农业工程学报
農業工程學報
농업공정학보
2014年
15期
206-213
,共8页
优化%数据处理%监测%生长特征%特征优化%局部保持投影%二维主成分分析
優化%數據處理%鑑測%生長特徵%特徵優化%跼部保持投影%二維主成分分析
우화%수거처리%감측%생장특정%특정우화%국부보지투영%이유주성분분석
optimization%data processing%monitoring%growth features%feature optimization%locality preserving projection%two-dimensional principal component analysis
由于现有的用于农业作物生长监测数据的特征优化方法-局部保持投影（locality preserving projection, LPP）只保留局部信息，同时存在未考虑样本类别信息导致特征提取时误分类，准确率与数据优化效率并不理想。针对上述问题，提出了改进型 LPP 方法，并将其用于作物生长特征的优化。首先将样本利用二维主成分分析（two-dimensional principal component analysis,2DPCA）进行初步降维，保留原样本数据中的整体空间信息；然后提出优化的2类子图-聚集子图和分离子图，用来描述不同类别数据之间的关联信息；然后提出优化的2类子图对不同类别数据间的远近关系进行描述；最后采用改进型LPP算法，将数据进一步投影到低维空间，提取样本的局部信息，完成样本特征优化。试验分析表明，改进型LPP具有很好的适应性，最高支持向量机（support vector machine, SVM）分类准确率能够达到96%以上，使精度达到最高的最优维数比主成分分析（principal component analysis, PCA）和二维主成分分析2种算法降低了25%以上，同时算法运行效率比PCA与2DPCA算法提升32.4%与8.3%，整体性能比基本LPP算法更为优越，能够适应农作物多维数据的优化处理。研究结果为现代精准农业信息监测过程中的数据处理与分析提供了参考。
由于現有的用于農業作物生長鑑測數據的特徵優化方法-跼部保持投影（locality preserving projection, LPP）隻保留跼部信息，同時存在未攷慮樣本類彆信息導緻特徵提取時誤分類，準確率與數據優化效率併不理想。針對上述問題，提齣瞭改進型 LPP 方法，併將其用于作物生長特徵的優化。首先將樣本利用二維主成分分析（two-dimensional principal component analysis,2DPCA）進行初步降維，保留原樣本數據中的整體空間信息；然後提齣優化的2類子圖-聚集子圖和分離子圖，用來描述不同類彆數據之間的關聯信息；然後提齣優化的2類子圖對不同類彆數據間的遠近關繫進行描述；最後採用改進型LPP算法，將數據進一步投影到低維空間，提取樣本的跼部信息，完成樣本特徵優化。試驗分析錶明，改進型LPP具有很好的適應性，最高支持嚮量機（support vector machine, SVM）分類準確率能夠達到96%以上，使精度達到最高的最優維數比主成分分析（principal component analysis, PCA）和二維主成分分析2種算法降低瞭25%以上，同時算法運行效率比PCA與2DPCA算法提升32.4%與8.3%，整體性能比基本LPP算法更為優越，能夠適應農作物多維數據的優化處理。研究結果為現代精準農業信息鑑測過程中的數據處理與分析提供瞭參攷。
유우현유적용우농업작물생장감측수거적특정우화방법-국부보지투영（locality preserving projection, LPP）지보류국부신식，동시존재미고필양본유별신식도치특정제취시오분류，준학솔여수거우화효솔병불이상。침대상술문제，제출료개진형 LPP 방법，병장기용우작물생장특정적우화。수선장양본이용이유주성분분석（two-dimensional principal component analysis,2DPCA）진행초보강유，보류원양본수거중적정체공간신식；연후제출우화적2류자도-취집자도화분리자도，용래묘술불동유별수거지간적관련신식；연후제출우화적2류자도대불동유별수거간적원근관계진행묘술；최후채용개진형LPP산법，장수거진일보투영도저유공간，제취양본적국부신식，완성양본특정우화。시험분석표명，개진형LPP구유흔호적괄응성，최고지지향량궤（support vector machine, SVM）분류준학솔능구체도96%이상，사정도체도최고적최우유수비주성분분석（principal component analysis, PCA）화이유주성분분석2충산법강저료25%이상，동시산법운행효솔비PCA여2DPCA산법제승32.4%여8.3%，정체성능비기본LPP산법경위우월，능구괄응농작물다유수거적우화처리。연구결과위현대정준농업신식감측과정중적수거처리여분석제공료삼고。
Nowadays, the evaluation for crop growth is based on various growth characteristics, which often brings a huge amount of information processing. Furthermore, the complex information can not directly reflect some key features of crops. Thus, the feature extraction and optimization plays an important role in the process. In this paper, the locality preserving projection (LPP) is used to achieve the dimensionality reduction of high dimensional data while keeping the invariance of its internal local structure. After being projected via the algorithm, the adjacent sample is able to maintain the original neighboring state while the original distant samples don’t keep the old state. Obviously, this result is not satisfactory for data optimization. In order to strengthen the effect of category separation, firstly, the dimension of sample data is preliminary reduced by using two-dimensional principal component analysis (2DPCA) to retain the spatial information. Secondly, two sub-graphs (gathered sub-graph and separated sub-graph) instead of original nearest neighbor graph are used to describe the relationship between homogeneous and heterogeneous data. The gathered sub-graph is improved on the basis of K-nearest neighbor graph. The addition of category information makes the homogeneous non-adjacent samples stay closer to each other after projection. The separated sub-graph is constructed to solve the problem that the application of K-nearest neighbor graph may reduce the accuracy of classification when the data are projected into low-dimensional space. Then the optimized global matrix and the improved objective function are provided to design the complete optimization method for feature extraction. Through the above steps, the category information of sample data is added for LPP algorithm. Finally, the feature parameters set are obtained by improved LPP algorithm to extract local information of samples. The data of crop growth features are further projected to low-dimensional space. The final extracted information is able to replace the original sample data without losing the data which can reflect the key information of sample set. In order to evaluate the performance of improved LPP algorithm to achieve dimensionality reduction and optimizing for crop growth characteristics, a set of data from cabbage was chosen as test sample. In the process of dimensionality reduction from 30 to 10 using different algorithms (PCA, 2DPCA, LPP and improved LPP), the improved LPP has higher overall performance with less running time, which is only longer than Basic LPP algorithm. By analyzing the performance of improved LPP algorithm for dimensionality reduction, the data of some cabbage and lettuce were chosen as test data. The contrast experiments using different algorithms (PCA, 2DPCA, LPP and improved LPP) for dimensionality reduction were carried out, and all the test data in the database achieved dimensionality reduction via the above-mentioned algorithms. Meanwhile, it accomplished data classification by SVM after accomplishing dimensionality reduction. The experiments show that the improved LPP algorithm has better adaptability, and the highest SVM classification accuracy rate of this method can reach up to 96%. Compared with other methods, the improved LPP has superior performances in terms of multidimensional data analysis and optimization. The method has good prospects, and is able to meet the demands for the information perception of new agriculture as well as the optimization of crop growth characteristic parameters.
