In this paper, we propose a lightweight salient object detection framework called Multi-Constraint Coupling optimization Network (MCONet) to address the conflict between model scale and inference ability, which can learn more knowledge with fewer parameters through embedding feature priors. Specifically, we build a lightweight encoder as the backbone network to represent the image, and then use two parallel decoders to infer salient mask features and salient edge features respectively. Besides, we fuse the output features of different decoders by a convolutional block attention module (CBAM) module. In addition, we adopt a multi-constraint coupling optimization strategy to increase the soft constraints in the training phase, and improve the prior guidance of the edge to the inference results. Experimental results on 5 public benchmark datasets show that the proposed MCONet can reach comparable even better performance of state-of-the-art lightweight salient object detection models. © 2023 IEEE.