Automated identification of tree species from images of the bark, leaves or needles

In this thesis a method for the automated identification of tree species from images of leaves, needles and bark is presented. In contrast to the current state of the art for the identification from leaf images, no segmentation of the leaves is necessary. Furthermore the proposed method is able to handle damaged and overlapping leaves. This is done by extracting keypoints in the image which allow the computation of local descriptors. With the help of a trained set of descriptors, a histogram of the occurrence of these descriptors is generated which is then classified using a one-vs-all SVM. For images of the bark the same method is used which allows to build a system which can automatically distinguish between images of bark and leaves without calculating new features. Detected features within the needle images can be used for a classification. The needle endings are used to distinguish between fir and spruce needles. The proposed method is evaluated on three datasets which were provided by “Österreichische Bundesforste AG” (“Austrian federal forests”). The first dataset consists of 134 images of the leaves of the 5 most common Austrian broad leaf trees, the second dataset containing 275 images of the needles of the 6 most common Austrian conifers. The last dataset is containing images of the 1183 bark of these 11 trees. The classification rate for the leaf dataset was 93.6% and the classification rate of the bark dataset was 69.7%.