中国农业机械化科学研究院集团有限公司 主管

北京卓众出版有限公司 主办

基于CNN-BiGRU的石榴树叶绿素含量预测方法

Prediction approach of chlorophyll content in pomegranate trees based on CNN-BiGRU

  • 摘要: 叶绿素含量作为深入了解果树生长和健康状态的关键参数,对于指导果园管理决策具有重要作用。然而,在大规模果园环境下,快速且准确地获取整个果园的叶绿素含量数据是一项重大挑战。为此,利用无人机遥感平台结合深度学习算法,提出了一种新的解决方案。通过多光谱无人机采集石榴树冠层RGB及多光谱图像,利用图像处理技术提取RGB图像颜色特征、纹理特征和多光谱图像植被指数等参数并建立不同数据集。在此基础上,结合地面实测叶绿素数据,构建一种结合双向门控循环单元(BiGRU)与卷积神经网络(CNN)的深度融合网络模型CNN-BiGRU,并且与原始CNN和随机森林(RF)进行试验对比。结果表明,组合模型在预测石榴树叶绿素含量方面的效果明显好于其他模型,尤其是在使用特征融合集建模时,其决定系数高达0.9737、均方根误差低至0.8233。该精度满足对石榴树叶绿素含量的精准预测,为大面积果园管理提供实用参考。

     

    Abstract: As a key parameter for an in-depth understanding of growth and health status of fruit trees, chlorophyll content plays an important role in guiding orchard management decisions.However, in a large-scale orchard environment, quickly and accurately obtaining chlorophyll content data for entire orchard is a major challenge.Therefore, a new solution was proposed by using a UAV remote sensing platform combined with deep learning algorithms.RGB and multispectral images of pomegranate tree canopy were collected by multispectral UAV, and an image processing technique was used to extract parameters such as RGB image color features, texture features, and multispectral image vegetation index and establish different datasets.On this basis, combined with ground-measured chlorophyll data, a deep fusion network model CNN-BiGRU combining bidirectional gated recurrent unit(BiGRU)and convolutional neural network(CNN)was constructed and experimentally compared with original CNN and random forest (RF).Experimental results showed that combined model was significantly better than other models in predicting chlorophyll content of pomegranate trees, especially when using feature fusion set modeling, with a determination coefficient as high as 0.9737 and a root mean square error as low as 0.8233.This accuracy met accurate prediction of chlorophyll content of pomegranate trees, providing a practical reference for large-scale orchard management.

     

/

返回文章
返回