A Computational Approach to Heat Detection in Radio Frequency Transmitters Using Convolutional Neural Networks

The purpose of this work was to implement a technique that will monitor overheating temperatures on the digital television transmission system. These overheating temperatures cause transmitters to switch off, resulting in the loss of signal transmission This was achieved by integrating convolutional neural network technique with thermal imaging technology. The application of this technique was implemented using the MATLAB software. This involved the investigation and the tuning of different parameters of convolutional neural networks (CNN) to identify the best option for the model. The thermal images were utilized to train and then test the convolutional neural network. The prediction precision was found to be influenced by the parameter options (Layers, Validation frequency, Filters). This means that any degree within these setting possibilities is feasible if the accuracy is good. The performance accuracy results from training to testing of this model varied from 96.9% to 100%. This indicates that an image’s temporal relationships can be successfully captured by applying associated parameters to a convolutional neural network. The results have shown that CNN performs better where there is a lot of trained data. The architecture provides a much precise fitting to an image dataset due to the decreased number of relevant parameters and the reusability of weights. The results of this work will be used on remote radio frequency transmitter sites for heat detection application.