基于PCA-LSTM小麦叶片水分检测研究

Study on wheat leaf moisture detection based on PCA-LSTM

为实现小麦叶片含水量精准检测,该文提出基于主成分分析(Principal Component Analysis,PCA)和长短时记忆网络(long Short-Term Memory,LSTM)构建预测模型对小麦叶片含水量进行精准检测。试验以济麦22号小麦为研究对象,均匀划分其种植区域并灌溉不同水量以作叶片水分梯度分析;然后采集叶片图像并提取叶片R、G、B颜色特征值,运用灰度梯度共生矩阵提取均值、灰度熵、逆差距、梯度熵、混合熵、惯性、等15个纹理特征值,形状特征值提取面积、宽、伸缩率;最后基于主成分分析对提取特征值进行降维处理,并分别应用BP神经网络、支持向量机(Support Vector Machine,SVM)进行检测对比。试验证明:该模型平均绝对误差、平均绝对百分误差、均方根误差分别为0.002、0.002和0.011,均优于传统预测方法,具有良好的预测性能和泛化能力。由此表明,该方法切实可行,能够满足小麦叶片水分预测的实际需要,为后期研发智能灌溉系统提供坚实理论指导。

In order to accurately detect the moisture content of wheat leaves,this paper proposes a prediction model based on Principal Component Analysis(PCA)and Long Short-Term Memory(LSTM)to accurately determine the moisture content of wheat leaves.The Jimai 22 wheat was used as the research object,which the planting area was evenly divided and the water content was irrigated for leaf water gradient analysis.Then the leaf image was extracted to extract the R,G and B color eigenvalues,and the gray gradient co-occurrence matrix which was used to extract 15 texture eigenvalues,such as mean,gray entropy,gradient entropy,mixed entropy,inertia and inverse gap.The inverse gap,extracted area,width and expansion ratio was extracted as the shape feature value.Finally,the extracted feature values are reduced in dimension based on principal component analysis,and BP neural network and Support Vector Machine(SVM)were used for detection and comparison.The experimental results show that the average relative error,root mean square error and average absolute error of the model are 0.002,0.002 and 0.011,respectively,which are better than the traditional prediction methods,and have good prediction performance and generalization ability.This indicates that the method is feasible and can meet the actual needs of wheat leaf moisture prediction,and provides solid theoretical guidance for the later development of intelligent irrigation systems.