A saliency object detection method based on deep feature extraction module and loss function
Ruixin Li, Fuan Dong, Xiaoning Wu, Yarong Li, Feng Ming Luo, Hongwei Wang · 2025
This article proposes a saliency object detection method that integrates multiple features and multiple loss functions to address unclear and incomplete target boundaries in real complex scenes. It introduces a deep feature extraction module (DFEM) to obtain depth feature information and reduce critical depth information loss. It also presents a jump scale feature fusion module (2SFM) that uses traditional encoding-decoding structure as the basic network for model training, improving feature selection and generalization ability. Finally, it proposes a new mixed loss function to optimize the model, solving the problem of inconsistent detection regions in different loss functions. The experimental results demonstrate the effectiveness and practicality of the proposed method in saliency object detection.