Sun Jun, Wu Xiao-Hong, Zhang Xiao-Dong, Gao Hong-Yan
Jiangsu Provincial Key Laboratory of Modern Agricultural Equipment and Technology, Jiangsu University, Zhenjiang 212013, China.
Guang Pu Xue Yu Guang Pu Fen Xi. 2013 Feb;33(2):522-6.
In order to conduct rational management of watering lettuce, the model of detecting lettuce leaves' moisture was built. First of all, the hyperspectral images of lettuce leaves were acquired and simultaneously the moisture proportions of leaves were measured. Meanwhile, hyperspectral images were analyzed and the characteristic bands of lettuce leaves' moisture were found. Then the images in characteristic bands were processed and the image features of lettuce leaves' moisture were computed. The image features highly relevant to moisture were obtained through correlation analysis. Furthermore, due to the possible correlation among image features, the principal components of the images were extracted by principal components analysis and were used as BP neural network's inputs to establish PCA-ANN model. At the same time, other models were constructed by using BP neural network and traditional MLR (multiple liner regression) method respectively. Prediction examinations of the three models were made based on the same sample data. The experimental results show that the average prediction error of PCA-ANN prediction model of tillering stage reaches 9.323% which is improved compared with BP-ANN and MLR prediction models.
为了对生菜灌溉进行合理管理，构建了生菜叶片水分检测模型。首先，采集生菜叶片的高光谱图像，并同步测量叶片的水分比例。同时，对高光谱图像进行分析，找出生菜叶片水分的特征波段。然后对特征波段的图像进行处理，计算出生菜叶片水分的图像特征。通过相关性分析得到与水分高度相关的图像特征。此外，由于图像特征之间可能存在相关性，利用主成分分析法提取图像的主成分，并将其作为BP神经网络的输入，建立PCA - ANN模型。同时，分别采用BP神经网络和传统的多元线性回归（MLR）方法构建其他模型。基于相同的样本数据对这三种模型进行预测检验。实验结果表明，分蘖期PCA - ANN预测模型的平均预测误差为9.323%，与BP - ANN和MLR预测模型相比有了改进。
