Forest classification method based on convolutional neural networks and sentinel-2 satellite imagery

The objective of this study is to develop a classification method based on convolutional neural network (CNN) and Sentinel-2 satellite imagery including the spectral feature, spectral index and spatial feature together as an input to answer forest monitoring problem. This research also used contextual information on Indonesia National Standard Agency's document for Land cover classification as a baseline for feature extraction to get the appropriate classifier feature. The test set was located in Semarang, Central Java, Indonesia. The research workflow consists of defining forest class based on Indonesia National Standard Agency for Land cover classification, extracting optical image features based on contextual information of the forest class definition, extracting image features from the Sentinel-2 satellite image, and classifying image object features using CNN classifier. Image segmentation produced 1,211 segments/objects by using eCognition software. Subsequently, these objects were used as a dataset. Overall accuracy was used to evaluate the performance of the classification result. The result showed the classification method results in this study yielded high overall accuracy (97.66%) when using CNN with the image features like NDVI, Brightness, GLCM homogeneity and Rectangular fit. Small improvement of overall accuracy was also achieved when it was compared to GBT with an overall accuracy of 95.50%.