Modern Lineal Colour Based Quality Grading
Lineal, i.e. longitudinal quality measurement has been used for many years in secondary woodworking, where the pieces move on lineal conveyors. In sawmills lineal measurement has been used in automated edgers since 1980’s. What kind of a machine would a modern, lineal, colour based quality grader be today? This article describes one such system, the OptiGrader.
The OptiGrader lineal system is based on a 2048 pixel colour line scan camera. The field of view of the camera is divided into four sectors, such as illustrated in Figure 1. The number of pixels used in each sector is controlled with software and with a simple mirror arrangement. In practise about half of the total pixels is used directly to scan the top face and with a mirror one side of a board. With the mirror arrangement the other half of the pixels is used to scan simultaneously the bottom face and the second side.
Figure 1. A lineal colour line scan camera based quality qrader, which measures simultaneously all surfaces while the board moves through the scanner.
The benefit of lineal measurement is that the measurement of all surfaces takes place simultaneously and the measured board moves in controlled manner on a conveyor. Robust and reliable devices to measure length and thickness are also easy to integrate in a lineal system. Because the scanning distance to the top face and one side are short, usually about half a metre, the position errors of detected features due to system vibration remain at worst within one millimetre. For the bottom face the maximum location error is less than four millimetres. As the one and only camera measures simultaneously all surfaces, the vibration of the camera does not cause classification errors, just a potential position error.
The range of lineal measurement is in width about 500mm and in thickness 75mm. Thus the detector elements 3*2048 pixels (2048 pixels for each colour) is divided in the total circumference of approximately 1500mm of the sawn piece (1150mm + gaps required by the image processing). Therefore the optical resolution in width direction is 0.7mm. The longitudinal resolution depends on the conveyor speed. When the speed is 5m/s and the camera exposure time one millisecond, the longitudinal resolution will be 5mm. The amount of 0.7*5mm2 colour data pixels from a top face of a six metre long and 300mm wide board will be 500,000. Because one and the same camera collects simultaneously data from all other surfaces, too, the total amount of colour data is easily over one million.
The quality and amount of light, the increasing and improvement of which is difficult in transverse systems, can be increased and controlled more easily in lineal systems. Lineal systems are compact, the illuminated area is smaller and the number of supervised, ageing light sources is smaller. It is far easier to construct a diffuse light source in the measuring ‘tunnel’ of a lineal system than to build a six metre long stable light source system for a transversal system. Also the implementation of white calibration is much simpler in a lineal system because there is only one camera.
Figure 2. Principle of a prism based 2048 pixel, true colour line scan camera.
Thanks to good light intensity in lineal measuring systems it is possible to use prism based special line scan cameras, which detect colour information from the same physical spot with all three colours, such as in Figure 2. The advantage of this technique is that the colour characteristics of wood are measured simultaneously and from the same exact spot with all three main colours. Thus the classification of knots and other features bases always on correct colour data.
The disadvantage of a lineal colour grader is that its capacity is at best about 50 full-length boards per minute. Additionally, at least still today, the longitudinal movement of boards must be changed into transversal for trimming and sorting bins. On the other hand the system has only one camera, and it is far simpler and easier to use than transversal measuring systems. When required, one can install several lineal systems parallel. Thus one can increase capacity to the desired level, and in case of disturbances the whole line does not stop, just the line through the faulty device. Increase of capacity is possible in 50 boards per minute steps, one device at the time.
Figure 3. Lineal colour grader OptiGrader and an insert of its measuring tunnel.
Lineal OptiGrader quality grading may be used both in old and new sorting lines, Figure 3. A stack from drying is unloaded in usual manner, but after separation the boards are fed with a sideways movement through a lineal measuring system. After value and quality optimisation and decision of trimming details, the boards are returned to transversal movement towards trimmer and sorting bins.
Determination of Quality
Most quality grading programs used in transversal graders operate on rule-based classifiers. In these systems detection and classification of defects and other surface features is based on parameters and thresholding. During start-up of the system and its tuning in operation depends on a human operator, who tries to adjust numerous parameters, which depend on each other, in order to find a combination that gives the desired outcome in grading. Setting and tuning of these parameters requires deep knowledge of the system and wood behaviour, and a lot of time, and constant supervision of the system. The major problem of rule-based systems, particularly in colour grading, is their poor adaptability to changes in wood material (colour, species, effects of seasons) and in imaging environment (ageing of light sources, dust, ambient light).
Because the lineal OptiGrader scanner produces high quality image data it is possible to use quality grading and teaching programs, which base on self organising map (SOM) technique and semi-supervised visual training. Teaching of a SOM based grading system is a compromise between supervised and unsupervised teaching methods. Successful operation of an automated grading requires a map of taught defects. In SOM map teaching the samples do not require any labelling, thus making the teaching fast and simple. Defects are taught for each species and for green and dry wood, as needed. Samples are used to teach different shades of sound wood colour, variations of annual growth rings, and various defects, such as sound and dead knots, pitch and bark pockets, shakes, and discoloration. Teaching is similar to teaching a new person to grade. The most important feature of the SOM maps is that similar defects self-organise themselves close to each other in nodes, which are near to each other also on a two-dimensional map. When the teaching of the selected samples is finished and the SOM program has organised their images in groups with similar features, the operator tells to the systems which groups represent which type of defects or surface qualities, see Figure 4. When the teaching is finished the program operates automatically and is ready for use.
When the OptiGrader system detects a defect it reasons, based on similarities with the taught SOM map, into which defect class the defect belongs to. In order to perform the value optimisation the software need a database, in which the sawmill defines the monetary value criteria for each sawn timber size and quality. Since the grader system is connected to sawmill data net, the value tables for sawn timber can be up-dated in accordance with the market situation in order to optimise value recovery and turnover.
Figure 4. A self organised map (SOM) of defects and wood surface features, where the defects are on the right side and sound wood surface on the left. Each node (square) contains a large number of characterising samples, which are on top of each other in the node.
The use of modern, SOM-based visual inspection and control systems requires scanning systems, which operate reliably. If detection and classification of defects does not succeed, the system does not operate correctly. Therefore the first SOM-based systems rely on lineal scanning technology, such as OptiGrader, where the sawn timber colour information is most reliable.