Optimization and Evaluation of YOLO Model on Small Dataset Based on Fine Tuning

Zhengqun Sun

doi:10.54097/d2p2wb29

Authors

Zhengqun Sun School of Software Technology, Dalian University of Technology, Dalian, 116620, China

DOI:

https://doi.org/10.54097/d2p2wb29

Keywords:

Computer vision, Target detection, Artificial intelligence.

Abstract

In the current era of rapid development of artificial intelligence, object detection is one of the core technologies of computer vision. The degree of accuracy is an important criterion to judge the quality of models. Aimed to improve the degree of accuracy of You Only Look Once (YOLO) models, this experiment adopted the transfer learning strategy, through backpropagation and gradient descent, fine-tuning the model parameters under the condition of a small learning rate. This experiment trained data on a small dataset to enhance the efficiency of learning. The result showed that the degree of accuracy of each model has slightly improved. For example, the mAP50 of the YOLOv8l model has been raised from 0.8625 to 0.8798, which is a 2% rise. Although the increase was relatively small, the result is in line with expectations, which verifies the feasibility of this method, and this is a highly generalized method to improve the accuracy of the model’s object detection.

References

[1] Yan X, Zhang Y, Chen H, et al. GLIF Global–Local interaction fusion based UAV small target detection domain shift suppression network. Infrared Physics and Technology, 2026, 152: 106206.

[2] Girma A, Homaifar A, Mahmoud M N. ADP-Net: Adaptive point network with multi-scale attention mechanism for small object detection. Neurocomputing, 2026, 659: 131381.

[3] Wang Y. Research on Object Re-identification and Tracking in Multi-camera Scenarios. Hangzhou: Hangzhou Dianzi University, 2025.

[4] Varghese R, S. M. YOLOv8: A Novel Object Detection Algorithm with Enhanced Performance and Robustness. In: 2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS). Chennai, India, 2024: 1-6.

[5] Mao S, Wang W. A Review of YOLO Series Object Detection Algorithms Based on Deep Learning. Journal of Yan’an University (Natural Science Edition), 2024, 43(02): 88-95.

[6] Zhou H, Jin S, Zhou L, et al. Classification and recognition of camellia oleifera fruit in the field based on transfer learning and YOLOv8n. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2023, 39(20): 159-166.

[7] Wang C, Guo X, Ma C, et al. A Flat Peach Bagged Fruits Recognition Approach Based on an Improved YOLOv8n Convolutional Neural Network. Horticulturae, 2025, 11(11): 1394.

[8] Kim G, Jung H, Lee S. Spatial reasoning for few-shot object detection. Pattern Recognition, 2021, 120: 108118.

[9] Xie W, Bai X, Liu M, et al. Transformer-based multi-scale feature fusion for real-time CT bone metastasis detection. Bone, 2026, 203: 117729.

[10] Alruwaili M, Siddiqi M H, Atta M N, et al. Deep learning and ubiquitous systems for disabled people detection using YOLO models. Computers in Human Behavior, 2024, 154: 108150.