[论文解读] Landslide Segmentation with U-Net: Evaluating Different Sampling Methods and Patch Sizes

Lucas Pedrosa Soares, Helen Cristina Dias+1 more|arXiv (Cornell University)|Jul 13, 2020

Landslides and related hazards参考文献 29被引用 35

一句话总结

本研究评估补丁大小、采样方法和输入数据集如何影响使用 U-net、RapidEye 图像与 ALOS PALSAR 导出的 DEM 的滑坡分割性能。随机采样和 DEM 增强提高泛化和准确性。

ABSTRACT

Landslide inventory maps are crucial to validate predictive landslide models; however, since most mapping methods rely on visual interpretation or expert knowledge, detailed inventory maps are still lacking. This study used a fully convolutional deep learning model named U-net to automatically segment landslides in the city of Nova Friburgo, located in the mountainous range of Rio de Janeiro, southeastern Brazil. The objective was to evaluate the impact of patch sizes, sampling methods, and datasets on the overall accuracy of the models. The training data used the optical information from RapidEye satellite, and a digital elevation model (DEM) derived from the L-band sensor of the ALOS satellite. The data was sampled using random and regular grid methods and patched in three sizes (32x32, 64x64, and 128x128 pixels). The models were evaluated on two areas with precision, recall, f1-score, and mean intersect over union (mIoU) metrics. The results show that the models trained with 32x32 tiles tend to have higher recall values due to higher true positive rates; however, they misclassify more background areas as landslides (false positives). Models trained with 128x128 tiles usually achieve higher precision values because they make less false positive errors. In both test areas, DEM and augmentation increased the accuracy of the models. Random sampling helped in model generalization. Models trained with 128x128 random tiles from the data that used the RapidEye image, DEM information, and augmentation achieved the highest f1-score, 0.55 in test area one, and 0.58 in test area two. The results achieved in this study are comparable to other fully convolutional models found in the literature, increasing the knowledge in the area.

研究动机与目标

• 评估补丁大小（32x32、64x64、128x128）对滑坡分割精度的影响。
• 评估采样方法（随机 vs 规则网格）对模型泛化的影响。
• 测试数据增强和以 DEM 作为输入通道对 U-net 性能的影响。
• 在两个测试区域中，使用标准分割指标比较模型结果。

提出的方法

• 使用 RapidEye 图像和 ALOS PALSAR DEM 作为输入训练 288 个 U-net 模型。
• 以三种尺寸（32x32、64x64、128x128）对补丁输入进行采样，方法为随机或规则网格。
• 使用 Albumentations 进行数据增强（旋转和翻转）。
• 在两个 1024x1024 的测试区域上评估模型，指标包括精确度、召回率、F1-score 和 mIoU。
• 使用二元交叉熵损失和 Adam 优化器；训练 200 轮，并基于验证损失进行早期保存。

实验结果

研究问题

• RQ1 补丁大小和采样方法如何影响 U-net 的滑坡分割精度？
• RQ2 相比仅使用光学数据，添加 DEM 通道是否能提高模型性能？
• RQ3 数据增强是否在不同地物覆盖环境中提升模型泛化能力？
• RQ4 在测试区域中，数据集、采样方法和补丁大小的哪一组合能在精确度、召回率和 IoU 之间取得最佳平衡？

主要发现

• 128x128 随机图块，结合 RapidEye+DEM+增强，在 F1-score 上达到最高：0.55（测试区域 1）和 0.58（测试区域 2）。
• 使用 32x32 图块的模型在召回率上更高，但由于更多假阳性导致精确度较低。
• 64x64 图块在泛化方面的结果介于 32x32 和 128x128 之间。
• DEM 作为输入通道在所有测试区域提高了精度。
• 在两个测试区域，随机采样在平均 mIoU 上优于规则网格采样。