In recent years, numerous deep learning (DL) detectors for optical remote sensing target identification have emerged. Nevertheless, the majority of the latest detectors are predominantly engineered to enhance accuracy, often neglecting the critical equilibrium among accuracy, feasibility, and detection speed. This deficiency hinders the practical implementation of these detectors, especially in embedded systems. Building upon YOLOv8, we introduce BRS-YOLO, a distinctive convolutional network. Firstly, a novel, lightweight and efficient attention mechanism convolutional module has been introduced in the backbone. This module guides the framework to focus on the aimed object and its image coordinate information while suppressing background information, thereby reducing computational costs. Secondly, a feature fusion module is designed to combine multi-scale data and improve the ability to recognize densely overlapped small objects. Finally, Inner-SIoU overcomes the limitations of existing methods in terms of generalization ability. Experimental results on the DOTA dataset illustrate that the function of BRS-YOLO surpasses that of YOLOv8, with mAP increasing from 56.0% to 60.9%, and Params decreasing by 17.7%. Compared to state-of-the-art detectors, BRS-YOLO achieves an optimal equilibrium among accuracy, feasibility, and detection speed.