SES-YOLOv8n is an improved object detection algorithm for autonomous driving based on YOLOv8n. The paper proposes a new network to address the challenge of balancing real-time detection and high accuracy in practical applications. The proposed model enhances the backbone network by replacing the SPPF module with SPPCSPC to improve feature fusion across different scales. An efficient multi-scale attention module (EMA) is introduced into the C2F module to enhance perception in critical areas and feature extraction efficiency. The SPD-Conv module replaces part of the convolution modules in the backbone network to retain more feature information and improve accuracy and learning ability. Experimental results on the KITTI and BDD100K datasets show that the improved model achieves average accuracies of 92.7% and 41.9%, which are 3.4% and 5.0% higher than the baseline model, demonstrating significant improvements. The model can perform real-time image processing in general scenes while maintaining high detection accuracy. Autonomous driving technology is becoming a practical solution due to advances in science and technology and AI. Object detection is crucial for identifying targets in complex environments. Deep learning has enabled computers to understand images similarly to humans. Object detection methods are divided into two-stage and one-stage methods. The YOLO family is popular due to its accuracy and lightweight design. The latest YOLOv8 series shows excellent performance. This study aims to enhance and optimize the YOLOv8 network to improve object detection ability. The paper introduces three key improvements: replacing the SPPF module with SPPCSPC to better capture target features, adding an EMA attention mechanism to enhance feature expression, and replacing downsampling with SPD-Conv to retain more feature information. These changes improve the model's accuracy and detection ability.SES-YOLOv8n is an improved object detection algorithm for autonomous driving based on YOLOv8n. The paper proposes a new network to address the challenge of balancing real-time detection and high accuracy in practical applications. The proposed model enhances the backbone network by replacing the SPPF module with SPPCSPC to improve feature fusion across different scales. An efficient multi-scale attention module (EMA) is introduced into the C2F module to enhance perception in critical areas and feature extraction efficiency. The SPD-Conv module replaces part of the convolution modules in the backbone network to retain more feature information and improve accuracy and learning ability. Experimental results on the KITTI and BDD100K datasets show that the improved model achieves average accuracies of 92.7% and 41.9%, which are 3.4% and 5.0% higher than the baseline model, demonstrating significant improvements. The model can perform real-time image processing in general scenes while maintaining high detection accuracy. Autonomous driving technology is becoming a practical solution due to advances in science and technology and AI. Object detection is crucial for identifying targets in complex environments. Deep learning has enabled computers to understand images similarly to humans. Object detection methods are divided into two-stage and one-stage methods. The YOLO family is popular due to its accuracy and lightweight design. The latest YOLOv8 series shows excellent performance. This study aims to enhance and optimize the YOLOv8 network to improve object detection ability. The paper introduces three key improvements: replacing the SPPF module with SPPCSPC to better capture target features, adding an EMA attention mechanism to enhance feature expression, and replacing downsampling with SPD-Conv to retain more feature information. These changes improve the model's accuracy and detection ability.