This document presents a two-stream hybrid network approach for salient object detection (SOD) inspired by human visual perceptual saturation, combining both unsupervised and supervised methods. The method employs a polyharmonic neural network with random weights to refine saliency maps and construct a positive feedback loop for improved training efficacy. Experimental results demonstrate that this novel framework surpasses existing SOD methods, enhancing accuracy and robustness across diverse visual scenarios.