Human chorionic gonadotropin (HCG) is an essential biomarker for the evaluation and diagnosis of early pregnancy, multiple pregnancies, and ectopic pregnancies. However, the accuracy of test strip interpretation is often compromised by inconvenient and uncomfortable professional testing, the black-box nature of AI-based detection methods, and variations in image quality caused by mobile photography and lighting conditions. To address these challenges, we propose 2SLOD-HCG, a novel network for test strip concentration detection. Our approach introduces an enhanced spatial pyramid pooling (SPP) module to better integrate multi-scale receptive field information and incorporates an elastic variational cross-FPN structure augmented with lightweight transformer blocks to strengthen global feature perception. Furthermore, a SimAM attention mechanism is applied to highlight critical local features. These improvements collectively enhance the network’s ability to capture both fine-grained and global contextual information. We constructed a dataset of 50,000 augmented test strip images collected under three lighting conditions and four mobile photography scenarios. The results demonstrate that 2SLOD-HCG achieves superior accuracy and robustness compared to existing YOLO-based baselines, particularly in detecting the small color-developing regions of test strips.