Historic building wood component tree species in-situ identification method and device, storage medium and equipment

Abstract

The invention discloses an ancient building wood component tree species in-situ identification method and device, a storage medium and equipment, and belongs to the technical field of wood identification. Comprising the steps that S1, a modern wood specimen classification feature reference data set is established, and the modern wood specimen classification feature reference data set comprises cross section image samples of modern wood; s2, establishing a convolutional neural network classification model, and training the convolutional neural network classification model by adopting the modern wood specimen classification feature reference data set; and S3, obtaining a cross section image of a to-be-identified ancient building wood component, and inputting the cross section image into the trained convolutional neural network classification model to obtain a tree species to which the ancient building wood component belongs. According to the method, in-situ rapid identification of the tree species of the ancient building wood component is realized by utilizing the constructed modern wood specimen image database and algorithm model, the problems that the number of samples of the ancient building wood component is small and data requirements of convolutional neural network construction are difficult to meet and the like are solved, and in-situ, rapid and accurate identification of the tree species of the ancient building wood component is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of wood identification, and in particular to a method, a device, a storage medium and equipment for in-situ identification of tree species of ancient building wood components. Background Art

[0002] Affected by the characteristics of wood itself, environmental and human factors, ancient wooden structures that have been in service for hundreds of years will suffer different types and degrees of damage, and the safety performance of the structure will be reduced. According to the national standard GB / T 50165-2020 "Technical Standard for Maintenance and Reinforcement of Ancient Building Wood Structures", when maintaining and reinforcing the wooden structures of ancient buildings, their original shape, structure, raw materials and craftsmanship should be preserved. When it is necessary to repair or replace the original wooden components, the wood used should be of the same species as the original components as much as possible. Therefore, accurately obtaining information on the tree species used for the wooden components of ancient buildings is the basis and prerequisite for the maintenance and protection of the wooden structures of ancient buildings.

[0003] The basic principle of wood anatomical identification is to distinguish between species based on the fine structural differences between wood species by comparing the sample to be tested with a wood specimen containing complete tree taxonomy and collection information. At present, the identification of tree species of wooden components of ancient buildings generally adopts the wood anatomical method, which requires sampling and destruction of wooden components, and taking the samples back to the laboratory for sample preparation, softening, slicing, staining, microscopic observation, and feature comparison and analysis. The process is cumbersome and the identification cycle is long. At the same time, for the wooden structure of ancient buildings, if all components are sampled, it will increase the workload on site, and it is difficult to achieve large-volume destructive sampling for some ancient wooden structures due to their unique historical value. In addition, this method mainly relies on the professional knowledge and experience of the identification personnel, and the results of tree species identification are greatly affected by human subjective factors.

[0004] In recent years, new wood identification technologies based on computer technology have been gradually developed in the field of wood species identification. They are mainly based on collecting wood surface images and predicting identification results based on convolutional neural networks. The relevant technology can not only realize on-site in-situ wood species identification, avoiding the complex steps of sampling and bringing back to laboratory experiments; at the same time, it can realize tree species identification in a short time, greatly shortening the detection cycle. Therefore, computer vision methods have great application potential in the field of tree species identification of ancient building wooden components.

[0005] Computer vision methods are data-driven and require a large amount of data to train neural network models. However, due to the unique cultural value of ancient wooden structures and the scarcity of wooden components, the available image data of ancient wooden components is extremely limited, which makes it difficult to meet the actual requirements of computer vision recognition technology for the amount of image data. Summary of the invention

[0006] In order to solve the defects of the prior art, the present invention provides a method, device, storage medium and equipment for in-situ identification of tree species of wooden components of ancient buildings, which solves the problem that the sample size of wooden components of ancient buildings is small and it is difficult to meet the data requirements for building convolutional neural networks, and realizes on-site, rapid and accurate identification of tree species of wooden components of ancient buildings.

[0007] The present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a method for in-situ identification of tree species of ancient building wooden components, the method comprising:

[0009] S1: establishing a modern wood specimen classification feature reference dataset, wherein the modern wood specimen classification feature reference dataset includes cross-section image samples of modern wood;

[0010] S2: establishing a convolutional neural network classification model, and using the modern wood specimen classification feature reference data set to train the convolutional neural network classification model;

[0011] S3: Obtain a cross-sectional image of the ancient building wooden component to be identified and input it into a trained convolutional neural network classification model to obtain the tree species to which the ancient building wooden component belongs.

[0012] Further, the S1 includes:

[0013] S11: Obtain cross-section image samples collected from accurately named modern wood specimens;

[0014] Wherein, the cross-section image samples include macroscopic structural images and microscopic structural images of modern wood cross-sections;

[0015] S12: performing data enhancement on the cross-section image sample;

[0016] S13: performing data cleaning on the cross-section image samples to remove abnormal data, and remove data whose collection location is located in the pith area of ​​the wood, the boundary area between young wood and mature wood, and the growth ring width is less than a set width or the difference between the widths of adjacent growth rings is greater than a set difference;

[0017] S14: adding a tree species label to each cross-section image sample to obtain the classification feature reference dataset of the modern wood specimens.

[0018] Further, the S2 includes:

[0019] S21: construct a convolutional neural network classification model, wherein the convolutional neural network classification model includes three parallel convolutional neural networks, each of which includes an input layer, a convolution layer, a pooling layer, an activation layer, and a fully connected layer;

[0020] S22: dividing the modern wood specimen classification feature reference data set into a training set and a validation set, training the convolutional neural network classification model according to a sample-level k-fold cross-validation method, and obtaining a test accuracy of the trained convolutional neural network classification model;

[0021] S23: Determine whether the test accuracy reaches a set accuracy threshold; if so, use all cross-section image samples included in the modern wood specimen classification feature reference data set to train the convolutional neural network classification model.

[0022] Further, the S3 includes:

[0023] S31: Acquire a cross-sectional image collected from a position to be processed of a wooden component of an ancient building to be identified;

[0024] The ratio of the deteriorated area in the latewood region of the position to be processed to the area of ​​the entire latewood region should be less than 30%, and the surface area of ​​the position to be processed does not include the wood pith area, the boundary area between young wood and mature wood, the area where the width of the growth ring is less than the set width, or the area where the difference in the width of adjacent growth rings is greater than the set difference; the position to be processed is subjected to surface sanding treatment using sandpaper of a set mesh; the ratio of the width of the growth ring included in the position to be processed to the width of the field of view of the image collector should be less than 1:3, and the field of view of the image collector should include at least one complete growth ring;

[0025] S32: The obtained cross-section image is input into a trained convolutional neural network classification model to obtain the tree species to which the wooden components of the ancient building belong.

[0026] In a second aspect, the present invention provides an in-situ identification device for tree species of ancient building wooden components, the device comprising:

[0027] A modern wood data set construction module is used to establish a modern wood specimen classification feature reference data set, wherein the modern wood specimen classification feature reference data set includes cross-section image samples of modern wood;

[0028] A model training module, used to establish a convolutional neural network classification model, and train the convolutional neural network classification model using the modern wood specimen classification feature reference data set;

[0029] The tree species recognition module is used to obtain the cross-sectional image of the wooden component of the ancient building to be identified and input it into the trained convolutional neural network classification model to obtain the tree species to which the wooden component of the ancient building belongs.

[0030] Furthermore, the modern wood dataset construction module includes:

[0031] A first acquisition unit is used to acquire a cross-section image sample collected from an accurately named modern wood specimen;

[0032] Wherein, the cross-section image samples include macroscopic structural images and microscopic structural images of modern wood cross-sections;

[0033] A data enhancement unit, used for performing data enhancement on the cross-section image samples;

[0034] A data cleaning unit is used to clean the cross-section image samples, remove abnormal data, and remove data whose collection location is located in the pith area of ​​the wood, the boundary area between young wood and mature wood, and the growth ring width is less than a set width or the difference between the widths of adjacent growth rings is greater than a set difference;

[0035] The data labeling unit is used to add a tree species label to each cross-section image sample to obtain the modern wood specimen classification feature reference data set.

[0036] Furthermore, the model training module includes:

[0037] A model building unit, used to build a convolutional neural network classification model, wherein the convolutional neural network classification model includes three parallel convolutional neural networks, each of which includes an input layer, a convolution layer, a pooling layer, an activation layer, and a fully connected layer;

[0038] A cross-validation unit, used to divide the modern wood specimen classification feature reference data set into a training set and a validation set, train the convolutional neural network classification model according to a sample-level k-fold cross-validation method, and obtain the test accuracy of the trained convolutional neural network classification model;

[0039] The training unit is used to determine whether the test accuracy reaches a set accuracy threshold. If so, all cross-section image samples included in the modern wood specimen classification feature reference data set are used to train the convolutional neural network classification model.

[0040] Furthermore, the tree species identification module includes:

[0041] The second acquisition unit is used to acquire a cross-sectional image collected from a to-be-processed position of a wooden component of an ancient building to be identified;

[0042] The ratio of the deteriorated area in the latewood region of the position to be processed to the area of ​​the entire latewood region should be less than 30%, and the surface area of ​​the position to be processed does not include the wood pith area, the boundary area between young wood and mature wood, the area where the width of the growth ring is less than the set width, or the area where the difference in the width of adjacent growth rings is greater than the set difference; the position to be processed is subjected to surface sanding treatment using sandpaper of a set mesh; the ratio of the width of the growth ring included in the position to be processed to the width of the field of view of the image collector should be less than 1:3, and the field of view of the image collector should include at least one complete growth ring;

[0043] The tree species identification unit is used to input the obtained cross-section image into a trained convolutional neural network classification model to obtain the tree species to which the wooden components of the ancient building belong.

[0044] In a third aspect, the present invention provides a computer-readable storage medium for in-situ identification of tree species of wooden components of ancient buildings, comprising a memory for storing processor-executable instructions, which, when executed by the processor, implement the steps of the in-situ identification method of tree species of wooden components of ancient buildings described in the first aspect.

[0045] In a fourth aspect, the present invention provides a device for in-situ identification of tree species of wooden components of ancient buildings, characterized in that it comprises at least one processor and a memory storing computer-executable instructions, and when the processor executes the instructions, it implements the steps of the in-situ identification method of tree species of wooden components of ancient buildings described in the first aspect.

[0046] The present invention has the following beneficial effects:

[0047] The present invention uses the constructed modern wood specimen image database and algorithm model to realize the in-situ rapid identification of the tree species of ancient building wood components, avoiding the complex processes of destructive sampling and laboratory experiments in traditional wood anatomical identification methods, and at the same time solving the problem that the small number of samples of ancient building wood components is difficult to meet the data requirements of convolutional neural network construction, and provides a new way for the in-situ, rapid and accurate identification of tree species of ancient building wood components. In addition, the present invention standardizes the image collection of ancient building wood components, ensuring the availability of images collected from ancient building wood components and the accuracy of identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the in-situ identification method of the tree species of ancient building wooden components of the present invention;

[0049] Figure 2 Schematic diagram of sample-level 5-fold cross validation;

[0050] Figure 3 The confusion matrix and sample confidence comparison diagram of the image data of the ancient building wooden components with intact materials; (a) is the confusion matrix, and (b) is the sample confidence;

[0051] Figure 4 This is an example diagram of the deterioration characteristics of wooden components of ancient buildings, where ad represents the cracking deterioration level (level 1-4), and eh represents the decay deterioration level (level 1-4);

[0052] Figure 5 This is a schematic diagram of the simulation of the deterioration characteristics of wooden components of ancient buildings;

[0053] Figure 6 Schematic diagram of confidence comparison analysis of samples of ancient building wood components with decay and deterioration characteristics;

[0054] Figure 7 Schematic diagram of confidence comparison analysis of samples of ancient building wood components with cracking and deterioration characteristics;

[0055] Figure 8 It is a schematic diagram of the in-situ identification device for tree species of ancient building wooden components of the present invention. DETAILED DESCRIPTION

[0056] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0057] The embodiment of the present invention provides a method for in-situ identification of tree species of ancient building wooden components. Figure 1 As shown, the method includes:

[0058] S1: Establish a reference dataset of classification characteristics of modern wood specimens. The reference dataset of classification characteristics of modern wood specimens includes cross-section image samples of modern wood.

[0059] The cross-section image samples may include macroscopic structural images and microscopic structural images of the cross-section of modern wood; the macroscopic structural image may be an image collected under a magnifying glass, and the microscopic structural image may be an image collected under an optical microscope.

[0060] This step uses modern wood specimens as library samples to establish a reference dataset containing a series of wood cross-section image samples. All samples in this reference dataset come from accurately named modern wood specimens with accurate tree species labels.

[0061] In an example, the specific implementation of this step includes:

[0062] S11: Obtain a sample of cross-section images collected from accurately named modern wood specimens.

[0063] For example, an iWood wood image acquisition device can be used to acquire cross-section structural images from modern wood specimens, or an optical microscope can be used to acquire cross-section structural images from modern wood specimens. The acquired wood cross-section images include macroscopic structural images and microscopic structural images, and the number of specimens of each "genus" should be no less than 60.

[0064] S12: Perform data enhancement on cross-section image samples.

[0065] Data augmentation processing includes image rotation, image scaling, image mirroring and / or image cropping operations to obtain more image information and enrich the sample diversity of the reference dataset.

[0066] S13: Perform data cleaning on the cross-section image samples to remove abnormal data, and remove data whose collection location is in the pith area of ​​the wood, the boundary area between young wood and mature wood, the growth ring width is less than the set width (i.e., the growth ring is too narrow), or the difference in width between adjacent growth rings is greater than the set difference.

[0067] S14: Add tree species tags to each cross-section image sample as labels to obtain a reference dataset of classification features of modern wood specimens.

[0068] S2: Establish a convolutional neural network classification model and use the modern wood specimen classification feature reference dataset to train the convolutional neural network classification model.

[0069] Specifically, this step includes:

[0070] S21: Construct a convolutional neural network classification model. The convolutional neural network classification model includes three parallel convolutional neural networks. Each convolutional neural network includes an input layer, a convolution layer, a pooling layer, an activation layer, and a fully connected layer. Improved strategies such as these improve recognition accuracy.

[0071] S22: Divide the modern wood specimen classification feature reference dataset into a training set and a validation set, train the convolutional neural network classification model according to the sample-level k-fold cross-validation method, and obtain the test accuracy of the trained convolutional neural network classification model.

[0072] For example, using the 5-fold cross validation method, the operation is as follows:

[0073] Using the 5-fold cross-validation method, the reference data set is divided into a training set and a test set at a ratio of 8:2 based on the sample. After the division, there should be 5 types of training and test data, and the test data and training data are from different samples. Sample-level 5-fold cross-validation is as follows Figure 2 shown.

[0074] The five training sets are used as inputs to the convolutional neural network model for training, and feature extraction is performed in the convolution layer, activation layer and pooling layer, and feature classification is performed through the fully connected layer. The parameters of the neural network are updated according to the loss function. The weight with the highest accuracy of the validation set after each iteration during the training process is used as the test weight. The test weights are tested on the five test sets respectively, and the test accuracy of the five weights can be obtained.

[0075] S23: Determine whether the test accuracy reaches the set accuracy threshold. If so, use all cross-section image samples included in the modern wood specimen classification feature reference data set to train the convolutional neural network classification model.

[0076] Calculate the average of the five test accuracies as the test accuracy. If the average accuracy is less than 80%, data supplementation or redesign of the convolutional neural network should be performed. If the average accuracy is greater than 80%, use all the data in the above reference dataset as the training set to train the model to ensure the generalization performance of the model.

[0077] S3: Obtain a cross-sectional image of the ancient building wooden component to be identified and input it into a trained convolutional neural network classification model to obtain the tree species to which the ancient building wooden component belongs.

[0078] The present invention directly identifies image data of ancient building wood components based on a constructed modern wood specimen image database and verifies its good generalization through unknown samples, thus effectively avoiding the difficulty of computer vision methods requiring the establishment of a large-scale database.

[0079] As an improvement, the S3 includes:

[0080] S31: Acquire a cross-sectional image collected from a position to be processed of a wooden component of an ancient building to be identified.

[0081] Compared with modern wood materials, the surface of wooden components of ancient buildings has defects such as decay, aging, cracking, and crushing, which greatly limits the image acquisition of wooden components. Therefore, in the process of image acquisition of ancient building components, the availability of on-site data acquisition must be strictly guaranteed, especially the standardized screening of wooden component surface treatment and image acquisition area.

[0082] To this end, this patent regulates the image acquisition of ancient building wood components, as follows:

[0083] The location to be treated should be located at a suitable treatable location of the wooden components of the ancient building, avoiding as much as possible damaged areas such as surface cracks, crushing, decay, and specific areas that are difficult to identify. The ratio of the deteriorated area in the late wood area to the overall late wood area should be less than 30%.

[0084] Furthermore, the surface area of ​​the position to be processed does not include the pith area of ​​the wood, the boundary area between young wood and mature wood, the area where the width of the growth ring is less than the set width, or the sapwood area where the difference in width between adjacent growth rings is greater than the set difference, etc.

[0085] Before collection, the position to be processed needs to be sanded with sandpaper of set mesh. For example, use 80 / 180 / 240 / 400 / 600 / 800 mesh sandpaper to sand the surface area of ​​the position to be processed, until the cross-sectional cells are clearly exposed and the surface of the area is flat and has no obvious scratches. The sanding depth is 1-5mm, and the sanding area should be slightly larger than the field of view of the collector.

[0086] Finally, use a collector to collect cross-section images. The collector parameters should be consistent with the collector used in the modern specimen reference dataset. The ratio of the growth ring width to the collector field of view width should be less than 1:3, and the field of view should contain at least one complete growth ring.

[0087] S32: The obtained cross-section image is input into a trained convolutional neural network classification model to obtain the tree species to which the wooden components of the ancient building belong.

[0088] After the above-mentioned in-situ identification of tree species of ancient building wood components is achieved, in order to verify the effectiveness of the identification method provided by the present invention, the accuracy of the method of the present invention is evaluated. The specific process is as follows:

[0089] 1. First, select the evaluation indicators.

[0090] The performance of the model is evaluated by accuracy and sample recognition rate. Accuracy is a common evaluation indicator in the field of image recognition, which is specifically expressed as:

[0091]

[0092] TP is the positive sample predicted by the model as positive, TN is the negative sample predicted by the model as negative, FP is the negative sample predicted by the model as positive, and FN is the positive sample predicted by the model as negative.

[0093] The sample recognition rate is expressed as:

[0094]

[0095] When the test confidence of a sample is greater than 70% or 90%, it is considered a correct sample. The sample confidence is expressed as:

[0096]

[0097] 2. In order to verify the effectiveness of the recognition method provided by the present invention, the present invention conducted a series of experiments on the estimated results for intact and degraded material data. First, a cross-section data set of wood from common pine tree species used for wooden components of ancient Chinese buildings was constructed, including 38,953 images from 481 wood specimens from four genera, including Pinus, Larix, Picea and Abies. Four algorithms were established, namely ResNet50, SeResNet50, RepVGG-B2 and RepLKNet-31B. Under the data set division based on specimens, RepLKNet-31B is the optimal model, and the average accuracy of its five test results is 96.89%, indicating that the constructed model has good generalization, so the above 38,953 images are placed in the RepLKNet-31B model for training.

[0098] 4050 wood cross-section image data from 60 samples of ancient building wood components with intact materials were used for testing and analysis. The results showed that the identification results of the in-situ identification method of ancient building wood components based on convolutional neural networks were consistent with the identification results of wood anatomy experts at 96.67%. 90% of the samples had a confidence level higher than 90%, and 93.33% of the samples had a confidence level higher than 70% ( Figure 3 ). The recognition model constructed by this patent can realize the rapid and accurate recognition of the cross-sectional image data of intact wood components.

[0099] 3. According to preliminary statistics, decay and cracking along the grain are the most common types of deterioration in wooden components of ancient buildings. Therefore, the recognition performance of the corresponding model was tested for wood image data with two types of deterioration, cracking and decay. Generally, decay deterioration types are more likely to occur in the early wood area of ​​the wood, while cracking deterioration types can occur in both the early wood and late wood areas ( Figure 4 ). This patent collected 530 and 860 images of decay and cracking of wood components, respectively, and classified them according to the deterioration area (see Table 1 below). Since there are few actual data of deterioration area greater than 30% in the cracking type, this type of data was simulated. The simulation results are as follows: Figure 5 The test results show that since the latewood area is relatively complete, the decay and deterioration type has little effect on the model recognition accuracy (see Table 2 below). Figure 6As shown in the figure, the confidence of samples with decay and deterioration is close to the image recognition accuracy of intact materials. However, due to the incomplete characteristics of latewood, the cracking and deterioration characteristics have a greater impact on the model accuracy. When the cracking area increases, the model accuracy decreases significantly. When the cracking area is greater than 30%, the image confidence of the cracking feature is significantly lower than that of the intact material data ( Figure 7 ). Therefore, when collecting images of wooden components of ancient buildings, the ratio of latewood deterioration area to the overall latewood area should be kept below 30%.

[0100] Table 1 Classification of deterioration characteristics of wood components

[0101]

[0102] Table 2 Test results of different levels of degradation data

[0103]

[0104] In summary, the present invention uses the constructed modern wood specimen image database and algorithm model to realize the in-situ rapid identification of the tree species of ancient building wood components, avoiding the complex processes of destructive sampling and laboratory experiments in traditional wood anatomical identification methods, and at the same time solving the problem that the small number of samples of ancient building wood components is difficult to meet the data requirements of convolutional neural network construction, and provides a new way for the in-situ, rapid and accurate identification of tree species of ancient building wood components. In addition, the present invention standardizes the image collection of ancient building wood components to ensure the accuracy of identification.

[0105] The embodiment of the present invention also provides an in-situ identification device for tree species of ancient building wooden components, such as Figure 8 As shown, the device comprises:

[0106] The modern wood data set construction module 1 is used to establish a modern wood specimen classification feature reference data set, and the modern wood specimen classification feature reference data set includes cross-section image samples of modern wood.

[0107] Model training module 2 is used to establish a convolutional neural network classification model and train the convolutional neural network classification model using a modern wood specimen classification feature reference data set.

[0108] The tree species identification module 3 is used to obtain the cross-sectional image of the wooden component of the ancient building to be identified and input it into the trained convolutional neural network classification model to obtain the tree species to which the wooden component of the ancient building belongs.

[0109] As an improvement, the aforementioned modern wood dataset building blocks include:

[0110] The first acquisition unit is used to acquire cross-section image samples collected from accurately named modern wood specimens.

[0111] Among them, the cross-section image samples include macroscopic structural images and microscopic structural images of modern wood cross-sections.

[0112] The data enhancement unit is used to perform data enhancement on cross-section image samples.

[0113] The data cleaning unit is used to clean the cross-section image samples, remove abnormal data, and remove data whose collection location is in the pith area of ​​the wood, the boundary area between young wood and mature wood, and the growth ring width is less than the set width or the difference between the widths of adjacent growth rings is greater than the set difference.

[0114] The data labeling unit is used to add a tree species label to each cross-section image sample to obtain a reference data set of classification characteristics of modern wood specimens.

[0115] Specifically, the model training module includes:

[0116] The model building unit is used to build a convolutional neural network classification model. The convolutional neural network classification model includes three parallel convolutional neural networks. Each convolutional neural network includes an input layer, a convolution layer, a pooling layer, an activation layer and a fully connected layer.

[0117] The cross-validation unit is used to divide the modern wood specimen classification feature reference data set into a training set and a validation set, train the convolutional neural network classification model according to the sample-level k-fold cross-validation method, and obtain the test accuracy of the trained convolutional neural network classification model.

[0118] The training unit is used to determine whether the test accuracy reaches the set accuracy threshold. If so, all cross-section image samples contained in the modern wood specimen classification feature reference data set are used to train the convolutional neural network classification model.

[0119] As an example, the tree species identification module includes:

[0120] The second acquisition unit is used to acquire a cross-sectional image collected from a position to be processed of a wooden component of an ancient building to be identified.

[0121] Among them, the ratio of the deteriorated area in the latewood area of ​​the position to be treated to the area of ​​the entire latewood area should be less than 30%, the surface area of ​​the position to be treated does not include the wood pith area, the boundary area between young wood and mature wood, the area where the growth ring width is less than the set width, or the area where the difference in width between adjacent growth rings is greater than the set difference; the position to be treated is surface sanded with sandpaper of set mesh; the ratio of the growth ring width included in the position to be treated to the field of view width of the image collector should be less than 1:3, and the field of view of the image collector should include at least one complete growth ring.

[0122] The tree species identification unit is used to input the acquired cross-section image into the trained convolutional neural network classification model to obtain the tree species to which the wooden components of the ancient building belong.

[0123] The present invention uses the constructed modern wood specimen image database and algorithm model to realize the in-situ rapid identification of the tree species of ancient building wood components, avoiding the complex processes of destructive sampling and laboratory experiments in traditional wood anatomical identification methods, and at the same time solving the problem that the small number of samples of ancient building wood components is difficult to meet the data requirements of convolutional neural network construction, and provides a new way for the in-situ, rapid and accurate identification of tree species of ancient building wood components. In addition, the present invention standardizes the image collection of ancient building wood components to ensure the accuracy of identification.

[0124] The device provided in the above embodiment, its implementation principle and the technical effect produced correspond to the embodiment of the above method one by one. For the sake of brief description, the part not mentioned in the embodiment of the device can refer to the corresponding content in the embodiment of the above method. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the modules and units described in the device can refer to the corresponding process in the embodiment of the above method, and will not be repeated here.

[0125] The method for in-situ identification of tree species of wooden components of ancient buildings described in the above embodiments provided by the present invention can implement business logic through a computer program and record it on a storage medium, and the storage medium can be read and executed by a computer to achieve the effect of the scheme described in the method embodiment of this specification. Therefore, the embodiment of the present invention also provides a computer-readable storage medium for in-situ identification of tree species of wooden components of ancient buildings, including a memory for storing processor-executable instructions, and when the instructions are executed by the processor, the steps of the method for in-situ identification of tree species of wooden components of ancient buildings including the above embodiments are implemented.

[0126] The storage medium may include a physical device for storing information, which is usually a medium that digitizes the information and then stores it in an electrical, magnetic or optical manner. The storage medium may include: a device that stores information in an electrical energy manner, such as various memories, such as RAM, ROM, etc.; a device that stores information in a magnetic energy manner, such as a hard disk, a floppy disk, a magnetic tape, a magnetic core memory, a magnetic bubble memory, a USB flash drive; a device that stores information in an optical manner, such as a CD or a DVD. Of course, there are other readable storage media, such as quantum memory, graphene memory, etc.

[0127] The storage medium described above may also include other implementation methods according to the description of the method embodiment. The implementation principle and technical effects produced by this embodiment are the same as those of the aforementioned method embodiment. For details, please refer to the description of the relevant method embodiment, and no further description will be given here.

[0128] The embodiment of the present invention also provides a device for in-situ identification of tree species of wooden components of ancient buildings, which may be a separate computer, or may include an actual operating device using one or more of the methods or one or more of the embodiments of this specification. The device for in-situ identification of tree species of wooden components of ancient buildings may include at least one processor and a memory storing computer executable instructions, and when the processor executes the instructions, any one or more steps of the in-situ identification method of tree species of wooden components of ancient buildings are implemented.

[0129] The device described above may also include other implementation methods according to the description of the method embodiment. The implementation principle and technical effects produced by this embodiment are the same as those of the aforementioned method embodiment. For details, please refer to the description of the relevant method embodiment, and no further description will be given here.

[0130] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for in-situ identification of tree species of ancient building wood components, characterized in that: The method comprises: S1: establishing a modern wood specimen classification feature reference dataset, wherein the modern wood specimen classification feature reference dataset includes cross-section image samples of modern wood; S2: establishing a convolutional neural network classification model, and using the modern wood specimen classification feature reference data set to train the convolutional neural network classification model; S3: Obtain a cross-sectional image of the ancient building wooden component to be identified and input it into a trained convolutional neural network classification model to obtain the tree species to which the ancient building wooden component belongs.

2. The in-situ identification method of tree species of ancient building wood components according to claim 1 is characterized in that: The S1 includes: S11: Obtain cross-section image samples collected from accurately named modern wood specimens; Wherein, the cross-section image samples include macroscopic structural images and microscopic structural images of modern wood cross-sections; S12: performing data enhancement on the cross-section image sample; S13: performing data cleaning on the cross-section image samples to remove abnormal data, and remove data whose collection location is located in the pith area of ​​the wood, the boundary area between young wood and mature wood, and the growth ring width is less than a set width or the difference between the widths of adjacent growth rings is greater than a set difference; S14: adding a tree species label to each cross-section image sample to obtain the classification feature reference dataset of the modern wood specimens.

3. The in-situ identification method of tree species of ancient building wood components according to claim 2 is characterized in that: The S2 includes: S21: construct a convolutional neural network classification model, wherein the convolutional neural network classification model includes three parallel convolutional neural networks, each of which includes an input layer, a convolution layer, a pooling layer, an activation layer, and a fully connected layer; S22: dividing the modern wood specimen classification feature reference data set into a training set and a validation set, training the convolutional neural network classification model according to a sample-level k-fold cross-validation method, and obtaining a test accuracy of the trained convolutional neural network classification model; S23: Determine whether the test accuracy reaches a set accuracy threshold; if so, use all cross-section image samples included in the modern wood specimen classification feature reference data set to train the convolutional neural network classification model.

4. The method for in-situ identification of tree species of ancient building wood components according to claim 3 is characterized in that: The S3 includes: S31: Acquire a cross-sectional image collected from a position to be processed of a wooden component of an ancient building to be identified; The ratio of the deteriorated area in the latewood region of the position to be processed to the area of ​​the entire latewood region should be less than 30%, and the surface area of ​​the position to be processed does not include the wood pith area, the boundary area between young wood and mature wood, the area where the width of the growth ring is less than the set width, or the area where the difference in the width of adjacent growth rings is greater than the set difference; the position to be processed is subjected to surface sanding treatment using sandpaper of a set mesh; the ratio of the width of the growth ring included in the position to be processed to the width of the field of view of the image collector should be less than 1:3, and the field of view of the image collector should include at least one complete growth ring; S32: The obtained cross-section image is input into a trained convolutional neural network classification model to obtain the tree species to which the wooden components of the ancient building belong.

5. An in-situ identification device for tree species of ancient building wooden components, characterized in that: The device comprises: A modern wood data set construction module is used to establish a modern wood specimen classification feature reference data set, wherein the modern wood specimen classification feature reference data set includes cross-section image samples of modern wood; A model training module, used to establish a convolutional neural network classification model, and train the convolutional neural network classification model using the modern wood specimen classification feature reference data set; The tree species recognition module is used to obtain the cross-sectional image of the wooden component of the ancient building to be identified and input it into the trained convolutional neural network classification model to obtain the tree species to which the wooden component of the ancient building belongs.

6. The in-situ identification device for tree species of ancient building wood components according to claim 5 is characterized in that: The modern wood dataset building blocks include: A first acquisition unit is used to acquire a cross-section image sample collected from an accurately named modern wood specimen; Wherein, the cross-section image samples include macroscopic structural images and microscopic structural images of modern wood cross-sections; A data enhancement unit, used for performing data enhancement on the cross-section image samples; A data cleaning unit is used to clean the cross-section image samples, remove abnormal data, and remove data whose collection location is located in the pith area of ​​the wood, the boundary area between young wood and mature wood, and the growth ring width is less than a set width or the difference between the widths of adjacent growth rings is greater than a set difference; The data labeling unit is used to add a tree species label to each cross-section image sample to obtain the modern wood specimen classification feature reference data set.

7. The in-situ identification device for tree species of ancient building wood components according to claim 6 is characterized in that: The model training module includes: A model building unit, used to build a convolutional neural network classification model, wherein the convolutional neural network classification model includes three parallel convolutional neural networks, each of which includes an input layer, a convolution layer, a pooling layer, an activation layer, and a fully connected layer; A cross-validation unit, used to divide the modern wood specimen classification feature reference data set into a training set and a validation set, train the convolutional neural network classification model according to a sample-level k-fold cross-validation method, and obtain the test accuracy of the trained convolutional neural network classification model; The training unit is used to determine whether the test accuracy reaches a set accuracy threshold. If so, all cross-section image samples included in the modern wood specimen classification feature reference data set are used to train the convolutional neural network classification model.

8. The in-situ identification device for tree species of ancient building wood components according to claim 7 is characterized in that: The tree species identification module comprises: The second acquisition unit is used to acquire a cross-sectional image collected from a to-be-processed position of a wooden component of an ancient building to be identified; The ratio of the deteriorated area in the latewood region of the position to be processed to the area of ​​the entire latewood region should be less than 30%, and the surface area of ​​the position to be processed does not include the wood pith area, the boundary area between young wood and mature wood, the area where the width of the growth ring is less than the set width, or the area where the difference in the width of adjacent growth rings is greater than the set difference; the position to be processed is subjected to surface sanding treatment using sandpaper of a set mesh; the ratio of the width of the growth ring included in the position to be processed to the width of the field of view of the image collector should be less than 1:3, and the field of view of the image collector should include at least one complete growth ring; The tree species identification unit is used to input the obtained cross-section image into a trained convolutional neural network classification model to obtain the tree species to which the wooden components of the ancient building belong.

9. A computer-readable storage medium for in-situ identification of tree species of ancient building wooden components, characterized in that: It comprises a memory for storing processor executable instructions, and when the instructions are executed by the processor, the steps of the method for in-situ identification of tree species of wooden components of ancient buildings as described in any one of claims 1-4 are implemented.

10. A device for in-situ identification of tree species of ancient building wood components, characterized in that: The method comprises at least one processor and a memory storing computer executable instructions, wherein when the processor executes the instructions, the steps of the method for in-situ identification of tree species of wooden components of ancient buildings as claimed in any one of claims 1 to 4 are implemented.

Images