Existing salient object detection methods typically depend on large, pretrained backbone networks for feature extraction. While this enhances performance, their large size and high number of parameters make them less practical for widespread use in real-world applications. In contrast, lightweight backbone networks are smaller and have fewer parameters, yet they typically fall short in delivering strong feature extraction and precise localization. To address this issue, this paper introduces FocusNet, an innovative lightweight solution. By cascading lightweight networks through the FOCUS module, we simulate the human eye’s focusing process, dividing localization and feature extraction into two independent and sequential steps, significantly enhancing the feature extraction capabilities. Additionally, utilizing the Ascending Feature Enhancement (AFE) strategy, we progressively saturate the localization of salient objects from shallow to deep layers, significantly improving localization accuracy. Extensive experiments on five public datasets demonstrate that our method, while maintaining a low parameter (3.15M), performs comparably to methods based on large backbone networks. The source code will be released at https://github.com/YinYinOvO/FocusNet .