Abstract The SARS‐COV‐2 virus, which causes COVID‐19 disease, continues to threaten the whole world with its mutations. Many methods developed for COVID‐19 detection are validated on the data sets generally including severe forms of the disease. Since the severe forms of the disease have prominent signatures on X‐ray images, the performance to be achieved is high. To slow the spread of the disease, effective computer‐assisted screening tools with the ability to detect the mild and the moderate forms of the disease that do not have prominent signatures are needed. In this work, various pretrained networks, namely GoogLeNet, ResNet18, SqueezeNet, ShuffleNet, EfficientNetB0, and Xception, are used as feature extractors for the COVID‐19 detection with severity level analysis. The best feature extraction layer for each pre‐trained network is determined to optimize the performance. After that, features obtained by the best layer are selected by following a wrapper‐based feature selection strategy using the features ranked based on Laplacian scores. The experimental results achieved on two publicly available data sets including all the forms of COVID‐19 disease reveal that the method generalized well on unseen data. Moreover, 66.67%, 90.32%, and 100% sensitivity are obtained in the detection of mild, moderate, and severe cases, respectively.

The development of systems that can predict epilepsy seizures in real time offers great hope for epilepsy patients. These systems aim to prevent accidents that patients may experience due to loss of consciousness during seizures. Therefore, systems that can predict epileptic seizures should both work in real time and be designed to maintain the daily activities of the patient. In this case, a system with as few electrodes as possible should be developed. In this study, it is aimed to choose the most appropriate electrode in predicting epileptic seizures. Channel selection is made according to two parameters and its effect on seizure prediction is examined. The first parameter is the difference in variance between preictal and interictal; The other parameter is the weighted average sensitivity (WAS). The Rusboosted Tree ensemble classification is used to calculate WAS. The prediction process is carried out with the method we proposed in the previous study. For performance evaluation, prediction accuracy, sensitivity (SEN) and false alarm rates per hour (FPR) are calculated. The prediction performance for the channel selected according to the variance difference results are 69%, 70.9% and 0.054 respectively and the for the channel selected according to WAS results are 69%, 71.8% and 0.031 respectively.

Multispectral imaging is one of the most important Earth observation techniques in remote sensing. Although their advantages, multispectral imaging systems continuously capture images resulting in enormous data volumes. To this end, some of the multispectral satellites use JPEG2000 based compression methods. However, the images that are compressed at low bit-rates contain compression artifacts and may present low-performances in remote sensing applications such as target detection and classification. In this paper, we propose the usage of three single image super-resolution methods, SRResNet, EDSR, and WDSR, for the resolution enhancement of JPEG2000 compressed images. First, the multispectral image is subsampled at factor of 4 along both spatial axes, and then the resulting image is compressed with JPEG2000. Finally, super-resolution methods are performed to improve the resolution and reduce the compression artifacts. Experiments were carried out on the Onera dataset shared by IEEE GRSS. The results are compared in terms of quality metrics such as signal-to-noise ratios, mean spectral angle, maximum spectral angle, and maximum absolute difference. Experimental results demonstrate that the proposed approaches provide higher quality metrics and better visual performance compared to bicubic upsampling.

In this paper, the hardware architecture for two-point non-uniformity correction (TPNUC) and bad pixel replacement (BPR) algorithms are presented based on field-programmable gate arrays (FPGA) for infrared focal plane arrays (IRFPA). An efficient hardware architecture modeled using C++ in the High-Level Synthesis (HLS) tool is presented. The design is tested on an FPGA fabricated at a 16 nm technology node. The design achieves a maximum frequency of 300 MHz and one pixel per clock. A thermal camera development platform (FullScale USB3A) with a resolution of 640×480 is used as the source for the raw video. The simulation results from MATLAB and FPGA posed close similarities.

Automated lung segmentation has importance because it gives clues about several diseases to the experts. It is the step that comes before further detailed analyses of the lungs. However, segmentation of the lungs is a challenging task since the opacities and consolidations are caused by various lung diseases. As a result, the clarity of the borders of the lungs may be lost which makes the segmentation task difficult. The presence of various medical equipment such as cables in the image is another factor that makes segmentation difficult. Therefore, it is a necessity to develop methods that can handle such situations. Learning the most useful patterns related to various diseases is possible with deep learning methods. Unlike conventional methods, learning the patterns improves the generalization ability of the models on unseen data. For this purpose, a deep segmentation framework including ensembles of pre-trained lightweight networks is proposed for lung region segmentation in this work. The experimental results achieved on two publicly available data sets demonstrate the effectiveness of the proposed framework.

DCT based transform techniques are popular in image compression. In this paper, luminance transform is applied to improve the compression performance of 3-D discrete cosine transform (3D-DCT) in hyperspectral images. The proposed scheme consists of two main steps. Firstly, luminance transform is performed on spectral band groups taking the first band image in a group as the reference. The aim of using luminance transform is to reduce the brightness and contrast difference within spectral band groups. Secondly, compression is performed by 3D-DCT followed by entropy encoding. The performance of the proposed approach is compared to 3D-DCT in terms of signal-to-noise ratio (SNR) and mean spectral angle (MSA). It is observed that applying luminance transform before 3D-DCT provides better results especially at low bit-rates.

Compression of hyperspectral images is an important topic for transmission and storage of data. There are several compression approaches proposed in the literature. Performance analysis of these methods is generally measured by image quality metrics. However, image quality metrics are not capable of determining compression performance for a specific application area. In this paper, popular compression approaches JPEG2000, PCA+JPEG2000, DWT+JPEG2000, 3D-SPECK, and 3D-TARP are evaluated in terms of unmixing, anomaly detection, target detection, and classification performances. Experimental evaluations are carried out on four hyperspectral datasets, and obtained results are interpreted.

Hyperspectral imaging provides increased capability for many image processing tasks with respect to standard imaging systems. One of such tasks in hyperspectral image processing is change detection, which aims to detect the differences occurring between images acquired from the same scene at different times. In this paper, a panoramic hyperspectral imaging system is used to capture multitemporal hyperspectral data, and novel multi-band Census Transform (MCT) is proposed for change detection on these data. Experimental results validate the performance of the proposed method for the utilized acquisition system.

Hyperspectral imaging provides high spectral resolution and thereby improved classification, detection, and recognition capabilities with respect to standard imaging systems. However, hyperspectral images generally have low spatial resolution, varying from a few to tens of meters, resulting from technical limitations such as platform data storing capacity and satellite-to-ground transmission bandwidth. Spectral unmixing provides information on pixels in terms of abundances of pure spectral signatures, without providing spatial distribution at subpixel level. Multisensor image fusion approaches can provide such information but require an additional image with higher spatial resolution that is acquired in similar conditions with the hyperspectral image. In this letter, a novel spatial resolution enhancement method using fully constrained least squares (FCLS) spectral unmixing and spatial regularization based on modified binary particle swarm optimization is proposed to achieve spatial resolution enhancement in hyperspectral images, without using an additional image with higher spatial resolution. The proposed method has a highly parallel nature with respect to its counterparts in the literature and is fit to be adapted to field-programmable gate array architecture.

Spectral unmixing is the process of identifying pure spectral signatures, called endmembers, from a hyperspectral data, and then expressing each pixel vector in terms of the fractional abundances of these endmembers. Most of the endmember extraction methods in the literature use only the spectral information, whereas the spatial composition of the data is disregarded. Spatial preprocessing methods, that are motivated by the assumption that endmembers are more likely to be located in homogeneous regions instead of transition areas, can alleviate this drawback and hence increase the performance. However, such a preprocessing approach generally results in a failure of extracting anomalous endmembers which can be of importance for many applications. In this paper, a preprocessing approach that guides the endmember extraction process to homogenous regions while retaining the anomaly points, by combining spatial preprocessing with anomaly detection, is proposed.

This paper proposes to use empirical mode decomposition (EMD) with spectral gradient enhancement to increase the classification accuracy of hyperspectral images with support vector machine (SVM) classification. Recently, it has been shown that higher hyperspectral image classification accuracy can be achieved by using 2-D EMD that is applied to each hyperspectral band separately to obtain the intrinsic mode functions (IMFs) of each band, while the sum of the IMFs are used as feature data in the SVM classification process. In the previous approach, IMFs have been summed directly, i.e., with equal weights. It is shown in this paper, that it is possible to significantly increase the classification accuracy by using appropriate weights for the IMFs in the summation process. In the proposed approach, the weights of the IMFs are obtained so as to optimize the total absolute spectral gradient, and a genetic algorithm-based optimization strategy has been adopted to obtain the weights automatically in this way. While the 2-D EMD basically provides spatial processing, the proposed method further incorporates spectral enhancement into the process. It is shown that a significant increase in hyperspectral image classification accuracy can be achieved using the proposed approach.

This paper proposes an empirical mode decomposition (EMD) based approach with spectral gradient enhancement for hyperspectral image classification using support vector machines (SVM). In a previous study, it has been shown that using the sum of intrinsic mode functions (IMFs), obtained by applying two-dimensional (2D) EMD to each hyperspectral band, increases the classification accuracies significantly. In this paper, it is shown that using optimum weights for the IMFs, instead of the equal weight approach of the previous study, results in increased classification accuracies. The weights for the IMFs are obtained by a genetic algorithm (GA) based optimization strategy which aims to maximize spectral gradient and hence incorporate spectral processing with the spatial processing of 2D EMD.

In this study we propose an automated fine registration of EO-1 Hyperion and IKONOS imagery. An intensity based registration that is area-based and pixelwise is performed to register given images of divergent spatial and spectral resolution. Two similarity measures that are commonplace in image registration; NCC and NMI, and an operation that is particular to image restoration; CTO is adopted as an error measure as a novelty in image registration. We are convinced with the performance and efficiency of CTO compared to other two common methods of intensity-based registration.

This paper presents a comparison of the classification performance of some vector machine based classification methods, namely, Import Vector Machines (IVM), Support Vector Machines (SVM) and Relevance Vector Machines (RVM), for hyperspectral images. Evaluation is carried out in terms of the number of vectors and classification accuracies. Furthermore, novel to this paper, Discriminative Random Field method with Graph Cut algorithm is applied to the probabilistic classification output of IVM based hyperspectral classification results, and it is shown that this approach significantly increases classification accuracies.

This letter proposes the use of denoising in conjunction with 2-D empirical mode decomposition (2D-EMD) of hyperspectral image bands for higher classification accuracy. Initially, 2D-EMD is performed to hyperspectral image bands for decomposition into intrinsic mode functions (IMFs). Then, denoising is applied to the first IMF of each band because this IMF includes local high-spatial-frequency components. Features reconstructed as the sums of lower order IMFs are then used for classification. Support vector machine classification is used as a classification approach in this letter. Experimental results show that the proposed technique can provide a higher classification accuracy.

This paper proposes Empirical Mode Decomposition (EMD) followed by wavelet shrinkage denoising in hyperspectral image classification to improve classification accuracy. EMD decomposes signals into several Intrinsic Mode Functions (IMFs) and a final residue. In this paper, firstly, EMD is applied to each hyperspectral image band separately to obtain the IMFs of all image bands. Then, the first IMF of each band is applied to wavelet shrinkage denoising, as this IMF includes all local high spatial frequency components. The sums of lower order IMFs are then used to reconstruct hyperspectral image bands that are used as new features for classification. Support Vector Machine (SVM) based classification is used as classification approach in this paper. Experimental results show the effectiveness of the proposed approach.

This paper presents an automatic scratch detection and removal approach for archive film sequences. The proposed detector mainly exploits temporal coherency of candidate scratch positions, which are obtained using an automatic scratch detection method proposed in the literature. In the restoration stage, both spatial and temporal information are employed. The proposed edge priority based scratch removal algorithm successfully removes scratch effects from archive film sequences.

This paper targets hard-cut detection for archive film, i.e., mainly black-and-white videos from the beginning of the last century, which is a particularly difficult task due to heavy visual degradations encountered in the sequences. A robust hard-cut detection system based on modified phase correlation is presented. Phase-correlation-based hard-cut detection is carried out using spatially sub-sampled video frames, and a candidate hard-cut is indicated in the case of low correlation. A double thresholding approach consisting of a global threshold used in conjunction with an adaptive local threshold is used to detect candidate hard-cuts. For uniformly colored video frames the phase correlation is extremely sensitive to noise and visual defects. Mean and variance based simple heuristic false removal at uniformly colored video frames is used at the final stage to prevent false detections in such cases. The paper provides a through theoretical analysis to show the usefulness of spatial sub-sampling. Furthermore through experimental results are presented for visual defects encountered in archive film material, to present the effectiveness of the proposed approach.