Convolutional neural networks (CNNs) are known for their ability to learn and recognize various objects and patterns, which makes them be powerful feature extractors. While 2D convolutions used in CNNs are effective for extracting features from images, their large convolutional kernels, such as \(7\times 7\) kernel, often result in more parameters. This makes them unsuitable for object detection task on devices with limited computational resources. However, 1D convolutions with fewer parameters also lead to a performance degradation of detection model. To address this issue, we build an efficient multi-scale cross 1D convolutional block (EMC1DB) and a enhanced compact inverted block (ECIB) in YOLOv10-N (a lightweight version of YOLOv10). Additionally, in order to address the problem of inaccurate bounding box localization in object detection, we introduce a modified bounding box regression loss named Anchored IoU loss (AIoU), which guides the model to precisely locate the bounding boxes. By integrating EMC1DB, ECIB, and AIoU into YOLOv10-N, we propose an enhanced YOLOv10-N. Experimental results on multiple datasets demonstrate that the enhanced YOLOv10-N achieves superior performance compared to the original model. In addition, we also extend EMC1DB and AIoU to YOLOv7-T (a lightweight version of YOLOv7), and experimental results show that the proposed method also performs well. The code can be available at https://github.com/Latitude9527/EMC1DB-AIoU