| 摘要: |
| 针对传统水稻倒伏监测方法以人工进行现场测量耗时耗力且受主观影响较大的问题，利用无人机成本低廉、操作简单以及分辨率高的优势，以黑龙江省佳木斯市七星农场水稻种植基地的水稻倒伏区域为研究对象，对无人机遥感图像结合改进DeepLabV3+模型的水稻倒伏识别方法进行研究。结果表明:1)与其他方法相比，改进DeepLabV3+网络模型取得了更高的准确率和更快的识别速度；2)改进DeepLabV3+网络模型对水稻倒伏图像测试集的准确率为0.99，Kappa系数为0.98，像素准确率0.99；召回率0.99；平衡F分数为0.99；水稻完全倒伏状态识别的交并比为0.96，3种水稻不同倒伏状态识别的平均交并比为0.97。无人机搭载RGB相机载荷平台拍摄遥感图像结合改进DeepLabV3+深度学习模型可以精确地对水稻倒伏进行识别，为大面积、高效率、低成本水稻倒伏监测识别研究提供了一种方法。 |
| 关键词: 水稻 倒伏 无人机遥感 深度学习 |
| DOI：10.11841/j.issn.1007-4333.2022.02.14 |
| 投稿时间：2021-06-21 |
| 基金项目:中央高校基本科研业务费项目(2572018BH02) |
|
| Rice lodging recognition method based on UAV remote sensing combined with the improved DeepLabV3+ model |
|
MU Taoyang,ZHAO Wei,HU Xiaoyu,LI Dan
|
| (College of Information and Computer Engineering, Northeast Forestry University, Harbin 150040, China) |
| Abstract: |
| In response to the problem of traditional rice lodging monitoring methods are time-consuming, labor-intensive and subjectively affected by manual field measurement, the advantages of low cost, simple operation and high resolution of drones are used. The rice planting base of Qixing Farm in Jiamusi City, Heilongjiang Province is taken as the research object, and the rice lodging recognition method of the UAV remote sensing image combined with the improved DeepLabV3+ model is studied. The results show that: 1)Compared with other methods, the improved DeepLabV3+ network model achieves higher accuracy and faster recognition speed in this research method. 2)The improved DeepLabV3+ network model has an accuracy rate of 0. 99 on the test set of rice lodging images, the Kappa coefficient is 0. 98, the pixel accuracy rate is 0. 99, the recall rate is 0. 99; the balanced F score is 0. 99, the cross-combination ratio for the recognition of the complete lodging state of rice is 0. 96, and the average cross-combination ratio for the recognition of the different lodging states of the three rice varieties is 0. 97. In conclusion, the UAV equipped with RGB camera load platform to take remote sensing images combined with the improved DeepLabV3+ deep learning model can accurately identify rice lodging, providing a method for large-area, high-efficiency, low-cost rice lodging monitoring and recognition research. |
| Key words: rice lodging UAV deep learning |
