Small Object Image Detection Based on YOLOV5 Improved Algorithm
Download as PDF
Mawei Chen, Peng Yang, Zhongguo Liu
For long-distance aerial images with blurring, unclear features, and small volume. The existing algorithms are not ideal for detecting small targets.This article proposes an improved algorithm model for YOLOV5. Firstly, we improved the Backbone layer of YOLOV5.Adjust the original fast pyramid layer SPPF to an SPPFP layer. Then, we added a CA attention mechanism to improve the recognition of small targets.Secondly, in the Neck layer, adjust the road force aggregation network PANet to BiFPN. Finally, we adjust the loss function CIOU of the original network to EIOU, improve the robustness and generalization ability of the model, and accelerate the convergence speed of the network. The improved model mAP in this article has increased by 10%, indicating that the performance of our proposed model has been effectively improved and has certain application prospects.
Small Goals, YOLOV5, CA Attention Mechanism, BiFPN, Effective Improvement