Abstract

The increasing area of Plastic-Mulched Farmland (PMF) is aggravating the conflict between agricultural development and environmental protection. The spatial distribution of PMF requires an effective and economic technique. However, most works are currently carried out on pixel-level algorithms, which leads inevitably to mixed spectral errors. In this connection, PMF has been mapped with Pléiades and Radarsat-2 data combining object-based image analysis (OBIA) and Random Forest (RF). At first, through visual interpretation, the outcomes of various segmenting scenarios were used to select the optimum segmentation parameters. The spectral characteristics, textural and geometric features were then extracted and tailored to the best PMF mapping function subset. Finally, we map the PMF using the optimized object-level feature subset based on RF. The results show that the ability of Pléiades data to map PMF in Northern China is higher than that of Radarsat-2. The overall mapping accuracy achieved is 90.27%. In general, the precision and reliability of the mapping are the product of extensive structural data and object-level features that can reduce the reliance on spectral data.