The present invention relates to a procedure for deter- mining the properties of a moving object, such as a log, as defined in the preamble of claim 1.

In prior art, methods are known whereby logs are ob¬ served visually or optically in order to sort them ac- cording to their quality. In a method based on visual inspection, the person performing the sorting directs the logs to different piles on the basis of visual ob¬ servation. However, this method does not reveal the in¬ ternal properties of the logs. In optical measurement of dimensions, the measurement is taken from the surface of the bark, which means that variations in bark thickness may result in considerable errors in the determination of the dimensions and volume of the log. These methods for the inspection of logs are mainly focused on easur- ing log thickness, and the measurement data is communi¬ cated to a sorter, who directs the logs manually to ap¬ propriate piles according to this information. These methods generally provide no other information about the logs. According to investigations, another drawback is that when the sorting is done by a human sorter, only about half of the logs are sorted fairly correctly with regard to the optimal result.

A further drawback with the above methods is that, even if metal detectors are used, it is not possible to iden¬ tify all foreign objects, such as rocks and non-ferrous metals, that may be present in the logs. Therefore, such objects remain inside the log and may cause damage in the equipment used for further processing of the logs.

Finnish patent application no. FI893938 (corresponding to US patent no. US5023805) presents a method based on three-projection X-ray photography, known in itself in prior art. In this method, from each radiographic pro- jection, the knot terminations are first determined via a longitudinal reconstruction of the log, whereupon knot vectors matching these points are calculated. The weak¬ ness of this method consists in the fact that the termi¬ nations cannot be determined sufficiently accurately and unambiguously from real logs. Among the reasons for this are overlapping knots and the moisture of fresh wood, which obliterates parts of the knot.

The object of the present invention is to eliminate the drawbacks of the methods described above and to achieve a reliable and effective procedure for determining the properties of logs relating to their quality. The proce¬ dure of the invention is characterized by what is pre¬ sented in the characterization part of claim 1. Other embodiments of the invention are characterized by what is presented in the other claims.

The operation of the procedure and the measuring and data processing equipment of the invention is based on wood-related knowledge defined by wood quality and on radiological application of said knowledge. The proce¬ dure comprises a radiological, adaptive expert system based on a knowledge of wood. The procedure can also be applied to other objects or materials moved as bulk goods.

The procedure of the invention has the advantage that it enables the internal defects of bulk goods moving at process speed on a conveyor line to be measured and identified using only few projections. This allows reli-

able determination of quality properties of logs moving at sawing speed. The measurements of the log can also be taken from the log surface beneath the bark, so that the true dimensions of the wooden part of the log can be ac- curately measured. Instead of making use of the knot outlines in the pictures in a longitudinal reconstruc¬ tion of the log as in the above-mentioned patent speci¬ fication, the procedure of the invention employs the principles of fuzzy logic to locate, by means of a re- construction formed in a direction perpendicular to the longitudinal axis and utilizing layered slices, three- dimensional objects in which the knot mass is concen¬ trated.

The procedure can be used to determine internal and ex¬ ternal properties of logs. The external properties in¬ clude log length, diameter, conicity and ellipticity as well as bends, multiple crookedness, crooked-growth and volume. One of the advantages of the invention is that the diameter, conicity and volume measurements can be determined from a log with the bark on it for the log without bark. Thus, accurate measurements of the useful wood portion are obtained.

Internal defects of the log include resin pockets, rot¬ ten spots, cavities and clefts and also foreign objects, such as rocks and ferrous and other metals. The proce¬ dure provides thorough and reliable information about the knots and knot clusters as well as their quality in- side the log. The procedure also reveals variations in density and moisture of the wood. By sorting the logs by quality as provided by the invention, healthy knots, dry and rotten areas and their transition zones can be de¬ termined.

An important feature in respect of measurement and costs is a fast log analysis achieved at a relatively low cost. It is possible to increase the intensity of X- radiation and the computing power to produce a faster analysis, but this may easily lead to excessive addi¬ tional costs. In the solutions according to the inven¬ tion, the intensity of X-radiation and the computing power to be used are optimized and thus sufficient speed and accuracy are achieved at a relatively low cost. One of the factors contributing to this is that the amount of measurement data can be significantly reduced as com¬ pared with prior-art solutions.

In the following, the invention is described in detail by the aid of an embodiment example by referring to the attached drawings, in which

Fig. 1 presents a system of cylindric coordinates, which is a handy way to describe a knot in a log, Fig. 2 shows how a log is divided into volumetric ele¬ ments in cylindric coordinates. Fig. 3 illustrates the geometry of the log raying proc¬ ess. Fig. 4 presents an example of the array sums of a ra- diographic projection, from which the locations of knots can be calculated, Fig. 5 represents the principle of filtering out the effect of stemwood from a radiographic projec¬ tion, Fig. 6 presents an example of an evidence graph for volumetric elements by sectors, Fig. 7 presents a simplified illustration of the pas¬ sage of an X-ray through a log divided into sec¬ tors and circles, and Fig. 8 presents a table showing the measurement results

obtained in the case illustrated by Fig. 7.

The procedure involves the use of tomography. In medical tomography, a problem of the same type has been solved in which an object is X-rayed from many directions and the internal structure of the object is calculated from the projections. However, the number of projections is 500 - 1000. It is not possible to take as many X- rayograms of a saw log, but in practice a few, e.g. three projections must suffice. In the procedure of the invention, a log 5 moving at sawing speed is ra¬ diographed by means of only a few, e.g. three radio- graphic devices, such as X-ray apparatuses, emitting a radiation capable of penetrating matter, and the picture data is stored by means of detector arrays 8, one or more detector arrays being used for each X-radiation source 4. It has been established in practice that as few as three projections are sufficient to provide enough information to allow the quality properties of a log to be measured with a good accuracy.

The procedure is not concerned with reconstructing the log from the pictures pixel for pixel as in earlier practice, but instead use is made of a knowledge of the typical geometry, density and other properties of the trunk, knots and the associated anomalies as well as the interdependencies of said properties. Typically, the pictures are analyzed to detect objects having the shape of knots or other anomalies, or parts of such objects. These are processed in a system of cylindric coordinates divided into discrete volumetric elements.

The process of determining the properties of a log or a corresponding object by the method of the invention can be divided into three main parts, which will be de-

scribed in greater detail later on.

The first part: Preliminary processing of the measure¬ ment data is performed using the pixel-specific inten- sity data obtained from the radiograph in each X-raying direction, on the basis of a knowledge of wood. The analysis is based on the relative attenuation differ¬ ences caused by internal objects in the log. For each wood quality, the radiological relative differences, i.e. attenuation differences, of the boundary sur¬ faces/values between normal wood and internal objects in the wood can be defined. These relative differences are compared both in the transverse plane and in the direc¬ tion of tree growth. The differences are relative from one tree to another and in the same tree depending on its moisture. These divergent areas of interest are fur¬ ther studied using more exact calculation methods. Thus, the measurement data relating to normal wood need not be processed further, so the processor power will be suffi- cient for real-time quality sorting of logs advancing at process speed. Using simple logic deduction rules, anomalies are identified as positive or negative anoma¬ lies and the boundary surface of the anomaly is deter¬ mined.

The second part: Objects detected in the log are identi¬ fied and their position is ascertained using the other measuring directions. The objects (knot, rotten spot, rock, etc.) are identified by making use of a wood type specific knowledge on the basis of their location, size and relative X-ray attenuation.

The third part: Based on wood type specific knowledge and radiographic appearance, a semi-empiric simple mathematical model or representation has been developed

for objects in the wood and is applied to an object de¬ tected and identified in an area of interest, so the size and quality of the object can be determined.

Fig. 1 presents a system of cylindric coordinates, con¬ sisting of an angle of rotation α, radius r and longitu¬ dinal axis z. A slice of log trunk is a cylindrical body in which the core runs along the longitudinal axis z. Knots start at the core and grow towards the surface with an upward gradient β and a spread angle γ. Each knot lies within a sector that contains no other knots, so each knot can be described with a conical model. A knot may contain both healthy 10 and rotten 11 wood. As illustrated by Fig. 2, a log can be divided into circles 1, slices 2 and sectors 3, and a section comprising each of these constitutes one volumetric element, whose posi¬ tion is defined in the cylindric coordinates. The slice 2 thickness in the lengthways direction of the log rep¬ resents the width of the detector elements 9 of the de- tector array 8 in the lengthways direction of the log and therefore the longitudinal log section exposed to radiography at a time.

Fig. 3 illustrates the geometry used in the radiological apparatus, showing only one X-ray for the sake of sim¬ plicity. When the description deals with the rays emit¬ ted by one X-radiation source 4 and the detector ele¬ ments 9 receiving them, which together form a detector array 8, one radiographic projection is being referred to. Therefore, as the procedure comprises the use of three X-radiation sources and a detector array corre¬ sponding to each of these, it can be said that the meas¬ urement ultimately takes place in three radiographic projections. The X-radiation sources 4 and the corre- sponding detector arrays 8 are placed at an angle of

120° relative to each other and so disposed around the path of the log that the log will pass between the X- radiation sources 4 and the corresponding detector ar¬ rays 8. Thus, the X-rays emitted by the X-radiation sources penetrate the log and, depending on the proper¬ ties of the log, are attenuated in different ways on their way to the detector arrays, whose detector ele¬ ments 9 receive the radiation thus attenuated. Each de¬ tector array 8 consists of a series of detector elements 9 placed in a curved arrangement around the log, all of the detector elements being located at equal distances from the corresponding X-radiation source 4 and in the same plane perpendicular to the log movement as the X- radiation source. The number of radiographic projections used may be greater or smaller than three as needed.

The log 5 lies on a conveyor surface 7, where it is ex¬ posed to X-radiation from an X-radiation source 4. The radiation penetrating the log 5 is received by a detec- tor element 9. The geometry is described by the distance dl between the X-radiation source 4 and the centre line of the log 5, the distance d2 between the X-radiation source 4 and the conveyor surface 7 and the distance d3 between the X-radiation source 4 and the detector array 8. Distance dl depends on the log radius R as follows: dl=d2-R. The detector elements 9 of the detector array 8 are indexed each one separately. The detector element 9 receives information about a sector element 6 of the log, but also for the entire distance covered by the ray. The information consists of X-ray attenuation data.

A knot model is created by utilizing a knowledge of the typical geometry and density of knots and stemwood. Be¬ low are a few rules: - The cross-section of the trunk is roughly elliptical.

The size of the cross-section can be estimated as be ^¬ ing the mean value of the diameters of the three ra¬ diographic projections. The largest one of the diame ^¬ ters of the radiographic projections is used to de- fine the circle that contains the cross-section of the reconstructed image.

- All knots start from the core of the trunk. The knot is a cone which is described by an angle of rotation α, an upward gradient β, a spread angle γ and a radial length r. The upward gradient β has certain predeter¬ mined values in degrees.

- All knots in a cluster of knots start from about the same point. Adjacent knots can not lie side by side, but a minimum value has been defined for the rota- tional distance between knots.

- The density of the trunk varies from the core towards the external surface. Typical densities of sapwood, heartwood and knots have been defined experimentally.

Before the calculation, the centre of the log must be brought exactly to the centre of the calculation coordi¬ nates. This is done by moving the image until the log centre coincides with the centre of the coordinates. The log centre again is obtained by determining the edges of the log from the radiographic projection by thresholding and then calculating the log diameter from the edge data obtained.

An X-ray penetrating a log undergoes greater attenuation when passing through a knot than when passing through other, softer wood material. By examining the rays re¬ ceived by the detector elements 9, it is possible to ob¬ tain hints, which at this stage constitute unreliable individual pieces of information, indicating that the samples represented by certain pixel groups or detector

element groups might contain knot mass. By combining the images and hints indicating the presence of knots from all radiographic projections, a certain truth value is obtained for the volumetric element. By combining the truth values of adjacent volumetric elements, a truth value is obtained for the assumption that the sector is part of a knot. The truth values are assigned values in the range -1 ... +1. The values -1, 0 and +1 may be de¬ fined verbally as meaning "absolutely no", "undefined" and "absolutely yes".

In the processing of log data, calculation time is saved by focusing exclusively on those parts of the log that contain knots or other anomalies. From the radiation re- ceived by the detector array 8, array sums are calcu¬ lated, from which the positions of knots can be deter¬ mined: since knots cause a greater attenuation than nor¬ mal wood, the array sum for an image array containing knots is greater than for an adjacent array containing no knots. According to the invention, the aim is to lo¬ cate those parts of the log which produce an increased array sum, which may contain knots. Fig. 4 presents an example of the variations of the array sums in the lon¬ gitudinal direction of the log. In graph (a), the posi- tions of knot clusters and also the log thickness are clearly visible.

In graph (b) , the variation of log thickness has been filtered out. By performing median filtering on the start and end coordinates of the knot clusters, graph (c) is obtained. After the positions of knot clusters have been determined, the effect of stemwood is filtered out from the radiographic projections, so that only changes caused by knots and other anomalies remain. Fig. 5 presents a longitudinal section of a log without stem-

wood filtering (a) and another graph representing the same section with the effect of stemwood filtered out (b) . The graph represents a longitudinal stripe of the log as seen by one detector element. In other words, this is a vertical stripe picked from a two-dimensional image. The log images contain a sufficient number of such stripes side by side.

Eligible filtering methods include e.g. average or me- dian filtering. The filtering is performed by observing a series of points p(i) consisting of N measuring points, i.e. index (i) is assigned values 1...N. The fil¬ tering compensates local variations, thus permitting larger entities to appear more clearly. In the case of the present invention, the filtering is performed to eliminate the effect of stemwood from the image, so what remains is the image produced by knots and other anoma¬ lies.

In average filtering, a new value q(i) is calculated for each point as follows:

3=ι-m

To calculate a filtered value for a point i, an average is determined by considering m points on both sides of the point i. The number m is so selected that the varia- tions to be filtered cover a length shorter than m points.

Correspondingly, in median filtering, points p(i-m) p(i+m) are similarly considered to calculate a filtered value for points i. The numeric values of the points are ordered in sequence according to magnitude and the mid¬ dle one is selected, called the median of this number

series. Median filtering involves more computing work than average filtering, but it is not sensitive to the effects of individual large anomalies.

After the effect of stemwood portions has been filtered out from the radiographic projections, each knot cluster is processed separately as follows:

a) The three filtered radiographic projections are projected back to a system of 3-D coordinates by making use of back projection coefficients calcu¬ lated beforehand. The coefficients take the known geometric properties of knots and trunk into ac¬ count. This process divides the knot mass into volumetric elements. b) The value of each volumetric element is indicative of the density of the wood in the element. Using experimental parameters, the density values can be converted into evidence values, which give a prob- ability as to whether the volumetric element is part of a knot. c) By combining the evidences of individual volumet¬ ric elements, truth values indicating possible knot sectors are obtained. Fig. 6 shows an example of a graph representing the truth values of log sectors. It can be seen from the graph that sectors 5, 12, 19 and 32 may contain knots. d) Back projection as described under item a) is re¬ peated, but this time only for selected rotational angles. In this way, side projections of the knots are obtained. From these projections, approximate upward gradient and spread angle values are now calculated.

The back projection and associated coefficients will be

now described by referring to Fig. 7 and 8. The basic idea of the procedure of the invention derives from the fact that a knot starts from the core of the trunk and grows regularly expanding towards the trunk surface. Therefore, the calculation is advantageously performed using cylindric coordinates because the shape of a knot resembles a sector. Fig. 7 shows only 12 sectors to sim¬ plify the matter, whereas the system actually uses more sectors. Since the average knot width is 20°, a knot may occupy a space extending across two or three sectors. In the calculation, each radiographic projection is first processed separately. The measurement results are then processed to eliminate the effect of stemwood, leaving only the values representing knots. Other anomalies are not considered at this point in this description.

In Fig. 7, the sectors are numbered 1...12 and the circles 1...4. The figure shows one ray, which is emitted by the X-radiation source 4 and received by one 9 of the detec- tor elements of a detector array. In the table in Fig. 8, the corresponding detector element is defined as pixel h. It is assumed that the ray has been attenuated during its passage through the log by an amount corre¬ sponding to ten units of knot mass, i.e. p(h)=10. The attenuation caused by a unit of knot mass has been de¬ termined experimentally beforehand, as stated before. However, a single measurement as described above is not sufficient to indicate where the knot is or whether there is only one knot or several knots. Still, it fol- lows from the measurement geometry that only certain elements of the cross-section are to be considered. From Fig. 7 it can be seen that the ray passes through ele¬ ments (1,2), (2,2), (2,3), (3,3), (3,4), (11,3), (11,4), (12,2) and (12,3). The first number in the element coor- dinates indicates the sector while the second number in-

dicates the circle. Thus, the attenuation information received at pixel h can only come from these elements, from one or more of them.

As the contribution of each element to the attenuation obtained as a measurement result is not yet known at this point, it is assumed that the attenuation is evenly distributed throughout the passage of the ray in the log. In the case of our example, the distance travelled by the ray in the log is 73.50 mm. The measured ten units of knot mass is now divided among the above- mentioned elements in proportion to the distances trav¬ elled by the ray in each element. For example, the back projection coefficient for element (1,2) is the distance of ray travel in element (1,2) divided by the total travelling distance of the ray in the log, i.e. 8.50mm/73.50mm=0.12. As the attenuation value was 10, the projection result obtained for element (1,2) will be 10*0.12=1.2 (Fig. 8 shows a more precise reading). The coefficients c(h,i,j) used for the division have been calculated in advance and placed in a table as shown in Fig. 8 as explained above.

A complete table naturally contains the coefficients for all pixel, sector and circle values h, i and j . Most of the coefficients have a zero value because each ray only passes through a few elements.

The term 'back projection' here means that each radio- graphic projection is returned via computation to the two-dimensional section from which it was produced. In the table in Fig. 8, back projection has been performed with only one radiographic projection and only one de¬ tector element (pixel h) . When the calculation is per- formed with all the radiographic projections and detec-

tor elements, i.e. with all the values obtained, and the results are summed for each element, then for each sec¬ tor element a numeric value describing the knot mass contained in it will be obtained. If a high numeric value is obtained, then the element is likely to be part of a knot. When several high numeric values fall within the same sector, this further corroborates the notion that the sector contains knot mass. The numeric values are combined via a method called evidential inference. When the evidence or truth value for a sector exceeds a certain predefined threshold, the sector is accepted as a knot sector.

Because a large proportion of the knot mass obtained via the first back projection process seems to be spread even into sectors having no knots, the back projection process has to be repeated. This time, all sectors that are not regarded as knot sectors are omitted by setting their coefficients c(h,i,j) to zero. In this way, the knot mass can be placed exclusively in actual knot sec¬ tors.

As stated above, the size and direction of individual knots can be characterized in terms of radial length r and angles α, β and γ. These parameters can be used to calculate the assumed positions and areas of knots on the sawn surface. This makes it possible to obtain an advance estimate of the value of the log as timber, and in further processes even to optimize the sawing posi- tion on the basis of the knot data.

It is obvious to a person skilled in the art that the invention is not limited to the example described above, but that different embodiments of the invention can be varied within the scope of the following claims.