Consideration of the possibility of using fractal analysis to improve the results of a convolutional neural network to increase the noise immunity of signals

Background. It is widely known that traditional signal processing approaches suffer from low efficiency in the presence of strong noise effects. Therefore, it is becoming relevant to use modern machine learning technologies and nonlinear analysis methods to solve the problem of recovering noisy data. Aim. Investigation of the possibilities of increasing the resistance of convolutional neural networks to interference by integrating fractal analysis, which allows to identify hidden patterns and improve the quality of recognition of noisy signals. Methods. Combining fractal analysis with deep learning methods. The fractal properties of the signals were analyzed, which made it possible to create effective algorithms for preprocessing data before feeding it to the network input. Data transformation was considered in order to enhance the important features of the signals and reduce the impact of noise. Results. Theoretically, an improvement in the quality of signal recognition and filtering has been determined when using the proposed method. It is considered that the integration of fractal analysis significantly improves the convolutional network’s ability to recover noisy data, as well as reduces the impact of external interference. Conclusion. The use of fractal analysis in conjunction with convolutional neural networks is a promising area for further research, since it is possible to increase the noise immunity of signal processing systems. Additionally, new opportunities are opening up for efficient analysis of complex data in conditions of high noise levels.