eigenfunctions to complete the Laplacian. The algorithm described in [10] and [5] is: 1) Randomly select k data points to form the set A, while the rest of data forms the set B consisting of n data points. The best utilization of Nyström has n >> k, so the sample set is much smaller then the rest of the data.…”

“…One hyperspectral image is captured every five seconds. This data set has been studied in other works such as [3], [4], [5]. Prior work on hyper spectral plume detection using other sensors includes [6] (MWIR) and [7] (HYDICE).…”