 This study compared the performance of various image classification algorithms for the same dataset. The dataset was composed of Landsat Thematic Mapper, TM, images from Guangzhou City, China. The algorithms included both unsupervised and supervised methods, as well as several machine learning algorithms that had become popular in remote sensing during the last 20 years. The authors evaluated each algorithm's performance using two different experiments, one based on pixel level decisions and another based on segmentation. They found that the algorithms generally performed similarly, but some were better able to handle insufficient training samples. Overall, the results suggest that the use of these algorithms can be beneficial for remote sensing applications. This article was authored by Tseng Songli, Jia Wang, Lei Wang, and others.