Log identification system and log identification method

Abstract

To provide a log identification system and a log identification method that enable log traceability without attaching an electronic tag or a similar device to a log.SOLUTION: A log identification system 1 comprises an image processing unit 2, a registration unit 3, and an identification unit 4. The image processing unit 2 calculates the frequency spectrum contained within the irregular waveform representing the irregularity of a butt end contour, based on a butt end image that is an image of a butt end of a log M. The registration unit 3 registers each log M into a database by associating frequency spectrum FS1, FS2, FS3, ... calculated by the image processing unit 2 with information of a first location A1, A2, A3, ... based on the butt end images of the log M captured at the first location A1, A2, A3, .... The identification unit 4 uses the frequency spectrum FS calculated by the image processing unit 2 based on the butt end images of the log M recorded at a second location B1, B2, B3, B4, ..., which is different from the first location A1, A2, A3, ..., to identify the log M corresponding to the frequency spectrum FS by referring to a database 5.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a log identification system and a log identification method. [Background technology]

[0002] As a measure against illegal logging, efforts are being made to make it mandatory to confirm the legality of timber and to promote the use of legally harvested timber. Timber traceability is essential to confirm the legality of timber and improve the reliability of legally harvested timber, and the identification and management of logs using identification information is important. To date, timber traceability technology using electronic tags has been devised (for example, see Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Yohei Soehara and three others, "Study on the architectural timber database, Part 2: Verification of the timber traceability system in Nagano and Niigata prefectures," [online], July 2014, Research Report of the Hokuriku Branch of the Architectural Institute of Japan, [Retrieved March 23, 2024], Internet <URL:https: / / www.shinshu-u.ac.jp / faculty / engineering / chair / arch009 / old / doc / reserch / traceability / pdf2.pdf> Summary of the Invention [Problem to be solved by the invention]

[0004] However, using electronic tags requires the work of attaching the tags to the logs, and additional costs are incurred, such as the cost of manufacturing the tags.Furthermore, there is a risk that the tags may be stolen and used fraudulently on other logs.

[0005] The present invention has been made in light of the above-mentioned circumstances, and aims to provide a log identification system and a log identification method that can achieve log traceability without attaching electronic tags or the like to the logs. [Means for solving the problem]

[0006] In order to achieve the above object, a log identification system according to a first aspect of the present invention comprises: an image processing unit that calculates a frequency spectrum included in an uneven waveform that represents the unevenness of the contour of the buttock based on an image of the buttock of the log; a registration unit that associates the frequency spectrum calculated by the image processing unit with information about the first location based on an image of the end grain of the log captured at the first location and registers the information in a database for each log; an identification unit that refers to the database and identifies a log corresponding to the frequency spectrum calculated by the image processing unit based on an image of an end grain of a log captured at a second location different from the first location; Equipped with.

[0007] The image processing unit a region extraction unit that generates a cross-section region image by extracting a cross-section region, which is a region of the cross-section excluding the bark, from the cross-section image using a segmentation model that has been trained in advance; a contour extraction unit that extracts a contour of the end grain region from the end grain region image; A waveform generating unit that rotates a half line starting from the center of gravity of the butt end region around the center of gravity and generates a concave-convex waveform with the rotation angle of the half line as the horizontal axis and the distance from the center of gravity to the intersection of the half line and the contour as the vertical axis; a frequency conversion unit that performs frequency conversion on the concave-convex waveform to generate the frequency spectrum, This may also be the case.

[0008] the waveform generating unit standardizes the amplitude of the concave and convex waveform based on a predetermined standard; the frequency conversion unit performs frequency conversion on the amplitude-standardized concave-convex waveform. This may also be the case.

[0009] the waveform generating unit generates the concave-convex waveform using the half-lines passing through the pixels constituting the contour of the end grain region image as sampling points, and performs resampling by interpolating the sampling points so that the difference in rotation angle between adjacent sampling points becomes equal; the frequency conversion unit performs frequency conversion on the resampled concave-convex waveform. This may also be the case.

[0010] the identification unit refers to the database and identifies the first location associated with the frequency spectrum using a k-nearest neighbor algorithm. This may also be the case.

[0011] A log identification method according to a second aspect of the present invention comprises: A log identification method executed by an information processing device, comprising: an image processing step of calculating a frequency spectrum included in a concave-convex waveform representing the concave-convex contour of the end grain based on an end grain image, which is an image of the end grain of the log; a registration step of associating the frequency spectrum calculated in the image processing step with information on the first location based on an image of the end grain of the log captured at the first location and registering the information in a database for each log; an identification step of referring to the database and identifying a log corresponding to the frequency spectrum using the frequency spectrum calculated in the image processing step based on an image of an end grain of a log captured at a second location different from the first location; Includes: [Effects of the Invention]

[0012] According to the present invention, traceability of logs can be achieved without attaching electronic tags or the like to the logs. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a block diagram showing the functional configuration of a log identification system according to an embodiment of the present invention. [Figure 2] 2 is a block diagram showing the functional configuration of an image processing unit that constitutes the log identification system of FIG. 1. FIG. [Figure 3] 1A is a diagram showing an example of an image of the buttocks of a log, FIG. 1B is a diagram showing an example of an image of the buttocks region, and FIG. 1C is a diagram showing an example of an image of the buttocks contour. [Figure 4] 10A, 10B, and 10C are schematic diagrams showing an example of a process for extracting the contour of the butt end. [Figure 5] Graph (A) shows an example of a concave-convex waveform, and graph (B) shows an example of a standardized concave-convex waveform. [Figure 6] 10A is a diagram showing an example of the rotation angle interval between sampling points, and FIG. 10B is a graph showing an example of linear interpolation. [Figure 7] 10 is a graph showing an example of a frequency spectrum of the concave-convex waveform of the contour of the butt end. [Figure 8] FIG. 1 is a schematic diagram illustrating an example of the k-nearest neighbor algorithm. [Figure 9] FIG. 2 is a block diagram showing the hardware configuration of the log identification system of FIG. 1. [Figure 10] 2 is a flowchart of a log identification process executed by the log identification system of FIG. 1. [Figure 11] FIG. 10 is a diagram showing an example of an evaluation result of log identification. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In each drawing, the same or equivalent parts are denoted by the same reference numerals. In the following embodiments, the terms "have," "include," or "contain" also mean "consist of" or "consist of."

[0015] A mountain yard is a place where wood scattered in a logging area after forests have been felled or cut down and processed is temporarily stored for processing or the next transportation. At mountain yard A1 to A3, which is the first location shown in Figure 1, trees are felled or cut down and processed using a processor or similar device to produce logs M. A processor is a large machine that removes branches from felled trees, measures them, and cuts them into bucks. The tree species can be, for example, cedar, but is not limited to this.

[0016] The logs M are distributed and used at various locations, such as sawmills B1 and B2, plywood factory B3, and biomass power plant B4, which are second locations. Log identification system 1, which serves as an information processing device according to this embodiment, identifies the yard from which the logs M sent to the sawmills B1 and B2, plywood factory B3, and biomass power plant B4, which are second locations, were produced, i.e., the first location from which the logs M were produced.

[0017] <Information generation section> In the log identification system 1, an information generation unit 10 is provided at each of the first and second locations. The information generation unit 10 captures an image of the butt end (see FIG. 3(A)), which is the cut surface of the log M. For example, the information generation unit 10 provided at the mountain yard A1 captures an image of the butt end of the log M produced at the mountain yard A1.

[0018] The information generating unit 10 generates transmission data 6 by adding information about the first location to an image of the end grain of the log M. This information may include, for example, information about the forest owner or location information about the first location. A registration / identification flag is also added to the transmission data 6. The information generating unit 10 installed at the first location sets the registration / identification flag to "register." The information generating unit 10 installed at the second location sets the registration / identification flag to "identify."

[0019] For example, the information generating unit 10 installed in the mountain yard A1 captures an image of a log M produced in the mountain yard A1, adds information about the mountain yard A1 (owner and location information about the mountain yard A1, hereinafter referred to as "information A1" as appropriate) to the image, and generates transmission data 6 with the registration / identification flag set to "registered." Also, the information generating unit 10 installed in the sawmill B1 generates an image of a log M placed in the sawmill B1, adds information about the sawmill B1 (owner and location information) to the image, and generates transmission data 6 with the registration / identification flag set to "identified." Transmission data 6 is generated for each log M.

[0020] The information generating unit 10 can be realized, for example, by a mobile terminal 20 with a camera (see FIG. 9). An image of the butt end of the log M is captured by the mobile terminal 20, and application software installed on the mobile terminal 20 adds information about the first location A1, A2, A3, ... or the second location B1, B2, B3, B4, ... and a registration / identification flag to the captured image of the butt end, thereby generating transmission data 6. However, the information generating unit 10 may also be configured with cameras and information processing devices installed at the first and second locations.

[0021] Furthermore, the log identification system 1 includes an image processing unit 2, a registration unit 3, a recognition unit 4, and a database 5 to identify the logs M. The image processing unit 2, the registration unit 3, the recognition unit 4, and the database 5 can be realized by a server computer 30 (see FIG. 9).

[0022] <Image processing unit> The image processing unit 2 receives the transmission data 6 transmitted from the information generation unit 10. Based on the image of the buttock of the log M included in the transmission data 6, the image processing unit 2 determines the frequency spectrum contained in the uneven waveform representing the unevenness of the buttock's contour. The shape of the buttock is not a perfect circle, and the shape of the contour changes depending on the growing environment, weather conditions, differences in the growth rate of the annual ring width, knots, etc. The unevenness of the buttock's contour changes due to the above-mentioned influences even in trees with the same genetic makeup. The log identification system 1 utilizes the fact that the buttock's contour unevenness is unique to each log, like a fingerprint, to identify the log M based on the frequency spectrum contained in the uneven waveform representing the contour unevenness.

[0023] <Registration Department> The registration unit 3 associates the frequency spectra FS1, FS2, FS3, ... calculated by the image processing unit 2 with the first points A1, A2, A3, ... based on the end grain images, which are images of the end grains of the logs M captured at the first points A1, A2, A3, ..., and registers them in the database 5. As a result, pairs of the first points A1, A2, A3, ... and the frequency spectra FS1, FS2, FS3, ... are registered in the database 5.

[0024] For example, when transmission data 6 is sent from information generation unit 10 of mountain yard A1, image processing unit 2 determines frequency spectrum FS1 contained in the uneven waveform representing the unevenness of the contour of the end grain, based on the image included in the transmission data 6. Since this transmission data 6 has information about mountain yard A1 and a registration / identification flag (registration) added, image processing unit 2 outputs frequency spectrum FS1 and the information about mountain yard A1 to registration unit 3. Registration unit 3 associates the information about mountain yard A1 (shown as information A1 in FIG. 1) with frequency spectrum FS1 and registers them in database 5. Similarly, information (A2) about mountain yard A2 and a frequency spectrum FS2 included in the uneven waveform representing the unevenness of the contour of the buttocks of logs M produced at mountain yard A2 are associated and registered, and information (A3) about mountain yard A3 and a frequency spectrum FS3 included in the uneven waveform representing the unevenness of the contour of the buttocks of logs M produced at mountain yard A3 are associated and registered. Each set of information is assigned a number.

[0025] <Identification section> The identification unit 4 uses the frequency spectrum FS calculated by the image processing unit 2 based on the end grain image, which is an image of the end grain of the log M taken at second points B1, B2, B3, B4, etc. different from the first points A1, A2, A3, etc., to refer to the database 5 and identify the log M corresponding to the frequency spectrum FS.

[0026] For example, when transmission data 6 is sent from the information generating unit 10 of a sawmill B1, the image processing unit 2 calculates a frequency spectrum FS included in the concave-convex waveform representing the concave-convex contour of the buttocks based on the image included in the transmission data 6. Since this transmission data 6 is accompanied by information about the sawmill B1 and a registration / identification flag (identification), the image processing unit 2 outputs the frequency spectrum FS and the information about the sawmill B1 to the identification unit 4. The identification unit 4 refers to the database 5 and uses the frequency spectrum FS calculated by the image processing unit 2 based on the buttock images, which are images of the buttocks of the log M captured at second points B1, B2, B3, B4, etc., different from the first points A1, A2, A3, etc., to identify the log M corresponding to the frequency spectrum FS.

[0027] For example, among the frequency spectra FS1, FS2, FS3, ... registered in the database 5, the log M associated with the frequency spectrum closest to the frequency spectrum FS is searched for, and the first location where the log M was produced is identified. For example, if the frequency spectrum FS is closest to the frequency spectrum FS1, the first log M associated with the frequency spectrum FS1 is searched for, and the lumber yard A1 where the log M was produced is identified. This achieves the traceability of the log M.

[0028] <Detailed configuration of the image processing unit> As shown in Fig. 2, the image processing unit 2 includes a region extraction unit 11, a contour extraction unit 12, a waveform generation unit 13, and a frequency conversion unit 14. Note that, hereinafter, the image of the butt end of the log M (see Fig. 3(A)) included in the transmission data 6 transmitted from the information generation unit 10 will also be referred to as butt end image Im1. The horizontal coordinate axis of butt end image Im1 is set to X [pixel], and the vertical coordinate axis is set to Y [pixel].

[0029] <Area extraction part> The region extraction unit 11 extracts a buttock region, which is the region of the buttock excluding the bark, from the buttock image Im1 using a segmentation model 40 that performs pre-trained semantic segmentation. Note that the buttock image Im1 is pre-trimmed into a square and resized to a predetermined size, for example, 512 x 512 pixels, before being input to the segmentation model 40. FIG. 3(B) shows an buttock region image Im2 that indicates the extracted buttock region PA. In FIG. 3(B), the white region with high brightness represents the buttock region PA, and the black region with low brightness represents the background.

[0030] <Segmentation> Segmentation is the problem of classifying an image at the pixel level. Semantic segmentation is the task of assigning object regions and object names (here, end grain) at the pixel level for multiple objects contained in a single image. Semantic segmentation divides an image into regions by object type, but it is not possible to divide an image into regions for individual objects of the same class. Therefore, in the log identification system 1, the object contained in the end grain image Im1 is limited to a single end grain.

[0031] The region extraction unit 11 also detects the buttock area PA excluding the bark (the outline area of ​​the outermost annual rings). There are two classes to classify: the background area and the buttock area PA. The reason for extracting the buttock area PA excluding the bark is that the bark peels easily, and if the buttock area PA is included as the target, the outline of the buttock will easily change.

[0032] The segmentation model 40 is generated by pre-learning. Prior to this pre-learning, for example, images of the end grains of multiple logs M are prepared, and supervised learning data is generated in which the pixels of the images are assigned attributes indicating whether they represent end grain or background. The segmentation model 40 is trained based on this learning data, and the region extraction unit 11 uses the pre-learned segmentation model 40 to generate an end grain region image Im2 shown in FIG. 3(B) in which an end grain region PA has been generated from the end grain image Im1 of the log M.

[0033] <Contour extraction section> The contour extraction unit 12 extracts the contour PO of the buttock area PA extracted by the area extraction unit 11. For example, pixels corresponding to the buttock outline PO shown in FIG. 3(C) are extracted from the buttock area PA of the buttock area image Im2 shown in FIG. 3(B). As shown in FIG. 4(A), the pixels corresponding to the extracted buttock outline PO are composed of the pixels located on the outermost edge among the pixels that make up the buttock area PA. FIG. 3(C) shows an example of a contour image Im3 that shows the outline PO.

[0034] <Waveform generation section: Extraction of concave and convex waveforms by rotating the radius around the center of gravity> The waveform generating unit 13 calculates the center of gravity G of the buttock area PA. The outline PO and its center of gravity G are shown in Figure 4(B). The center of gravity G is the area inside the outline PO that is filled in, i.e., the center of gravity G of the buttock area PA. The X coordinate of the center of gravity G is calculated by dividing the sum of (pixel brightness values ​​x X coordinates) in the buttock area PA by the sum of brightness values, and the Y coordinate of the center of gravity G is calculated by dividing the sum of (pixel brightness values ​​x Y coordinates) in the buttock area PA by the sum of brightness values.

[0035] As shown in Fig. 4(B), the waveform generation unit 13 draws a ray L starting from the center of gravity G to any one pixel (reference point RF) corresponding to the contour PO. The waveform generation unit 13 sets the position connecting the center of gravity G and the reference point RF as the start position and rotates the ray L counterclockwise around the center of gravity G within the image as shown in Fig. 4(B). Then, when the ray L intersects with the pixel corresponding to the contour PO, the waveform generation unit 13 samples the distance between the pixel corresponding to the contour PO and the center of gravity G, and generates a point group of sampling points S1 with the distance as the vertical axis.

[0036] Figure 5(A) shows a concave-convex waveform W1, which is a point cloud of sampling points S1. The concave-convex waveform W1 shown in Figure 5(A) is a waveform that expresses the concave-convex of the end grain contour PO, and is calculated by calculating the distance from the center of gravity G to the pixels corresponding to the contour PO over one circumference. As shown in Figure 5(A), the vertical axis of the concave-convex waveform W1 represents the distance in pixels from the center of gravity G to the pixels that make up the end grain contour PO, and the horizontal axis represents the number of pixels (sampling points S1) that correspond to the contour PO that have been passed so far. Euclidean distance is used to calculate the distance.

[0037] Incidentally, a log M has a center of annual rings called the pith. In this embodiment, the center of gravity G, rather than the pith, is used as the distance reference for calculating the concave-convex waveform. One reason for this is that once the contour PO is extracted, the center of gravity G can be determined as a single point, eliminating the need to additionally detect the pith. The main reason is that generating the concave-convex waveform W1 based on the center of gravity G more accurately represents the concave-convex of the contour PO. The position of the pith in the buttock region PA is biased. For this reason, if the concave-convex waveform W1 is generated based on the distance from the pith to the pixel corresponding to the contour PO, the distance from the center of gravity G to the contour PO will be shorter on the contour PO closer to the pith, while the distance from the center of gravity G to the contour PO will be longer on the opposite side, resulting in the frequency spectrum containing frequency components due to the bias in the pith that are unrelated to the concave-convex of the contour PO.

[0038] <Standardization> The waveform generator 13 standardizes the amplitude of the concave-convex waveform W1 (the group of sampling points S1) using a predetermined standard. Specifically, the waveform generator 13 sorts the generated concave-convex waveform W1 starting from the position where the wave height is smallest and standardizes the amplitude to generate the concave-convex waveform W2. The concave-convex waveform W2 converted from the concave-convex waveform W1 in FIG. 5(A) is shown in FIG. 5(B). A log M has no definition of up, down, left, or right, and its orientation changes as it rotates during transportation. If the orientation of the end grain in the end grain image changes, the starting position of the concave-convex waveform W1 also changes. Therefore, by sorting the concave-convex waveform W1 starting from the smallest value, even if the orientation of the end grain in two end grain images of the same log M taken at different times is different, the concave-convex waveform W2 extracted from each end grain image can be easily compared by aligning the starting position of the concave-convex waveform W1 and generating the concave-convex waveform W2.

[0039] The uneven waveform W1 is standardized to ignore changes in the size of the end grain shown in the end grain image Im1 (see Figure 3(A)). The distance on the vertical axis of the uneven waveform W1 is the distance from the center of gravity G to the pixel corresponding to the outline PO, calculated in pixel units. However, even for the same end grain, if the size of the end grain shown in the end grain image Im1 is different, the distance from the center of gravity G to the outline PO will also be different. Therefore, the waveform generation unit 13 standardizes the uneven waveform W1 to generate the uneven waveform W2. In this way, even if the size of the end grain shown in two images taken at different times of the same log M is different, the distance from the center of gravity G to the outline PO can be made uniform. Note that standardization is performed using the average value and standard deviation of the uneven waveform W1 as a predetermined criterion, and aligning these average values ​​and standard deviations to the same level.

[0040] As shown in FIG. 5(A), the horizontal axis of the concave-convex waveform W1 represents the number of pixels corresponding to the contour PO through which the half-line L passed during rotation. However, the number of pixels constituting the contour PO varies depending on the size of the end grain image Im1. The waveform generator 13 converts this horizontal axis from the number of pixels constituting the contour PO shown in FIG. 5(A) to the rotation angle of the half-line L shown in FIG. 5(B). Note that the waveform generator 13 may also sample the rotation angle θ of the half-line L when sampling the sampling point S1, and use the rotation angle θ as the horizontal axis of the concave-convex waveform W1.

[0041] <Resampling> As shown in Figure 6(A), the rotation angle θ at which each sampling point S1 is obtained is not evenly spaced. In the concave-convex waveform W1 shown in Figure 5(A), the horizontal axis represents the number of pixels and is expressed at equal intervals, but by setting the horizontal axis of the concave-convex waveform W2 to the rotation angle θ, it is possible to express the concave-convex waveform W2 with accurate positional relationship within the image. Note that the concave-convex waveform W2 shown in Figure 5(B) is actually a waveform formed by connecting the sampling points S1 using linear interpolation.

[0042] In this way, as shown in Figure 6(B), the waveform generation unit 13 standardizes and converts the horizontal axis into a rotation angle θ, linearly interpolates the sampling points S1 to generate the above-mentioned concave-convex waveform W2, and then resamples the concave-convex waveform W2 so that the rotation angle θ, which represents the sampling interval, becomes uniform, thereby generating a concave-convex waveform W3 composed of the sampling points S2.

[0043] Linear interpolation is a process of linearly interpolating the empty spaces between discrete data. Resampling is the process of re-taking samples from linearly interpolated waveform data at a new interval. In Figure 6(B), sampling points S1 are shown as 11 large dots. The waveform generation unit 13 interpolates between the sampling points S1 with line segments to generate 15 new sampling points S2 at the new interval.

[0044] The waveform generator 13 performs linear interpolation resampling of the concave-convex waveform W2 so that the number of samples at the sampling points S2 is 1024. The reason for the resampling number being 1024 is that the number of pixels corresponding to the end grain contour PO in the 512 x 512 pixel end grain image was approximately 1000, and because the fast Fourier transform requires a sampling number of 2 n . Equally spaced sampling points S2 of the concave-convex waveform W3 enable fast Fourier transform signal processing. It is also very important that the number of samples for the concave-convex waveform W3 can be fixed between images. As described below, log M is identified based on distance in a feature space, where each frequency component of the frequency spectrum is an element. However, distance calculation is impossible if the number of dimensions of each frequency component of the frequency spectrum is not consistent. Therefore, it is desirable to fix the number of samples for the concave-convex waveform W3 to 1024.

[0045] <Frequency conversion section> The frequency conversion unit 14 performs frequency conversion on the concave and convex waveform W3 formed by the sampling points S2 to generate a frequency spectrum FS. The frequency spectrum FS is obtained by decomposing the time signal into frequency components and arranging the intensities of the frequency components in an easily visible manner.

[0046] The frequency converter 14 converts the concave-convex waveform W3 into a frequency spectrum FS using a fast Fourier transform. Fast Fourier transform is an algorithm that performs discrete Fourier transform processing at high speed. The frequency converter 14 uses fast Fourier transform to decompose a signal into several frequency components and represent the magnitude of each component as a spectrum. Figure 7 shows an example of the concave-convex waveform W3 converted into a frequency spectrum FS. Since the concave-convex contour PO is unique to the log M, the log M can be identified by determining the frequency components that make up the concave-convex contour PO as a feature. The concave-convex waveform W3 resampled using linear interpolation is a discrete signal with 1024 samples. Therefore, the number of dimensions of the frequency spectrum obtained by fast Fourier transforming it is 512. This frequency spectrum FS is vector data in a 512-dimensional feature space that indicates the amplitude of each frequency component from DC to 511 frequencies.

[0047] <Details of the identification unit> The log identification system 1 identifies the log M using the frequency spectrum (vector of 512 - dimensional frequency components in the frequency spectrum of the uneven waveform W3) obtained as described above. Returning to FIG. 1, the identification unit 4 refers to the database 5 and uses the k - nearest neighbor method to identify the log M associated with the frequency spectrum. The identification unit 4 searches for the log M using the frequency spectrum FS as a feature amount and identifies the first location where the log M was produced.

[0048] <k - nearest neighbor method> The k - nearest neighbor method is one of the algorithms of supervised learning. The k - nearest neighbor method is a typical example of lazy learning, and it is a method that can use the training data set as it is without learning a discriminant function from the training data. As shown in FIG. 8, the k - nearest neighbor method finds the k training data that are closest (most similar) to the data point to be classified from the training data set based on the selected distance metric, and determines the class label of the data point by majority voting among the k nearest neighbors.

[0049] In the k - nearest neighbor method used in the log identification system 1 according to the present embodiment, the value of k is set to 1, and the Euclidean distance is used as the distance metric. A set of information on the first locations A1, A2, A3, ··· and the frequency spectra FS1, FS2, FS3, ··· of the uneven waveform W3 obtained from the photographed end - face image Im1 can be registered in the database 5 as a training data set. The identification unit 4 generates the frequency spectrum FS of the uneven waveform W3 based on the end - face image Im1 photographed at the second location, and searches for the log M of the class that is closest (most similar) to the frequency spectrum FS among the frequency spectra FS1 to FS3 registered in the database 5 in the vector space.

[0050] 8, based on the end grain image Im1 captured at the second location, the identification unit 4 calculates the distance (similarity) in feature space between the frequency spectrum FS of the concave-convex waveform W3 of the end grain contour PO and the frequency spectra FS1, FS2, FS3, ... of No. 0, No. 1, No. 2, ... registered in the database 5. Among these, the distance (similarity) in feature space with the frequency spectrum of No. 0 is 0.05, the smallest, so the log M corresponding to the frequency spectrum of No. 0 is searched for, and the first location (A1) where that log M was produced is identified.

[0051] <Hardware configuration> The log identification system 1 shown in Fig. 1 is realized, for example, by a mobile terminal 20 and a server computer 30 having the hardware configuration shown in Fig. 9 executing a software program. Specifically, the mobile terminal 20 includes a CPU (Central Processing Unit) 21, which is a processor that controls the entire device, a main memory 22 such as RAM (Random Access Memory), an external memory 23 consisting of non-volatile memory such as flash memory or a hard disk, a camera 27 that takes images of the end grain, a communication interface 26 that communicates data with the server computer 30, and an internal bus 28 that connects these.

[0052] The program 29 is loaded from the external memory 23 into the main memory 22 and executed by the CPU 21. This realizes the functions of the information generation unit 10. When executing the program 29, the CPU 21 performs data communication with an external computer via the communication interface 26 as necessary. In this embodiment, the program 29 executed by the CPU 21 includes the program of the information generation unit 10.

[0053] The functions of the information generation unit 10 can be implemented in a computer system consisting of one or more computers including one or more processors and one or more storage devices including non-transitory storage media. The multiple computers realize the functions of the information generation unit 10 while communicating via an interconnected communication network. For example, some of the functions of the information generation unit 10 may be implemented in one computer, and other parts may be implemented in other computers. The functions of the information generation unit 10 may also be realized by a cloud computer.

[0054] The image processing unit 2, registration unit 3, and identification unit 4 shown in Fig. 1 are realized, for example, by a server computer 30 having the hardware configuration shown in Fig. 9 executing a software program. Specifically, the server computer 30 includes a CPU (Central Processing Unit) 31, a main memory 32 of the CPU 31, an external memory 33 that stores a program 39, an operation unit 34 that is a device such as a keyboard and a mouse, a display 35 that is a display device such as a CRT (Cathode Ray Tube) or a liquid crystal monitor, a communication interface 36 that performs data communication with other computers, and an internal bus 38 that connects these.

[0055] The program 39 is loaded from the external memory 33 into the main memory 32 and executed by the CPU 31. This realizes the server computer 30. When executing the program 39, the CPU 31 performs data communication with an external computer via the communication interface 36 as necessary. In this embodiment, the program 39 executed by the CPU 31 includes programs for the image processing unit 2, the registration unit 3, the recognition unit 4, and the database 5.

[0056] <Log identification method> Next, the operation of the log identification system 1, i.e., the log identification method, will be described. As shown in Fig. 10, the image processing unit 2 waits until it receives transmission data 6 (step S1; No). The information generation units 10 installed at the first points A1, A2, A3, ... and the second points B1, B2, B3, B4, ... generate transmission data 6 including an image of the log M, information about the point, and a registration / identification flag, and transmit the transmission data 6 to the image processing unit 2.

[0057] Upon receiving the transmission data 6 (Step S1; Yes), the image processing unit 2 obtains a frequency spectrum included in an uneven waveform that represents the unevenness of the contour PO of the buttock based on an image of the buttock of the log M (Steps S2 to S5: image processing step). In the image processing step, the region extraction unit 11 first uses a segmentation model 40 that has been trained in advance to generate an buttock area image Im2 in which the buttock area PA, which is the area of ​​the buttock excluding the bark, is extracted from the buttock area image Im1 (Step S2; region extraction step). Next, the contour extraction unit 12 extracts pixels corresponding to the contour PO of the buttock area PA from the buttock area image Im2 (Step S3; contour extraction step). Next, the waveform generation unit 13 draws a ray L between the center of gravity G of the end grain area PA and any one of the pixels corresponding to the contour PO, and rotates the ray L within the image around the center of gravity G. The waveform generation unit 13 samples the rotation angle θ of the ray L when the ray L intersects with the pixel and the distance between the pixel intersecting the ray L and the center of gravity G, and generates a concave-convex waveform W3 with the rotation angle θ as the horizontal axis and the distance as the vertical axis (step S4; concave-convex data generation step). Next, the frequency conversion unit 14 performs frequency conversion on the concave-convex waveform W3 to generate a frequency spectrum (step S5; frequency spectrum generation step).

[0058] Next, the image processing unit 2 determines whether the registration / identification flag is "registration" (step S6). If it is registration (step S6; Yes), the image processing unit 2 outputs the frequency spectrum and information about the first location to the registration unit 3. The registration unit 3 associates the frequency spectrum calculated by the image processing unit 2 with the first location based on the end grain image, which is an image of the end grain of the log M captured at the first location, and registers it in the database 5 (step S7; registration step). After step S7 is completed, the log identification system 1 returns to step S1.

[0059] If it is identification rather than registration (step S6; No), the image processing unit 2 outputs the frequency spectrum and information about the second location to the identification unit 4. The identification unit 4 refers to the database 5 using the frequency spectrum FS calculated by the image processing unit 2 based on the end grain image Im1, which is an image of the end grain of the log M captured at the second location different from the first location, and identifies the log M corresponding to the frequency spectrum FS, and ultimately the first location (step S8; identification step). After step S7 is completed, the log identification system 1 returns to step S1.

[0060] <Evaluation results> The log identification system 1 according to this embodiment was evaluated.

[0061] (Evaluation of pre-training in segmentation) The 250 end grain images used for pre-learning of segmentation model 40, which extracts the end grain region, were divided into 200 training end grain images and 50 verification end grain images, and these were used for pre-learning of segmentation model 40. The end grain region PA was enclosed in end grain image Im1 using operational input to generate a polygon for the end grain region PA, and annotation was performed by converting this information into an image that would become training data.

[0062] The segmentation model 40 used Unet from Segmentation Models Pytorch:0.3.3. The region extraction unit 11 inputs a 512 x 512 pixel end grain image Im1 into the segmentation model 40, outputting a 512 x 512 pixel end grain region image Im2. IoU (Intersection over Union) was used as the evaluation index for semantic segmentation. IoU is an index that represents the degree of overlap between the actual end grain region PA and the estimated end grain region PA. The average IoU for the end grain image Im1 used by the segmentation model 40 to identify log M was approximately 0.9941, demonstrating that the end grain region PA can be detected with high accuracy.

[0063] (Evaluation of log M identification) We evaluated the identification of log M using the frequency spectrum FS of the uneven waveform W3 of the butt end contour PO. Figure 11 shows the results of a comparison between the distance in the frequency spectrum FS between the same log M and the distance in the frequency spectrum FS between different logs M in the feature space of the frequency spectrum. As shown in Figure 11, the distance in the feature space of the frequency spectrum between the same logs M is almost 0, and they match well. In contrast, the distance in the frequency spectrum FS between different logs M is large. This demonstrates that log M can be identified with high accuracy using the frequency spectrum FS of the uneven waveform W3 of the butt end contour PO.

[0064] (summary) (1) As described above in detail, the log identification system 1 according to this embodiment includes an image processing unit 2, a registration unit 3, and an identification unit 4. The image processing unit 2 calculates a frequency spectrum FS included in an uneven waveform W3 representing the unevenness of the contour PO of the end grain, based on an end grain image Im1, which is an image of the end grain of the log M. The registration unit 3 associates the frequency spectra FS1, FS2, FS3, etc. calculated by the image processing unit 2 based on the end grain image Im1 of the log M captured at first points A1, A2, A3, etc. with information about the first points A1, A2, A3, etc., and registers them in a database 5 for each log M. The identification unit 4 refers to the database 5 using the frequency spectrum FS calculated by the image processing unit 2 based on the end grain image Im1 of the log M captured at second points B1, B2, B3, B4, etc., different from the first points A1, A2, A3, etc., to identify the log M corresponding to the frequency spectrum FS. In this way, the frequency spectra FS1, FS2, FS3, ... of the uneven waveform (end grain contour unevenness) representing the end grain contour PO obtained from the log M at the first points A1, A2, A3, ... are stored in database 5 in association with the information of the first points A1, A2, A3, ..., and the frequency spectrum FS of the end grain contour unevenness is obtained again from the log M at the second points B1, B2, B3, B4, ..., and a matching log M in database 5 can be searched for, thereby achieving traceability of the log M. As a result, traceability of the log M can be achieved without attaching an electronic tag or the like to the log M.

[0065] Traceability of logs M using the log identification system 1 according to this embodiment is inexpensive because it does not require the use of electronic tags or the like. Furthermore, the end grain contour irregularities of the log M, which are biometric characteristics, are difficult to rewrite, just like human fingerprints, so the log identification system 1 can achieve highly reliable timber traceability. It is expected that inexpensive and highly reliable timber traceability using the log identification system 1 according to this embodiment will contribute to compliance with the Clean Wood Act, which requires confirmation of timber legality and separate management, and to measures against illegal logging.

[0066] According to the log identification system 1 of this embodiment, the log M is identified using the frequency spectrum FS of the uneven waveform W3 of the buttock contour PO, making it possible to identify the log M with high accuracy regardless of the size and orientation of the log M in the buttock image Im1.

[0067] (2) In the log identification system 1 according to this embodiment, the image processing unit 2 includes a region extraction unit 11 that uses a pre-trained segmentation model 40 to generate an end grain region image Im2 in which an end grain region PA, which is the region of the end grain excluding the bark, is extracted from an end grain image Im1, a contour extraction unit 12 that extracts an outline PO of the end grain region PA from the end grain region image Im2, a waveform generation unit 13 that rotates a half line L starting from the center of gravity G of the end grain region PA around the center of gravity G to generate a concave-convex waveform W3 in which the horizontal axis represents the rotation angle θ of the half line L and the vertical axis represents the distance from the center of gravity G to the intersection of the half line L and the outline PA, and a frequency conversion unit 14 that performs frequency conversion on the concave-convex waveform W3 to generate a frequency spectrum FS. In this way, the end grain region PA can be accurately extracted using the pre-trained segmentation model 40, regardless of changes in the bark.

[0068] Furthermore, the log identification system 1 according to this embodiment generates a concave-convex waveform W3 of the outline PO centered at the center of gravity G, so there is no need to detect the pith of the end grain. Because the center of gravity G is less biased relative to the outline PO than the pith, obtaining the concave-convex waveform W3 centered at the center of gravity G makes it less likely that low frequency components due to central bias will be included in the frequency spectrum FS of the concave-convex waveform W3 than obtaining the concave-convex waveform W3 centered at the pith. This improves the search accuracy using the frequency spectrum FS.

[0069] (3) In the log identification system 1 according to this embodiment, the waveform generator 13 standardizes the amplitude of the concave-convex waveform W1 of the contour PO according to a predetermined standard. The frequency converter 14 performs frequency conversion on the amplitude-standardized concave-convex waveform W3. This allows the frequency spectrum FS to be calculated with high accuracy from the concave-convex waveform W3 of the contour PO of the end grain, regardless of differences in the size of the logs M for each end grain image Im1.

[0070] (4) In the log identification system 1 according to this embodiment, the waveform generator 13 generates a concave-convex waveform W2 using half-lines L passing through the pixels that make up the contour PO of the buttock region image Im2 as sampling points, and then performs resampling by interpolating the sampling points S1 so that the difference in rotation angle between adjacent sampling points is equal, generating a concave-convex waveform W3. The frequency converter 14 performs frequency conversion on the resampled concave-convex waveform W3. This reduces errors in the frequency spectrum FS due to unevenness in the sampling points S1, making it possible to accurately calculate the frequency spectrum FS from the concave-convex waveform W3 of the buttock contour PO.

[0071] (5) In the log identification system 1 according to this embodiment, the identification unit 4 refers to the database 5 and identifies the first points A1, A2, A3, ... associated with the frequency spectrum FS using the k-nearest neighbor method. In this way, it is possible to identify the log M from the frequency spectrum FS without performing pre-learning.

[0072] In the above embodiment, the log identification system 1 is assumed to include a single image processing unit 2. However, this is not limited to this. As with the information generating unit 10, an image processing unit 2 may be provided for each of the first location and the second location.

[0073] In the above embodiment, the rotation direction of the half line L, which is rotated around the center of gravity G to generate the concave-convex waveform W3 from the contour PO of the end grain, is uniformly counterclockwise. However, the rotation direction of the half line L may also be uniformly clockwise.

[0074] In the above embodiment, the first point is the mountain yard where the log M is produced, but this is not limited to this. There are no particular restrictions on the first and second points as long as they are places where the log M is produced or distributed. For example, an intermediate point in the transportation of the log M may be set as the first or second point, or if the log M is to be exported to a foreign country, the port where the export takes place may be set as the first or second point.

[0075] If multiple end grains are captured in an image of the end grain, the images of the end grains may be extracted one by one from the image, and the background removed to generate each end grain image Im1. Also, if the end grain image Im1 is captured from a direction in which the end grain surface is tilted relative to its normal direction, the distortion due to the capture direction of the end grain image Im1 may be corrected and the image may be converted into an image captured in the normal direction.

[0076] The hardware and software configurations of the log identification system 1 are merely examples and can be changed and modified as desired.

[0077] The core processing parts of the mobile terminal 20 and the server computer 30, which are composed of CPUs 21, 31, main memories 22, 32, external memories 23, 33, operation units 24, 34, displays 25, 35, communication interfaces 26, 36, and internal buses 28, 38, etc., can be realized using an ordinary computer system rather than a dedicated system. For example, the log identification system 1 that executes the above-described processes may be configured by storing and distributing a computer program for executing the above-described operations on a computer-readable recording medium (such as a flexible disk, CD-ROM, or DVD-ROM), and installing the computer program on a computer. Alternatively, the log identification system 1 may be configured by storing the computer program in a storage device of a server device on a communication network such as the Internet, and then downloading the program into an ordinary computer system.

[0078] When the functions of the log identification system 1 are realized by sharing the functions between an OS (operating system) and an application program, or by cooperation between the OS and the application program, only the application program portion may be stored in a recording medium or storage device.

[0079] It is also possible to superimpose a computer program on a carrier wave and distribute it over a communications network. For example, the computer program may be posted on a bulletin board system (BBS) on the communications network and distributed over the network. The computer program may then be started and executed under the control of an operating system in the same way as any other application program, thereby enabling the above-mentioned processing to be performed.

[0080] This invention allows various embodiments and modifications without departing from the broad spirit and scope of this invention. Furthermore, the above-described embodiments are intended to explain this invention and do not limit the scope of this invention. That is, the scope of this invention is defined by the claims, not the embodiments. Various modifications made within the scope of the claims and the meaning of the invention equivalent thereto are considered to be within the scope of this invention. [Industrial Applicability]

[0081] The present invention can be applied to the traceability of logs in circulation. [Explanation of symbols]

[0082] 1 Log identification system (information processing device), 2 Image processing unit, 3 Registration unit, 4 Identification unit, 5 Database, 6 Transmission data, 10 Information generation unit, 11 Region extraction unit, 12 Contour extraction unit, 13 Waveform generation unit, 14 Frequency conversion unit, 20 Mobile terminal, 21 CPU, 22 Main memory, 23 External memory, 24 Operation unit, 25 Display, 26 Communication interface, 27 Camera, 28 Internal bus, 29 Program, 30 Server computer, 31 CPU, 32 Main memory, 33 External memory, 34 Operation unit, 35 Display, 36 Communication interface, 38 Internal bus, 39 Program, 40 Segmentation model, A1, A2, A3 Logging yard (first location), B1, B2 Sawmill (second location), B3 Plywood factory (second location), B4 Biomass power plant (second location), G Center of gravity, Im1 End grain image, Im2 end grain area image, Im3 contour image, L half line, M log, PA end grain area, PO contour, S1, S2 sampling points, W1, W2, W3 uneven waveform

Claims

1. an image processing unit that calculates a frequency spectrum included in an uneven waveform that represents the unevenness of the contour of the buttock based on an image of the buttock of the log; a registration unit that associates the frequency spectrum calculated by the image processing unit with information about the first location based on an image of the end grain of the log captured at the first location and registers the information in a database for each log; an identification unit that refers to the database and identifies a log corresponding to the frequency spectrum calculated by the image processing unit based on an image of an end grain of the log captured at a second location different from the first location; A log identification system comprising:

2. The image processing unit a region extraction unit that generates a cross-section region image by extracting a cross-section region, which is a region of the cross-section excluding the bark, from the cross-section image using a segmentation model that has been trained in advance; a contour extraction unit that extracts a contour of the end grain region from the end grain region image; A waveform generating unit that rotates a half line starting from the center of gravity of the butt end region around the center of gravity and generates a concave-convex waveform with the rotation angle of the half line as the horizontal axis and the distance from the center of gravity to the intersection of the half line and the contour as the vertical axis; a frequency conversion unit that performs frequency conversion on the concave-convex waveform to generate the frequency spectrum, The log identification system of claim 1 .

3. the waveform generating unit standardizes the amplitude of the concave and convex waveform based on a predetermined standard; the frequency conversion unit performs frequency conversion on the amplitude-standardized concave-convex waveform. The log identification system of claim 2.

4. the waveform generating unit generates the concave-convex waveform using the half-lines passing through the pixels constituting the contour of the end grain region image as sampling points, and performs resampling by interpolating the sampling points so that the difference in rotation angle between adjacent sampling points becomes equal; the frequency conversion unit performs frequency conversion on the resampled concave-convex waveform. The log identification system of claim 2.

5. the identification unit refers to the database and identifies the first location associated with the frequency spectrum using a k-nearest neighbor algorithm. A log identification system according to any one of claims 1 to 4.

6. A log identification method executed by an information processing device, comprising: an image processing step of calculating a frequency spectrum included in a concave-convex waveform representing the concave-convex contour of the end grain based on an end grain image, which is an image of the end grain of the log; a registration step of associating the frequency spectrum calculated in the image processing step with information on the first location based on an image of an end grain of the log captured at the first location and registering the information in a database for each log; an identification step of referring to the database and identifying a log corresponding to the frequency spectrum calculated in the image processing step based on an image of an end grain of a log captured at a second location different from the first location; Log identification methods including.

Images