Abstract:Partial discharges are known as indicators of degradation of insulation systems.The reliability and selectivity of methods to detect internal partial discharges in the covered conductors are dictated by the level of background noise. The background noise distorts the pattern of partial discharges (PD-pattern) and decreases the capability of detection methods to recognize the features of PD-pattern corresponding to the degradation of an insulation system. This paper proposes a deep learning based framework with novel implementation of frequency and phase attention layers to detect partial discharge pattern on insulated overhead conductors.The introduced phase and frequency attention layers finds the discriminative regions responsible for PD activity in the spectograms of the signals.