Comparison of machine learning classifiers using different satellite images for land use land cover mapping

Knowledge of land-use land-cover (LULC) change is essential in a number of fields based on the use of Earth observations. Accurate land use and cover data are essential for effective land-use planning. With the advancement of remote-sensing (RS) techniques satellites provide data at various spatial and temporal scales. Recently, the application of machine-learning algorithms on remotely-sensed images for LULC mapping has been attracting considerable attention. The comparison show a few studies that have been carried out to identify the best suited and accurate algorithm among used machine-learning classifiers for LULC mapping. It has been found that RF generally provide better accuracy for the LULC classification compared to other machine-learning techniques. However, the sensor characteristics (spatial and temporal resolution) processing software, machine-learning model set-up, training samples and input parameters also determine the accuracy of LULC classification.

土地利用与土地覆被（Land-Use Land-Cover, LULC）变化的相关认知，在诸多依托地球观测技术的研究领域中至关重要。精准的土地利用与覆被数据，是开展高效土地利用规划的必要前提。随着遥感（Remote-Sensing, RS）技术的进步，卫星能够提供不同空间与时间尺度的观测数据。近年来，将机器学习算法应用于遥感影像以开展LULC制图的研究，受到了广泛关注。已有多项对比研究旨在从已应用的机器学习分类器中，筛选出最适配且精度最优的算法以用于LULC制图。现有研究表明，相较于其他机器学习技术，随机森林（Random Forest, RF）在LULC分类任务中通常能取得更优的分类精度。但传感器特性（空间与时间分辨率）、处理软件、机器学习模型搭建、训练样本以及输入参数等因素，同样会决定LULC分类的精度。