摘要
本研究采集夏玉米拔节、抽雄、灌浆和成熟4个时期的无人机可见光和多光谱影像,提取和筛选植被指数与纹理特征参数,构建植被指数与纹理特征融合变量,采用反向传播神经网络、随机森林和支持向量机3种机器学习方法构建玉米产量预测模型。结果表明:相较于单一类型参数,融合植被指数和纹理特征进行产量预测模型精度更高;3种机器学习方法中以随机森林构建的玉米产量预测模型效果最好,且最佳预测时期为灌浆期(籽粒水泡期);综合评价建模和验证结果,基于多光谱影像植被指数与纹理特征融合变量和随机森林方法构建的模型玉米产量预测效果最佳。遥感信息的多特征融合与机器学习方法的搭配能够挖掘和利用更多信息并提高玉米产量预测的精度和鲁棒性。
This study investigates the integration of spectral and texture features for pridicting maize yield using remote sensing techniques.Visible light and multispectral images captured by drone were obtained at four key growth stages of summer maize:jointing,tasseling,filling,and maturity.Vegetation indexes and texture feature parameters were extracted and used to formulate spectral and texture feature fusion variables.Three machine learning methods(a back propagation neural network,a random forest,and a support vector machine)were employed to develop predictive models for predicting maize yield.The results showed that the accuracy of maize yield prediction models incorporating both spectral and texture feature fusion was superior to that of models using only spectral or texture features.Among the three machine learning methods,the random forest yielded the best maize prediction model.Among the four growth stages,the model for the filling stage(blister)had the highest accuracy.An overall evaluation of the results showed that the model obtained using the fused vegetation indexes and texture feature variables extracted from the multispectral images using the random forest method predicted maize yield with the highest accuracy.The fusion of multiple remote sensing features and the application of machine learning methods allow the extraction and utilization of more information,thereby improving the accuracy and robustness of yield prediction.
作者
马元花
汪乐印
张祯鑫
郑大圣
叶玉澜
崔志峰
杜冰笑
寸玉洁
李军
王瑞
MA Yuanhua;WANG Leyin;ZHANG Zhenxin;ZHENG Dasheng;YE Yulan;CUI Zhifeng;DU Bingxiao;CUN Yujie;LI Jun;WANG Rui(College of Agronomy,Northwest A&F University,Yangling Shaanxi 712100,China)
出处
《西北农业学报》
CAS
CSCD
北大核心
2024年第10期1827-1838,共12页
Acta Agriculturae Boreali-occidentalis Sinica
基金
陕西省自然科学基础研究计划(2023-JC-QN-0192)。
关键词
玉米
无人机遥感
可见光
多光谱
植被指数
纹理特征
产量
Maize
UAV remote sensing
RGB
Multi-spectral
Vegetation index
Texture features
Yield
