Paheding Sidike

Trends in oil spill detection via hyperspectral imaging

In hyperspectral imaging, pixels of interest generally incorporate information from disparate components which requires quantitative decomposition of these pixels to extract desired information. Since hyperspectral sensors collect data in hundreds of spectral bands, it is essential to perform spectral unmixing to identify the spectra of all endmembers in the pixel in order to ascertain the fractional abundances of pure target spectral signatures. By extracting desired spectral signature from high-dimensional remotely sensed hyperspectral imagery, one can detect and identify objects in vast geographical regions. While numerous algorithms were developed for target detection in hyperspectral imagery, a unified and synergistic approach to evaluate the performance of these algorithms for oil spill detection in ocean environment is yet to be done. Consequently, in this paper, we investigate and compare the performance of five most widely used target detection algorithms for the identification and tracking of surface and subsurface oil spills in ocean environment. Test results using real life oil spill based hyperspectral image datasets show that the spectral fringe-adjusted joint transform correlation technique and the constrained energy minimization technique yield better results compared to alternate techniques.

State Preserving Extreme Learning Machine for face recognition

Extreme Learning Machine (ELM) has been introduced as a new algorithm for training single hidden layer feed-forward neural networks (SLFNs) instead of the classical gradient-based algorithms. Based on the consistency property of data, which enforce similar samples to share similar properties, ELM is a biologically inspired learning algorithm with SLFNs that learns much faster with good generalization and performs well in classification applications. However, the random generation of the weight matrix in current ELM based techniques leads to the possibility of unstable outputs in the learning and testing phases. Therefore, we present a novel approach for computing the weight matrix in ELM which forms a State Preserving Extreme Leaning Machine (SPELM). The SPELM stabilizes ELM training and testing outputs while monotonically increases its accuracy by preserving state variables. Furthermore, three popular feature extraction techniques, namely Gabor, Pyramid Histogram of Oriented Gradients (PHOG) and Local Binary Pattern (LBP) are incorporated with the SPELM for performance evaluation. Experimental results show that our proposed algorithm yields the best performance on the widely used face datasets such as Yale, CMU and ORL compared to state-of-the-art ELM based classifiers.

A fast single-image super-resolution via directional edge-guided regularized extreme learning regression

The goal of super-resolution (SR) is to increase the spatial resolution of a low-resolution (LR) image by a certain factor using either single or multiple LR input images. This paper presents a machine learning-based approach to reconstruct a high-resolution (HR) image from a single LR image. Inspired by the human visual cortex system, which is sensitive to high-frequency (HF) components in an image, we aim to model this concept by training a neural network to estimate the missing HF components that contain structural details. In our method, various directional edge responses at each pixel are considered to obtain more complete HF information and then a regularized extreme learning regression model is trained using a set of LR and HR images. Finally, the trained system is applied to a LR image to generate HR image. The experimental results confirm the effectiveness and efficiency of the proposed scheme in comparison with the state-of-the-art SR methods.

State Preserving Extreme Learning Machine: A Monotonically Increasing Learning Approach

Extreme Learning Machines (ELM) has been introduced as a new algorithm for training single hidden layer feedforward neural networks instead of the classical gradient-based approaches. Based on the consistency property of data, which enforces similar samples to share similar properties, ELM is a biologically inspired learning algorithm that learns much faster with good generalization and performs well in classification tasks. However, the stochastic characteristics of hidden layer outputs from the random generation of the weight matrix in current ELMs leads to the possibility of unstable outputs in the learning and testing phases. This is detrimental to the overall performance when many repeated trials are conducted. To cope with this issue, we present a new ELM approach, named State Preserving Extreme Leaning Machine (SPELM). SPELM ensures the overall training and testing performance of the classical ELM while monotonically increases its accuracy by preserving state variables. For evaluation, experiments are performed on different benchmark datasets including applications in face recognition, pedestrian detection, and network intrusion detection for cyber security. Several popular feature extraction techniques, namely Gabor, pyramid histogram of oriented gradients, and local binary pattern are also incorporated with SPELM. Experimental results show that our SPELM algorithm yields the best performance on tested data over ELM and RELM.

Spectral unmixing of hyperspectral data for oil spill detection

Spectral unmixing is a popular tool for remotely sensed hyperspectral data interpretation and classification. It aims at identifying the spectra of all endmembers in the scene to find the fractional abundances of pure spectral signatures in each mixed pixel collected by an imaging spectrometer. Complete spectral unmixing exploits the theory that the reflectance spectrum of any pixel is the result of linear combinations of the spectra of all endmembers inside that pixel and simply solves a set of l linear equations for each pixel, where l is the number of bands in the image. But often the estimation of all the endmember signatures may be difficult due to the unavailability of pure spectral signatures in the original data, or inadequacy of spatial resolution. For such cases, partial unmixing can be used where only the user chosen targets need to be mapped and the unmixing equations are partially solved. Like complete unmixing, a pixel value in the output image of partial unmixing is proportional to the fraction of the pixel that contains the target material. In this paper, we study the partial spectral unmixing problem under the light of recent theoretical results published in those areas. Our experimental results, which are conducted using real hyperspectral data sets collected by the NASA Jet Propulsion Laboratory’s Airborne Visible Infrared Imaging Spectrometer (AVIRIS) and spectral libraries publicly available; indicate the potential of partial unmixing techniques in the task of accurately characterizing the mixed pixels using the library spectra. Furthermore, we provide a comparison of complete spectral unmixing and partial spectral unmixing for the oil spill detection in the sea.

Intrusion detection in aerial imagery for protecting pipeline infrastructure

We present an automated mechanism that can detect and issue warnings of machinery threat such as the presence of construction vehicles on pipeline right-of-way. The proposed scheme models the human visual perception concepts to extract fine details of objects by utilizing the corners and gradient histogram information in pyramid levels. Two real-world aerial image datasets are used for testing and evaluation.

A modular approach for key-frame selection in wide area surveillance video analysis

This paper presents an efficient preprocessing algorithm for big data analysis. Our proposed key-frame selection method utilizes the statistical differences among subsequent frames to automatically select only the frames that contain the desired contextual information and discard the rest of the insignificant frames. We anticipate that such key frame selection technique will have significant impact on wide area surveillance applications such as automatic object detection and recognition in aerial imagery. Three real-world datasets are used for evaluation and testing and the observed results are encouraging.

Automatic segmentation of carcinoma in radiographs

A strong emphasis has been made on making the healthcare system and the diagnostic procedure more efficient. In this paper, we present an automatic detection technique designed to segment out abnormalities in X-ray imagery. Utilizing the proposed algorithm allows radiologists and their assistants to more effectively sort and analyze large amount of imagery. In radiology, X-ray beams are used to detect various densities within a tissue and to display accompanying anatomical and architectural distortion. Lesion localization within fibrous or dense tissue is complicated by a lack of clear visualization as compared to tissues with an increased fat distribution. As a result, carcinoma and its associated unique patterns can often be overlooked within dense tissue. We introduce a new segmentation technique that integrates prior knowledge, such as intensity level, color distribution, texture, gradient, and shape of the region of interest taken from prior data, within segmentation framework to enhance performance of region and boundary extraction of defected tissue regions in medical imagery. Prior knowledge of the intensity of the region of interest can be extremely helpful in guiding the segmentation process, especially when the carcinoma boundaries are not well defined and when the image contains non-homogeneous intensity variations. We evaluate our algorithm by comparing our detection results to the results of the manually segmented regions of interest. Through metrics, we also illustrate the effectiveness and accuracy of the algorithm in improving the diagnostic efficiency for medical experts.

Logarithmic fringe-adjusted joint transform correlation

Abstract. An improved fringe-adjusted joint transform correlation (FJTC) technique is proposed, which employs a new real-valued filter, called the logarithmic fringe-adjusted filter. The Fourier-plane image subtraction technique is applied to the joint power spectrum before applying the inverse Fourier transform to generate better correlation output. The proposed technique yields better correlation output compared to alternate optical pattern recognition techniques when the target is embedded in noise-corrupted input scenes. Test results are presented to verify the performance of the proposed technique.

New wide-area surveillance techniques for protection of pipeline infrastructure

There are millions of miles of pipes buried along the length and breadth of the United States. The areas through which these pipelines run cannot be used for other activities. Furthermore, machinery—such as construction equipment and heavy vehicles—are major threats to pipeline infrastructure. Monitoring is therefore required to know whether a pipeline’s right-of-way is threatened at any time. Rapid advances in sensor technologies have enabled the use of high-end video acquisition systems to monitor the right-of-way of pipelines and have generated a huge amount of data. It is, however, very costly to employ analysts to scan through this data and to identify right-of-way threats. An automated mechanism that can detect these threats and issue warnings is therefore warranted. Several object detection and recognition algorithms have been proposed previously.1–5 Car body edges, edges of a front windshield, and shadows can be used as features for car detection.6 An alternative simple vehicle detection algorithm involves exploring four elongated edge operators.7 A top-down matching method has also been developed for vehicle detection from high-resolution aerial imagery.8 On-line boosting, with an interactive training framework, is another option for automatic car detection.9 A feature-based approach to car detection uses scale-invariant transform features and an affinity propagation algorithm.10 We previously designed a three-stage pattern recognition framework to detect construction equipment in various lighting conditions and different object orientations using monogenic signal representation.11, 12 The majority of these techniques, however, are either computationally expensive or unable to deal with the complex environments associated with aerial Figure 1. A sample result of the local textural features-based segmentation (LTFS) algorithm. (a) The original red, green, blue (RGB) image. (b) The LTFS output. The yellow circle indicates the threat object that has been identified.