Due to the rising costs of farming and an aging agrarian population, large-scale cultivated lands have been abandoned in China's rural areas, especially in remote mountainous areas. Remote sensing technology can quickly extract abandoned farmland and reveal its formation laws, and this is of great significance to food security and to the economical and intensive utilization of cultivated land. Xingning city in Guangdong, China was selected as the research area. Based on Landsat 8 images, Google Earth images, and digital elevation models, we developed an archival classification method, which we used to extract abandoned farmland in Xingning according to the variation of crop life cycles. On this basis, we reveal the spatial distribution characteristics, natural ecology, and production conditions of abandoned farmland, and we propose differentiated countermeasures. The results show that our archival classification method extracted abandoned farmland with 92% accuracy, while the accuracy of the classification supervised by the artificial neural network was 80%. Abandoned farmland on the plains of Xingning is mainly distributed around the town center and the industrial park, while that in the mountainous area is mainly located on ridges and saddles. From the point of view of natural ecology, abandoned farmland in the study area is generally located on the ridge and saddle area with a large slope close to timberland. From the perspective of agricultural production, abandoned farmland is mainly restricted by field roads and irrigation conditions. Most of the abandoned farmland is limited by both natural ecology and production conditions, and so it should be treated by classification. We expect that our results can provide a reference for the extraction of the abandoned farmland and a basis for analyzing its formation mechanism.