A fast pith detection for computed tomography scanned hardwood logs

Abstract

Tree and log evaluation prior to processing is traditionally conducted by visual inspection of the outside surface. This approach has obvious limitations, because many internal features cannot be directly observed. Just recently, computed tomography-based (CT) approaches allowed visualization and analysis of the internal wood structure. The data gathering reveals the internal structure in a form of successive 2D discrete images in limited resolutions in both spatial and value domains. However, the size of scan data can be very large (gigabytes) and their processing using traditional approaches can be time-intensive. There is a need for classification and quantification of internal log defects in real time to keep up with processing speed at modern mills. The aim of this study was to develop a real-time pith detection from CT-scanned log data. The speed necessary for real-time processing is achieved in two ways: first, by adaptive method that uses precise detection only when necessary; and second, by parallel processing power of graphic processors (GPU) that are more suitable for parallel data processing of large datasets than classical central processing units (CPU). The input of our system is a set of 2D images that were collected during the CT scanning and the output is a set of locations within the slice that have been identified as pith. Results of our algorithm tested on data from North American species of Black Cherry, Black Walnut, Hard Maple, Red Oak, White Oak, and Yellow Poplar show that on average, the algorithm found pith with precision of 4.2 mm as compared to manual pith detection. The GPU acceleration by using CUDA enables processing speed of about 0.003 s per image with high precision. This makes the developed algorithm suitable for an industrial application in hardwood sawmills and veneer slicing operations.