Search academic texts

Information × Registration Number 0222U004908, 0121U111107 , R & D reports Title Increasing the resolution of the infrared image using a convolutional neural network popup.stage_title Head Varfolomieiev Anton Yu., Кандидат технічних наукYahanov Petro O., Кандидат технічних наук Registration Date 17-11-2022 Organization National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" popup.description2 Relevance. The problem of increasing the resolution of infrared sensors arises primarily because of the critical nature of their application (military, medicine, etc.) and the low resolution compared to other types of cameras. One of the most modern approaches of solving this problem is the use of neural networks. Therefore, the development of completely new and improvement of already existing methods of increasing the resolution of infrared images based on neural networks is relevant. The aim of the work is to improve the existing methods based on convolutional neural networks to increase the resolution of infrared images, to create algorithmic and software solutions for their implementation. In order to achieve the goal, the following tasks were solved in the work: 1) analysis of existing methods of increasing image resolution and selection of a basic model for further improvement - the BCLSR neural network; 2) development of HCNNSR neural network, its simulation and comparative analysis with BCLSR. The object of research is the process of increasing the resolution of infrared images. The subject of research is set of methods of increasing the resolution of infrared images using artificial neural networks. Research methods are convolutional neural networks. The scientific novelty of the obtained results is that: 1) For the first time, the BCLSR neural network was proposed to be used to increase the resolution of infrared images. 2) Developed the HCNNSR neural network as an improvement of the BCLSR. The proposed neural network is almost twice (1.9 times) faster and restores the image more accurately (by the PSNR parameter, the difference reaches 0.063 dB). The practical value of the obtained results is determined by the created algorithmic and software solutions for the implementation of the proposed neural network. The HCNNSR software model was created in the Python programming language using the Tensorflow library in the Google Collaboratory environment. Product Description popup.authors Yaroshenko Maksym Oleksandrovych popup.nrat_date 2022-11-17 Close