基于改进全卷积神经网络的缓控释制剂检测方法研究  

Research on the detection method of sustained-release formulations based on improved Fully Convolutional Neural Network

作　　者：樊晨[1] 冯飞艳 郭建刚 贾毅 FAN Chen;FENG Feiyan;GUO Jiangang;JIA Yi(Xi’an Vocational and Technical College,Xi’an 710077,China;Shaanxi Kanghui Pharmaceutical Co.,Ltd.,Xianyang Shaanxi,712021,China;Xi’an Kelikang pharmaceutlcal technology CO.,LTD,Xi’an 710076,China)

机构地区：[1]西安职业技术学院,西安710077 [2]陕西康惠制药股份有限公司,陕西咸阳712021 [3]西安科力康医药科技有限公司,西安710076

出　　处：《自动化与仪器仪表》2024年第12期42-45,51,共5页Automation & Instrumentation

基　　金：西安职业技术学院2022年科技项目《缓控释制剂的质量评价》(2022YB11)。

摘　　要：深度学习的发展对药品称量检测有着重要意义,由于缓控释制剂的特点,需要对药品的称量严格控制,而传统药品称量检测方法并不能精准地对药品进行称量。因此,此次研究首先对药品图像成像存在大量噪音的问题,提出了一种基于高斯滤波和双边滤波的图像处理方法,然后提出了全卷积神经网络来构建模型,并引入可变形卷积对模型进行改进,从而构建了一种改进全卷积神经网络模型来对缓控释制剂进行检测。实验结果表明,当验证集尺寸为600时,U神经网络、全卷积神经网络、改进全卷积神经网络算法模型的识别准确率分别为0.40、0.42、0.71。研究结果表明,所提出的改进全卷积神经网络算法模型表现出优于其他模型的性能。The development of deep learning is of great significance for drug weighing and testing.Due to the characteristics of sustained-release formulations,strict control of drug weighing is required,and traditional drug weighing and testing methods cannot accurately weigh drugs.Therefore,this study first addresses the issue of a large amount of noise in drug image imaging and proposes an image processing method based on Gaussian filtering and bilateral filtering.Then,a fully convolutional neural network is proposed to construct the model,and deformable convolution is introduced to improve the model.Thus,an improved fully convolutional neural network model is constructed to detect sustained-release formulations.The experimental results show that when the validation set size is 600,the recognition accuracy of the U neural network,fully convolutional neural network,and improved fully convolutional neural network algorithm models are 0.40,0.42,and 0.71,respectively.The research results indicate that the proposed improved fully convolutional neural network algorithm model performs better than other models.

关 键 词：全卷积神经网络 缓控释制剂 药品检测 可变形卷积 高斯滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]