Metric learning aims to measure the similarity among samples while using an optimal distance metric for learning tasks. Metric learning methods, which generally use a linear projection, are limited in solving real-world problems demonstrating non-linear characteristics. Kernel approaches are utilized in metric learning to address this problem. In recent years, deep metric learning, which provides a better solution for nonlinear data through activation functions, has attracted researchers’ attention in many different areas. This article aims to reveal the importance of deep metric learning and the problems dealt with in this field in the light of recent studies. As far as the research conducted in this field are concerned, most existing studies that are inspired by Siamese and Triplet networks are commonly used to correlate among samples while using shared weights in deep metric learning. The success of these networks is based on their capacity to understand the similarity relationship among samples. Moreover, sampling strategy, appropriate distance metric, and the structure of the network are the challenging factors for researchers to improve the performance of the network model. This article is considered to be important, as it is the first comprehensive study in which these factors are systematically analyzed and evaluated as a whole and supported by comparing the quantitative results of the methods.

Ultrasound offers a safe, non-invasive, and inexpensive way of imaging. However, due to its natural intrinsic imaging characteristics, it produces poor quality images with low resolution (LR) compared to other medical imaging modalities. Various image enhancement techniques have been extensively studied to overcome these shortcomings. Super-resolution (SR) is one of these methods, which endeavor to obtain high resolution (HR) images from LR images while enlarging them. Numerous studies have already utilized different SR techniques in various stages of ultrasound imaging (USI). Unlike other studies, which aimed at obtaining SR in the pre-processing phase or early stages of the post-processing phase of USI, we achieved SR on B-mode ultrasound images, which is the last stage of USI. We constructed a deep convolutional neural network (CNN) and trained it with a very large dataset of B-mode ultrasound images for the scale factors 2, 3, 4, and 8. We evaluated the performance of our proposed model quantitatively with eight image quality measures. The quantitative results revealed that our algorithm is much more successful than other methods at each magnification factor. Furthermore, we also verified that there is a statistically significant difference between our approach and others. Besides, qualitative analysis of the reconstructed images also confirms that it produces much better quality HR images than other methods in terms of the human visual system.

Localization and mapping technologies are of great importance for all varieties of Unmanned Aerial Vehicles (UAVs) to perform their operations. In the near future, it is planned to increase the use of micro/nano-size UAVs. Such vehicles are sometimes expendable platforms, and reuse may not be possible. Compact, mounted and low-cost cameras are preferred in these UAVs due to weight, cost and size limitations. Visual simultaneous localization and mapping (vSLAM) methods are used for providing situational awareness of micro/nano-size UAVs. Fast rotational movements that occur during flight with gimbal-free, mounted cameras cause motion blur. Above a certain level of motion blur, tracking losses exist, which causes vSLAM algorithms not to operate effectively. In this study, a novel vSLAM framework is proposed that prevents the occurrence of tracking losses in micro/nano-UAVs due to the motion blur. In the proposed framework, the blur level of the frames obtained from the platform camera is determined and the frames whose focus measure score is below the threshold are restored by specific motion-deblurring methods. The major reasons of tracking losses have been analyzed with experimental studies, and vSLAM algorithms have been made durable by our studied framework. It has been observed that our framework can prevent tracking losses at 5, 10 and 20 fps processing speeds. vSLAM algorithms continue to normal operations at those processing speeds that have not been succeeded before using standard vSLAM algorithms, which can be considered as a superiority of our study.

Background: Accurate and fast image registration (IR) is critical during surgical interventions where the ultrasound (US) modality is used for image-guided intervention. Convolutional neural network (CNN)-based IR methods have resulted in applications that respond faster than traditional iterative IR methods. However, general-purpose processors are unable to operate at the maximum speed possible for real-time CNN algorithms. Due to its reconfigurable structure and low power consumption, the field programmable gate array (FPGA) has gained prominence for accelerating the inference phase of CNN applications. Methods: This study proposes an FPGA-based ultrasound IR CNN (FUIR-CNN) to regress three rigid registration parameters from image pairs. To speed up the estimation process, the proposed design makes use of fixed-point data and parallel operations carried out by unrolling and pipelining techniques. Experiments were performed on three US datasets in real time using the xc7z020, and the xcku5p was also used during implementation. Results: The FUIR-CNN produced results for the inference phase 139 times faster than the software-based network while retaining a negligible drop in regression performance of under 200 MHz clock frequency. Conclusions: Comprehensive experimental results demonstrate that the proposed end-to-end FPGA-based accelerated CNN achieves a negligible loss, a high speed for registration parameters, less power when compared to the CPU, and the potential for real-time medical imaging.

The exploration of underwater environments for applications like environmental monitoring, scientific research, and surveillance has propelled the significance of underwater wireless navigation. Light waves have emerged as a promising solution, offering the potential to achieve the required data rates and propagation speeds. However, underwater optical wireless navigation faces challenges, particularly limited range. This research investigates a novel FMCW (frequency-modulated continuous wave)-based photonic radar system’s efficacy in detecting underwater vehicles across diverse salinity levels and distances. Numerical simulation evaluations reveal distinct signal-to-noise ratios (SNR) and detected power peaks corresponding to varying salinity levels, demonstrating the system’s sensitivity. At 5 g/L salinity, the detected power peaked at −95 dBm, decreasing to −105 dBm at 15 g/L. SNR analysis indicates robust detection within a 4 m range, with challenges emerging at extended ranges and higher salinity. Despite these challenges, the system shows promise for near-range underwater navigation, contributing to sustainable marine exploration by enhancing the accuracy and efficiency of underwater monitoring systems. This advancement aligns with the goals of sustainable development by supporting the protection of marine ecosystems, promoting scientific understanding of underwater environments, and aiding in the sustainable management of marine resources.

In this study, a new method is presented for face recognition by using DCT-faces that are calculated by the 3 dimensional discrete cosine transform (3D DCT). In this method, firstly, face images in the database are collected on top of each other and then 3D DCT is applied to this 3D structure. Then 2D Inverse DCT of each coefficient layer is calculated separately. After this operation, DCT-faces are found which are very similar to eigenfaces. By taking the inner product between DCT-faces and original face images, feature vectors are calculated to be used in face recognition. Experimental results show that, the face recognition ratio of DCT-faces is close to that of eigenface method. Moreover, the proposed method has a lower time complexity and lower memory complexity.