Deep Learning Based Supervised Image Classification Using UAV Images for Forest Areas Classification

Applications of unmanned aerial vehicles (UAVs) based remote sensing is increasing rapidly due to their advanced accessibility, capability for fast and easy deployment, capability for miniaturization of sensors and efficient collection of remotely-sensed data from relatively low altitudes. Recently, UAV data sets have been found to be quite useful for forest feature identification due to their relatively high spatial resolution. Several machine learning algorithms have been broadly used for remotely-sensed image classification. In remote sensing image classification, deep learning based methods can be considered quite effective techniques as they have achieved promising results. In this study, we have used deep learning based supervised image classification algorithm and images collected using UAV for classification of forest areas. The deep learning algorithm stacked Auto-encoder has been found to have tremendous potential regarding image classification and the assessment of forest coverage area. Our experimental results show that deep learning method provides better accuracy compared to other machine learning algorithms. Cross-validation showed that the overall accuracy of the deep learning method is about 93%. This study highlights the essential role that UAV observations and deep learning could play in the planning and management of forest areas which are often under the threat of deforestation and forest encroachment.