化学分析实验中的行为识别研究

Research on Action Recognition in Chemical Analysis Experiments

为了提高化学实验室的智能化水平,本文对双流卷积网络进行研究,提出一种应用在化学分析实验领域的行为识别网络。本文网络将输入视频分为RGB和光流,通过改进的EfficientNetv2网络加深网络结构提取特征信息,同时探究双流网络的融合位置,改进损失函数,确定网络的最佳策略。经验证在自制的数据集中改进的双流卷积神经网络模型准确率可以达到92.4%,相比常规的行为识别算法具有更高的识别率。

In order to improve the intelligence level of the chemical laboratory, this paper studies the two-stream convolutional network and proposes an action recognition network applied in the field of chemical analysis experiments. In this paper, the network divides the input video into RGB and optical flow, and uses the improved EfficientNetv2 network to deepen the network structure to extract feature information. At the same time, it explores the fusion position of the dual-stream network, improves the loss function, and determines the best strategy for the network. It has been verified that the accuracy rate of the improved two-stream convolutional neural network model in the self-made dataset can reach 92.4%, which is higher than the conventional behavior recognition algorithm.