A Lightweight and Efficient Algorithm for Detecting Violations of Dress by Power Operators Based on the Improved RT-DETR

To address the challenges of high computational complexity, significant background interference, and inconsistent object scales in dress code violation detection for power operators, we propose a novel lightweight algorithm named ICSS-RTDETR.Initially, the iRMB_Block module integrates convolution with inverted residual moving blocks, replacing the Basic_Block structure in traditional backbones. This modification reduces computational effort and enhances feature extraction efficiency. Subsequently, the CGA-AIFI module combines the cascaded group attention mechanism with the intra-scale feature interaction module to improve the model's focus on key feature regions. Finally, we introduce the Slimneck-SSFF architecture as the cross-scale feature fusion structure to enhance adaptability to targets of varying scales.Extensive experiments conducted on our private DEDV dataset demonstrate that the proposed ICSS-RTDETR significantly improves both detection accuracy and speed, outperforming the baseline RT-DETR and other popular methods.