Monitoring Of The Equipment Based On Deep Learning

With the continuous development of electronic information technology,the computer room equipment is constantly updated and increased.If the computer room equipment failure,it will directly affect the normal operation of the whole system.In this paper,according to the situation of unattended computer room,the monitoring system of the computer room equipment is designed,which can monitor the running state of the equipment in real time by analyzing the pictures of the room.The core of our mission here is the object detection.Object detection in the field of artificial intelligence has been in-depth study,and in the industry has been widely used.As the object detection is widely used,there are many challenges and problems in object detection.The traditional object detection algorithm adopts morphological statistics algorithm,because the need to set the algorithm parameters accords to the environment,resulting in poor adaptability of the algorithm.We use the selective search to extract the region of interest in the image.It needs searching the global image,so extraction speed is very slow.Therefore,we use convolution neural network for object detection tasks.To solve these problems,we design a suitable convolution neural network to extract the region of interest,and then design a convolution neural network to identify the target object in the region of interest.We design a class of error function to update the whole network weight to improve the accuracy of classification;In addition,we design a position coordinate error function,in the training time to constantly update the network weight to realize linear position regression to improve the accuracy of location prediction.We can verify that through experiments the accuracy and detection speed of this method can meet our requirements.In addition,we have studied the improvement of the convolution neural network and found that the convolution layer of the convolution neural network has the phase symmetry characteristic,so we can reduce the output characteristic of the first few layers by half,half from the mirror.Through the above methods it can reduce the numbers of network weights,which can improve the training and testing speed.