基于CapsNet的中国手指语识别

传统的手指语识别采用卷积神经网络的方法,模型结构单一,在池化层会丢弃很多信息。Capsule(胶囊)是在神经网络中构建和抽象出的子网络,每个胶囊都专注于一些单独的任务,又能保留图像的空间特征。分析了中国手语中手指语的特征,构建并扩展了手指语图片训练集,试图用CapsNet(胶囊网络)模型完成手指语识别的任务,对比了不同参数下CapsNet的识别率,并与经典的GoogLeNet卷积网络作对比。实验结果表明,CapsNet在手语识别任务上能达到较好的识别效果。

Chinese finger language recognition using CapsNet

Traditional finger-language recognition adopted the method of convolution neural network(CNN), leading to the structure of the model was single and a lot of information would be discarded in the pooling layer. Capsules are kinds of constructed and abstracted subnetworks in neural networks, and meanwhile each Capsule focuses on individual tasks and preserving spatial features of the image. This paper analyzed characteristics of finger language in Chinese sign language, and constructed and expanded training set of finger language pictures, it tried to solve the task of finger language recognition by using CapsNet. Compared the CapsNet recognition rate under different parameters and compared with the classic GoogLeNet convolution network, experimental results show that CapsNet can achieve better recognition effect in the task of sign language recognition.