Classification of commercial grade lumber requires a visual inspection of the milled board to determine if the board meets attributes of the various grades of lumber. The lower the number of anomalies, such as knots, the higher the grade of the piece. Knot wood and clear wood radiate and absorb heat at different rates, thereby allowing for the development of a computerized thermal recognition system to detect knot anomalies. This dissertation investigated the use of high power (6 kW) quartz infrared halogen lamp heaters, and high power radio frequency (35 kW) to treat the piece to be inspected. The thermal response was obtained by an infrared image (512 by 480 pixels). Twelve species of wood were investigated. For this dissertation, a computerized thermal recognition system was developed using the optimal derived images to produce a technique to detect anomalies in images. The thermal recognition system to identify anomalies used a soft computing architecture for edge detection in this noisy environment. The principle of soft computing techniques is to arrive at a near-best solution with imprecise, incomplete, and marginal information by adopting a coherent strategy. This methodology recognizes that an optimum solution cannot be obtained but that a near or best-available solution may be just as workable when dealing with real systems. The technology described here evolves an object such that it becomes a replica of the object visualized and constantly compares the view and the image until it is fully-grown. In this research, this approach is referred to as "seedlet technology". Seedlet technology uses soft computational techniques to detect anomalies in images. The seedlet system consists of a discrete cosine transform (DCT) filter, a neural network, a seedlet, and a genetic algorithm. The DCT filter forms a preprocessing module to reduce noise on the image. The neural network provides information about the anomaly in which rules for the seedlet can be developed. In addition, the neural network forms an intelligent selective low-pass filter of the image. The seedlet then grows according to derived rules and by using information provided by the neural network. The seedlets remove noise on the image, and identify the approximate location of anomalies on the image. The genetic algorithm then manipulates parameters of the seedlets to optimize the location of the anomaly. At this point, the location of the anomaly has been determined. This technique was successfully applied for locating knot anomalies in wood.
| Identifer | oai:union.ndltd.org:MSSTATE/oai:scholarsjunction.msstate.edu:td-4889 |
| Date | 13 December 2002 |
| Creators | Patton, Michael Dean |
| Publisher | Scholars Junction |
| Source Sets | Mississippi State University |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Theses and Dissertations |
Page generated in 0.0017 seconds