基于树冠面积与苹果数量的果树产量预估方法  

A CNN-LSTM prediction method for apple tree yield based on canopy area and number of apple

作　　者：李会宾 史云[1] 张保辉[1] 李丹丹 Li Huibin;ShiYun;Zhang Baohui;Li Dandan(Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences,Beijing 100081,China;School of Software Engineering,Tongji University,Shanghai,201804,China)

机构地区：[1]中国农业科学院农业资源与农业区划研究所,北京100081 [2]同济大学软件学院,上海201804

出　　处：《中国农业信息》2022年第1期11-18,共8页China Agricultural Informatics

基　　金：中国农业科学院基本科研业务费专项(Y2021XK07)。

摘　　要：【目的】苹果产量的早期预测对于市场供需关系、果农、消费者都有重要的影响,准确预估苹果产量具有重要的理论和应用意义。【方法】文章提出了一种利用卷积神经网络和长短期记忆网络进行苹果产量预测的方法。利用由果树图像数据中获得的果树苹果数量,和由果园无人机图像获取的果树树冠面积作为特征信息,再经过该文提出的卷积—长短期记忆网络,获得每棵果树的预估产量。【结果】该文使用5种评价标准,分别是预测误差概率密度、预测绝对误差、均方误差(MSE)、均方根误差(RMSE)、平均绝对误差(MAE),经过实验验证,该文所述方法的均方根误差RMSE最优可以达到9.16。【结论】该文所述方法依据果树的图像与果园的无人机图像,可以较好地预估苹果产量,为大规模果园产量预测提供技术支撑。[Purpose]Apple is an important cash crop in the world,the early prediction of apple yield has an important impact on the market supply,demand relationship,fruit farmers and consumers.Accurate prediction of apple yield has important theoretical and application significance.[Method]To address this problem,the current research methods mainly include linear modeling,Markov and other traditional methods,and some neural network methods with simple structures.[Result]In this paper,a neural network with convolutional layers and longshort term memory network layers are used to predict apple yield.The number of apples obtained from the fruit tree image data and the tree crown area obtained from the orchard UAV image were used as characteristic information,and then the estimated yield of each fruit tree was obtained through the CNN-LSTM model proposed in this paper.[Conclusion]The experimental results show that the method proposed in this paper can predict apple yield well.

关 键 词：苹果 产量预测 神经网络 图像识别 

分 类 号：S661.1[农业科学—果树学] TP183[农业科学—园艺学] TP391.41[自动化与计算机技术—控制理论与控制工程]