Start Date: 08/01/2006
End Date: 08/01/2008

Developing an automated system to detect defects on hardwood logs and stems has been recognized as significant research priority for nearly 20 years. Several studies estimate the value gained from improved sawing decision-making due to automated defect detection to be as much as 11 percent. Given the large volume of hardwoods sawn, there is enormous potential for improved value recovery.  Several approaches have been attempted for hardwood log defect detection.  The most significant efforts have focused on using X-ray/CT and MRI approaches.  However, despite the expenditure of millions of dollars on these efforts, no commercial system has been developed to date.  Further, the test-bed X-ray/CT and MRI systems suffer from high costs, slow scanning speed (several minutes to several hours per log), sensitivity to log moisture, and require adequate shielding for the high energy levels needed to penetrate average-sized logs.

In contrast to the issues of X-ray/CT and MRI systems, laser-based systems are safe, fast, inexpensive (a fraction of the cost), and are already employed in many sawmills for collecting shape data for accurate log positioning.  In 2001, the USDA Forest Service began a cooperative research work agreement with Virginia Tech to locate defects on barked hardwood logs (red oak and yellow-poplar) using high-resolution variants of an industrial log scanner.  The use of the scanner was donated by USNR/Perceptron for the project.  A total of 159 yellow-poplar and red oak logs were scanned.   Results have shown that defects can be accurately located and roughly classified on the logs using novel processing methods and high resolution laser scans.  Although the data from the scanner was at a relatively high resolution, areas of missing data and outliers complicated the analyses.  Due to these complications, robust estimation theory was applied and an existing generalized M-Estimator (GME; a statistical approach based on maximum likelihood concepts for estimating a model) was modified to identify and remove outlying data.  A series of computer programs were developed to process the data.  Using a sample of 15 logs with 68 observed defects, 63 defects were correctly identified by the detection software while only 5 defects were missed.  Ten defects that did not exist were erroneously detected.

Grant Institution: USDA Forest Service

Amount: $50,000

People associated with this grant:

Cliff Shaffer