Content published in the media significantly impacts the perception of the situation in the world and the formation of opinions on current events. The Internet, in particular, has become a medium where information can be published easily and quickly. More and more information portals are being created from which we learn about the reality around us. Unfortunately, most do not use primary and reliable resources but only from sources that have already published information, especially photos. There are also situations where fake news and supporting images are created intentionally to evoke negative emotions or spread disinformation. Such actions may impact global politics and the economy and, in specific cases, threaten local or international security. Modern scientific analyses focus primarily on detecting fake news and fake images. However, a more effective method is to prevent the publication of fake news. We propose a method for authorizing content and images using blockchain technology. The presented method has mechanisms that protect against manipulation attempts, which has been confirmed in research.

Recent years have shown how online news can raise public doubts about the actions taken by governments, construct alternative narratives of events, especially in times of crisis, and even, as reported by sources, poison democratic elections and incite riots. False news spread on the Internet threatens entire societies, public opinion, views, and even democracy. The sources of false information have become weapons of violence, persecution, and even blackmail. So far, a reliable, fully automated system for detecting false news has not yet been developed. New solutions are emerging to support the fight against fake news. Often, an element of such news is an image, which can also defame or mislead. This paper presents a tool for detecting false or modified facial images.

The interpretation and understanding of images using deep learning algorithms in monitoring systems is an important area of research in computer vision. In this paper, we propose a novel approach for interpreting and understanding images of human behaviour captured from cameras and drones in monitoring systems. In the work, we presented a pedestrian detection method. This is the first element of a larger system of interpreting the behaviour of traffic participants. We begin by pre-processing the images to remove noise and unwanted artefacts. We then use labelled and unlabelled data to train a deep learning algorithm to accurately classify and predict human behaviour. To account for imprecise information, we incorporate probabilistic models into the algorithm to allow it to make predictions with uncertainty. To evaluate the performance of our proposed approach, we use standard metrics, such as precision, recall, and F1 score, and compare our results to existing methods. We also identify the limitations of existing approaches and propose new methods for interpreting and understanding images that can improve their accuracy and reliability. Our proposed approach has the potential to provide a more accurate and robust method for interpreting and understanding images of human behaviour in monitoring systems. It can be used in various applications like traffic monitoring, security systems, and surveillance. Our results demonstrate the importance of incorporating imprecise information into deep learning algorithms to improve their performance and highlight the potential for future research in this area.

When retrieving data from IoT devices, errors in time series data often occur due to interference. Data with errors cannot be processed or saved in databases and warehouses because it causes data inconsistency, conflicting with Big Data principles. Built-in mechanisms in databases are not always able to fix incorrect data. Manually correcting data for extensive collections is too time-consuming and costly. Therefore, there is a need for automatic data cleansing, especially time series. This article proposes our data cleaning method for time series based on a moving average. Test results show a slight improvement in the signal-to-noise ratio.