RU2815187C1 - Method of determining and classifying defects of wood material - Google Patents

Abstract

FIELD: various technological processes; control devices.

SUBSTANCE: invention relates to process control and relates to a method for determining and classifying wood material defects. Method involves formation of three illumination zones: an upper light zone, a side light zone and a lower light zone. Method also includes video recording of wood surface using video cameras using all three illumination zones, classification of defects features on each frame using simultaneously all three illumination zones, constructing a classification model and training the neural network to determine defect features using data on feature classification for each of the three illumination zones. At the same time the upper light zone is intended for determining the signs of defects caused by the change in the colour of the wood surface, side light zone is intended for determination of signs of defects caused by change of geometry of wood surface, zone of lower light is intended for determination of signs of defects caused by through damage of wood.

EFFECT: high quality of determining and identifying defects due to use of a large number of unique identifiers.

8 cl, 7 dwg, 2 tbl

Description

Field of technology

The invention relates to a method for identifying and classifying defects in wood material, which is intended for early detection of defects and timely notification to the operator of a wood processing line. The method identifies and analyzes defects in real time, which allows operators to promptly take measures to eliminate the production of low-quality products. Thus, the method of determining and classifying defects in wood material is a tool for intermediate control, allowing to achieve high quality, reduce production costs and losses due to defects. Using this method makes it possible to analyze the quality of manufactured products much faster, reducing time resources for data processing. Flexible data analysis can significantly increase the efficiency (increasing quality and reducing raw material costs for defects) of the production line.

Technical problem

Wood materials, for example, veneer or blanks cut from a wooden log, are made from tree species such as oak, pine, birch, and alder. During the production and processing of wood materials, various difficult-to-detect defects may occur, for example: eyes, healthy knots, fused knots, fused sprouts, veins, partially fused knots, pin knots, dark unfused sprouts, non-through holes, pocket, false core, spotting, chemical stains, biological damage, rot, closed cracks, knots falling out and holes from them: split cracks, through middle holes, through edge holes, rough (not smooth) peeling, blockages, risks. These defects are difficult to detect because they require a lot of attention from the operator. The technical problem is to create a method for detecting the above difficult to detect defects in wood material.

State of the art

The closest analogue of the invention is RF patent 2 795 680, IPC G01B 11/16 (2006.01), G01N 33/46 (2006.01), publ. 05.05.2023, which describes a device for determining difficult-to-detect defects in veneer lumber after peeling (“unsmooth (rough) peeling,” “risk,” and “clogging”), on the basis of which the final assessment of the quality of lumber is made. The disadvantage of the analogue is the limitation on the number of unique identifiers (signs) of defects and, consequently, the low quality of determination and classification of defects.

The essence of the invention

When applying the proposed method, a new non-obvious technical result is achieved, namely, by combining the results of determining signs of defects in wood material, using all three lighting zones (top light, side light, bottom light), the number of unique identifiers (signs) of defects increases and Consequently, the quality of determination and classification of defects in wood material increases.

These interrelated technical results are achieved by the fact that, according to the proposed invention, the method for determining and classifying defects in wood material includes the steps:

Formation of three backlight zones: top light zone, side light zone, bottom light zone, while

the overhead light zone is designed to identify signs of defects caused by changes in the color of the wood surface,

the side light zone is designed to identify signs of defects caused by changes in the geometry of the wood surface,

the low light zone is designed to identify signs of defects caused by through damage to wood,

Video recording of the wood surface using video cameras, using all of the above three illumination zones, while on each frame the classification of signs of defects is carried out using simultaneously all of the above three illumination zones;

Construction of a classification model and training of a neural network to determine signs of defects using data on the classification of signs for each of the specified three backlight zones;

Prediction by a trained neural network of the presence or absence of defects based on video recording of the wood surface.

In one embodiment, the video recording speed of the wood surface is from 2 to 40 frames per second.

In one embodiment, the overhead light zone is formed by a luminous flux directed from top to bottom onto the surface of the wood, creating illumination to detect signs of defects caused by changes in the color of the wood surface. Examples of defects caused by changes in the color of the wood surface are: eyes, healthy knots, fused knots, fused sprouts, veins, partially fused knots, pin knots, dark unfused sprouts, non-through corks, pocket, false core, spotting, chemical stains, biological damage, rot.

In one embodiment, a side light zone is formed by a light flux directed along the surface of the wood to create a contrasting cut-off line to identify signs of defects caused by changes in the geometry of the wood surface. An example of defects caused by changes in the geometry of the wood surface are: rough (not smooth) peeling, blockages, risks.

In one embodiment, the lower light zone is formed by a light flux directed from bottom to top on the opposite side of the wood to create gaps and identify signs of defects caused by through damage to the wood. An example of defects caused by through damage to wood are: closed cracks, falling out knots and holes from them: divergent cracks, through middle holes, through edge holes.

In one embodiment, the video recording speed in the specified range is changed depending on the number of signs of defects to be classified, namely, when the number of signs decreases, the video recording speed is reduced, and when the number of signs increases, the video recording speed is increased.

In one embodiment, the illumination zones are located sequentially in the direction of movement of the wood, in which the light fluxes are formed by a directional light source and/or a linear diffused light source.

Brief description of drawings

FIG. 1 - Block diagram of determination and classification of defects in wood material with the formation of three illumination zones.

FIG. 2 - Example of a “Knot” defect under different lighting conditions. On the left - in natural light. The central photo is the overhead light zone. The boundaries of the defect are clearer, the distribution of shades of the color of the defect is visible. On the right is the side light zone. Allows you to highlight changes in the surface of the defect itself and the area around the defect.

FIG. 3 - Example of a “Knot” defect. On the left there is natural light, on the right there is overhead light. In the right photo, the boundaries of the defect are clearer; you can see a change in color inside the knot caused by the beginning processes of wood rotting. In this zone the defect can be classified as “Rot”.

FIG. 4 - Example of the defect “Rough (unsmooth) peeling”. On the left in natural light, on the right when using the side light zone.

FIG. 5 - Example of a “Crack” defect. In natural light (left), this defect cannot be accurately classified. With illumination organized in the low light zone (right), the classification accuracy is 100%.

FIG. 6 - Example of a “Falled out knot” defect. On the left is natural light - the defect cannot be accurately classified. With illumination organized in the low light zone (right), the classification accuracy is 97%.

FIG. 7 - An example of detection of defects on the surface of wood by a trained neural network is shown. An image is shown on the display device, on which defects are highlighted and the types of defects are assigned to them. The upper right side of the screen shows the percentage of area of each defect recognized by the neural network. Rough peeling - 0.15%, Knot -0.75%, Blockage - 0%, Risk -1.25%

TABLE 1 - Results of laboratory work on the accuracy of determination and classification of defects in wood materials (see graphic part).

TABLE 2 - Results of laboratory work on the accuracy of determination and classification of defects on wood material depending on the number of frames, the speed of wood movement, the neural network model and the complexity of defects (see graphic part).

Carrying out the invention

The invention can be implemented as follows. In the area before the automatic sorting of the drying line, industrial cameras are installed on a separate frame (linear camera - 1 pc., matrix camera - 1 pc.), light sources to form three illumination zones, according to the proposed method, a display device is installed to visualize detected defects and display information about defects on the surface of wood from a trained neural network.

Building a classification model and training a neural network to determine the signs of defects using data on the classification of signs for each of the specified three highlight zones is as follows. The classification model is built on the basis of the classification of features for each of the specified three illumination zones. The neural network localizes defects on the wood surface and transmits an array of received data to the classification model. The classification model identifies what type a given defect should be classified as. The neural network generates an output image in which defects are highlighted and the types of defects are assigned to them.

Examples of implementation of the invention

Example 1

In fig. Figure 2 shows an example of determining the “Knot” defect under different lighting conditions. On the left - in natural light. Central photo - Top light zone. The boundaries of the defect are clearer, the distribution of shades of the color of the defect is visible. On the right - when using the side light zone. Allows you to highlight changes in the surface of the defect itself and the area around the defect. Combining the results of defect determination obtained in the top and side light zones allows you to improve the quality of defect determination due to a larger number of unique features - identifiers.

Example 2

In fig. Figure 3 shows an example of determining the “Knot” defect. On the left in natural light, on the right is the overhead light zone. In the right photo, the boundaries of the defect are clearer; you can see a change in color inside the knot caused by the beginning processes of wood rotting. In this zone the defect can be classified as “Rot”.

Example 3

In fig. Figure 4 shows an example of determining the defect “Rough (non-smooth) peeling”. Left in natural light, right when using Side Light Zone.

Example 4

In fig. Figure 5 shows an example of defining a “Crack” defect. In natural light (left), this defect cannot be accurately classified (right). When using the Low Light Zone, the classification accuracy is 100%.

Example 5

An example of determining the defect “Falled out knot”. On the left in natural light - the defect cannot be accurately classified. On the right, using Low Light Zone, classification accuracy is 97%.

As can be seen from Examples 1-5, the presence of 3 lighting zones significantly facilitates the early detection of lumber defects and timely notification of the peeling line operator. When developing the implementation of the method, experiments were carried out using traditional devices for determining changes in the surface of a material (triangulation), illumination with powerful light sources. When the light sources are located in the final form, implemented in the claimed method, in the illumination zone, defects become visually distinguishable, have clear boundaries and can be classified by the eye even without the use of video cameras, which would not have been possible without lighting.

Thus, when using the method according to the claimed invention, the quality of determining difficult-to-detect defects in wood material improves.

Claims (15)

1. A method for determining and classifying defects in wood material, including the steps: formation of three backlight zones: top light zone, side light zone, bottom light zone, while the overhead light zone is designed to identify signs of defects caused by changes in the color of the wood surface, the side light zone is designed to identify signs of defects caused by changes in the geometry of the wood surface, the low light zone is designed to identify signs of defects caused by through damage to wood; video recording of the wood surface using video cameras using all of these three illumination zones, while on each frame the classification of signs of defects is carried out using simultaneously all of these three illumination zones; building a classification model and training a neural network to determine signs of defects using data on the classification of signs for each of the specified three backlight zones; execution by a trained neural network of predicting the presence or absence of defects based on video recording of the wood surface. 2. The method according to claim 1, in which the speed of video recording of the wood surface is from 2 to 40 frames per second. 3. The method according to claim 1, in which the overhead light zone is formed by a light flux directed from top to bottom onto the surface of the wood, forming illumination to detect signs of defects caused by changes in the color of the wood surface. 4. The method according to claim 1, in which the side light zone is formed by a light flux directed along the surface of the wood to create a contrasting cut-off line to identify signs of defects caused by changes in the geometry of the wood surface. 5. The method according to claim 1, in which the lower light zone is formed by a light flux directed from bottom to top to the opposite side of the wood to create gaps and identify signs of defects caused by through damage to the wood. 6. Method according to any one of paragraphs. 3-5, in which light fluxes are formed by a directional light source and/or a linear source of scattered light. 7. The method according to claim 1, in which the video recording speed in the specified range is changed depending on the number of signs of defects to be classified, namely, when the number of signs decreases, the video recording speed is reduced and when the number of signs increases, the video recording speed is increased. 8. The method according to claim 1, in which the lighting zones are located sequentially in the direction of wood movement.