RT-DETR based Lightweight Design and Optimization of Thermal Infrared Object Detection for Resource-Constrained Environments

In the domains of military, security, and related fields, thermal infrared object detection has found widespread applications. By directing our focus toward the design and optimization of lightweight infrared object detection network models to reduce computational demands and enhance computational efficiency. This paper selects the RT-DETR object detection network as the foundation and enhances its backbone network to achieve algorithm lightweighting. Simultaneously, based on RepC3, we propose a novel CSP module named FMACSP to improve the feature fusion module of the RT-DETR model, enabling more effective information fusion and reducing model complexity through lightweight strategies. Experimental results demonstrate that the proposed improvements significantly alleviate the computational burden of the model while maintaining detection performance, making it more suitable for resource-constrained environments in the field of thermal infrared image processing and object detection.