基于改进人工神经网络的植物叶面积测定

叶面积作为植物光合作用的重要指标,是研究作物及林木生产力的基础.采用L-M算法和贝叶斯规则相结合的网络训练模式,以毛竹叶面积为研究对象,综合优化其人工神经网络结构,构建最优的叶面积预测模型.研究结果显示,模型的最佳预测变量为叶片宽度和叶片长度变量组合,而增加叶片形状指数未提高叶面积预测模型精度;所建神经网络模型性能好、预测精度高,决定系数达0.992,平均相对预测误差为4.28％,可以准确估测毛竹叶面积.

Improved Artificial Neural Network for Determination of Plant Leaf Area

Leaf area is an essential indicator of photosynthesis for the study of crop and forest productivity. The Levenberg-Marquardt back-propagation optimization algorithm was coupled with Bayesian regulation to train the artificial neural network (ANN), and the predictive model was developed to determinate rapidly and accurately Moso bamboo leaf area. The results showed that the best input variables were the combination of leaf width and leaf length for ANN model, whereas the leaf shape index did not significantly affect the variability of leaf area. The optimization ANN model possessed with excellent performance and predictable accuracy, with the high determination coefficient of 0.992 and mean relative prediction error of 4.28%. The ANN model would be allowed for estimating accuracy the leaf area of Moso bamboo.