Search

In this paper, a real-time unmanned aerial vehicles (UAVs) detection framework is proposed for GPU embedded applications. To achieve this, this paper proposed a new modified model based on You Only Look Once (YOLO) to detect multi-UAV in aerial images with a complex background. The proposed CNN architecture uses five inception module and dilated convolution with two different factors to increase detection accuracy of YOLOv3 tiny, while preserving its computational time. To further improve the detection accuracy, Generalized Intersection over Union loss is replaced with the bounding box regression loss in the original YOLOv3 tiny loss function. To obtain higher frame rate, scalable kernel correlation filter (sKCF) algorithm is integrated to the detection model. More specifically, the proposed UAV detection method is used to initialize the sKCF tracker at every $$n\mathrm{th}$$ frame. Thus, the detected UAVs can be detected in intermediate frames with a low memory footprint. The proposed model and compared models are trained and tested on a variety of training and test airborne videos, captured under different outdoor scenarios. The average precision of the proposed model is 0.8265 and achieves 32 FPS performance on Jetson-TX2. The results show that the proposed model which is widened with inception module using dilated convolutions has a very high performance in terms of accuracy and speed even when it is compared with deep CNN architectures such as Darknet 53.

In this paper, a real-time deep learning-based framework for detecting and tracking Unmanned Aerial Vehicles (UAVs) in video streams captured by a fixed-wing UAV is proposed. The proposed framework consists of two steps, namely intra-frame multi-UAV detection and the inter-frame multi-UAV tracking. In the detection step, a new multi-scale UAV detection Convolutional Neural Network (CNN) architecture based on a shallow version of You Only Look Once version 3 (YOLOv3-tiny) widened by Inception blocks is designed to extract local and global features from input video streams. Here, the widened multi-UAV detection network architecture is termed as FastUAV-NET and aims to improve UAV detection accuracy while preserving computing time of one-step deep detection algorithms in the context of UAV-UAV tracking. To detect UAVs, the FastUAV-NET architecture uses five inception units and adopts a feature pyramid network to detect UAVs. To obtain a high frame rate, the proposed method is applied to every nth frame and then the detected UAVs are tracked in intermediate frames using scalable Kernel Correlation Filter algorithm. The results on the generated UAV-UAV dataset illustrate that the proposed framework obtains 0.7916 average precision with 29 FPS performance on Jetson-TX2. The results imply that the widening of CNN network is a much more effective way than increasing the depth of CNN and leading to a good trade-off between accurate detection and real-time performance. The FastUAV-NET model will be publicly available to the research community to further advance multi-UAV-UAV detection algorithms.

Anomaly detection in Digital Subscriber Line (DSL) networks is a vital task to immediately detect unusual network behavior caused by cyber security threats, faulty hardware or software. Generally, to make this process automatic, state-of-the-art methods use machine learning techniques to analyze data collected from either Customer Premises Equipment (CPE) or from devices in Access Network. In contrast to the existing methods, this paper utilizes network traffic data collected from multiple static Broadband Network Gateways (BNGs) which are core network devices at Network Service Provider (NSP). To automatically detect anomalies in BNG traffic data, a new framework is proposed which consists of three steps: data acquisition, feature extraction, and modeling anomalies using a random forest-based framework. The proposed method is compared with state-of-the-art methods and the results show the effectiveness and robustness of our method.

This paper proposes a real-time video analysis algorithm to read the resistance value of a resistor using a color recognition technique. To achieve this, firstly, a nonlinear filtering is applied to input video frame to smooth intensity variations and remove impulse noises. After that, a photometric invariants technique is employed to transfer the video frame from RGB color space to Hue-Saturation-Value (HSV) color space, which decreases sensitivity of the proposed method to illumination changes. Next, a region of interest is defined to automatically detect resistor’s colors and then an Euclidean distance based clustering strategy is employed to recognize the color bars. The proposed method provides a wide range of color classification which includes twelve colors. In addition, it utilizes relatively low computational time which makes it suitable for real-time mobile video applications. The experiments are performed on a variety of test videos and results show that the proposed method has low error rate compared to the other resistor color code recognition mobile applications.

In this paper, a novel unsupervised change detection method is proposed to automatically detect changes between two cloud-contaminated Landsat images. To achieve this, firstly, a photometric invariants technique with Stationary Wavelet Transform (SWT) are applied to input images to decrease the influence of cloud and noise artifacts in the change detection process. Then, mean shift image filtering is employed on the sub-band difference images, generated via image differencing technique, to smooth the images. Next, multiple binary change detection masks are obtained by partitioning the pixels in each of the smoothed sub-band difference images into two clusters using Fuzzy c-means (FCM). Finally, the binary masks are fused using Markov Random Field (MRF) to generate the final solution. Experiments on both semi-simulated and real data sets show the effectiveness and robustness of the proposed change detection method in noisy and cloud-contaminated Landsat images.

In this paper, a novel algorithm for unsupervised change detection in multitemporal multispectral images of the same scene using parallel binary particle swarm optimization (PBPSO) is proposed. The algorithm operates on a difference image, which is created by using a novel fusion algorithm on multitemporal multispectral images, by iteratively minimizing a cost function with PBPSO to produce a final binary change-detection mask representing changed and unchanged pixels. Each BPSO of parallel instances is run on a separate processor and initialized with a different starting population representing a set of change-detection masks. A communication strategy is applied to transmit data in between BPSOs running in parallel. The algorithm takes the full advantage of parallel processing to improve both the convergence rate and detection performance. We demonstrate the accuracy of the proposed method by quantitative and qualitative tests on semisynthetic and real-world data sets. The semisynthetic results for different levels of Gaussian noise are obtained in terms of false and miss alarm (MA) rates between the estimated change-detection mask and the ground truth image. The proposed method on the semisynthetic data with high level of Gaussian noise obtains the final change-detection mask with a false error rate of 1.50 and MA error rate of 14.51.

A new unsupervised approach based on a hybrid wavelet transform and Fuzzy Clustering Method (FCM) with Multiobjective Particle Swarm Optimization (MO-PSO) is proposed to obtain a binary change mask in Landsat images acquired with different atmospheric conditions. The proposed method uses the following steps: (1) preprocessing, (2) classification of preprocessed image, and (3) binary masks fusion. Firstly, a photometric invariant technique is used to transform the Landsat images from RGB to HSV colour space. A hybrid wavelet transform based on Stationary (SWT) and Discrete Wavelet (DWT) Transforms is applied to the hue channel of two Landsat satellite images to create subbands. After that, mean shift clustering method is applied to the subband difference images, computed using the absolute-valued difference technique, to smooth the difference images. Then, the proposed method optimizes iteratively two different fuzzy based objective functions using MO-PSO to evaluate changed and unchanged regions of the smoothed difference images separately. Finally, a fusion approach based on connected component with union technique is proposed to fuse two binary masks to estimate the final solution. Experimental results show the robustness of the proposed method to existence of haze and thin clouds as well as Gaussian noise in Landsat images.

In this paper, a novel method for generation of multiple description (MD) wavelet based image coding is proposed by using Multi-Objective Evolutionary Algorithms (MOEAs). Complexity of the multimedia transmission problem has been increased for MD coders if an input image is affected by any type of noise. In this case, it is necessary to solve two different problems which are designing the optimal side quantizers and estimating optimal parameters of the denoising filter. Existing MD coding (MDC) generation methods are capable of solving only one problem which is to design side quantizers from the given noise-free image but they can fail reducing any type of noise on the descriptions if they applied to the given noisy image and this will cause bad quality of multimedia transmission in networks. Proposed method is used to overcome these difficulties to provide effective multimedia transmission in lossy networks. To achieve it, Dual Tree-Complex Wavelet Transform (DT-CWT) is first applied to the noisy image to obtain the subbands or set of coefficients which are used as a search space in the optimization problem. After that, two different objective functions are simultaneously employed in the MOEA to find pareto optimal solutions with the minimum costs by evolving the initial individuals through generations. Thus, optimal quantizers are created for MDCs generation and obtained optimum parameters are used in the image filter to remove the mixed Gaussian impulse noise on the descriptions effectively. The results demonstrate that proposed method is robust to the mixed Gaussian impulse noise, and offers a significant improvement of optimal side quantizers for balanced MDCs generation at different bitrates. © 2018 Elsevier B.V.

This paper proposes a new unsupervised satellite change detection method, which is robust to illumination changes. To achieve this, firstly, a preprocessing strategy is used to remove illumination artifacts and results in less false detection than traditional threshold-based algorithms. Then, we use the corrected input data to define a new fitness function based on the difference image. The purpose of using Self-Adaptive Hybrid Particle Swarm Optimization-Genetic Algorithm (SAPSOGA) is to combine two meta-heuristic optimization algorithms to search and find the feasible solution in the NP-hard change detection problem rapidly and efficiently. The hybrid algorithm is employed by letting the GA and PSO run simultaneously and similarities of GA and PSO have been considered to implement the algorithm, i.e. the population. In the SAPSOGA employed, in each iteration/generation the two population based algorithms share different amount of information or individual(s) between themselves. Thus, each algorithm informs each other about their best optimum results (fitness values and solution representations) which are obtained in their own population. The fitness function is minimized by using binary based SAPSOGA approach to produce binary change detection masks in each iteration to obtain the optimal change detection mask between two multi temporal multi spectral landsat images. The proposed approach effectively optimizes the change detection problem and finds the final change detection mask.

International audience; Mapping an endometrial tissue seen in Transvaginal Ultrasound (TVUS) to a Magnetic Resonance (MR) image is an important approach to improve the accuracy of diagnosis and preoperative surgery planning. To date, no technical methods solve this problem. In a clinical setting where only one 2D TVUS image is available, methods in the state of the art require the selection of a 2D MR slice and use 2D/2D deformable image registration. This is a limitation since the standard TVUS and MR imaging techniques used for diagnosing endometriosis are 2D and 3D, respectively. In this work, we introduce a novel deformable slice-to-volume registration process to overcome this limitation. To this end, we register a 2D TVUS image to a 3D MR data in order to transform the TVUS planar image to a curved 2D surface in MR volume. The proposed method facilitates the transfer of two types of information from a 2D TVUS image to the MR volume: 1) the location and shape of small implants and 2) the implants' depth of infiltration in the host tissue. Our TVUS-MR registration method uses contour to surface correspondences through a novel variational one-step deformable Iterative Closest Point (ICP) method. We demonstrate the accuracy of the proposed method by quantitative and qualitative tests on semi-synthetic and clinical data sets. Experimental error analysis shows high registration accuracy, illustrating the robustness of the method.

We propose a multi-modal image registration and fusion method to cope with the limitations of Magnetic Resonance (MR) and Transvaginal Ultrasound (TVUS) imaging in observing abdominal endometrial implants. The proposed method facilitates the transfer of two types of information from a 2D TVUS image to a 2D MR slice: (1) the location and shape of small implants and (2) the implants' depth of infiltration in the host tissue. Our registration uses point correspondences sampled on user provided, corresponding organs' contours. Experimental results are provided for 5 patient datasets, showing that the proposed method outperforms the classical Thin-Plate Spline (TPS).