Pointer-type instruments continue to play an important role in monitoring key operating parameters in industrial environments such as power systems and chemical plants. However, current vision-based reading techniques frequently struggle in terms of accuracy and real-time performance, especially when deployed on edge devices with low processing resources. This study presents an enhanced YOLOv8-based framework that surpasses existing algorithms in terms of accuracy and real-time performance, while maintaining a lightweight design and suitability for deployment on edge devices. A new Context-Aware Spatial-Channel Attention (CASCA) mechanism is suggested, which efficiently captures spatial context and channel dependencies while improving the network’s capacity to discern instrument-specific features. To improve computational efficiency, the model uses structural pruning and replaces standard convolutions with Ghost convolutions, resulting in a smaller model size and lower inference complexity while maintaining accuracy. Furthermore, a hybrid approach combining U2Net-based fine segmentation and Hough transform-based line detection is used to accurately recover the pointer’s orientation and scale. Comprehensive experiments show that the suggested technique achieves a detection accuracy of 96.1% while boosting inference speed, validating its appropriateness for real-time, high-precision sensor data in industrial settings.
Zhao et al. (Wed,) studied this question.