冰川是最重要的淡水储存库之一,精确识别冰川和监测冰川的变化对于了解气候变化和水资源管理具有重要意义。基于Landsat 8影像,以喀喇昆仑区域为研究对象,利用单波段阈值法、雪盖指数法、非监督分类、监督分类和U-Net卷积神经网络提取冰川边界,并以交并比和混淆矩阵对冰川边界提取结果进行精度评定。结果表明,非监督分类和单波段阈值法对于表碛覆盖型冰川以及阴影中冰川存在严重的漏分现象,易将薄雪覆盖的山地错分为冰川,K-means的提取效果最差,交并比为57.69%,Kappa系数为0.57。监督分类方法对于表碛覆盖型冰川的提取效果有明显改善,但对于阴影中的冰川的提取效果不佳,提取结果的Kappa系数均为0.70以上。雪盖指数法可以有效提取阴影中的冰川,但易将大面积冰川中的非冰川区域错分为冰川,交并比为74.49%,Kappa系数为0.76。U-Net卷积神经网络能够较完整地提取冰川边界,精度要明显高于其他分类方法,重叠面积最接近地面真值面积,其交并比为88.57%,Kappa系数为0.90。U-Net卷积神经网络虽然表现较好,但是对于极小面积冰川仍存在漏分,后续研究可通过改进网络结构来提高精度...

Wildfires have caused natural environmental damage that has contributed to deforestation, consequently demonstrating a significant influence on atmospheric emissions. Wildfires occur frequently in South Korea, especially during the spring season. This study assessed post-wildfires areas in Gangneung, South Korea, on April 11, 2023, which were generated by implementing remote sensing technology and statistical analysis. Remote sensing and classification techniques, including PlanetScope, have been developed for identifying wildfire-damaged areas. The method for classifying post-wildfire mapping estimation includes the utilization of deep learning approaches, especially using the U-Net architecture. Therefore, the assessment of wildfire severity can be conducted using Sentinel-2 and Sentinel-5P imagery in addition to an analysis of the vegetation type and air pollutant within the affected region. In the present study, Sentinel-2 imagery was to generate spectral indices, including the differenced normalized burn ratio (dNBR), differenced normalized difference moisture index (dNDMI), differenced soil adjusted vegetation index (dSAVI), and differenced normalized vegetation index (dNDVI). Sentinel-5P imagery was utilized to produce carbon monoxide (CO) column number densities. The estimation of wildfire areas was conducted using a PlanetScope classified image with the U-Net classifier, which was evaluated based on the overall accuracy value of 95% and kappa accuracy of 0.901. The wildfire severity level was shown by dNBR, which was correlated with the parameters, including RBR, dNDMI, dSAVI, dNDVI, and CO. The statistical analysis demonstrated a significant and positive correlation between the wildfire severity and the parameters. Moreover, the average of vegetation indices (NDMI, SAVI, and NDVI) before and after a wildfire were found to decrease by vegetation type, including 17.55% in mixed barren land areas, 17.49% in other grasses, 24.71% in mixed forest land, 22.48% in coniferous land, 13.48% in fields, and 4.29% in paddy fields. On the basis of the results, these estimates can be employed to identify the level of damage caused by wildfires to vegetation and air quality.

冰川变化会对当地的气候环境、水资源环境产生重要影响，随着遥感技术的发展，通过遥感图像进行冰川提取成为相关研究的主要手段，相比于人工目视解释法会出现的耗时长、效率低、主观因素大等问题，深度学习有着一定的优势。该文基于传统U-Net语义分割网络进行冰川分割，但因受限于冰川训练集缺失，真彩色图像在冰川地区进行分割会有较大的干扰，无法凸显冰川的特征，冰川分割效率较低。因此，利用冰川的矢量数据，基于Landsat 8遥感卫星图像，建立成对的假彩色冰川分割训练集，充分利用遥感多波段图像的优势，强化冰川特征信息。同时，通过添加不同波段组合的假彩色图像，丰富冰川的分割信息，并利用Inception v1深度学习模块将两种特征信息进行融合，提升冰川分割的准确性。实验结果表明，所提方法可以有效分割出冰川范围，相比于其他深度学习方法，分割准确性有了一定的提高。