深度连续卷积神经网络模型构建与性能分析

为了提升卷积神经网络特征提取能力,设计了一种基于连续卷积的深度卷积神经网络模型.该模型采用小尺度的卷积核来更细致地提取局部特征,并借助连续的两个卷积层增加模型的非线性表达能力,结合Dropout技术降低神经元之间的相互依赖,利用抑制网络过拟合对模型进行优化.人脸表情、手写数字字符和彩色图像的目标识别实验表明,在图像较为复杂时,该模型在识别的准确性和泛化性能上比手工特征提取方法及一般的2、3层卷积结构具有明显的优势.

Model construction and performance analysis for deep consecutive convolutional neural network

In order to improve the feature extraction capability of convolutional neural network, a deep convolutional neural network model based on the consecutive convolution was proposed. The small-scaled convolutional kernels in the model were adopted to precisely extract the local features, and the nonlinear expression capability of the model was improved with the help of two continuous convolutional layers. In addition, the mutual dependency between neurons was reduced with the Dropout technology, and the model was optimized through restraining the network over-fitting. The objective recognition experiments of facial expressions, handwritten numeric characters and color images show that when the images are complicated, the proposed model has obvious advantages in the aspects of both recognition accuracy and generalization capability, compared with the manual feature extraction method and general two-layer and three-layer convolutional structures.