Imaging photoplethysmographic as a physiological signal monitoring technology has significant advantages in achieving non-contact heart rate detection, particularly in terms of convenience and safety. However, it still faces challenges from multiple interference factors regarding accuracy and robustness. Existing research has primarily focused on the impact of motion interference and complex lighting fluctuations on heart rate detection, but solutions for video blur issues under substantial occlusion conditions remain insufficient. To address this, this study proposes an innovative hybrid network model, iPPGYdrNet, which can extract high-quality iPPG signals from high frame rate videos with severe facial occlusion. By employing a frame-splitting dual-stream method to expand the signal channels, a significant improvement in heart rate extraction accuracy is achieved during the signal analysis phase. To validate the model’s performance, a relevant dataset was established and testing was completed, with results indicating an average absolute error of 1.80 and a root mean square error of 2.13 under severe occlusion and blur conditions, while the signal quality measurement standard root mean square error reached 0.96. Experimental results demonstrate that iPPGYdrNet can still provide highly reliable heart rate estimates under multiple interferences such as occlusion, motion interference, and blur, offering a novel solution for heart rate extraction in complex environments.