Abstract: The power communication equipment room is equipped with a wide variety of devices arranged in a disorganized manner. Traditional manual inspection consumes substantial manpower and time, frequently leading to inspection omissions. To address this issue, high-precision real-time automatic visual identification techniques are adopted in this work. The proposed algorithm optimizes computational efficiency through multi-level feature extraction and enhancement as well as multi-scale feature fusion, substantially improving the image detection performance of room devices. Comparative experiments are carried out under variable lighting, occlusion scenarios and mixed device categories, with the model reaching a recognition accuracy of 96.5%. Significant performance gains are observed for small-sized and occluded equipment, which enables the system to fulfill real-time monitoring demands and sustain stable performance during long-running operations. Comprehensive tests demonstrate that the presented approach improves both the recognition accuracy and adaptability for communication room devices, and can supply dependable support for efficient equipment operation and real-time condition monitoring.