基于卷积神经网络的中草药识别算法研究

Research on a Convolutional Neural Network-Based Algorithm for Herbal Medicine Recognition

目的:对基于卷积神经网络的中草药识别算法进行研究,以提高中草药识别效率和准确率。方法:首先,自构建中草药图像数据集,利用Z-score标准化和图像归一化等方法对图像进行预处理,采用图像分割方法消除复杂背景对识别的影响;然后,搭建中草药识别网络模型——HMRNet,该网络模型由多个卷积层和全连接层组成,并引入Flatten层和Droupout层以提高识别效率;最后,基于该模型设计出中草药识别系统。结果:相较于VGGNet、AlexNet、GoogLeNet和ResNet50主流网络模型,HMRNet在识别性能上提高明显,达到了97.3%的识别准确率。结论:基于卷积神经网络的中草药识别系统界面简洁、操作方便,且能达到较好的中草药识别效果,可为中草药知识的普及和应用提供技术支持。

Objective:To study a convolutional neural network(CNN)-based algorithm for herbal medicine recognition to improve recognition efficiency and accuracy.Methods:First,a self-constructed herbal medicine image dataset was used.Image preprocessing was performed using methods such as Z-score normalization and image normalization,and image segmentation was employed to eliminate the influence of complex backgrounds on recognition.Then,an herbal medicine recognition network model,HMRNet,was built.This network model consists of multiple convolutional layers and fully connected layers,and introduces Flatten and Dropout layers to enhance recognition efficiency.Finally,a herbal medicine recognition system was designed based on this model.Results:Compared with mainstream network models such as VGGNet,AlexNet,GoogLeNet,and ResNet50,HMRNet showed significant improvement in recognition performance,achieving an accuracy rate of 97.3%.Conclusion:The CNN-based herbal medicine recognition system has a simple interface,is easy to operate,and achieves good recognition results.It can provide technical support for the popularization and application of herbal medicine knowledge.