农业工程学报
農業工程學報
농업공정학보
Transactions of the Chinese Society of Agricultural Engineering
2015年
18期
240-246
,共7页
徐赛%陆华忠%周志艳%吕恩利%杨径
徐賽%陸華忠%週誌豔%呂恩利%楊徑
서새%륙화충%주지염%려은리%양경
无损检测%水果%模型%电子鼻%成熟阶段%模糊C均值聚类%k最近邻%神经网络
無損檢測%水果%模型%電子鼻%成熟階段%模糊C均值聚類%k最近鄰%神經網絡
무손검측%수과%모형%전자비%성숙계단%모호C균치취류%k최근린%신경망락
nondestructive examination%fruits%models%electronic nose%mature stage%fuzzy C-means%k-nearest neighbor%probabilistic neural network
为了无损快速监测荔枝成熟阶段,该文提出了一种基于电子鼻技术的果园荔枝成熟阶段监测方法,采用 PEN3电子鼻获取挂果约25 d到果实成熟过程中6个成熟阶段荔枝样本的仿生嗅觉信息并同步获取了各成熟阶段荔枝的3项物理特征(果实直径、果实质量与果实可溶性固形物含量)。根据不同成熟阶段荔枝物理特征变化可知,荔枝果实直径与果实质量2项物理指标在挂果约32 d~39 d,以及53 d~60 d增长较快,可溶性固形物含量在挂果约32 d前无法测量,53 d~60 d阶段增长速度较慢。提取各样本电子鼻采样数据75 s时刻的各传感器响应值作为特征值后,采用载荷分析(loadings)进行传感器阵列优化,优选了传感器R2、R4、R6、R7、R8、R9和R10的响应数据进行后续分析。将优化后的传感器响应数据进行归一化处理。采用线性判别分析(linear discriminant analysis,LDA)进一步提取特征信息,降低数据中包含的冗余信息。LDA对荔枝成熟阶段的分类识别效果不佳。为进一步探究电子鼻监测果园荔枝成熟阶段的可行性,采用模糊C均值聚类分析(fuzzy C means clustering,FCM)、k最近邻函数分析(k nearest neighbor,KNN)和概率神经网络(probabilistic neural network,PNN)进行模式识别。研究结果表明,FCM对果园荔枝成熟阶段识别的正确率为89.17%。采用KNN与PNN建立识别模型后,KNN与PNN识别模型对训练集的回判正确率均为100%,对测试集的识别率均为96.67%,具有较好的分类识别效果。试验证明了采用电子鼻进行果园荔枝成熟度监测的可行性,为果园水果品质的实时监测提供参考。
為瞭無損快速鑑測荔枝成熟階段,該文提齣瞭一種基于電子鼻技術的果園荔枝成熟階段鑑測方法,採用 PEN3電子鼻穫取掛果約25 d到果實成熟過程中6箇成熟階段荔枝樣本的倣生嗅覺信息併同步穫取瞭各成熟階段荔枝的3項物理特徵(果實直徑、果實質量與果實可溶性固形物含量)。根據不同成熟階段荔枝物理特徵變化可知,荔枝果實直徑與果實質量2項物理指標在掛果約32 d~39 d,以及53 d~60 d增長較快,可溶性固形物含量在掛果約32 d前無法測量,53 d~60 d階段增長速度較慢。提取各樣本電子鼻採樣數據75 s時刻的各傳感器響應值作為特徵值後,採用載荷分析(loadings)進行傳感器陣列優化,優選瞭傳感器R2、R4、R6、R7、R8、R9和R10的響應數據進行後續分析。將優化後的傳感器響應數據進行歸一化處理。採用線性判彆分析(linear discriminant analysis,LDA)進一步提取特徵信息,降低數據中包含的冗餘信息。LDA對荔枝成熟階段的分類識彆效果不佳。為進一步探究電子鼻鑑測果園荔枝成熟階段的可行性,採用模糊C均值聚類分析(fuzzy C means clustering,FCM)、k最近鄰函數分析(k nearest neighbor,KNN)和概率神經網絡(probabilistic neural network,PNN)進行模式識彆。研究結果錶明,FCM對果園荔枝成熟階段識彆的正確率為89.17%。採用KNN與PNN建立識彆模型後,KNN與PNN識彆模型對訓練集的迴判正確率均為100%,對測試集的識彆率均為96.67%,具有較好的分類識彆效果。試驗證明瞭採用電子鼻進行果園荔枝成熟度鑑測的可行性,為果園水果品質的實時鑑測提供參攷。
위료무손쾌속감측려지성숙계단,해문제출료일충기우전자비기술적과완려지성숙계단감측방법,채용 PEN3전자비획취괘과약25 d도과실성숙과정중6개성숙계단려지양본적방생후각신식병동보획취료각성숙계단려지적3항물리특정(과실직경、과실질량여과실가용성고형물함량)。근거불동성숙계단려지물리특정변화가지,려지과실직경여과실질량2항물리지표재괘과약32 d~39 d,이급53 d~60 d증장교쾌,가용성고형물함량재괘과약32 d전무법측량,53 d~60 d계단증장속도교만。제취각양본전자비채양수거75 s시각적각전감기향응치작위특정치후,채용재하분석(loadings)진행전감기진렬우화,우선료전감기R2、R4、R6、R7、R8、R9화R10적향응수거진행후속분석。장우화후적전감기향응수거진행귀일화처리。채용선성판별분석(linear discriminant analysis,LDA)진일보제취특정신식,강저수거중포함적용여신식。LDA대려지성숙계단적분류식별효과불가。위진일보탐구전자비감측과완려지성숙계단적가행성,채용모호C균치취류분석(fuzzy C means clustering,FCM)、k최근린함수분석(k nearest neighbor,KNN)화개솔신경망락(probabilistic neural network,PNN)진행모식식별。연구결과표명,FCM대과완려지성숙계단식별적정학솔위89.17%。채용KNN여PNN건립식별모형후,KNN여PNN식별모형대훈련집적회판정학솔균위100%,대측시집적식별솔균위96.67%,구유교호적분류식별효과。시험증명료채용전자비진행과완려지성숙도감측적가행성,위과완수과품질적실시감측제공삼고。
Mature stage monitoring can provide significant scientific instruction for the management of litchi orchard. However, nowadays, any research based on mature stage monitoring in orchard has not been reported yet. Given that this paper proposed a monitoring method of litchi orchard mature stage based on electronic nose. We used electronic nose (PEN3) to sample litchis which were in 6 different mature stages (s1, s2, s3, s4, s5 and s6) from about 25 days after it fruited to maturity, and measured 3 physical characteristics of litchi fruits (fruits’ size, fruits’ weight and fruits’ soluble solid content). According to the changes of litchi’s physical characteristics in different mature stages, the 2 physical indices (fruit size and weight) of litchi from the 30th to the 39th day and from the 53th to the 60th day after it fruited were increasing comparatively faster than other stages. That was to say, the litchi fruit normally grew fast in the 2 periods. In addition, the soluble solid content of litchi grew slowly from the 53th to 60th day after it fruited and could not be tested before the 32th day after it fruited. After extracting each sensor’s response value in stable time (75 s), we used loading analysis (Loadings) for sensors optimization, and kept sensors (R2, R4, R6, R7, R8, R9 and R10) for the next analysis. Loadings results also showed that R7, R4 and R6 were comparatively more sensitive than other sensors when identifying the volatile of litchi, which provided a reference for the next research when exploring especial instrument for litchi quality detection based on bionic olfaction mechanism. Then, unitary processing was used for the noise reduction of the sensor’s response value. At last, we used linear discriminant analysis (LDA) for further extraction of feature information to decrease the redundant information. In addition, LDA could not detect the mature stage of litchi in orchard effectively. LDA classification results showed that the sample points in s2 and s3 were overlapped by each other, which had poor classification effect. The sample points in s5 and s6 were not overlapped by each other, but the distance between them was close, which may easily cause the confusion in practical monitoring of fruit mature stage. For further research the feasibility of electronic nose application for litchi mature stage monitoring in orchard, fuzzy c means clustering (FCM) method, k-nearest neighbor (KNN) method and probabilistic neural network (PNN) method were used for pattern recognition. The experimental results showed that the accuracy of FCM for litchi mature stage monitoring in orchard was 89.17%. The classification effects of s2 and s3 were undesirable, and the mature stages s5 could not be absolutely distinguished from s6. After building up KNN and PNN detection model, their accuracies of training set were all 100%, and their accuracies of test set were both 96.67%, which had good effect for litchi mature stage monitoring in orchard. By comparing electronic nose analysis results with physical characteristics changes, we could infer that the accumulation speed of litchi’s inner compositions had inverse correlation with the size growing speed of litchi fruit. That meant when the size of litchi fruit grew faster, the accumulation speed of litchi’s inner compositions was slower. Otherwise, the accumulation speed of litchi’s inner compositions was faster, and the classification effect was better. This research proves the feasibility of using electronic nose for litchi mature stage monitoring in orchard, and provides the reference for fruit quality and situation monitoring in orchard in the future.