Method and system for board grade classification and cutting based on multi-modal image recognition

Abstract

The disclosure provides a multi-modal image recognition-based board grade classification and cutting method and system, relating to the technical field of machine learning, which comprises preprocessing a board image; detecting a circular positioning point for board cutting, obtaining a coarse positioning image of the cut board; performing vertical direction projection according to the coarse positioning image, obtaining a mask image of the projection occlusion area; extracting board features from the mask image, inputting the extracted feature vector into a support vector machine for classification, and obtaining a classification type; performing maximum inscribed rectangle algorithm calculation on the non-defect board after classification, obtaining the maximum cutting area in the board area, and then obtaining the cutting position; using an improved Mask R-CNN algorithm to perform pixel-level positioning on the board, then performing maximum inscribed rectangle algorithm to obtain the maximum cutting area in the board area, and determining the cutting method for board segmentation by using the area difference between the two and a certain threshold size relationship. The disclosure improves the yield of the board.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of machine learning, in particular to a wood board grade classification and cutting method and system based on multi-modal image recognition. BACKGROUND

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] In the wood board processing industry, wood board defect rejection and wood board grade classification are indispensable important links, and the work efficiency and accuracy of the two are directly related to the wood yield rate. However, in the current wood board grading and cutting process, automation is not yet realized. Instead, the wood is first cut into wood boards. If there are many defects in the wood board, it is eliminated, otherwise the wood board is graded. Finally, the wood board is automatically cut to obtain the largest area of board. Now, artificial calibration cutting position is generally used. In the whole wood board grading and wood board cutting process, more workers are needed for manual operation, which is low in efficiency, high in cost, and the worker training period is relatively long. SUMMARY

[0004] In order to solve the above problems, the present disclosure proposes a wood board grade classification and cutting method and system based on multi-modal image recognition, which automatically grades and automatically divides the cutting area of the wood board, improves the wood grading efficiency and the yield rate.

[0005] According to some embodiments, the present disclosure adopts the following technical solutions:

[0006] The wood board grade classification and cutting method based on multi-modal image recognition comprises:

[0007] Obtaining a wood board image to be processed, and pre-processing the wood board image;

[0008] Detecting a circular positioning point of wood board cutting, obtaining a coarse positioning image of the cut wood board, and performing vertical direction projection on the coarse positioning image to obtain a mask image of a projection occlusion area;

[0009] The wood board feature extraction is performed on the mask image, the extracted feature vector is input into a support vector machine for classification, and a classification type is obtained; the non-defect wood board after classification is calculated by using a maximum inscribed rectangle algorithm, the maximum cutting area in the wood board area is obtained, and then the cutting position is obtained; and then an improved Mask R-CNN algorithm is used, a ResNeSt network is used to replace the backbone network of Mask R-CNN, the feature representation capability is enhanced, the network can better capture the local and global information of the wood board, the wood board is positioned at a pixel level, then the maximum inscribed rectangle algorithm is performed, the maximum cutting area in the wood board area is obtained, and the cutting method is determined according to the area difference value and the threshold size relationship to segment the wood board.

[0010] According to some embodiments, the present disclosure adopts the technical scheme as follows:

[0011] The wood board grade classification and cutting system based on multi-modal image recognition comprises:

[0012] A wood board image preprocessing module is configured to acquire a wood board image to be processed and perform preprocessing on the wood board image.

[0013] A wood board occlusion area processing module is configured to detect a circular positioning point of wood board cutting, acquire a rough positioning image of the cut wood board, perform vertical direction projection on the rough positioning image, and acquire a mask image of a projection occlusion area.

[0014] A wood board grading module is configured to perform wood board feature extraction on the mask image, input the extracted feature vector into a support vector machine for classification, and obtain a classification type.

[0015] A wood board segmentation module is configured to perform maximum inscribed rectangle algorithm calculation on the non-defect wood board after classification, obtain the maximum cutting area in the wood board area, and then obtain the cutting position; use an improved Mask R-CNN algorithm, use a ResNeSt network to replace the backbone network of Mask R-CNN, enhance the feature representation capability, so that the network can better capture the local and global information of the wood board, perform pixel-level positioning on the wood board, then perform the maximum inscribed rectangle algorithm, obtain the maximum cutting area in the wood board area, and determine the cutting method for wood board segmentation according to the area difference value and the threshold size relationship.

[0016] According to some embodiments, the present disclosure adopts the technical scheme as follows:

[0017] A non-transitory computer readable storage medium is configured to store computer instructions, and the computer instructions are executed by a processor to implement the wood board grade classification and cutting method based on multi-modal image recognition.

[0018] According to some embodiments, the present disclosure adopts the technical solutions as follows:

[0019] An electronic device, comprising: a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the method for classifying and cutting wood board based on multi-modal image recognition.

[0020] Compared with the prior art, the present disclosure has the beneficial effects as follows:

[0021] The method of the present disclosure can quickly find the image positioning point and coarsely position the wood board according to the positioning point, reducing the interference of other areas on the subsequent process. In addition, if the positioning point cannot be found for several times in succession, it is prompted to clean the positioning point; the occluded position in the wood board image can be quickly removed, reducing the influence on the image cutting position in the subsequent process; the classification of the wood board can be quickly realized; the cutting position can be accurately provided using multiple algorithms, and a full-process and full-automatic processing method for the wood board image can be realized in the wood processing field, which can greatly reduce the manual operation steps and increase the classification efficiency and yield of the wood board. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings, which form a part of the present disclosure, are used to provide further understanding of the present disclosure, and the schematic embodiments of the present disclosure and the description thereof are used to explain the present disclosure, and do not constitute an improper limitation on the present disclosure.

[0023] Figure 1 The accompanying drawings, which form a part of the present disclosure, are used to provide further understanding of the present disclosure, and the schematic embodiments of the present disclosure and the description thereof are used to explain the present disclosure, and do not constitute an improper limitation on the present disclosure.

[0024] Figure 2 The accompanying drawings, which form a part of the present disclosure, are used to provide further understanding of the present disclosure, and the schematic embodiments of the present disclosure and the description thereof are used to explain the present disclosure, and do not constitute an improper limitation on the present disclosure.

[0025] Figure 3 The accompanying drawings, which form a part of the present disclosure, are used to provide further understanding of the present disclosure, and the schematic embodiments of the present disclosure and the description thereof are used to explain the present disclosure, and do not constitute an improper limitation on the present disclosure. DETAILED DESCRIPTION

[0026] The present disclosure will be further described below in combination with the accompanying drawings and embodiments.

[0027] It should be pointed out that the following detailed description is exemplary and is intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present disclosure belongs.

[0028] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present disclosure. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0029] Embodiment 1

[0030] In an embodiment of the present disclosure, a wood board grade classification and cutting method based on multi-modal image recognition is provided, comprising:

[0031] Step one: obtaining a wood board image to be processed, and pre-processing the wood board image;

[0032] Step two: detecting a circular positioning point of wood board cutting, obtaining a coarse positioning image of the cut wood board, and performing vertical direction projection according to the coarse positioning image to obtain a mask image of a projection occlusion area;

[0033] Step three: extracting wood board features from the mask image, inputting the extracted feature vectors into a support vector machine for classification to obtain a classification type, performing maximum inscribed rectangle algorithm calculation on the non-defect wood board after classification to obtain the maximum cutting area in the wood board area, and then using an improved Mask R-CNN algorithm, wherein the ResNeSt network is mainly used to replace the backbone network of Mask R-CNN to enhance the representation ability of features, so that the network can better capture the local and global information of the wood board, and perform pixel-level positioning on the wood board, and then perform maximum inscribed rectangle algorithm to obtain the maximum cutting area in the wood board area, and determine the cutting method for wood board segmentation by using the area difference value and a certain threshold size relationship.

[0034] As an embodiment, in step one, the wood board image to be processed is obtained, and a pre-processing operation is performed on the wood board image, including performing image preprocessing on the wood board image captured by an industrial camera, and respectively performing grayscale, binarization, and morphological processing on the image.

[0035] Specifically, after obtaining the wood board image captured by the industrial camera through the interface, first, grayscale operation is performed on the wood board image to obtain a two-channel grayscale image; then, fixed threshold binarization operation is performed on the grayscale image to obtain a binarized image; and finally, open-close operation is used to remove the noise points in the binarized image.

[0036] After the image is grayed and binarized, the positioning points and the wood board shooting image have high brightness. In the binarization process, the general background interference can be removed by increasing the binarization threshold. Then the Hough transform is used to detect the circular positioning points of the binarized image. Specifically, in step two, the method for obtaining the rough positioning image of the cut wood board comprises:

[0037] The circular positioning points of the wood board cutting are detected using the Hough transform, wherein the detected circular positioning points in the image are controlled based on the parameters of the gradient by controlling the distance between the circles, the minimum radius of the circles, and the maximum radius of the circles.

[0038] The parameter control based on the gradient includes: if the number of detected circular positioning points meets the set threshold number, the angles of adjacent circular positioning points are calculated in sequence, and if the angles of the circular positioning points are all within the threshold, it is normal; if the number of detected circular positioning points is less than the threshold number, the circular positioning points are prompted to be cleaned, and if it is greater than the threshold number, the camera parameters are adjusted, and the current baffle position is recorded.

[0039] Specifically, the Hough transform is used to detect the circular positioning points of the binarized image, wherein the detected circular positioning points in the binarized image are controlled based on the parameters of the gradient by controlling the distance between the circles, the minimum radius of the circles, the maximum radius of the circles, and the detection method based on the gradient. If the number of detected positioning points is equal to 4, the angles of adjacent positioning points are calculated in sequence, and if the angles of the positioning points are all within the threshold, it is normal, otherwise an abnormality needs to be prompted. If the number of detected positioning points is less than 4, the positioning points are prompted to be cleaned by the industrial control system. If the number of detected positioning points is greater than 4, the industrial control system is prompted to record the abnormality. In addition, by setting the camera parameters and adjusting the industrial control software, the position of the lower baffle can be recorded.

[0040] After the adjustment of the industrial control system and the camera, the system runs stably and the positioning points are obtained accurately during normal use. After this module, the positions of the positioning points and the lower baffle are mainly recorded, and the wood board image is cut based on the positions to obtain the rough positioning image of the wood board. The purpose is to reduce the interference of the surrounding complex background on the subsequent processing flow.

[0041] Due to the shielding of the cutting tool to the wood board, the industrial camera captures the wood board image with shielding at the top and bottom. As shown in Figure 2 the left and right pressing plates, the wood board shielding area processing module automatically removes the influence of the shielding, automatically whitens the shielding area according to the wood board trend, and obtains the position information of the shielding area and the image after whitening the wood board shielding area. The lower baffle and the left and right pressing plates serve to fix the wood board, and the four positioning points above can assist in fixing the wood board and facilitate the image algorithm to locate the wood board.

[0042] Specifically, for the obtained rough positioning image of the wood board, the process of obtaining the mask image of the projection occlusion region according to the vertical direction projection of the rough positioning image is: according to the position information of the wood board circular positioning point, the vertical direction projection is performed to obtain the starting position and the ending position of the occlusion region in the x-axis direction of the image; then the horizontal direction projection is performed on the intercepted rough positioning image to obtain the starting position and the ending position in the y-axis direction, and finally the mask image of the occlusion region is obtained, and then the occlusion region is white-processed according to the wood board trend to obtain the position information of the occlusion region and the image after the white processing of the occlusion of the wood board.

[0043] First, the nearby region is intercepted according to the position information to obtain an image I_cut, and the image I_cut is subjected to grayscale and binarization processing; then the vertical direction projection is performed on the binarized image to obtain the starting position X_start and the ending position X_end of the occlusion region in the x-axis direction of the image; then the horizontal direction projection is performed on the intercepted binarized image to obtain the starting position Y_start and the ending position Y_end in the y-axis direction, and finally the mask image of the occlusion region is obtained, and the mask position in the original image is subjected to white processing.

[0044] This step can prevent the interference of the occlusion region in the wood board grading and wood board cutting process, and finally obtain the position (X_start, X_end, Y_start, Y_end) of the occlusion region and the image after the white processing of the occlusion of the wood board.

[0045] For a wood board, the defect is the first part to be processed. The defects mainly include knots, discoloration, borer eyes, decay, etc., which can reduce the local strength of the wood board and increase the cracking probability of the wood board. The so-called multi-modal image recognition technology refers to the fusion or integration of multiple image features, the use of the characteristics and unique advantages of different image recognition technologies, and the combination of fusion technology, so that the wood board grading process is more accurate, thereby improving the overall performance of the system. Using image processing technology and some extraction feature rules set by artificial, and using machine learning method to classify the extracted features of wood board grade, the wood board image is divided into three grades of whether there is a defect, straight wood board and non-straight wood board.

[0046] As an embodiment, the process of wood board grading includes:

[0047] First, the obtained image of the occluded area is white, and then the features of the wood board to be extracted are calculated, including the width of the wood board, the height of the wood board, the area of the wood board, the average pixel value of the wood board, the total area of the circular shape in the wood board, the number of circular shapes in the wood board, the overall bending degree of the wood board, the width difference of the wood board section, the bending degree of the grain, the number of grains, the grain spacing, the number of grains, the horizontal projection histogram, the vertical projection histogram, etc. The above features are normalized to form a feature vector. The constructed data set and svm algorithm are used to classify the feature vector, and finally three types of images of defective, straight-grained wood board and non-straight-grained wood board are obtained.

[0048] For a three-classification problem, three SVM models are trained using the One-vs-all strategy, each model corresponding to one class and the combination of all other classes.

[0049] Specifically: for the first class, mark the samples of the first class as positive samples, and mark the samples of the other two classes as negative samples, and then train an SVM model. For the second class, mark the samples of the second class as positive samples, and mark the samples of the other two classes as negative samples, and then train a second SVM model. Similarly, for the third class, mark the samples of the third class as positive samples, and mark the samples of the other two classes as negative samples, and then train a third SVM model.

[0050] As an embodiment, the process of cutting the wood board includes:

[0051] For the graded wood board, two methods are used to obtain the cutting area. By comparing the difference between the two different methods, the most reasonable cutting scheme is found to make the wood board cutting area most accurate and get greater yield. Finally, the cutting position in the wood board image is output.

[0052] In the industrial control condition, the shooting quality of the wood board image is high. For the obtained non-defect wood board, the cutting position is determined. First, method 1 is used, which is a traditional image processing algorithm. The maximum inscribed rectangle algorithm is used to calculate the wood board image without the influence of the removed shielding area, to obtain the maximum cutting area S1 in the wood board area, and then the cutting position is obtained. Then method 2, that is, the Mask R-CNN algorithm, is used. The improved Mask R-CNN algorithm is used to realize instance segmentation of the wood board. The main point is to use the ResNeSt network to replace the backbone network of Mask R-CNN, to enhance the representation ability of the features, so that the network can better capture the local and global information of the wood board. The data set is mainly obtained by method 1 and manual labeling. After the model is trained and converged, the mask of the wood board image can be obtained using the model. The wood board is positioned at the pixel level, and then the maximum inscribed rectangle algorithm is used to obtain the maximum cutting area S2 in the wood board area. If the area difference obtained by the two methods is less than the threshold t1, the cutting position in method 1 is used. In method 1, the above steps can obtain a binary wood board image, in which the wood board cutting surface is white and the background is black. The maximum inscribed rectangle of the binary image is obtained to obtain four rectangular vertices, which are the cutting positions. If it is greater than the threshold t1 and less than the threshold t2, it is determined that the wood board mask image region has small average gray value, large edge bending degree and many holes, and the cutting position of method 2 is used. The cutting position obtained by method 2 is the same as that obtained by method 1. The difference between the two methods is that the processing process of the wood board edge image is different. Method 1 uses a traditional image processing method, and method 2 uses a deep learning model to classify whether the pixel points in the image belong to the wood board. The two methods finally obtain a binary wood board image. Finally, the maximum inscribed rectangle method is used to obtain the vertex, which is the cutting position. Otherwise, the cutting position of method 1 is used. If the area difference of the two is greater than the threshold t2, it is returned to be abnormal, and neither of the two methods is used.

[0053] Embodiment 2

[0054] In an embodiment of the present disclosure, a wood board grade classification and cutting system based on multi-modal image recognition is provided, comprising:

[0055] A preprocessing module is configured to acquire a wood board image to be processed and pre-process the wood board image.

[0056] A wood board grading module is configured to detect a circular positioning point of wood board cutting, acquire a rough positioning image of the cut wood board, and perform vertical direction projection on the rough positioning image to acquire a mask image of a projection shielding area.

[0057] Wood board feature extraction is performed on the mask image, and the extracted feature vector is input into a support vector machine for classification to acquire a grading type.

[0058] The wood board segmentation module is used for calculating the maximum inscribed rectangle algorithm of the non-defective wood board after grading, obtaining the maximum cutting area in the wood board area, and then obtaining the cutting position, and using the improved Mask R-CNN algorithm, wherein mainly the ResNeSt network is used to replace the backbone network of Mask R-CNN, the feature representation capability is enhanced, the network can better capture the local and global information of the wood board, the pixel-level positioning of the wood board is performed, then the maximum inscribed rectangle algorithm is performed to obtain the maximum cutting area in the wood board area, and the area difference value and the certain threshold size relationship are used to determine the cutting method to segment the wood board.

[0059] Embodiment 3

[0060] In an embodiment of the present disclosure, a non-transitory computer readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the steps of the wood board grade classification and cutting method based on multi-modal image recognition.

[0061] Embodiment 4

[0062] In an embodiment of the present disclosure, an electronic device is provided, comprising a processor, a memory, and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the steps of the wood board grade classification and cutting method based on multi-modal image recognition.

[0063] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks. Figure 1 The functions specified in one or more flows and / or blocks.

[0064] These computer program instructions can also be loaded onto a computer or other programmable data processing device to cause a series of operation steps to be performed on the computer or other programmable data processing device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing device provide a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks. Figure 1steps of the functions specified in the one or more blocks.

[0065] Although the specific embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings, the present disclosure is not limited thereto, and various modifications or changes can be made to the present disclosure by those skilled in the art without departing from the technical scope of the present disclosure.

Claims

1. A method for board grade classification and cutting based on multi-modal image recognition, characterized in that, The method comprises the following steps: An image of a wood board to be processed is acquired, and the wood board image is preprocessed; A circular positioning point of wood board cutting is detected, and a coarse positioning image of the cut wood board is acquired; A vertical direction projection is performed according to the coarse positioning image, and a mask image of a projection occlusion area is acquired; The process of performing the vertical direction projection according to the coarse positioning image to acquire the mask image of the projection occlusion area comprises the following steps: performing a vertical direction projection according to position information of the circular positioning point of the wood board to obtain a starting position and an ending position of the occlusion area in an x-axis direction of the image; then performing a horizontal direction projection on the intercepted coarse positioning image to obtain a starting position and an ending position in a y-axis direction, and finally obtaining the mask image of the occlusion area; then the occlusion area is white-processed according to a wood board trend to obtain position information of the occlusion area and an image after the occlusion area of the wood board is white-processed. Wood board features are extracted from the mask image, the extracted feature vectors are input into a support vector machine for classification, and a classification type is acquired; a maximum inscribed rectangle algorithm is used to calculate a maximum cutting area in a wood board region after classification of non-defective wood boards, and then a cutting position is obtained; an improved Mask R-CNN algorithm is used, a ResNeSt network is used to replace a backbone network of the Mask R-CNN, the feature representation capability is enhanced, the network can better capture local and global information of the wood board, pixel-level positioning of the wood board is performed, then a maximum inscribed rectangle algorithm is performed to obtain the maximum cutting area in the wood board region, and a cutting method is determined according to a certain threshold size relationship between area differences of the two to segment the wood board.

2. The multi-modal image recognition based board grade classification and cutting method of claim 1, wherein, The preprocessing mode comprises: respectively performing grayscale, binarization and morphological processing on the image. 3.The multi-modal image recognition based board grading and cutting method of claim 1, wherein, The method for acquiring the coarse positioning image of the cut wood board comprises the following steps: The circular positioning point of the wood board cutting is detected by using a Hough transform, wherein the circular positioning point detected in the image is controlled based on gradient parameters by controlling a distance between the circles, a minimum radius of the circles and a maximum radius of the circles. 4.The multi-modal image recognition based board grading and cutting method of claim 3, wherein, The parameter control based on the gradient comprises the following steps: if the number of the detected circular positioning points meets a set threshold number, the angles of adjacent circular positioning points are calculated in sequence, and if the angles of the circular positioning points are all within a threshold, it is normal; if the number of the detected circular positioning points is less than the threshold number, the circular positioning points are prompted to be cleaned, and if the number is greater than the threshold number, the camera parameters are adjusted, and the current baffle position is recorded. 5.The multi-modal image recognition based board grading and cutting method of claim 1, wherein, The defect types of the wood board include knots, discoloration, insect eyes and decay, and are classified into three grades of whether there is a defect, straight wood board and non-straight wood board. 6.The multi-modal image recognition based board grading and cutting method of claim 1, wherein, The wood board features include a wood board width, a wood board height, a wood board area, a wood board average pixel value, a total area of circular shapes in the wood board, a number of circular shapes in the wood board, a wood board overall bending degree, a wood board cross-section width difference, a bending degree of a wood grain and a corresponding number of wood grains, a wood grain spacing, a number of wood grains, a horizontal projection histogram, and a vertical projection histogram.

7. A board grade classification and cutting system based on multi-modal image recognition, characterized in that, The method comprises the following steps: A wood board image preprocessing module is configured to acquire an image of a wood board to be processed, and preprocess the wood board image; The wood board shielding area processing module is configured to detect a circular positioning point of wood board cutting, and acquire a coarse positioning image of the cut wood board; and perform vertical direction projection according to the coarse positioning image to acquire a mask image of a projection shielding area. The process of performing vertical direction projection according to the coarse positioning image to acquire the mask image of the projection shielding area includes the following steps: performing vertical direction projection according to position information of the wood board circular positioning point to obtain a starting position and an ending position of the shielding area in the x-axis direction of the image; then performing horizontal direction projection on the intercepted coarse positioning image to obtain a starting position and an ending position in the y-axis direction, and finally obtaining the mask image of the shielding area; then performing white setting on the shielding area according to a wood board trend to obtain position information of the shielding area and an image after white setting of the wood board shielding area. The wood board grading module is configured to perform wood board feature extraction on the mask image, input the extracted feature vector into a support vector machine for classification, and acquire a grading type. The wood board cutting module is configured to perform maximum inscribed rectangle algorithm calculation on a non-defect wood board after grading to obtain a maximum cutting area in a wood board area, and then obtain a cutting position; and an improved Mask R-CNN algorithm is used, in which a ResNeSt network is mainly used to replace a backbone network of the Mask R-CNN to enhance the representation ability of features, so that the network can better capture local and global information of the wood board, perform pixel-level positioning on the wood board, then perform maximum inscribed rectangle algorithm to obtain the maximum cutting area in the wood board area, and determine a cutting method for wood board segmentation by using a certain threshold size relationship between area differences of the two.

8. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is configured to store computer instructions, and the computer instructions are executed by a processor to implement the wood board grading and cutting method based on multi-modal image recognition according to any one of claims 1-6.

9. An electronic device, comprising: The electronic device includes a processor, a memory, and a computer program; the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the wood board grading and cutting method based on multi-modal image recognition according to any one of claims 1-6. ​

Images