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

Abstract

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