CN113674233A - Wood optimal selection saw visual detection method based on artificial intelligence - Google Patents
Wood optimal selection saw visual detection method based on artificial intelligence Download PDFInfo
- Publication number
- CN113674233A CN113674233A CN202110931750.0A CN202110931750A CN113674233A CN 113674233 A CN113674233 A CN 113674233A CN 202110931750 A CN202110931750 A CN 202110931750A CN 113674233 A CN113674233 A CN 113674233A
- Authority
- CN
- China
- Prior art keywords
- cutting
- point cloud
- knots
- characteristic signal
- signal sequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 239000002023 wood Substances 0.000 title claims abstract description 53
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 14
- 238000001514 detection method Methods 0.000 title abstract description 17
- 230000000007 visual effect Effects 0.000 title abstract description 8
- 238000005520 cutting process Methods 0.000 claims abstract description 173
- 108010076504 Protein Sorting Signals Proteins 0.000 claims abstract description 51
- 238000000034 method Methods 0.000 claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 claims abstract description 16
- 230000008859 change Effects 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000011179 visual inspection Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 description 6
- 239000007787 solid Substances 0.000 description 6
- 230000007547 defect Effects 0.000 description 5
- 238000005457 optimization Methods 0.000 description 4
- 230000035945 sensitivity Effects 0.000 description 3
- 239000002699 waste material Substances 0.000 description 3
- 238000002372 labelling Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000003776 cleavage reaction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000007017 scission Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30161—Wood; Lumber
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
Abstract
The invention relates to the technical field of artificial intelligence, in particular to a wood optimal selection saw visual detection method based on artificial intelligence. Detecting RGB images of wood to obtain a surrounding frame of a plurality of knots, and cutting the positions of the knots on the corresponding wood based on the surrounding frame to obtain an initial cutting characteristic signal sequence; acquiring a point cloud area and point cloud data corresponding to the bounding box based on the depth image to obtain a coefficient vector consisting of a dead knot coefficient; and obtaining a new cutting characteristic signal sequence by combining the initial cutting characteristic signal sequence and the coefficient vector, classifying the knots by using the new cutting characteristic signal sequence, training a deep neural network by using the knot images corresponding to different knot types to detect the knot types of the knots, and further adopting a corresponding cutting method. The nodes are subjected to feature description based on cutting load and point cloud features, more detailed automatic sample marking and classification are realized, the accuracy of deep neural network training is guaranteed, and the accuracy of optimal selection and cutting of the saw is improved.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a wood optimal selection saw visual detection method based on artificial intelligence.
Background
In the wood processing industry, the problems of high processing cost investment, uneven quality, low wood resource utilization rate, serious waste and the like of the processed solid wood board are caused by low detection efficiency of the surface defects of the solid wood board, limitation of manual batching and scribing and lack of intellectualization of an optimized sawing device.
The knots are the defects which have the greatest influence on the quality of the solid wood board, and are the most common defects. The nodes, i.e., knots, can be divided into articulations and dead knots. The wood tissue of the articulations is tightly connected with the surrounding wood without gaps, while the wood tissue of the dead knots is separated from the surrounding wood, so that the whole knots are sometimes detached to leave a hole. The existing solid wood board optimization processing system mainly takes the knot defect as a main detection target, and the optimization of the solid wood board is realized by designing a saw cutting processing system.
Most of the existing joint detection systems use a simple target detection deep neural network to realize the detection and positioning of the joint position and identify the articulated joints and the dead joints. The growth state and the texture of the knots are variable due to the random production texture of the wood, and further, the target detection algorithm using a specific object can only cover most of the cases and cannot cover some of the unusual cases, for example, a knot with high texture hardness exists in the movable joint, the included angle between the main shaft of the knot and the main shaft direction of the trunk is obvious, and cutter damage is easily caused when the cutting processing of the wood is carried out. Because the knot can not be directly marked by manually observing the acquired image, the marking difficulty of the image data is extremely high.
Because the samples corresponding to the nodes with the dead knots and the high hardness are fewer, and the nodes with the dead knots and the high hardness cannot be distinguished well by manpower, the sample marking difficulty of the deep neural network is large, the marking workload is increased, and then the optimal selection and cutting of the solid wood plate cannot be realized by the optimal selection saw system with high accuracy.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a visual detection method for wood optimal selection saw based on artificial intelligence, which adopts the following technical scheme:
the embodiment of the invention provides a visual detection method for a wood optimized saw based on artificial intelligence, which comprises the following specific steps:
collecting RGB images of wood, and obtaining a bounding box of a plurality of knots on the RGB images; cutting the positions of the knots corresponding to the wood according to the surrounding frame to obtain a cutting power value sequence in the cutting process, and acquiring an initial cutting characteristic signal sequence corresponding to the knots by using the cutting power value sequence;
acquiring a depth image corresponding to the RGB image, acquiring a point cloud area and point cloud data corresponding to the bounding box based on the depth image, and performing equidistant segmentation on the point cloud area to acquire a plurality of point cloud sub-areas, wherein the number of the point cloud sub-areas is matched with the length of the initial cutting characteristic signal sequence; acquiring a dead knot coefficient of each point cloud subregion from the point cloud data of each point cloud to obtain a coefficient vector corresponding to the point cloud subregion;
combining the coefficient vector and the initial cutting characteristic signal sequence to obtain a new cutting characteristic signal sequence, obtaining a cosine distance between two corresponding knots according to the two new cutting characteristic signal sequences, and clustering the knots by using the cosine distance to obtain a plurality of knot types; training a deep neural network according to the node images corresponding to different node types; the segment image is captured on the RGB image according to the bounding box;
and detecting the type of the knot according to the trained deep neural network so as to cut by a corresponding method.
Preferably, the method for cutting the positions of the knots on the wood according to the bounding box to obtain the cutting power value sequence in the cutting process comprises the following steps:
acquiring coordinate positions of two opposite angles of the surrounding frame, and taking two abscissa values in the coordinate positions as a starting point and an end point of a cutting process; obtaining the vertical coordinate of the cutter according to the two vertical coordinate values in the coordinate position;
and in the cutting process, obtaining a cutting power value corresponding to the power change of the cutter to form the cutting power value sequence.
Preferably, the method for obtaining an initial cutting characteristic signal sequence corresponding to the knot by using the cutting power value sequence includes:
calculating the average value of the cutting power values in the cutting power value sequence to obtain reference power; and acquiring the weight of each cutting power value, and acquiring a corresponding initial cutting characteristic signal from the weight, the reference power and the cutting power value so as to form the initial cutting characteristic signal sequence.
Preferably, the method for obtaining the weight includes:
the data constructed by the cutting power value and the reference power deviate from a weight model; the data deviation weight model is:
wherein W is the weight; psIs the cutting power value; pmeanIs the reference power.
Preferably, the method for obtaining the dead knot coefficient of each point cloud subregion from the point cloud data of each point cloud comprises:
acquiring a reference curve F of which the Z value changes along with the Y value when the point cloud subarea is flatZAnd (Y) Z, calculating the sum of the absolute values of the residual errors of the Z-axis point cloud data by combining the reference curve, the Y-axis point cloud data and the Z-axis point cloud data of each point cloud in the point cloud sub-region, and taking the sum of the absolute values of the residual errors as the dead knot coefficient of the point cloud sub-region.
Preferably, the method for obtaining a new cutting feature signal sequence by combining the coefficient vector and the initial cutting feature signal sequence includes:
and multiplying the coefficient vector with each corresponding dead knot coefficient in the initial cutting characteristic signal sequence and the initial cutting characteristic signal to obtain a new cutting characteristic signal so as to form the new cutting characteristic signal sequence.
Preferably, the knot types comprise a dead knot, a normal knot and a knot with high hardness, and the knot with high hardness means that an included angle between a main shaft of the knot and a main shaft of a trunk is obvious, and a cutter is easily damaged.
Preferably, the length of the initial cutting feature signal sequence of each of the nodes is made equal by means of nearest neighbor interpolation.
The embodiment of the invention at least has the following beneficial effects: the nodes are subjected to feature description based on cutting load and point cloud features, so that more detailed automatic sample marking and classification are realized, the accuracy of deep neural network training is ensured, and the accuracy of optimal selection and cutting of the saw is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flowchart illustrating steps of a method for visual inspection of an artificial intelligence based wood-preferred saw according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a point cloud sub-region after point cloud region segmentation provided in the embodiment of the present invention.
Detailed Description
To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, the following description, in conjunction with the accompanying drawings and preferred embodiments, provides a visual inspection method for wood optimization saw based on artificial intelligence, and the detailed implementation, structure, features and functions thereof are described in detail below. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The following describes a specific scheme of the wood optimization saw visual detection method based on artificial intelligence in detail with reference to the accompanying drawings.
Referring to fig. 1, a flowchart illustrating steps of a method for visual inspection of an artificial intelligence based wood-preferred saw according to an embodiment of the present invention is shown, the method comprising the steps of:
s001, collecting RGB (red, green and blue) images of the wood, and acquiring surrounding frames of a plurality of knots on the RGB images; and cutting the positions of the knots on the corresponding wood according to the bounding box to obtain a cutting power value sequence in the cutting process, and acquiring an initial cutting characteristic signal sequence of the corresponding knots by using the cutting power value sequence.
Specifically, an RGBD camera mounted on a reference timber preferable saw is used for collecting an RGB image of the timber, the RGB image is input into a target detection network to obtain a bounding box of a plurality of joints, and an implementer can use a traditional target detection network during defect detection, such as a YOLO series and a centrnet series.
It should be noted that when the reference timber is preferably used by a saw, the timber is moved to a reference position, which is given by the actual project, so as to map the specific position in the tool and timber image.
Acquiring coordinate positions of two opposite angles of the surrounding frame, taking two abscissa values in the coordinate positions as a starting point and an end point of the cutting process, and obtaining the ordinate of the cutter according to two ordinate values in the coordinate positions; and in the cutting process, obtaining a cutting power value corresponding to the power change of the cutter to form a cutting power value sequence.
As an example, in the embodiment of the present invention, a cutting power value sequence is obtained by analyzing a bounding box of a segment, and the specific process includes:
(1) obtaining the coordinates (x) of the upper left corner of the bounding boxl，yt) And the coordinates of the lower right corner (x)r，yb) And wood is made according to RGB mapVertically feeding the image into a reference wood optimized saw from top to bottom, obtaining a longitudinal coordinate of the cutter according to two longitudinal coordinate values in an upper left corner coordinate and a lower right corner coordinate, namely obtaining a longitudinal coordinate surrounding the center point of the frame according to the two longitudinal coordinate values
The ordinate of the centre point of the bounding box is the ordinate of the tool and cutting from this position means that the tool will pass the joint to the maximum.
(2) Since the tool of the reference timber preferred saw cuts the timber board from left to right, two abscissa values x in the coordinates of the upper left corner and the upper right cornerlAnd xrThe cutting process is divided into a cutting starting point and a cutting end point which correspond from left to right in the cutting process.
(3) Because the dead knot is a structure that the wood tissue and the surrounding wood are separated, the load of the cutter is small, and the power of the motor is reduced due to the fact that the load is small instantaneously. When the cutter meets a section with increased hardness, the rotating speed of the motor is reduced due to increased cutter load, and the constant speed controller increases power to increase the rotating speed, so that low-power reading can be generated when the cutter meets a dead knot, and high-power reading can be generated when the cutter meets the section with the higher hardness. The practitioner gives the expected power fluctuation range, recording from the cut start point xlTo the cutting end point xrThe cutting power value corresponding to the power change of the cutter during the cutting process is formed into a cutting power value sequence Ps＝Ps0...PsnWhere n is the number of values of the power variation acquired during the cutting process, this number being related to the acquisition frequency, and the magnitude of n depends on the cutting feed (cutting speed) and the acquisition frequency.
Preferably, the acquisition frequency in the embodiment of the invention is 50 Hz.
Further, after the cutting power value sequence is obtained, since the cutter load fluctuation is easy to occur in the cutting process, in order to avoid the accuracy of subsequent processing, the cutting power value sequence is used to obtain the initial cutting characteristic signal sequence of the corresponding joint, and the obtaining method comprises the following steps: calculating the average value of the cutting power values in the cutting power value sequence to obtain reference power; and acquiring the weight of each cutting power value, and acquiring a corresponding initial cutting characteristic signal from the weight, the reference power and the cutting power value so as to form an initial cutting characteristic signal sequence.
As an example, a specific process of acquiring an initial cutting feature signal sequence of a corresponding joint by using a cutting power value sequence in the embodiment of the present invention is as follows:
(1) calculating the average value of the cutting power values in the cutting power value sequence to obtain a reference power Pmean。
(2) In order to further enhance the signal characteristics of the cutting power value under different conditions, a data deviation weight model constructed by the cutting power value and the reference power obtains the weight of the cutting power value, and the data deviation weight model is as follows:
wherein W is a weight; psIs the cutting power value; pmeanIs the reference power.
It should be noted that when any cutting power value in the cutting power value sequence exceeds the reference power, a node with high hardness appears in the cutting process, and the node easily damages a cutter in the processing process, generally, it is considered that invisible damages of a transmission shaft, a cutting cutter and the like can occur when the cutter load exceeds a conventional load by several times, and meanwhile, the cutting material of the wood board can also be cracked, and the later result is serious, because the cutting power value in the cutting power value sequence is in a multiple relation with the reference power, the sensitivity of the cutting power value is far greater than the dead-end condition, the response value can reach about [1, 5], and the actual condition is determined by the performance of the cutting equipment; when any cutting power value in the cutting power value sequence is smaller than the reference power, dead knots appear in the cutting process, the knots are low in hardness and prone to produce invalid wood, and therefore a certain response value [1, 2] is needed for the situation.
And acquiring the weight corresponding to each cutting power value in the cutting power value sequence by using the data deviation weight model to form a weight sequence, wherein the data deviation weight model scales the cutting power value with smaller deviation from the reference power, amplifies the abnormal cutting power value and relatively plays roles of reducing noise and improving sensitivity.
(3) Based on the weight sequence, calculating an initial cutting characteristic signal sequence, for a sequence of cutting power values PsAnd the ith cutting power value corresponding to the weight sequence comprises:
wherein,
an initial cutting characteristic signal of the ith cutting power value; wiA weight for the ith cutting power value;
is the ith cutting power value in the sequence of cutting power values.
The function of acquiring the initial cutting characteristic signal is as follows: and further scaling the power change fluctuation signal based on the reference power by using the weight, so that the sensitivity of the hard nodes is improved, and dead nodes are distinguished. When a node with high hardness is encountered, continuous negative values appear in the initial cutting characteristic signal sequence due to the increase of the cutter load; otherwise, when the dead section is encountered, the cutter load is lower than the reference value, continuous positive values appear in the initial cutting characteristic signal sequence, and negative values hardly appear.
It should be noted that, because the nodes have different sizes and the number of samples of the initial cutting feature signal sequence is not necessarily the same, resampling is required to obtain the initial cutting feature signal sequence with a specific length, the length of the initial cutting feature signal sequence in the embodiment of the present invention is 32, and regardless of whether the length of the initial cutting feature signal sequence of each node is the same, the initial cutting feature signal sequence with a specific length is processed by using a nearest neighbor interpolation method
S002, obtaining a depth image corresponding to the RGB image, obtaining a point cloud area and point cloud data corresponding to the bounding box based on the depth image, carrying out equidistant segmentation on the point cloud area to obtain a plurality of point cloud sub-areas, wherein the number of the point cloud sub-areas is matched with the length of the initial cutting characteristic signal sequence; and acquiring a dead knot coefficient of each point cloud sub-area by the point cloud data of each point cloud to obtain a coefficient vector of the corresponding point cloud area.
Specifically, since the dead knot is a structure in which the wood tissue and the surrounding wood are separated, the surface of the dead knot is not as flat as a common board, and therefore, dead knot information can be analyzed based on the height variation variance of the point cloud on the surface of the measured board.
Acquiring a depth image corresponding to an RGB image, acquiring a point cloud area and point cloud data corresponding to a surrounding frame of each node based on the depth image, and performing equidistant segmentation based on an x-axis position of the point cloud area to obtain point cloud sub-areas as shown in FIG. 2, wherein the number of the point cloud sub-areas is matched with the length of an initial cutting characteristic signal sequence and is 32; referring to fig. 2, the point cloud sub-region C1 contains the leftmost point cloud data of the bounding box corresponding to the node, and so on, and the point cloud sub-region C32 contains the rightmost point cloud data of the bounding box corresponding to the node.
Processing the point cloud subregions in the point cloud region C ═ { C1.. C32} one by one:
(1) assume that the center of the dead center is located at the very center of the bounding box, and for the convenience of understanding, the embodiment of the present invention takes the point cloud subregion C16 as an example: taking the Y-axis point cloud data Y of each point cloud in the point cloud subregion C1616＝{Y1...YnZ and Z-axis point cloud data Z16＝{Z1...ZnN is the number of point clouds in the point cloud sub-region C16.
(2) Although the pose of the wood board observed by the camera is a term view, the wood board still inclines, so a reference curve F of which the Z value changes along with the Y value when the point cloud sub-region is flat is obtained by using a RANSAC (random sample consensus) straight line fitting methodZ(Y)＝Z。
(3) Sub-area of point cloudY-axis point cloud data Y of domain C1616＝{Y1...YnSubstituting into the reference curve FZ(Y) calculating the sum of absolute values of residuals of the z-axis point cloud data, and taking the sum of the absolute values of the residuals as a dead knot coefficient R of the point cloud sub-region C1616＝∑|FZ(Y16)-Z16|。
(4) Acquiring a dead knot coefficient of each point cloud subregion in the point cloud region C ═ { C1.. C32} by using the steps (1) to (3) to form a coefficient vector R ═ { R ═ R1...R32And normalized to the interval [1, 2] by normalizing the difference of the maxima]。
Step S003, combining the coefficient vector and the initial cutting characteristic signal sequence to obtain a new cutting characteristic signal sequence, obtaining cosine distances between two corresponding knots according to the two new cutting characteristic signal sequences, and clustering the knots by utilizing the cosine distances to obtain a plurality of knot types; training a deep neural network according to the node images corresponding to different node types; the segment image is cut out of the RGB image according to the bounding box.
Specifically, considering that a motor is approximately idle when a dead knot occurs, and because the angular momentum of a cutter is large and a certain hysteresis exists in the process of reducing the load of the cutter, a dead knot coefficient determined based on a point cloud sensing result can be used as compensation when the load of the cutter is reduced, so that the characteristic of load reduction is more obvious, and the reliability of a subsequent automatic labeling sample is improved
Further, a new cleavage feature signal sequence Pf' is a 32-dimensional feature descriptor, which can describe the load change of the cutter during the cutting process of the corresponding knot position on the wood. Based on the new cutting characteristic signal sequence Pf' calculating the cosine distance between two nodes, i.e. using the cosine similarity s to the new cutting characteristic signal of two nodesAnd (4) performing similarity calculation on the sequence, and obtaining a cosine distance d by using the cosine similarity through a cosine distance formula d-1-s. And clustering a plurality of knots on the RGB image by using the cosine distance to obtain a plurality of knot types.
Preferably, the knot types in the embodiment of the invention comprise a dead knot, a normal knot and a knot with high hardness, and the knot with high hardness means that an included angle between a main shaft of the knot and a main shaft direction of a trunk is obvious and a cutter is easy to damage.
Intercepting a node image of each node on the RGB image according to a surrounding frame of the node, training a deep neural network by utilizing the node images corresponding to different node types to obtain a node classifier Q of the deep neural network, wherein the training process comprises the steps of carrying out sample marking on the node based on the clustered node types, wherein a dead node is 1, a normal node is 2, and a node with higher hardness is 3; classifying the node images by using a deep neural network based on the labeling result; the loss function uses a cross-entropy loss function.
And step S004, detecting the type of the joint according to the trained deep neural network so as to cut by a corresponding method.
Specifically, the node sub-classifiers Q are deployed on a standard wood preferred saw, which is different from a reference wood preferred saw in that the reference wood preferred saw is equipped with an RGBD camera, and the standard wood preferred saw runs the node sub-classifiers Q trained based on data of the reference wood preferred saw, so that only an RGB camera needs to be equipped.
The implementer adjusts and gives a confidence threshold T so that the judgment result of the section type of the standard wood-preferred saw meets the expectation:
when the index value returned by the knot sub-classifier Q is 1, the occurrence of dead knots is indicated, and therefore the cutting positions of the standard wood preferable saw are set to be the top end y value and the bottom end y value of the surrounding frame respectively, and the dead knot waste cutting is realized; when the index value returned by the node classifier Q is 2, the normal node appears, and therefore the normal cutting can be carried out according to the actual cutting size; when the index value returned by the node classifier Q is 3, it means that a node having a large hardness is present, and thus it is possible to cut waste in a dead node disposal manner, or to reduce the cutting feed amount, and to use the reduced cutting feed amount as the cutting feed amount preset by the implementer, which can improve the material utilization and protect the cutter from damage.
Preferably, the cutting feed amount after the reduction of the embodiment of the invention is the standard cutting feed amount
It should be noted that, since the wood is fed along the y-axis direction of the bounding box, the cutting positions are set to the top y value and the bottom y value of the bounding box, and the distance between the top y value and the bottom y value of the bounding box also represents the length of the wood to be cut.
In summary, the embodiment of the present invention provides an artificial intelligence-based visual detection method for a wood-optimized saw, which detects RGB images of a wood to obtain a bounding box of a plurality of joints, and cuts a position of a corresponding joint on the wood by the bounding box to obtain an initial cutting characteristic signal sequence; acquiring a point cloud area and point cloud data corresponding to the bounding box based on the depth image to obtain a coefficient vector consisting of a dead knot coefficient; and obtaining a new cutting characteristic signal sequence by combining the initial cutting characteristic signal sequence and the coefficient vector, classifying the knots by using the new cutting characteristic signal sequence, training a deep neural network by using the knot images corresponding to different knot types to detect the knot types of the knots, and further adopting a corresponding cutting method. The nodes are subjected to feature description based on cutting load and point cloud features, more detailed automatic sample marking and classification are realized, the accuracy of deep neural network training is guaranteed, and the accuracy of optimal selection and cutting of the saw is improved.
It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (8)
1. An artificial intelligence based visual inspection method for wood-preferred saws, which is characterized by comprising the following steps:
collecting RGB images of wood, and obtaining a bounding box of a plurality of knots on the RGB images; cutting the positions of the knots corresponding to the wood according to the surrounding frame to obtain the cutting power value sequence in the cutting process, and acquiring an initial cutting characteristic signal sequence corresponding to the knots by using the cutting power value sequence;
acquiring a depth image corresponding to the RGB image, acquiring a point cloud area and point cloud data corresponding to the bounding box based on the depth image, and performing equidistant segmentation on the point cloud area to acquire a plurality of point cloud sub-areas, wherein the number of the point cloud sub-areas is matched with the length of the initial cutting characteristic signal sequence; acquiring a dead knot coefficient of each point cloud subregion from the point cloud data of each point cloud to obtain a coefficient vector corresponding to the point cloud subregion;
combining the coefficient vector and the initial cutting characteristic signal sequence to obtain a new cutting characteristic signal sequence, obtaining a cosine distance between two corresponding knots according to the two new cutting characteristic signal sequences, and clustering the knots by using the cosine distance to obtain a plurality of knot types; training a deep neural network according to the node images corresponding to different node types; the segment image is captured on the RGB image according to the bounding box;
and detecting the type of the knot according to the trained deep neural network so as to cut by a corresponding method.
2. The method according to claim 1, wherein the method of cutting the positions of the knots on the corresponding wood according to the bounding box to obtain the sequence of cutting power values in the cutting process comprises:
acquiring coordinate positions of two opposite angles of the surrounding frame, and taking two abscissa values in the coordinate positions as a starting point and an end point of a cutting process; obtaining the vertical coordinate of the cutter according to the two vertical coordinate values in the coordinate position;
and in the cutting process, obtaining a cutting power value corresponding to the power change of the cutter to form the cutting power value sequence.
3. The method according to claim 1 or 2, wherein the method for obtaining an initial cutting characteristic signal sequence corresponding to the knot using the cutting power value sequence comprises:
calculating the average value of the cutting power values in the cutting power value sequence to obtain the reference power; and acquiring the weight of each cutting power value, and acquiring a corresponding initial cutting characteristic signal from the weight, the reference power and the cutting power value so as to form the initial cutting characteristic signal sequence.
4. The method of claim 3, wherein the method of obtaining the weight comprises:
the data constructed by the cutting power value and the reference power deviate from a weight model; the data deviation weight model is:
wherein W is the weight; psIs the cutting power value; pmeanIs the reference power.
5. The method of claim 1, wherein the method of obtaining a dead-beat coefficient for each of the sub-regions of the point cloud from the point cloud data of each point cloud comprises:
acquiring a reference curve F of which the Z value changes along with the Y value when the point cloud subarea is flatZAnd (Y) Z, calculating the sum of the absolute values of the residual errors of the Z-axis point cloud data by combining the reference curve, the Y-axis point cloud data and the Z-axis point cloud data of each point cloud in the point cloud sub-region, and taking the sum of the absolute values of the residual errors as the dead knot coefficient of the point cloud sub-region.
6. The method of claim 1, wherein said combining said coefficient vector and said initial cut signature signal sequence to obtain a new cut signature signal sequence comprises:
and multiplying the coefficient vector with each corresponding dead knot coefficient in the initial cutting characteristic signal sequence and the initial cutting characteristic signal to obtain a new cutting characteristic signal so as to form the new cutting characteristic signal sequence.
7. The method of claim 1, wherein the types of knots include dead knots, normal knots and knots with high hardness, and the knots with high hardness are obvious in included angle between the main axis direction of the knots and the main axis direction of the trunk, and are easy to cause cutter damage.
8. The method of claim 1, wherein the length of the initial cut feature signal sequence of each of the nodes is made equal by nearest neighbor interpolation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110931750.0A CN113674233B (en) | 2021-08-13 | 2021-08-13 | Wood optimal selection saw visual detection method based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110931750.0A CN113674233B (en) | 2021-08-13 | 2021-08-13 | Wood optimal selection saw visual detection method based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113674233A true CN113674233A (en) | 2021-11-19 |
| CN113674233B CN113674233B (en) | 2022-05-20 |
Family
ID=78542812
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110931750.0A Active CN113674233B (en) | 2021-08-13 | 2021-08-13 | Wood optimal selection saw visual detection method based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113674233B (en) |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700417A (en) * | 2015-01-19 | 2015-06-10 | 湖南大学 | Computer image based automatic identification method of timber knot flaws |
| CN205538740U (en) * | 2016-01-29 | 2016-08-31 | 广州番禺职业技术学院 | Intelligence timber surface defect detection system |
| CN108171748A (en) * | 2018-01-23 | 2018-06-15 | 哈工大机器人（合肥）国际创新研究院 | A kind of visual identity of object manipulator intelligent grabbing application and localization method |
| CN108247764A (en) * | 2017-12-14 | 2018-07-06 | 北京木业邦科技有限公司 | Plank cutting method, device, electronic equipment and medium based on machine learning |
| CN108262809A (en) * | 2017-12-14 | 2018-07-10 | 北京木业邦科技有限公司 | Plank processing method, device, electronic equipment and medium based on artificial intelligence |
| CN108943179A (en) * | 2018-08-22 | 2018-12-07 | 慧泉智能科技（苏州）有限公司 | A kind of optimal cutling method for arithmetic for lumber surface defects' |
| JP6450489B1 (en) * | 2018-06-05 | 2019-01-09 | 後藤木材株式会社 | Plastic processed wood |
| CN109557104A (en) * | 2018-11-19 | 2019-04-02 | 临沂众为智能科技有限公司 | A kind of timber based on deep learning preferably saws intelligent detecting method |
| CN111325713A (en) * | 2020-01-21 | 2020-06-23 | 浙江省北大信息技术高等研究院 | Wood defect detection method, system and storage medium based on neural network |
| CN111862028A (en) * | 2020-07-14 | 2020-10-30 | 南京林业大学 | Wood defect detecting and sorting device and method based on depth camera and depth learning |
| CN111951223A (en) * | 2020-07-16 | 2020-11-17 | 西南林业大学 | Automatic image identification method and system for wood knot flaws |
| CN112083018A (en) * | 2020-09-11 | 2020-12-15 | 黑龙江省林业科学研究所 | Nondestructive detection method for tree knots |
| CN112129773A (en) * | 2020-08-20 | 2020-12-25 | 湖南大学 | Wood surface defect detection method, device, equipment, system and storage medium |
| CN112381861A (en) * | 2020-09-18 | 2021-02-19 | 南京航空航天大学 | Forest land point cloud data registration and segmentation method based on foundation laser radar |
| CN213165840U (en) * | 2020-07-22 | 2021-05-11 | 优铠(上海)机械有限公司 | Timber saw cuts correcting unit and preferred saw |
| CN113128405A (en) * | 2021-04-20 | 2021-07-16 | 北京航空航天大学 | Plant identification and model construction method combining semantic segmentation and point cloud processing |
-
2021
- 2021-08-13 CN CN202110931750.0A patent/CN113674233B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700417A (en) * | 2015-01-19 | 2015-06-10 | 湖南大学 | Computer image based automatic identification method of timber knot flaws |
| CN205538740U (en) * | 2016-01-29 | 2016-08-31 | 广州番禺职业技术学院 | Intelligence timber surface defect detection system |
| CN108247764A (en) * | 2017-12-14 | 2018-07-06 | 北京木业邦科技有限公司 | Plank cutting method, device, electronic equipment and medium based on machine learning |
| CN108262809A (en) * | 2017-12-14 | 2018-07-10 | 北京木业邦科技有限公司 | Plank processing method, device, electronic equipment and medium based on artificial intelligence |
| CN108171748A (en) * | 2018-01-23 | 2018-06-15 | 哈工大机器人（合肥）国际创新研究院 | A kind of visual identity of object manipulator intelligent grabbing application and localization method |
| JP6450489B1 (en) * | 2018-06-05 | 2019-01-09 | 後藤木材株式会社 | Plastic processed wood |
| CN108943179A (en) * | 2018-08-22 | 2018-12-07 | 慧泉智能科技（苏州）有限公司 | A kind of optimal cutling method for arithmetic for lumber surface defects' |
| CN109557104A (en) * | 2018-11-19 | 2019-04-02 | 临沂众为智能科技有限公司 | A kind of timber based on deep learning preferably saws intelligent detecting method |
| CN111325713A (en) * | 2020-01-21 | 2020-06-23 | 浙江省北大信息技术高等研究院 | Wood defect detection method, system and storage medium based on neural network |
| CN111862028A (en) * | 2020-07-14 | 2020-10-30 | 南京林业大学 | Wood defect detecting and sorting device and method based on depth camera and depth learning |
| CN111951223A (en) * | 2020-07-16 | 2020-11-17 | 西南林业大学 | Automatic image identification method and system for wood knot flaws |
| CN213165840U (en) * | 2020-07-22 | 2021-05-11 | 优铠(上海)机械有限公司 | Timber saw cuts correcting unit and preferred saw |
| CN112129773A (en) * | 2020-08-20 | 2020-12-25 | 湖南大学 | Wood surface defect detection method, device, equipment, system and storage medium |
| CN112083018A (en) * | 2020-09-11 | 2020-12-15 | 黑龙江省林业科学研究所 | Nondestructive detection method for tree knots |
| CN112381861A (en) * | 2020-09-18 | 2021-02-19 | 南京航空航天大学 | Forest land point cloud data registration and segmentation method based on foundation laser radar |
| CN113128405A (en) * | 2021-04-20 | 2021-07-16 | 北京航空航天大学 | Plant identification and model construction method combining semantic segmentation and point cloud processing |
Non-Patent Citations (5)
| Title |
|---|
| MINGYU GAO 等: "A Transfer Residual Neural Network Based on ResNet-34 for Detection of Wood Knot Defects", 《FORESTS 2021》 * |
| YU CHENG 等: "Weld Defect Detection and Image Defect Recognition using Deep Learning Technology", 《WELDING》 * |
| 范佳楠 等: "机器视觉在木材缺陷检测领域应用研究进展", 《世界林业研究》 * |
| 萧嘉奕 等: "基于Faster R-CNN模型的木材表面节子缺陷检测", 《木材工业》 * |
| 邵明伟 等: "基于深度学习的木材优选锯视觉检测算法", 《林业科学》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113674233B (en) | 2022-05-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110930357B (en) | In-service steel wire rope surface defect detection method and system based on deep learning | |
| WO2018107525A1 (en) | Detection method for mold of electric injection molding machine | |
| CN108181316B (en) | Bamboo strip defect detection method based on machine vision | |
| CN107085846A (en) | Surface Flaw image-recognizing method | |
| CN111753577A (en) | Apple identification and positioning method in automatic picking robot | |
| CN110110687B (en) | Method for automatically identifying fruits on tree based on color information and three-dimensional contour information | |
| CN112634237A (en) | Long bamboo strip surface defect detection method and system based on YOLOv3 improved network | |
| CN113012098B (en) | Iron tower angle steel punching defect detection method based on BP neural network | |
| CN115187803B (en) | Positioning method for picking process of famous tea tender shoots | |
| CN102269571A (en) | Method and device for measuring diameter of bamboo tube based on digital image processing technology | |
| CN108022232A (en) | A kind of aircraft surface rivet detection method | |
| CN111681222A (en) | Automatic detection and judgment method for edge breakage of tool mark and scribing machine applied by same | |
| CN113240693A (en) | Chip cutting control method, system, equipment and computer readable storage medium | |
| CN116258682A (en) | PCB solder paste defect detection method based on PSPNet and improved YOLOv7 | |
| CN114029943A (en) | Target grabbing and positioning method and system based on image data processing | |
| CN115953398A (en) | Defect identification method for strip steel surface | |
| CN113674233B (en) | Wood optimal selection saw visual detection method based on artificial intelligence | |
| CN117115196B (en) | Visual detection method and system for cutter abrasion of cutting machine | |
| CN113281115B (en) | Control method for intelligent tobacco leaf shearing | |
| CN107133958B (en) | Optical remote sensing ship slice segmentation method based on block particle size pre-judging balance histogram | |
| CN113989308A (en) | Polygonal target segmentation method based on Hough transform and template matching | |
| CN116309375B (en) | Method for detecting double-sided defects of solid wood plate and determining intelligent processing coordinates | |
| CN115131539B (en) | Aluminum template automatic identification and classification system based on machine vision | |
| Aleksi et al. | Features extraction and texture defect detection of sawn wooden board images | |
| Suppitaksakul et al. | Machine vision system for counting the number of corrugated cardboard |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |