The Dosimeter Calibration Laboratory (LCD) of the Nuclear Technology Development Center (CDTN) performs the calibration of radiation meters from several radioactive installations in Brazil. The meters are subjected to specific radiation fields while being remotely monitored by means a video camera. The operator then reads the measurement given by the instrument through the video. Most of the radiation meters calibrated by the LCD are analog, that is, their measurement is presented using a pointer. Computer vision is a technology that seeks to automatically recognize images and is suitible for this aplication.

The present work aims to apply Convolutional Neural Networks for Rotated Object Detecton to automate the reading of analog meters, making their calibration faster and less laborious for the operator. To train and test the neural network model, images were acquired from several analog meters calibrated in the LCD and they were annotated and subjected to data augmentation, such as rotation and noise application. The open source framework MMdetection and the Python programming language were used. Preliminary results showed good accuracy of the trained model and the great potential of the approach.

Acknowledgements: This work was supported by the Brazilian National Council for Scientific and Technological Development (CNPq, Project INCT/INAIS - 406303/2022-3 and CNPq, Process 308368/2022-3 -  Research Productivity Fellowship) and the Minas Gerais State Agency for Research and Development (FAPEMIG).