Automatic Reading Method for Offshore Platform Wellhead Instruments Based on an Improved RTDETR Model

A method for automatic reading of instrument parameters was proposed to satisfy the real-time, accuracy, and computational requirements of wellhead pointer instrument detection on offshore platforms. An Edge Fusion Cross-Stage Partial Module (EFCM) was developed, and the backbone network of the Real-Time Detection Transformer (RT-DETR) was restructured to enhance detection speed. A Matchability-Aware Loss (MAL) was introduced to address the inability of the original loss function to preserve matching quality information in both loss weights and supervised labels simultaneously. Results from ablation and model comparison experiments demonstrated that the optimized model offers significant improvements in lightweight design, detection speed, and accuracy. Evaluation experiments indicated that the proposed pointer instrument reading method accurately retrieves parameters from images, thereby meeting the monitoring and automatic reading requirements for offshore platform wellhead instruments.