Siamese Network with Phash for Video Vibration Detection

An efficient Simaese network(SiaNet) with Phash for video vibration detection in the paper were formed, such a network can transform the images to feature vectors and by caculating the distance between two consecutive picture vectors to judge whether the video is in stable or vibrating state. The proposed SiaNet takes the network with excellent feature extraction performance such as vgg as the base network branch, Uses Phash to preprocess the images that previous frame and current frame of the input network to screen out the simple samples, and adds the SPP layer before the decision layer to increase the applicability and robustness of the network. Then, a technique called Contrasive Loss of Loss function was used in this article to reduce the distance between positive sample pairs and increase the distance between negative sample pairs. Finally, some experiments for the the SiaNet on 16000 electric meter pictures which were collected in the subway station machine room are utilized to demonstrate the validity of the proposed SiaNet with Phash.