一种改进的残差网络手势识别方法

Gesture Recognition Method Based on Improved Residual Network

为了提高手势识别的准确性、鲁棒性以及收敛速度,提出一种基于改进残差网络和动态调整学习率的手势识别方法研究。改进原始残差块中的ReLU激活函数,通过降低改进后残差块与卷积核的数量来减少卷积层参数;对改进后的残差网络模型进行动态学习率的调节和动量的优化选择;将重建好的网络模型进行训练测试,验证手势识别的准确率。实验结果表明:改进后的残差块具有一定的抗干扰能力,残差网络模型sResnet I识别准确率平均提升20%;采用动态调整学习率,网络模型的预训练收敛时间减少15%。

In order to improve the accuracy,robustness and convergence speed of gesture recognition,a gesture recognition method is proposed based on an improved residual network and dynamic adjustment of learning rate.The ReLU activation function in the original residual block was improved to reduce the convolutional layer parameters by reducing the number of the improved residual block and the convolution kernel.And then the dynamic learning rate was adjusted and the momentum is optimized for the improved residual network model.Finally,the reconstructed network model was trained and tested to verify the accuracy of gesture recognition.The experimental results show that the improved residual block has a certain anti interference ability,and that the recognition accuracy of the residual network model sResnet-I is increased by 20%on average.By dynamically adjusting the learning rate,the pre training convergence time of the network model is reduced by 15%.