<p>Camouflaged personnel detection models are crucial for the unmanned combat mode of modern warfare. The high integration of camouflaged personnel with the environment makes recognition difficult, and existing models struggle to balance model complexity and recognition accuracy. To address this challenge, we propose a Lightweight Camouflaged Personnel Detection model (LCPD-DETR) to improve model effectiveness with lower computational cost. In the backbone, we integrate Dual Convolution (DualConv) into residual blocks to alleviate high computational redundancy and propose a Lightweight Multi-Scale Cascaded Block (LMSCBlock) to improve multi-scale feature extraction capability for high-level semantic features. In the neck stage, we design a Local–Global Attention (LGA) module to enhance the model’s ability to focus on local details of camouflaged personnel and improve the model’s computational efficiency. We present the Partial-Diverse-Branch Block (PDBBlock) for cross-scale feature fusion, thereby strengthening the ability of multi-scale feature fusion with lower computational resources. We evaluated our method separately on the Camouflaged People Dataset and the Military Camouflaged Personnel Dataset. The experimental results show that our model meets the requirements of lightweight and accurate detection, providing an efficient solution for military camouflaged personnel detection.</p>