基于多传感器人工嗅觉系统及机器学习模型的苹果种类识别方法  

A Multi-sensor Artificial Olfactory System and Machine Learning Model Based on Apple Species Recognition Method

作　　者：李书贤 周琪 樊亚楠 叶诗琪 赵志彪 张思祥[3] LI Shuxian;ZHOU Qi;FAN Yanan;YE Shiqi;ZHAO Zhibiao;ZHANG Sixiang(School of Automation and Electrical Engineering,Tianjin University of Technology and Education,Tianjin 300222,China;Tianjin Key Laboratory of Information Sensing and Intelligent Control,Tianjin University of Technology and Education,Tianjin 300222,China;School of Mechanical Engineering,Hebei University of Technology,Tianjin 300401,China)

机构地区：[1]天津职业技术师范大学自动化与电气工程学院,天津300222 [2]天津职业技术师范大学天津市信息传感与智能控制重点实验室,天津300222 [3]河北工业大学机械工程学院,天津300401

出　　处：《天津农业科学》2023年第7期63-70,81,共9页Tianjin Agricultural Sciences

基　　金：天津市研究生科研创新项目(2022SKYZ303);天津市青年项目基金(22JCQNJC01100);市教委科研计划项目(2021KJ014)。

摘　　要：基于多传感器人工嗅觉系统的苹果种类识别方法是将自行研发的便携式硬件和上位机算法相结合,目的是将市面上常见的外形相似的‘花牛’和‘阿克苏’苹果进行无损种类识别,降低检测成本。实施方法是根据实际情况选择传感器并设计电路对待测样本气味信息进行响应,下位机将采集到的信号传输至上位机的机器学习算法中进行模型训练。通过线性判别分析算法(Linear Discriminant Analysis,LDA)、逻辑回归算法(Logistic Regression,LR)、邻近算法(K-NearestNeighbor,KNN)、前馈神经网络算法(Back Propagation,BP)分类模型对气味信息数据进行计算并作出分类。最终得到LDA、LR、KNN、BP算法识别的准确率分别为86.83%、85.33%、91.26%、85.00%,通过stacking框架将以上4种算法模型进行融合,融合后算法识别的准确率最高为97.14%。与传统单预测模型相比,基于多模型融合的苹果识别方法精确度更高。研究结果表明,基于多传感器人工嗅觉系统可以直接通过气味对其种类进行识别,为苹果的无损分类做出有效的判断,可为受主观因素影响的的感官评价提供客观的理论依据。The method of apple variety recognition based on a multi-sensors artificial olfaction system combines a self-developed portable hardware device with a PC-based algorithm.The aim is to achieve non-destructive identification of visually similar apple varieties,such as‘Huaniu'and‘Akane',in the market,thereby reducing detection costs.The implementation method involved selecting sensors based on the actual conditions and designing circuits to respond to the odor information of the tested samples.The lower-level device transmitted the collected signals to the machine learning algorithm implemented on the upper-level PC for model training.The collected odor information data was processed and classified using classification models,including Linear Discriminant Analysis(LDA),Logistic Regression(LR),K-Nearest Neighbor(KNN),and Back Propagation(BP).The classification accuracy of LDA,LR,KNN and BP algorithms was determined to be 86.83%,85.33%,91.26%and 85.00%,respectively.The four algorithm models were then fused using the stacking framework,resulting in the highest accuracy of 97.14%.Compared to traditional single prediction models,the apple recognition method based on multi-model fusion achieved higher accuracy.The research results indicated that the multi-sensors artificial olfaction system can directly identify the variety of apples based on their odor,providing an effective means for non-destructive classification.This method could offer objective criteria for sensory evaluation that may be influenced by subjective factors.

关 键 词：多传感器 种类识别 机器学习 stacking融合算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]