中圖分類號(hào)： TP391.41
文獻(xiàn)標(biāo)識(shí)碼： A
DOI： 10.19358/j.issn.2096-5133.2021.03.006
引用格式： 何啟莉，何家峰，郭娟. 用于自動(dòng)視力檢測(cè)的手勢(shì)識(shí)別方法研究[J].信息技術(shù)與網(wǎng)絡(luò)安全，2021，40(3)：32-37，47.

Research on gesture recognition method for automatic vision detection

Abstract： For automatic vision detection systems, gesture recognition is a key issue, but the traditional convolutional neural network model to recognize gestures has problems such as over-fitting and large amount of calculation. This paper proposes a GR-Alexnet model, which adaptively modifies and optimizes the Alexnet network model. In order to speed up the calculation, three small convolution kernels of 7×7, 5×5, and 1×1 are used to replace the original 11×11 large convolution kernel, and delete the LRN layer and a fully connected layer; in order to reduce the over-fitting effect, a dropout optimization is added after each convolution. The LeNet model, the Alexnet model ,the VGG16 model and the GR-Alexnet model were used for comparative experiments on the same data set. Experiments show that the GR-Alexnet model has a certain improvement in recognition accuracy compared with the traditional model, can suppress the over-fitting phenomenon, and has a faster training speed.

Key words : automatic vision detection；ＯpenＣＶ；gesture recognition；Gesture Recognition AlexＮet(GR-AlexNet)