Improved YOLOv8 Algorithms for Object Detection

Abstrak

Current Over the past decade, deep neural networks have progressed rapidly, with computer vision consistently achieving new performance milestones and becoming a part of daily life. In target detection, the YOLO model stands out as a popular real-time detection algorithm, known for its speed, efficiency, and accuracy. This study focuses on enhancing the latest YOLOv8 model to improve small object detection and compares it with other YOLO versions. Using YOLOv8 as a foundational deep learning algorithm, we introduced several optimizations, including redefining the detection head, narrowing its perceptual field, and increasing the number of detection heads to better capture fine details in small objects. We then compared this optimized YOLOv8 model with established YOLO models, such as YOLOv3 and YOLOv5n. The experimental results indicate that our optimized model achieves higher accuracy in detecting small objects. This research offers a promising approach for small object detection with strong application potential. As this technology evolves, the YOLO algorithm is expected to remain a key solution in object detection, serving various real-time applications effectively.