Towards Automatic Trunk Classification on Young Conifers

Abstract

In the garden nursery industry providing young Nordmann firs for Christmas tree plantations, there is a rising interest in automatic classification of their products to ensure consistently high quality and reduce the cost of manual labor. This paper describes a fully automatic single-view algorithm for distinguishing between young fir trees having correctly developed top shoots and trunks, and trees having either multiple or missing top shoots. This is accomplished by an approach combining a Euclidian distance transform with a dynamic programming algorithm for finding optimal paths following the trunk. The classification performance of the method was investigated using a SVM and 10-fold stratified cross validation, resulting in a correct classification rate of 90.2% (101/112) when discriminating between trees having one top shoot and trees having multiple top shoots. Future work aims to improve the classification performance of the algorithm by incorporating color information into the data considered by the dynamic programming algorithm.