Pointer instruments, known for their powerful anti-interference capabilities, stable structure, and low production costs, are widely used in petrochemical production. These instruments require accurate readings for effective industrial measurement and data collection. To address the inefficiency and low precision of traditional manual reading methods, this paper proposes an intelligent single-pointer instrument reading method based on deep learning. In the instrument reading processing workflow, we introduced a brightness-adaptive histogram equalization algorithm, effectively enhancing image contrast and laying a solid foundation for subsequent analysis. For the target detection stage, we performed an in-depth optimization of the YOLOv11 model, significantly improving model performance through a normalization attention module (NAM). This module cleverly suppresses unimportant weights and applies a weight sparsity penalty strategy, achieving higher computational efficiency while maintaining the original performance. Experimentally, the model successfully reached 97.4% accuracy and 98.8% mean average precision (mAP50) at an intersection over union threshold of 0.50. Subsequently, we used an improved U2-Net algorithm to precisely extract pointer and instrument pixels, with the algorithm achieving 94.12% precision and 93.31% recall. To address complex lighting variations in industrial environments, we introduced γ image enhancement technology, significantly improving the detectability of dial information. Finally, through a geometric distance reading method, we achieved precise capture of instrument readings, ensuring stable and reliable acquisition of critical data in diverse industrial scenarios.