基于深度卷积网络的港口集装箱属性识别方法

Port Container Attribute Recognition Method Based on Deep Convolution Network

计算机目标识别技术在智慧港口建设中有广泛需求和应用,本文提出一种基于深度卷积神经网络的集装箱箱门及铅封识别方法,该方法充分利用卷积神经网络自身的频率选择特性以及产生平移、旋转、缩放不变性特征的能力,对卷积网络中间层的深度表征进行分析,提取与检测目标相关的特征图子集。通过对特征子集进行组合,产生能够检测目标的显著性分布特征图,并设计相应的目标函数。最后通过实验,对集装箱箱门、铅封等相关目标进行检测,取得良好效果,验证了该方法的有效性。

Computer target recognition technology has a wide range of needs and applications in the construction of smart ports.This paper proposes a container door and seal identification method based on deep convolutional neural network,which makes full use of the frequency selection characteristics of convolutional neural networks.And the ability to generate translation,rotation,and scaling invariant features,analyze the depth representation of the intermediate layer of the convolutional network,and extract a subset of the feature maps associated with the detection target.By combining the feature subsets,a feature map of the significance distribution that can detect the target is generated,and the corresponding objective function is designed.Finally,through experiments,the container door,seal and other related targets were tested,and good results were obtained.The effectiveness of the method was verified.