Improved Background Estimation for Gas Plume Identification in Hyperspectral Images

Abstract

Longwave infrared (LWIR) hyperspectral imaging can be used for many tasks in remote sensing, including detecting and identifying effluent gases by LWIR sensors on airborne platforms. Once a potential plume has been detected, it needs to be identified to determine exactly what gas or gases are present in the plume. During identification, the background underneath the plume needs to be estimated and removed to reveal the spectral characteristics of the gas of interest. Current standard practice is to use ``global" background estimation, where the average of all non-plume pixels is used to estimate the background for each pixel in the plume. However, if this global background estimate does not model the true background under the plume well, then the resulting signal can be difficult to identify correctly. The importance of proper background estimation increases when dealing with weak signals, large libraries of gases of interest, and with uncommon or heterogeneous backgrounds. In this paper, we propose two methods of background estimation, in addition to three existing methods, and compare each against global background estimation to determine which perform best at estimating the true background radiance under a plume, and for increasing identification confidence using a neural network classification model. We compare the different methods using 640 simulated plumes. We find that PCA is best at estimating the true background under a plume, with a median of 18,000 times less MSE compared to global background estimation. Our proposed K-Nearest Segments algorithm improves median neural network identification confidence by 53.2%.

Figures (10)

• Figure 1: Example of gas plume identification process: 1. False color LWIR image from the Los Angeles Basin with a simulated N$_2$O plume (shown in red). 2. ACE detection map for a reference N$_2$O signal at 280K and a globally estimated background. 3. Automatically created regions of interest (shown in white). 4. The various background estimation strategies once a ROI has been created, where each method is applied to a single example pixel in the ROI. Green shading indicates which pixels are being used in the estimation process with the exception of K-means++, which shows the cluster assignments, and K-nearest segments which shows the image segmentation boundaries. 5. Individually whitened ROI pixels, and the final averaged whitened spectral signature (black). Example was whitened using the K-nearest segments background estimation with 32 minimum-pixels, complete linkage, and background target separation. 6. A basic overview of the neural network identifier architecture. 7. Neural network prediction confidences. For reference, the top three predictions using global background estimation are None (4.4%), Calcium Sulfate (4.1%), and N$_2$O (3.5%).
• Figure 2: Diagram of the NNI architecture. The input is a 1-D vector, which is then passed through four squeeze-excite blocks, before being passed into a batch norm layer, and a final softmax layer for classification.
• Figure 3: Library reference absorption signatures ($\bm\alpha$) used to simulate gas plumes.
• Figure 4: Background estimation MSE aggregated across all 640 simulated plumes for each method. Green shaded area are the kernel density estimates of the data. The orange lines indicate the median values and the purple "x"s indicate the means. The dashed gray line at one indicates a method MSE equal to the corresponding Global MSE.
• Figure 5: Plot of the distribution of MSE scores for each method when aggregated by signal strength. The horizontal dashed purple line is the average over all plumes, and the dotted orange line is the median over all plumes. The green areas are the density estimates of the distributions. Note the different scalings on the y-axes.