In the field of medicine there is a need for the automatic detection of retinal disorders. Blindness in older persons is primarily caused by Central Retinal Vein Occlusion (CRVO). It results in rapid, irreversible eyesight loss, therefore, it is essential to identify and address CRVO as soon as feasible. Hemorrhages, which can differ in size, pigment, and shape from dot-shaped to flame hemorrhages, are one of the earliest symptoms of CRVO. The early signs of CRVO are, hemorrhages, however, so mild that ophthalmologists must dynamically observe such indicators in the retina image known as the fundus image, which is a challenging and time-consuming task. It is also difficult to segment hemorrhages since the blood vessels and hemorrhages (HE) have the same color properties also there is no particular shape for hemorrhages and it scatters all over the fundus image. A challenging study is needed to extract the characteristics of vein deformability and dilatation. Furthermore, the quality of the captured image affects the efficacy of feature Identification analysis. In this paper, a deep learning approach for CRVO extraction is proposed.

This paper presents a deep learning-based image texture recognition system. The methodology taken in this solution is formed in a bottom-up manner. It means we swipe a moving window through the image in order to categorize if a given region belongs to one of the classes seen in the training process. This categorization is done based on the Deep Neural Network (DNN) of fixed architecture. The training process is fully automated regarding the training data preparation, investigation of the best training algorithm, and its hyper-parameters. The only human input to the system is the definition of the categories for further recognition and generation of the samples (region markings) in the external application chosen by the user. The system is tested on road surface images where its task is to categorize image regions to a different road category (e.g. curb, road surface damage, etc.) and is featured with 90% and above accuracy.

Skin cancer is the most common form of cancer affecting humans. Melanoma is the most dangerous type of skin cancer; and early diagnosis is extremely vital in curing the disease. So far, the human knowledge in this field is very limited, thus, developing a mechanism capable of identifying the disease early on can save lives, reduce intervention and cut unnecessary costs. In this paper, the researchers developed a new learning technique to classify skin lesions, with the purpose of observing and identifying the presence of melanoma. This new technique is based on a convolutional neural network solution with multiple configurations; where the researchers employed an International Skin Imaging Collaboration (ISIC) dataset. Optimal results are achieved through a convolutional neural network composed of 14 layers. This proposed system can successfully and reliably predict the correct classification of dermoscopic lesions with 97.78% accuracy.

In recent years, deep learning and especially deep neural networks (DNN) have obtained amazing performance on a variety of problems, in particular in classification or pattern recognition. Among many kinds of DNNs, the convolutional neural networks (CNN) are most commonly used. However, due to their complexity, there are many problems related but not limited to optimizing network parameters, avoiding overfitting and ensuring good generalization abilities. Therefore, a number of methods have been proposed by the researchers to deal with these problems. In this paper, we present the results of applying different, recently developed methods to improve deep neural network training and operating. We decided to focus on the most popular CNN structures, namely on VGG based neural networks: VGG16, VGG11 and proposed by us VGG8. The tests were conducted on a real and very important problem of skin cancer detection. A publicly available dataset of skin lesions was used as a benchmark. We analyzed the influence of applying: dropout, batch normalization, model ensembling, and transfer learning. Moreover, the influence of the type of activation function was checked. In order to increase the objectivity of the results, each of the tested models was trained 6 times and their results were averaged. In addition, in order to mitigate the impact of the selection of learning, test and validation sets, k-fold validation was applied.

The pathologists follow a systematic and partially manual process to obtain histological tissue sections from the biological tissue extracted from patients. This process is far from being perfect and can introduce some errors in the quality of the tissue sections (distortions, deformations, folds and tissue breaks). In this paper, we propose a deep learning (DL) method for the detection and segmentation of these damaged regions in whole slide images (WSIs). The proposed technique is based on convolutional neural networks (CNNs) and uses the U-net model to achieve the pixel-wise segmentation of these unwanted regions. The results obtained show that this technique yields satisfactory results and can be applied as a pre-processing step for automatic WSI analysis in order to prevent the use of the damaged areas in the evaluation processes.

In the last few years, a great attention was paid to the deep learning Techniques used for image analysis because of their ability to use machine learning techniques to transform input data into high level presentation. For the sake of accurate diagnosis, the medical field has a steadily growing interest in such technology especially in the diagnosis of melanoma. These deep learning networks work through making coarse segmentation, conventional filters and pooling layers. However, this segmentation of the skin lesions results in image of lower resolution than the original skin image. In this paper, we present deep learning based approaches to solve the problems in skin lesion analysis using a dermoscopic image containing skin tumor. The proposed models are trained and evaluated on standard benchmark datasets from the International Skin Imaging Collaboration (ISIC) 2018 Challenge. The proposed method achieves an accuracy of 96.67% for the validation set .The experimental tests carried out on a clinical dataset show that the classification performance using deep learning-based features performs better than the state-of-the-art techniques.

The paper presents special forms of an ensemble of classifiers for analysis of medical images based on application of deep learning. The study analyzes different structures of convolutional neural networks applied in the recognition of two types of medical images: dermoscopic images for melanoma and mammograms for breast cancer. Two approaches to ensemble creation are proposed. In the first approach, the images are processed by a convolutional neural network and the flattened vector of image descriptors is subjected to feature selection by applying different selection methods. As a result, different sets of a limited number of diagnostic features are generated. In the next stage, these sets of features represent input attributes for the classical classifiers: support vector machine, a random forest of decision trees, and softmax. By combining different selection methods with these classifiers an ensemble classification system is created and integrated by majority voting. In the second approach, different structures of convolutional neural networks are directly applied as the members of the ensemble. The efficiency of the proposed classification systems is investigated and compared to medical data representing dermoscopic images of melanoma and breast cancer mammogram images. Thanks to fusion of the results of many classifiers forming an ensemble, accuracy and all other quality measures have been significantly increased for both types of medical images.

Thousands of low-power micro sensors make up Wireless Sensor Networks, and its principal role is to detect and report specified events to a base station. Due to bounded battery power these nodes are having very limited memory and processing capacity. Since battery replacement or recharge in sensor nodes is nearly impossible, power consumption becomes one of the most important design considerations in WSN. So one of the most important requirements in WSN is to increase battery life and network life time. Seeing as data transmission and reception consume the most energy, it’s critical to develop a routing protocol that addresses the WSN’s major problem. When it comes to sending aggregated data to the sink, hierarchical routing is critical. This research concentrates on a cluster head election system that rotates the cluster head role among nodes with greater energy levels than the others.We used a combination of LEACH and deep learning to extend the network life of the WSN in this study. In this proposed method, cluster head selection has been performed by Convolutional Neural Network (CNN). The comparison has been done between the proposed solution and LEACH, which shows the proposed solution increases the network lifetime and throughput.

Convolutional Neural Networks (CNN) have achieved huge popularity in solving problems in image analysis and in text recognition. In this work, we assess the effectiveness of CNN-based architectures where a network is trained in recognizing handwritten characters based on Latin script. European languages such as Dutch, French, German, etc., use different variants of the Latin script, so in the conducted research, the Latin alphabet was extended by certain characters with diacritics used in Polish language. To evaluate the recognition results under the same conditions, a handwritten Latin dataset was also developed. The proposed CNN architecture produced an accuracy of 96% for the extended character set. This is comparable to state-of-the-art results found in the domain of identifying handwritten characters. The presented approach extends the usage of CNN-based recognition to different variants of the Latin characters and shows it can be successfully used for a set of languages based on that script. It seems to be an effective technique for a set of languages written using the Latin script.

Multi-focus image fusion is a method of increasing the image quality and preventing image redundancy. It is utilized in many fields such as medical diagnostic, surveillance, and remote sensing. There are various algorithms available nowadays. However, a common problem is still there, i.e. the method is not sufficient to handle the ghost effect and unpredicted noises. Computational intelligence has developed quickly over recent decades, followed by the rapid development of multi-focus image fusion. The proposed method is multi-focus image fusion based on an automatic encoder-decoder algorithm. It uses deeplabV3+ architecture. During the training process, it uses a multi-focus dataset and ground truth. Then, the model of the network is constructed through the training process. This model was adopted in the testing process of sets to predict the focus map. The testing process is semantic focus processing. Lastly, the fusion process involves a focus map and multi-focus images to configure the fused image. The results show that the fused images do not contain any ghost effects or any unpredicted tiny objects. The assessment metric of the proposed method uses two aspects. The first is the accuracy of predicting a focus map, the second is an objective assessment of the fused image such as mutual information, SSIM, and PSNR indexes. They show a high score of precision and recall. In addition, the indexes of SSIM, PSNR, and mutual information are high. The proposed method also has more stable performance compared with other methods. Finally, the Resnet50 model algorithm in multi-focus image fusion can handle the ghost effect problem well.

In industrial drive systems, one of the widest group of machines are induction motors. During normal operation, these machines are exposed to various types of damages, resulting in high economic losses. Electrical circuits damages are more than half of all damages appearing in induction motors. In connection with the above, the task of early detection of machine defects becomes a priority in modern drive systems. The article presents the possibility of using deep neural networks to detect stator and rotor damages. The opportunity of detecting shorted turns and the broken rotor bars with the use of an axial flux signal is presented.

In this paper, we present an improved efficient capsule network (CN) model for the classification of the Kuzushiji-MNIST and Kuzushiji-49 benchmark datasets. CNs are a promising approach in the field of deep learning, offering advantages such as robustness, better generalization, and a simpler network structure compared to traditional convolutional neural networks (CNNs). Proposed model, based on the Efficient CapsNet architecture, incorporates the self-attention routing mechanism, resulting in improved efficiency and reduced parameter count. The experiments conducted on the Kuzushiji-MNIST and Kuzushiji-49 datasets demonstrate that the model achieves competitive performance, ranking within the top ten solutions for both benchmarks. Despite using significantly fewer parameters compared to higher-rated competitors, presented model achieves comparable accuracy, with overall differences of only 0.91% and 1.97% for the Kuzushiji-MNIST and Kuzushiji- 49 datasets, respectively. Furthermore, the training time required to achieve these results is substantially reduced, enabling training on nonspecialized workstations. The proposed novelties of capsule architecture, including the integration of the self-attention mechanism and the efficient network structure, contribute to the improved efficiency and performance of presented model. These findings highlight the potential of CNs as a more efficient and effective approach for character classification tasks, with broader applications in various domains.

For the proper operation of intelligent lighting, the precise detection of a human silhouette on the scene is necessary. Correctly adjusting the light beam divergence requires locating the detected figure in virtual three-dimensional coordinates in real time. The market is currently dominated by the markers systems. This paper is focused on the advanced solution of the markerless system of identifying and tracking characters based on deep learning methods. Analyses of the selected pose detection, holistic detection (including BalzePose and MoveNet models), and body segmentation (BlazePose and tfbodypix) algorithms are presented. The BlazePose model was implemented for both pose tracking and body segmentation in the markerless dynamic lighting and mapping system. This article presents the results of the accuracy analysis of matching the displayed content to a moving silhouette. An assessment of the illumination precision was done as the function of the movement speed for the system with and without delay compensation.

For brain tumour treatment plans, the diagnoses and predictions made by medical doctors and radiologists are dependent on medical imaging. Obtaining clinically meaningful information from various imaging modalities such as computerized tomography (CT), positron emission tomography (PET) and magnetic resonance (MR) scans are the core methods in software and advanced screening utilized by radiologists. In this paper, a universal and complex framework for two parts of the dose control process – tumours detection and tumours area segmentation from medical images is introduced. The framework formed the implementation of methods to detect glioma tumour from CT and PET scans. Two deep learning pre-trained models: VGG19 and VGG19-BN were investigated and utilized to fuse CT and PET examinations results. Mask R-CNN (region-based convolutional neural network) was used for tumour detection – output of the model is bounding box coordinates for each object in the image – tumour. U-Net was used to perform semantic segmentation – segment malignant cells and tumour area. Transfer learning technique was used to increase the accuracy of models while having a limited collection of the dataset. Data augmentation methods were applied to generate and increase the number of training samples. The implemented framework can be utilized for other use-cases that combine object detection and area segmentation from grayscale and RGB images, especially to shape computer-aided diagnosis (CADx) and computer-aided detection (CADe) systems in the healthcare industry to facilitate and assist doctors and medical care providers.

Specific emitter identification (SEI) is the process of identifying individual emitters by analyzing the radio frequency emissions, based on the fact that each device contains unique hardware imperfections. While the majority of previous research focuses on obtaining features that are discriminative, the reliability of the features is rarely considered. For example, since device characteristics of the same emitter vary when it is operating at different carrier frequencies, the performance of SEI approaches may degrade when the training data and the test data are collected from the same emitters with different frequencies. To improve performance of SEI under varying frequency, we propose an approach based on continuous wavelet transform (CWT) and domain adversarial neural network (DANN). The proposed approach exploits unlabeled test data in addition to labeled training data, in order to learn representations that are discriminative for individual emitters and invariant for varying frequencies. Experiments are conducted on received signals of five emitters under three carrier frequencies. The results demonstrate the superior performance of the proposed approach when the carrier frequencies of the training data and the test data differ.

This paper addresses the problem of part of speech (POS) tagging for the Tamil language, which is low resourced and agglutinative. POS tagging is the process of assigning syntactic categories for the words in a sentence. This is the preliminary step for many of the Natural Language Processing (NLP) tasks. For this work, various sequential deep learning models such as recurrent neural network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU) and Bi-directional Long Short-Term Memory (Bi-LSTM) were used at the word level. For evaluating the model, the performance metrics such as precision, recall, F1-score and accuracy were used. Further, a tag set of 32 tags and 225 000 tagged Tamil words was utilized for training. To find the appropriate hidden state, the hidden states were varied as 4, 16, 32 and 64, and the models were trained. The experiments indicated that the increase in hidden state improves the performance of the model. Among all the combinations, Bi-LSTM with 64 hidden states displayed the best accuracy (94%). For Tamil POS tagging, this is the initial attempt to be carried out using a deep learning model.

Convolutional neural networks have achieved tremendous success in the areas of image processing and computer vision. However, they experience problems with low-frequency information such as semantic and category content and background color, and high-frequency information such as edge and structure. We propose an efficient and accurate deep learning framework called the multi-frequency feature extraction and fusion network (MFFNet) to perform image processing tasks such as deblurring. MFFNet is aided by edge and attention modules to restore high-frequency information and overcomes the multiscale parameter problem and the low-efficiency issue of recurrent architectures. It handles information from multiple paths and extracts features such as edges, colors, positions, and differences. Then, edge detectors and attention modules are aggregated into units to refine and learn knowledge, and efficient multi-learning features are fused into a final perception result. Experimental results indicate that the proposed framework achieves state-of-the-art deblurring performance on benchmark datasets.

With the continuous development of bridge technology, the condition assessment of large bridges has gradually attracted attention. Structural Health Monitoring (SHM) technology provides valuable information about a structure's existing health, keeping it safe and uninterrupted use under various operating conditions by mitigating risks and hazards on time. At the same time, the problem of bridge underwater structure disease is becoming more obvious, affecting the safe operation of the bridge structure. It is necessary to test the bridge’s underwater structure. This paper develops a bridge underwater structure health monitoring system by combining building information modeling (BIM) and an underwater structure damage algorithm. This paper is verified by multiple image recognition networks, and compared with the advantages of different networks, the YOLOV4 network is used as the main body to improve, and a lightweight convolutional neural network (Lite-yolov4) is built. At the same time, the accuracy of disease identification and the performance of each network are tested in various experimental environments, and the reliability of the underwater structure detection link is verified.

Electrocardiography is an examination performed frequently in patients experiencing symptoms of heart disease. Upon a detailed analysis, it has shown potential to detect and identify various activities. In this article, we present a deep learning approach that can be used to analyze ECG signals. Our research shows promising results in recognizing activity and disease patterns with nearly 90% accuracy. In this paper, we present the early results of our analysis, indicating the potential of using deep learning algorithms in the analysis of both onedimensional and two–dimensional data. The methodology we present can be utilized for ECG data classification and can be extended to wearable devices. Conclusions of our study pave the way for exploring live data analysis through wearable devices in order to not only predict specific cardiac conditions, but also a possibility of using them in alternative and augmented communication frameworks.