基于深度学习的农业大棚检测

Agricultural Greenhouse Detection Method Based on Deep Learning

农业大棚的即时空间分布信息能为相关农业部门提供农业管理和资源分配的依据,具有重要意义。针对这一需求,对比分析了深度学习模型FCN,UNet和HRNet在高分辨率遥感影像大棚检测任务上的性能,对FCN,UNet和HRNet模型在大棚检测方面的效果进行定量和定性分析。通过对比发现UNet等网络模型正确率均可达到90%以上,交并比IOU达到85%以上,其中HRNet的效果最好,正确率和IOU分别为92.79%和87.32%。实验表明基于深度学习的大棚检测方法可为快速精确获取大棚分布信息提供技术支撑,具有实用价值。

It’s of great significance that real-time spatial distribution information of agricultural greenhouses can provide a basis for agricultural management and resource allocation for relevant agricultural departments.In view of this demand,we compared and analyzed the performance of various deep learning models FCN,UNet and HRNet in greenhouse detection task with high spatial resolution images,giving quantitative and qualitative analysis on the effects of various models in greenhouse detection.It is found that the accuracy of these models can reach more than 90%and the intersection ratio of IoU more than 85%,among which HRNet has the best result with 92.79%accuracy and 87.32%IoU.The experiment shows that the greenhouse detection method based on deep learning can provide technical support and practical value for quickly and accurately acquiring greenhouse distribution information.