Convolutional Neural Network-based Identification of Communication Room Indicator Status in Complex Environment

Recognizing and detecting the status indicators of communication room equipment is crucial to improving the security and stability of communication networks. Deep learning-based recognition methods have demonstrated higher accuracy and real-time performance for image recognition, with YOLOv5s being a small network-scale model that achieves high precision and real-time performance. It is particularly suitable for detecting indicator lights in computer room environments. However, target recognition faces significant challenges and limitations in complex production environments, such as physical phenomena like light reflection and refraction, which can make feature extraction and fusion difficult. In this paper, we address these challenges by enhancing the YOLOv5s network with a SimAM attention mechanism module. This improves the network's attention to critical information and enhances its feature fusion capability. Our goal is to improve the accuracy and recognition precision of the model in complex communication room environments. We conducted experiments on a self-built dataset based on the actual production environment in the server room. The results show that the improved YOLOv5s network achieved 94.7% recognition accuracy and 95.0% mAP-50, which is 1.6 and 1.3 percentage points higher than the baseline network YOLOv5s, respectively. The improved model showed better performance in the actual production environment of the complex communication server room.