Research of Forest Classification Based on Remote Sensing Image

The forest plays an important role in regulating climate and improving ecological carrying capacity. In view of heterogeneous mixed young afforestation, the object-based classification on rules method is used to identify the types of planted forest, combined with spectrum, texture and shape characteristics information. ESP tool was applied to obtain the best segmentation scale and the rule set was built for the classification. The overall accuracy for classifying species was around 58%. The class-specific producer’s accuracies ranged between 38% and 82% and the user’s accuracies was between 33% and 96%. Compared with MLC which simply relied on the spectral information (overall accuracy 43%, kappa 0.33), the method achieved much more improvement in overall accuracy of 15 points and increased kappa coefficient to 0.52.