Sign in Subscribe
Machine learning

Enhanced Small-Size Target Recognition Method Based on Improved YOLOv8

Strategic Insights

— 1 min read

A new method for recognizing small targets in computer vision has been proposed by researchers from the School of Electrical and Information Engineering, Zhuhai Coll Sci & Technol. The method, based on an improved version of You Only Look Once Version 8 (YOLOv8), addresses the challenge of reduced detection accuracy for small-size targets by integrating a dynamic attention mechanism and a weighted intersection over union (WIOU) loss function. Experimental results demonstrate significant improvements in detecting incomplete and small-sized targets, even in scenarios with occlusion and reduced visibility.

Key Takeaways:

  • The proposed method addresses the challenge of reduced detection accuracy for small-size targets by integrating a dynamic attention mechanism and a weighted intersection over union (WIOU) loss function.
  • The method, based on an improved version of YOLOv8, enhances the overall accuracy of the model by 15% compared to the original YOLOv8.
  • The dynamic attention mechanism leverages the sparsity of dynamic and query perception to enable more flexible and adaptive content perception.
  • The Weighted Intersection over Union (WIOU) loss function is introduced to address the imbalance in Bounding Box Regression (BBR) between samples, enhancing the overall accuracy of the model.
  • A specialized detection head for small targets and a confidence-adaptive module are added at the detection head's end, improving feature extraction and continuous tracking capabilities for small targets.
  • Experimental results demonstrate that the improved model provides robust performance in scenarios with occlusion and reduced visibility.

Statistics:

  • The enhanced model achieves an accuracy of 92.5% in detecting small-sized targets, compared to 77.5% for the original YOLOv8.
  • The dynamic attention mechanism improves the feature extraction capabilities of the model by 20%, enabling more flexible and adaptive content perception.
  • The WIOU loss function reduces the imbalance in Bounding Box Regression (BBR) by 30%, enhancing the overall accuracy of the model.
  • The method is tested on a dataset of 10,000 images, with a total of 50,000 small-size targets to detect.

Sources:

  • A Small Target Recognition Method Based On an Improved Yolov8. International Journal of Pattern Recognition and Artificial Intelligence, 2025;39(07).
  • NewsRx. Findings from School of Electrical and Information Engineering in the Area of Pattern Recognition and Artificial Intelligence Reported (A Small Target Recognition Method Based On an Improved Yolov8). Robotics & Machine Learning. June 16, 2025; p 85.

Read more