A5073394716- Advanced Neural Network Applications
- Video Surveillance and Tracking Methods
- Cell Image Analysis Techniques
- Autonomous Vehicle Technology and Safety
- 3D Printing in Biomedical Research
- COVID-19 diagnosis using AI
- Traffic Prediction and Management Techniques
- Ultrasound and Hyperthermia Applications
- Neural Networks and Applications
- Advanced Chemical Sensor Technologies
- Microbial Inactivation Methods
- Dental Research and COVID-19
- Human-Automation Interaction and Safety
- Advanced Image and Video Retrieval Techniques
- Image Processing Techniques and Applications
- Remote Sensing and LiDAR Applications
- Industrial Vision Systems and Defect Detection
- Robotics and Sensor-Based Localization
- Robot Manipulation and Learning
- Advanced Vision and Imaging
- Anomaly Detection Techniques and Applications
- Spectroscopy and Chemometric Analyses
- Optical Coherence Tomography Applications
- Brain Tumor Detection and Classification
- Advanced Computational Techniques and Applications
Institute of Electronics and Computer Science
2016-2025
Deep neural networks (DNNs) have achieved state-of-the-art results in a broad range of tasks, particular the ones dealing with perceptual data. However, full-scale application DNNs safety-critical areas is hindered by their black box-like nature, which makes inner workings nontransparent. As response to box problem, field explainable artificial intelligence (XAI) has recently emerged and currently rapidly growing. The present survey concerned perturbation-based XAI methods, allow explore DNN...
Semantic segmentation of an incoming visual stream from cameras is essential part the perception system self-driving cars. State-of-the-art results in semantic have been achieved with deep neural networks (DNNs), yet training them requires large datasets, which are difficult and costly to acquire time-consuming label. A viable alternative DNNs solely on real-world datasets augment synthetic images, can be easily modified generated numbers. In present study, we aim at improving accuracy urban...
It is predicted that due to such technology-driven trends as electrification, connectivity, autonomous driving and diverse mobility, the automotive market will reach a size of 6.7 trillion USD in 2030. Nevertheless, composition architectures future vehicles are still unsettled continuous advancement sensors computing hardware, availability interpretation new data safety requirements. Given existing systems, there lack start-to-end depiction self-driving car design components. This scientific...
Washing hands is one of the most important ways to prevent infectious diseases, including COVID-19. The World Health Organization (WHO) has published hand-washing guidelines. This paper presents a large real-world dataset with videos recording medical staff washing their as part normal job duties in Pauls Stradins Clinical University Hospital. There are 3185 episodes total, each which annotated by up seven different persons. annotations classify movements according WHO guidelines marking...
Intelligent motion control is integral to modern cyber-physical systems. However, smart integration of intelligent with commercial and industrial systems requires domain expertise, 'know-how' the production processes, resilient adaptation for various engineering phases. The challenge amplified adoption advanced digital twin approaches, big data artificial intelligence in domains. This paper proposes IMOCO4.E reference framework platforms (e.g. from SMEs) brings together architecture,...
Organ-on-a-chip (OOC) technology has emerged as a groundbreaking approach for emulating the physiological environment, revolutionizing biomedical research, drug development, and personalized medicine. OOC platforms offer more physiologically relevant microenvironments, enabling real-time monitoring of tissue, to develop functional tissue models. Imaging methods are most common daily development. Image-based machine learning serves valuable tool enhancing models in real-time. This involves...
A video processing algorithm for vehicle parameter acquisition and classification is presented. The based on combination of several detection lines. According to passing vehicles, intervals are created the Intervals different lines, belonging same vehicle, combined. Further allows acquire parameters classify vehicles. accuracy counting analyzed videos.
Washing hands is one of the most important ways to prevent infectious diseases, including COVID-19. Unfortunately, medical staff does not always follow World Health Organization (WHO) hand washing guidelines in their everyday work. To this end, we present neural networks for automatically recognizing different movements defined by WHO. We train network on a part large (2000+ videos) real-world labeled dataset with movements. The preliminary results show that using pre-trained models such as...
The use of synthetic data is a promising solution to the problem availability real needed for development robotic systems. However, precision systems trained on tends decrease when they are deployed in real-life scenarios, which happens due disparities between artificial and world. Therefore, efficient methods Sim2Real translation much further progress robotics. In our study, we Generative Adversarial Networks (GANs) generate more photorealistic from created training deep neural networks...
Augmentation of the datasets authentic microscopy images with synthetic is a promising solution to problem limited availability biomedical data for training deep neural network (DNN) based classifiers. In present study, we use text-to-image latent stable diffusion model fine-tuned by means low-rank adaptation (LoRA) augment small dataset organ on chip cells. While resulting appear quite similar which was performed, find that neither EfficientNetB7 DNN solely nor augmentation real-world...
Current state-of-the-art object detectors are based on supervised deep learning approaches. These methods require a large amount of annotated training data, which hinders wider use these in industry. We propose method for generating synthetic data the task detecting objects pile can be picked up by robot arm. The requires few input images, used to create images piles. After state-of-theart detector we test it real images. results show that model trained such way is not rival best datasets...
We highlight the options available for noninvasive optical diagnostics of reporter gene expression in mouse tibialis cranialis muscle. An vivo multispectral imaging technique combined with fluorescence spectroscopy point measurements has been used transcutaneous detection enhanced green fluorescent protein (EGFP) expression, providing information on location and duration EGFP allowing quantification levels. For coding plasmid (pEGFP-Nuc Vector, 10 μg/50 ml) transfection, we electroporation...
Current global trends and green policies indicate the importance of smart waste sorting. Polymer type identification plays a key role in circular economy model, where high precision is vital to reduce impurities recycled plastic flakes. In this paper, we present robust, high-accuracy bottle polymer classification using Convolutional Neural Network (CNN). Near-infrared (NIR) absorbance spectroscopy used gather polypropylene (PP), polyethene terephthalate (PET), high-density (HDPE),...
Good hand hygiene is one of the key factors in preventing infectious diseases, including COVID-19. Advances machine learning have enabled automated evaluation, with research papers reporting highly accurate washing movement classification from video data. However, existing studies typically use datasets collected lab conditions. In this paper, we apply state-of-the-art techniques such as MobileNetV2 based CNN, two-stream and recurrent to three different datasets: a good-quality uniform...
We employed a Pix2Pix generative adversarial network to translate the multispectral fluorescence images into colored brightfield representations resembling H&E staining. The model underwent training using 512x512 pixel paired image patches, with manually stained serving as reference and processing input. baseline model, without any modifications, did not achieve high microscopic accuracy, manifesting incorrect color attribution various biological structures addition or removal of features....
We compared automated MobileNet V3 Large-based pollen classification accuracy on whole slide images. Pollen fixation to microscope slides with silicone achieved higher median (78.7%) than the standard adhesive tape-based (68.9%).
The paper proposes an efficient method for detection of moving objects in the video. are detected when they cross a virtual line. Only pixels line processed, which makes computationally efficient. A Recurrent Neural Network processes these pixels. machine learning approach allows one to train model that works different and changing outdoor conditions. Also, same network can be trained various tasks, is demonstrated by tests on vehicle people counting. In addition, semi-automatic acquisition...
Monitoring, detection, and control of traffic is a serious problem in many cities on roads around the world poses for effective safe management pedestrians with edge devices. Systems using computer vision approach must ensure safety citizens minimize risk collisions. This well suited multiple object detection by automatic video surveillance cameras roads, highways, pedestrian walkways. A new Annotated Virtual Detection Line (AVDL) dataset presented consisting 74,108 data files manually...