| 引用本文: | 马世骄,房城泰,赵经华,等.基于无人机遥感的春玉米产量预测研究[J].灌溉排水学报,2025,44(1):43-49. |
| MA Shijiao,FANG Chengtai,ZHAO Jinghua,et al.基于无人机遥感的春玉米产量预测研究[J].灌溉排水学报,2025,44(1):43-49. |
|
| 摘要: |
| 【目的】适应现代农业发展作物产量快速预测的需求。【方法】以春玉米为研究对象，设置50%ETc（W1）、75%ETc（W2）、100%ETc（W3）、125%ETc（W4）、150%ETc（W5）5个水分梯度处理试验，基于无人机多光谱影像进行植被指数的构建，并利用皮尔逊相关系数法筛选模型输入变量。采用偏最小二乘法（Partial least squares，PLS）、随机森林回归（Random forest regression，RF）和粒子群算法（Particle swarm optimization，PSO）优化随机森林模型，分别对拔节期、抽雄期和灌浆期春玉米进行基于单一植被指数和多植被指数组合的产量估测，结合模型精度评价指标，最终确定研究区春玉米产量分布图。【结果】以多植被指数组合为输入变量的模型精度总体高于以单一植被指数为输入变量的模型精度，且PSO-RF模型在抽雄期的拟合效果最好。基于单一植被指数NDVI的PSO-RF模型验证集的R2为0.685，RMSE为1 792.71 kg/hm2，RPD为1.764。基于多植被组合的PSO-RF模型验证集的R2为0.806，RMSE为1 485.88 kg/hm2，RPD为2.032。研究区W3处理春玉米产量最高，平均产量为19 845.25 kg/hm2，W1产量处理最低，平均产量为12 054.52 kg/hm2。【结论】综上所述，多光谱无人机能够准确预测作物产量，可为实现精准农业提供技术支持。 |
| 关键词: 无人机；春玉米；产量；植被指数；抽雄期；随机森林回归 |
| DOI：10.13522/j.cnki.ggps.2024218 |
| 分类号: |
| 基金项目: |
|
| Predicting spring maize yield using UAV remote sensing |
|
MA Shijiao, FANG Chengtai, ZHAO Jinghua, LIU Feng, YANG Tingrui, YUAN Ruxin
|
|
1. College of Water Conservancy and Civil Engineering, Xinjiang Agricultural University, Urumqi 830052, China
2. Xinjiang Key Laboratory of Hydraulic Engineering Security and Water Disasters Prevention, Urumqi 830052, China
3. Corps Soil and Water Conservation and Water Resources Development Center, Urumqi 830002, China
|
| Abstract: |
| 【Objective】Predicting potential crop yield is critical not only to policy making but also to improving agricultural management to safeguard food production. In this paper, we investigate the feasibility of using unmanned aerial vehicles to predict crop yields.【Method】Spring maize was used as the model plant. The field experiment consisted of five soil moisture treatments by irrigating 50% (W1), 75% (W2), 100% (W3), 125% (W4) and 150% (W5) of estimated evapotranspiration. Multispectral images captured by an unmanned aerial vehicle (UAV) were used to construct the vegetation indices, and the pearson’s correlation coefficient method was used to screen the input variables for the yield prediction models. Partial least squares (PLS), random forest regression (RF), and particle swarm optimization (PSO) were used to develop and optimize yield prediction models based on the UAV images obtained at the nodulation, tasseling and filling stages of the maize growth. The grain yield was predicted using both single vegetation index and multi-vegetation indices, with the most accurate model used to generate a yield map of the studied area.【Result】Models using multiple vegetation indices were more accurate than models using single vegetation index. Among all models, the PSO-optimized random forest (PSO-RF) model was most accurate during the tasseling stage. Compared with the measured data, the PSO-RF model based on NDVI achieved an R2 of 0.685, RMSE of 1 792.71 kg/hm2, and RPD of 1.764, when using single index, while when using multi-indices, it achieved an R2 of 0.806, RMSE of 1 485.88 kg/hm2, and RPD of 2.032. The W3 produced the highest grain yield (19 845.25 kg/hm2), while the W1 resulted in the least yield (12 054.52 kg/hm2).【Conclusion】Multispectral UAV imagery combined with PSO-optimized random forest models offers a robust and accurate method for crop yield prediction. This method can serve as a valuable tool to improve agricultural management and optimize resource allocation. |
| Key words: drone; spring corn; yield; vegetation indices; male extraction period; random forest regression |
