深度学习在仿生眼监控系统中的应用

在视频监控系统中,由于受到复杂的背景、环境光线变化以及设备本身性能的限制,导致目标检测算法设计难度的加大,而传统的目标检测算法通常依赖于人工选择特征,难以从海量的数据中自动学习得到一个有效的分类器.基于深度学习算法,构建了一个卷积神经网络,并利用仿生眼视频监控系统中采集的人、车图像进行训练,在此基础上设计若干实验对深度学习网络特性进行分析,证明了训练集中各个类别样本的分布以及小样本训练的情况下对深度学习的训练结果会造成较大的影响.

Application of Deep Learning in Bionic Eye Video Surveillance System

In the video surveillance system,the difficulty of objective detection algorithm designed will markedly increase due to the complex background,the changes of ambient light and limitation of performance of the equipment itself.The traditional objective detection algorithm usually relies on artificial feature selection,it is hard to get an effective classifier from the learning of the vast amounts of data automatically.A convolutional neural network is set up based on depth learning algorithm.And a bionic eye video monitoring system is built then to collect images of people and vehicles as training.On this basis,several experiments are designed to analyze the characteristics of the depth learning network.It is proved that the distribution of different classes samples in the training set and the small samples training can result in great influence on training result of the depth learning.