Waqas Rasheed

Scene classification for aerial images based on CNN using sparse coding technique

ABSTRACT Aerial scene classification purposes to automatically label aerial images with specific semantic categories. However, cataloguing presents a fundamental problem for high-resolution remote-sensing imagery (HRRS). Recent developments include several approaches and numerous algorithms address the task. This article proposes a convolutional neural network (CNN) approach that utilizes sparse coding for scene classification applicable for HRRS unmanned aerial vehicle (UAV) and satellite imagery. The article has two major sections: the first describes the extraction of dense multiscale features (multiple scales) from the last convolutional layer of a pre-trained CNN models; the second describes the encoding of extracted features into global image features via sparse coding to achieve scene classification. The authors compared experimental outcomes with existing techniques such as Scale-Invariant Feature Transform and demonstrated that features from pre-trained CNNs generalized well with HRRS datasets and were more expressive than low- and mid-level features, exhibiting an overall 90.3% accuracy rate for scene classification compared to 85.4% achieved by SIFT with sparse coding. Thus, the proposed CNN-based sparse coding approach obtained a robust performance that holds promising potential for future applications in satellite and UAV imaging.

Automated visualization for epilepsy surgical evaluation

Epilepsy is a neurological illness which may be controlled by medication but not fully cured. In the case of pharmacoresistant epilepsy, seizures may prove fatal and surgery becomes an option. Compared with EEG (electro-encephalography), MEG (magenetoencephalography) offers greater accuracy in epilepsy localization owing to higher spatial resolution. However, streaming data from MEG system for surgical evaluation involves cumbersome processes, e.g. down-sampling, artifact removal (via Independent Component Analysis most of the time) and time/frequency plotting. This paper proposes an automated framework using open-source solutions to visualize epilepsy. The capability of the framework is demonstrated with a case of epilepsy surgery.

Foreground extraction for real-time crowd analytics in surveillance system

In this paper, we propose an adaptive background modeling algorithm for crowd surveillance system. We employed Approximate Median Method (AMM) along with the Phase congruency edge detector to develop the background model. The resulting foreground of the proposed model was obtained by applying a logical AND operation between binary maps of the (foreground) of the AMM image and the gradient information of the (Phase congruency edge detector) PC. Experimental results demonstrate that the proposed method is highly accurate while providing a processing speed of 24.8 fps allowing its implementation for real time application.

Novel framework for small-world network connectivity analysis for MEG data

MEG is an emerging clinical tool to gather insight on neuronal dysfunction inside the human brain. This paper analyzes data saved as FIF file format at a small-world level. Traditionally, operating cost is very high for artifact removal. Also, the absence of a standardization with availability of more than a dozen MEG recording device vendors makes it difficult to store and process data followed by its comparison. This paper proposes a framework to avoid computation overheads, while it takes care of MEG data file format and calculates coherence with ease and efficiency. Small-world network is analyzed and 3D visualization is obtained.

Measuring height of high-voltage transmission poles using unmanned aerial vehicle (UAV) imagery

ABSTRACT Aerial imagery is important in remote sensing applications. Unmanned aerial vehicle (UAV) has a wide range of applications in remote sensing and presents a substantial cost-effective solution when monitoring objects on the earth’s surface. Moreover, object detection and classification are important aspects of global information system, especially for remote sensing applications and power line monitoring, which are essential for the proper distribution of electricity to consumers. Manual inspection consumes much time and involves risk, especially in remote areas that host dangerous wildlife; hence, UAV-based approaches are more feasible for such monitoring. The authors propose an UAV approach that utilises a digital surface model and incorporates a stereo matching algorithm based on UAV stereo images. The proposed algorithm was based on a graph-cut (GC) algorithm that measured the disparity map. Results were compared with well-known algorithms; including, for example, global and local stereo matching algorithms. The proposed solution introduces and integrates ordering constraints along with a submodular energy minimisation function to/with the GC algorithm to enhance performance. The authors measured sensitivity and recall for all parameters against ground truth data for differently cropped images of 16 power poles. Results showed that the proposed model performed more accurately compared to extant methods.

Image classification based on sparse-coded features using sparse coding technique for aerial imagery: a hybrid dictionary approach

This work offers an approach to aerial image classification for use in remote sensing object recognition, image processing and computer vision. Sparse coding (SC) is used to classify unmanned-aerial-vehicle (UAV) and satellite images because SC representation can generalize a large dataset and improve the detection of distinctive features by reducing calculation time for feature matching and classification. Features from images are extracted based on the following descriptors: (a) Scale Invariant Feature Transform; (b) Histogram of Oriented Gradients; and (c) Local Binary Patterns. SC representation and local image features are combined to represent global features for classification. Features are deployed in a sparse model to store descriptor features using extant dictionaries such as (a) the Discrete Cosine Transform and (b) the Discrete Wavelet Transform. An additional two dictionaries are proposed as developed for the present work: (c) the Discrete Ridgelet Transform (DRT) and (d) the Discrete Tchebichef Transform. The DRT dictionary is constructed by using the Ricker wavelet function to generate finite Ridgelet transforms as basis elements for a hybrid dictionary. Different pooling methods have also been employed to convert sparse-coded features into a feature matrix. Various machine learning algorithms are then applied to the feature matrix to classify objects contained in UAV and satellite imagery data. Experimental results show that the SC model secured better accuracy rates for extracted discriminative features contained in remote sensing images. The authors concluded that the proposed SC technique and proposed dictionaries provided feasible solutions for image classification and object recognition.

Early visual analysis tool using magnetoencephalography for treatment and recovery of neuronal dysfunction

Functional neuroimaging modalities play an important role in deciding the diagnosis and course of treatment of neuronal dysfunction and degeneration. This article presents an analytical tool with visualization by exploiting the strengths of the MEG (magnetoencephalographic) neuroimaging technique. The tool automates MEG data import (in tSSS format), channel information extraction, time/frequency decomposition, and circular graph visualization (connectogram) for simple result inspection. For advanced users, the tool also provides magnitude squared coherence (MSC) values allowing personalized threshold levels, and the computation of default model from MEG data of control population. Default model obtained from healthy population data serves as a useful benchmark to diagnose and monitor neuronal recovery during treatment. The proposed tool further provides optional labels with international 10-10 system nomenclature in order to facilitate comparison studies with EEG (electroencephalography) sensor space. Potential applications in epilepsy and traumatic brain injury studies are also discussed.

Temporal comparison of ocular blood flow in quadrant divisions of optic nerve head before and after water drinking test

Water drinking test is used as stress test for glaucoma, as it produces changes in intra-ocular pressure. Laser speckle flowgraphy (LSFG) system is used to assess blood flow changes in four regions of the optic nerve head in this study. The LSFG system is a relatively new and simple blood flow measuring technique that displays optic nerve head vascular maps providing relative blood flow readings. The results show significant changes in blood flow of the four regions of the optic nerve head namely, superior, inferior, temporal and nasal after drinking water. These changes are important to further the current understanding of auto-regulation phenomenon after water drinking test in healthy subjects. This data can serve as an important baseline data for comparison with patients displaying glaucoma onset and progression.

Investigating eye-strain due to prolonged exposure to low resolution multimedia using LSFG

Gazing at computer screen for long hours may cause eye strain. The strain goes away after the eyes have been relaxed and kept away from the screen for some time; however, prolonged exposure may cause dryness and irritability. The eye-strain is monitored by a laser speckle flowgraphy (LSFG) system in this study. The LSFG system measures and displays choroidal vascular maps providing relative blood flow readings. The results show consistently high erythrocyte movement after watching low definition and low quality video for long time. Moreover, it is also observed that the blood flow tends to stabilize itself after relaxing the eyes for a couple of hours.

Threshold for computing generalized model of default mode network connectivity

Functional connectivity is becoming popular as a second opinion for neurosurgeons and specialists in order to decide on the need for surgical resection, or prescribing medication and appraise prognosis. Neuroimaging modalities such as fMRI, fNIRS, PET, and EEG provide functional connectivity estimation. MEG is the most recent trend in functional connectivity assessment research as it gives more accurate results. The magnetic signals are not disrupted by volume conduction, as in EEG. Besides a reasonable spatial resolution, it offers an extraordinary temporal resolution. However there is a need of a generalized model for default mode network connectivity using MEG. This paper presents a novel method for generating a generalized model and discusses significance of threshold levels in assessing synchronization of activity from various brain regions.