基于DBN图像识别的机房巡检系统设计研究

针对电力系统信息化问题提出了一套智能化机房巡检系统，硬件结构包括轨道车装置、云台及工业相机等，软件系统基于Windows平台进行开发，采用B/S三层架构。鉴于服务器信号灯本身较小并发光，且在一幅图像中往往分布有多个多种状态的信号灯等问题，给出了一种新的更加有效的信号灯图像特征——RGB极大比值（RGBMR）的提取算法。运用深度置信网络（DBN）对信号灯图像进行评估识别，RGBMR特征数据的一部分用于DBN模型的训练，另一部分则用于测试。大量的实验分析，以及与图像HSV空间特征和BPNN网络识别效果的对比研究证明，所提算法能更准确地识别信号灯状态图像，该系统能有效地应用于机房巡检中。

Study and Design of Computer Room Inspection System Based on DBN Image Recognition

Aiming at the informationization of power system,a set of intelligent server inspection system is presented.The hardware includes rail car device,cloud terrace and industrial camera etc.The software system is developed based on Windows platform and is of B/S three-layer architecture.In view of the fact that the server signal light is small and shine,and there always distributed with a plurality of various state lights in a single image and other issues,a new and more efficient abstract algorithm for signal light image feature-RGB maximum ratio (RGBMR) is put forward.Then a deep belief network (DBN) is used to recognize the images,a part of RGBMR data is used for training the DBN model,the other part is for testing.A series of experiment analysis and the comparative study on image HSV spatial features and BPNN network recognition prove that the proposed algorithm can identify the signal light state images more accurately,and the system can be effectively applied to computer room inspection.