Abstract: Existing text detection algorithms for complex natural scenes often suffer from high model complexity, making them challenging to deploy on edge embedded devices, while also tending to miss small-scale text. To address these issues, scene text detection algorithm called Center-enhanced Network (CENet) is proposed. To facilitate deployment on edge devices, CENet utilizes the lightweight ResNet18 as the backbone network. Center-Enhanced Module (CEM) is introduced to enhance the features of small-scale text, addressing the problem of small-scale text being easily overwhelmed by the background due to downsampling operations. Additionally, we incorporate a Shallow Feature-Augmented Module (SFAM) that leverages the rich semantics from high-level features to guide shallow features, thereby enhancing the detailed information of small-scale text in shallow layers. Ablation experiments on public datasets ICDAR2015 and MSRA-TD500 demonstrate the effectiveness of the CENet algorithm. Comparative experiments on the small-scale text dataset ICDAR2015 show that CENet improves accuracy and F-score compared to the mainstream DBNet algorithm.