Improvement of Remote sensing image target detection algorithm based on YOLO V5

Abstract YOLO V5 algorithm is the current target detection algorithm, but accuracy of boundary regression relatively low, and it cannot be used in scenarios with high requirement prediction frame intersection ratio. Therefore, this paper proposes an improved fast convergence model. The can avoid loss shallow semantic information, deepen pyramid depth increase layer by using cross-hierarchy method multiple feature fusion. By reducing network computation parameter number, lightweight processing realized, real-time performance caused model complexity avoided.