Object detection has made great progress in recent years along with the rapid development of deep learning. However, most current object detection networks cannot be used in the devices with limited computation power and memory resource, such as electronic chips, mobile phones, etc. To achieve an object detection network for the resource-constrained scenario, this paper proposes a reconstructed Network for lightweight object detection via Branch Merging (BMNet). BMNet introduces an innovative and efficient architecture named 2-way Merging Lightweight Dense Block (2-way MLDB), which merges the duplicate parts of two branches in a dense block of the backbone network to obtain multi-receptive field features with fewer parameters and computations. In addition, to alleviate the decrease of accuracy caused by drastically reduced parameter size, BMNet builds an FPN-like SSD based on an Attention Prediction Block (APB) structure. Through extensive experiments on two classic benchmarks (PASCAL VOC 2007 and MS COCO), we demonstrate that BMNet is superior to the most advanced lightweight object detection solutions such as Tiny SSD, MobileNet-SSD, MobileNetv2-SSD and Pelee in terms of parameter size, FLOPs and accuracy. Concretely, BMNet achieves 73.48% of mAP on PASCAL VOC 2007 dataset with only 1.49 M parameters and 1.51 B FLOPs, which is the latest result with relatively low resource requirements and without pre-training to date.