Recognition method for space target components in terahertz radar images based on improved YOLOv5

In the study of spatial target component recognition based on terahertz radar imaging using the traditional YOLOv5network, the recognition performance of the model decreases due to large overlapping areas of components in some samples and unclear imaging of small components. To address this issue, this paper proposes a typical component target recognition model, BoT-YOLO+, based on an improved YOLOv5 network architecture. On one hand, the proposed model enhances performance by introducing the BoTNet backbone architecture, which incorporates the attention mechanism from Transformers and improves the feature extraction capability for small components and thereby increasing the recognition rate of small components, without significantly increasing computational costs.