Classification of the tree for aerial image using a deep convolution neural network and visual feature clustering

In recent years, over-exploitation has led to the accelerated destruction of rural and natural environments for urban development, making an understanding of land use and land cover changes, one of the most urgently required and important tools for urban land planning. To this end, before any land planning begins, the distribution ratio of trees for a given piece of land is determined by calculating the area of that land covered by trees. This study proposes the use of supervised machine learning methods to classify treed areas and combines unsupervised color clustering techniques to achieve optimum classification results. First, the YOLO (you only look once) classification model is used to obtain tree features and location information. The 'K-means' and 'Flood fill algorithm' methods were tested with tree classification experiments, measuring precision rate, accuracy rate, and recall rate, with shape, illumination, and angle of tree species, and color differences affecting classification results.