Recognition of land use based on random forest algorithm Study area and data

Given remote sensing spectral data, building vector data, POI data and mobile phone positioning data, a 61-dimensional feature space was constructed in this study. In addition, the random forest algorithm was adopted to identify the urban land use in the built-up area of Changping District, Beijing, and this algorithm was compared with the recognition results achieved by the SVM algorithm and the FCM algorithm. As revealed from the study, the random forest land use identification method had the highest accuracy index, which achieved a total accuracy of 0.79 and a kappa coefficient of 0.75. In the random forest algorithm-based land use recognition algorithm, the recognition of all data sources was more accurate than the recognition of a single data source and two data sources combined. As revealed from the evaluation of the contribution of different data sources to land recognition results, the 5-dimensional features from building data, the 3-dimensional features from the geometric attributes of the plot, as well as the features from 8-10 h of the working day and 20-22 h of the rest day contribute the most to the result.