Studies on High-Resolution Remote Sensing Sugarcane Field Extraction based on Deep Learning

Sugarcane is one of the most important economic crops in Guangxi. For a long time, the sugarcane cultivated areas were estimated via sampling data statistics, while effective and accurate dynamic monitoring data keep absent. High spatial resolution is one of the advantages of high-resolution remote sensing images, through which the texture of sugarcane fields is found clear and unique; however, effective and accurate methods are lacking extracting them automatically in the past. In this paper, a novel deep learning method for sugarcane field extraction from high-resolution remote sensing images is proposed based on DeepLab V3+. It consists of blocks for multi-temporal remote sensing images fusion, which increases the ability of DCNN temporal factors processing. The experiment shows 94.32% extraction accuracy of sugarcane field. Also, its processing speed is superior to the traditional object-oriented extraction method, which solves the problems of low extraction accuracy and slow processing speed using traditional methods.