A method and system for building a library of kai wood carving resources based on internet information crawling

Abstract

The application discloses a kind of based on internet information crawling's Kai wood carving resource library construction method and system, it is related to data processing technical field, the method includes: setting up crawling factor, automatically crawls internet picture and classifies it, obtains picture classification cluster;Kai wood standard feature set is constructed and is used as training data, obtains standard identification submodel, determines network sharing weight, constructs twinborn network discriminant model;Determine matching standard feature set;Picture classification cluster is input into twinborn network discriminant model with matching standard feature set, obtains discriminant result;When discriminant result meets filing requirement, corresponding picture is stored in Kai wood resource library corresponding category resource set.The present application solves the technical problem that the accuracy of the Kai wood picture resources obtained in the prior art is low, resulting in poor quality of Kai wood resource library resources, achieves high accuracy through internet information crawling and screening Kai wood picture resources, and improves the technical effect of Kai wood resource library resource quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a method and system for constructing a Kai wood carving resource library based on internet information crawling. BACKGROUND

[0002] Internet information crawling is to collect data on the Internet through a web crawler, which is a program or script that automatically captures Web information according to certain rules.

[0003] Kai wood carving is a unique traditional handicraft art with unique artistic style and characteristics, and has a history of more than 2400 years, and has important historical value, so it needs to be protected and inherited, and the construction of Kai wood resource library needs to be done. However, due to its complex manufacturing process and rich types, there are currently high difficulty in screening Kai wood picture resources, resulting in low accuracy of screened pictures and poor quality of Kai wood resource library resources. SUMMARY

[0004] The present application provides a method and system for constructing a Kai wood carving resource library based on internet information crawling, which is used to solve the technical problem of low accuracy of Kai wood picture resources and poor quality of Kai wood resource library resources.

[0005] The first aspect of the present application provides a method for constructing a Kai wood carving resource library based on internet information crawling, the method comprising: setting a crawling factor, automatically crawling internet pictures based on the crawling factor, and constructing a picture dataset; classifying the picture dataset according to a preset classification feature to obtain a picture classification cluster, the preset classification feature being a Kai wood morphological feature; constructing a Kai wood standard feature set, the Kai wood standard feature set corresponding to the preset classification feature; obtaining a standard recognition sub-model according to the Kai wood standard feature set as training data, and determining a network sharing weight based on the standard recognition sub-model; constructing a twin network discrimination model based on the standard recognition sub-model and the network sharing weight; performing feature matching according to the picture classification cluster and the Kai wood standard feature set to determine a matching standard feature set; inputting the picture classification cluster and the matching standard feature set into the twin network discrimination model to obtain a discrimination result; and when the discrimination result meets the archiving requirements, storing the corresponding picture in the Kai wood resource library corresponding to the category resource set.

[0006] In a second aspect of the present application, an internet information crawling-based Kai wood carving resource library construction system is provided, which comprises: a picture dataset construction module, configured to set a crawling factor, automatically crawl internet pictures based on the crawling factor, and construct a picture dataset; a picture classification cluster obtaining module, configured to classify the picture dataset according to a preset classification feature, obtain a picture classification cluster, and the preset classification feature is a Kai wood morphological feature; a Kai wood standard feature set construction module, configured to construct a Kai wood standard feature set, and the Kai wood standard feature set corresponds to the preset classification feature; a standard recognition sub-model construction module, configured to obtain a standard recognition sub-model according to the Kai wood standard feature set as training data, and determine a network shared weight based on the standard recognition sub-model; a twin network discriminant model construction module, configured to construct a twin network discriminant model based on the standard recognition sub-model and the network shared weight; a matching standard feature set determination module, configured to determine a matching standard feature set by performing feature matching between the picture classification cluster and the Kai wood standard feature set; a discriminant result obtaining module, configured to input the picture classification cluster and the matching standard feature set into the twin network discriminant model to obtain a discriminant result; and a picture archiving module, configured to store a corresponding picture into a corresponding category resource set of a Kai wood resource library when the discriminant result meets archiving requirements.

[0007] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0008] The internet information crawling-based Kai wood carving resource library construction method provided in the present application relates to the technical field of data processing. By setting a crawling factor, internet pictures are automatically crawled and classified to obtain a picture classification cluster. A Kai wood standard feature set is constructed and used as training data to obtain a standard recognition sub-model, determine a network shared weight, and construct a twin network discriminant model. A matching standard feature set is determined. The picture classification cluster and the matching standard feature set are input into the twin network discriminant model to obtain a discriminant result. When the discriminant result meets archiving requirements, a corresponding picture is stored into a corresponding category resource set of a Kai wood resource library. The technical problem of poor resource quality of a Kai wood resource library due to low accuracy of obtained Kai wood picture resources in the prior art is solved. The technical effect of obtaining Kai wood picture resources with high accuracy through internet information crawling and screening is achieved, and the resource quality of the Kai wood resource library is improved. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor based on these drawings.

[0010] Figure 1 A flowchart of a method for constructing a Kai wood carving resource library based on internet information crawling provided by an embodiment of the present application is shown in the figure.

[0011] Figure 2 A flowchart of a method for constructing a Kai wood carving resource library based on internet information crawling provided by an embodiment of the present application is shown in the figure.

[0012] Figure 3 A flowchart of a method for constructing a Kai wood carving resource library based on internet information crawling provided by an embodiment of the present application is shown in the figure.

[0013] Figure 4 A flowchart of a system structure for constructing a Kai wood carving resource library based on internet information crawling provided by an embodiment of the present application is shown in the figure.

[0014] Explanation of reference signs: picture data set construction module 11, picture classification cluster obtaining module 12, Kai wood standard feature set construction module 13, standard identification sub-model construction module 14, twin network discrimination model construction module 15, matching standard feature set determination module 16, discrimination result obtaining module 17, picture archiving module 18. DETAILED DESCRIPTION

[0015] The present application provides a method for constructing a Kai wood carving resource library based on internet information crawling, which is used to solve the technical problem of poor quality of Kai wood resource library resources due to low accuracy of obtained Kai wood picture resources in the prior art.

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0017] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and in the above drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices.

[0018] Embodiment one

[0019] As Figure 1 shown, the present application provides a method for building a Kai wood carving resource library based on Internet information crawling, the method comprising:

[0020] S100: setting a crawling factor, automatically crawling Internet pictures based on the crawling factor, and constructing a picture data set;

[0021] Specifically, the web crawler is a program or script that automatically captures World Wide Web information according to certain rules, so the crawling factor is the collection rule for participating in Internet pictures. Based on the picture collection requirements of the Kai wood resource library, the crawling factor is set, and the pictures related to Kai wood on the Internet are automatically captured based on the crawling factor to construct a Kai wood picture data set.

[0022] S200: classifying the picture data set according to a preset classification feature to obtain a picture classification cluster, the preset classification feature being a Kai wood morphological feature;

[0023] Specifically, the pictures in the picture data set are classified according to a preset classification feature, the preset classification feature being a Kai wood morphological feature, including the shape of Kai wood, commonly used tool types, color, texture, texture, gloss, etc. Pictures with the same features are classified into the same picture classification set to obtain multiple picture classification sets, and the picture classification cluster is composed of multiple picture classification sets, which can be used as basic data for Kai wood picture resource extraction.

[0024] Further, as Figure 2 shown, the step S200 of the present application further comprises:

[0025] S210: inputting the pictures in the picture data set into a picture frame recognition model to obtain a picture frame recognition result;

[0026] S220: clustering based on the picture frame recognition result and the preset classification feature, and determining the classification result of the picture according to the clustering result;

[0027] S230: storing all pictures into the corresponding classification picture set, and constructing the picture classification cluster.

[0028] Specifically, the pictures in the picture data set are sequentially input into the frame recognition model, and the picture frame recognition result is obtained after the picture frame recognition by the frame recognition model. The frame recognition model is a model for recognizing the edge of the Kai wood instrument in the Kai wood picture, which can be constructed based on a semantic segmentation or watershed segmentation algorithm. The semantic segmentation is a deep learning algorithm for associating a label or category with each pixel of a picture, which can be used to identify a pixel set constituting a distinguishable category. By performing picture frame recognition, that is, recognizing the edge of the Kai wood instrument, the Kai wood instrument itself is separated from other backgrounds in the picture, and only the picture area of the Kai wood instrument itself is extracted as the image recognition area to exclude the interference of other backgrounds of the picture, thereby obtaining the picture frame recognition result.

[0029] Further, the recognition area of each picture is extracted based on the picture frame recognition result and the preset classification feature, and clustering is performed according to the preset classification feature. The clustering is a process of classifying data into different classes or clusters. The objects in the same cluster have great similarity, while the objects in different clusters have great dissimilarity. The pictures in the picture data set are classified into different clusters according to the preset classification feature, and the classification result of the picture, that is, the classification picture set, is obtained. All pictures are stored into the corresponding classification picture set, and the picture classification cluster is constructed by the multiple classification picture sets.

[0030] Further, the step S210 of the embodiment of the present application further includes:

[0031] S211: preprocessing the picture data set and setting a starting point;

[0032] S212: setting a distance constraint threshold according to the gray value distribution of the picture data set;

[0033] S213: performing liquid surface rising fitting based on the starting point. When the liquid surfaces corresponding to any starting point intersect, a watershed line is generated at the corresponding position. When the liquid surface rises to the maximum gray value of the image, the liquid surface fitting is stopped;

[0034] S214: performing image segmentation by using the watershed line to obtain the picture frame recognition result.

[0035] Specifically, binarization or grayscale processing is performed on all pictures in the picture data set in advance, the gray value of the pixel point on the image is set to 0 or 255, that is, the entire image presents a clear visual effect of only black and white, all pictures are converted into grayscale pictures, the image is simple and easy to highlight the target contour. Further, the pixel point with the smallest gray value is set as a starting point, a distance constraint threshold is set according to the gray value distribution of the pictures in the picture data set, the distance constraint threshold is a distance threshold of the pixel point to the corresponding starting point, liquid surface rising simulation is performed based on the starting point, if the distance of the pixel point to the corresponding starting point is less than the distance constraint threshold, these pixels are flooded, the liquid surface continues to rise until the corresponding liquid surface of any starting point intersects, a watershed line is generated at the corresponding position, that is, when the distance of the pixel point to the corresponding starting point is greater than or equal to the distance constraint threshold, a watershed line is set on these pixels, when the liquid surface rises to the maximum gray value of the image, all regions meet on the watershed line, the liquid surface fitting is stopped, the image is segmented by using the watershed line, and the picture is divided into a Chinese cypress utensil region and other interference regions, which are taken as the picture frame recognition result.

[0036] S300: Construct a Chinese cypress standard feature set corresponding to the preset classification feature;

[0037] Specifically, based on big data, the use fields of Chinese cypress and the display forms in each use field are obtained, such as being used as living utensils, furniture, building materials and the like, and then the appearance features of different Chinese cypress utensils from different fields are extracted, including the appearance features of logs, boards, house component wood carvings, handles and other living utensils, and the iconic features are extracted therefrom as Chinese cypress standard features, for example, the Chinese cypress bark is light reddish brown, the heartwood is dark greenish brown, the annual rings are obvious, the texture is tight and tough, and planing has a bright luster, and a plurality of Chinese cypress iconic features are used to construct a Chinese cypress standard feature set corresponding to the preset classification feature, which both contain a plurality of Chinese cypress shape features and can be used as a standard reference for subsequent Chinese cypress picture recognition.

[0038] S400: Obtain a standard recognition sub-model according to the Chinese cypress standard feature set as training data, and determine network sharing weights based on the standard recognition sub-model;

[0039] Specifically, the standard feature set of the Chinese fir is taken as training data, a standard recognition sub-model is constructed based on a BP neural network, and supervised training is performed using the standard feature set of the Chinese fir until the standard recognition sub-model converges and meets a preset accuracy requirement, so as to obtain the standard recognition sub-model used for recognizing standard features of the Chinese fir in a picture. The BP neural network is a multi-layer feedforward neural network trained according to an error back propagation algorithm, and does not need to determine a mathematical equation of a mapping relationship between input and output in advance, but learns a certain rule through training of itself, and obtains a result closest to an expected output value when a given input value is input. Further, based on the standard recognition sub-model, weight distribution of each parameter in the standard recognition sub-model is obtained as network shared weights, which can be used as a subsequent twin network discrimination model.

[0040] Further, the step S400 of the embodiment of the present application further includes:

[0041] S410: determining a classification cluster core area according to the picture classification cluster;

[0042] S420: determining a core discrimination area according to the picture frame recognition result and the classification cluster core area, and marking the picture.

[0043] Specifically, the core area of each picture classification cluster is determined according to the picture classification cluster, the core area is a key recognition area in the picture, including an area of a Chinese fir morphological feature set or an area where a landmark feature appears, such as an area where a texture of a Chinese fir board appears, a surface layer of Chinese fir furniture or a landmark pattern, and the like. The core discrimination area is determined and the picture is marked, so as to improve the efficiency and accuracy of feature extraction.

[0044] Further, the step S420 of the embodiment of the present application further includes:

[0045] S421: when the classification cluster core area is a multi-area, performing case library collection based on the picture classification cluster and the multi-area recognition area;

[0046] S422: performing weight calculation of each core area based on the classification cluster case library, and determining the weight of each area;

[0047] S423: setting an identification area level according to the weight of each area, and generating an area mark based on the identification area level.

[0048] Specifically, when the classification cluster core area is a multi-area, such as a multi-part furniture composed of a plurality of parts, each part has different morphological characteristics, then there are multiple core areas, based on the picture classification cluster, multi-area identification area, feature case collection, such as collecting the identification features of the chair leg, seat and back of the Chinese classical wooden chair, and the classification cluster case library is composed of a plurality of feature cases, and the weight distribution of each core area is based on the classification cluster case library to determine the weight of each area. For example, the seat area of the Chinese classical wooden chair is larger, and the feature recognition is relatively easy, so a larger weight is allocated, and the chair leg and seat area are smaller, so a smaller weight is allocated. Further, according to the weight of each area, the identification area level is set, the weight is larger, and the identification area level is higher, and the area mark is generated based on the identification area level, which is convenient for subsequent feature recognition in the area.

[0049] S500: Construct a twin network discriminant model based on the standard identification sub-model and the network shared weight;

[0050] Specifically, based on the standard identification sub-model and the network shared weight, a twin network discriminant model is constructed, which includes the standard identification sub-model and another identification sub-model which is completely identical to it, that is, two network models with shared weights. Two groups of feature recognition data can be output by inputting two image samples, one being a standard Chinese classical wooden sample and the other being an automatically crawled Internet picture sample. The similarity of the two groups of feature recognition data is compared to determine whether the automatically crawled Internet picture sample belongs to the Chinese classical wooden related picture, so as to improve the accuracy and efficiency of picture recognition.

[0051] S600: According to the picture classification cluster and the Chinese classical wooden standard feature set, feature matching is performed to determine the matching standard feature set;

[0052] Specifically, all pictures in the picture classification cluster are matched with the Chinese classical wooden standard feature set, and all standard features in the picture classification cluster that need to be identified are screened out, such as color, texture, glossiness, etc. The corresponding feature pictures are extracted from the Chinese classical wooden standard feature set to form a matching standard feature set, which is used for subsequent Chinese classical wooden standard feature recognition analysis, and the feature recognition analysis result is used to judge the similarity of the pictures.

[0053] S700: Input the picture classification cluster and the matching standard feature set into the twin network discriminant model to obtain a discriminant result;

[0054] Further, as shown in Figure 3 , the step S700 of the embodiment of the application further includes:

[0055] S710: Feature recognition analysis is performed on the matching standard feature set by a standard recognition sub-model in the twin network discriminant model to obtain standard feature analysis data;

[0056] S720: Feature recognition analysis is performed on each picture in the picture classification cluster by a sample recognition sub-model in the twin network discriminant model to obtain sample feature analysis data;

[0057] S730: Based on the standard feature analysis data and the sample feature analysis data, a feature loss value is obtained by loss function calculation;

[0058] S740: When the feature loss value reaches a preset threshold, the discriminant result is obtained as being different.

[0059] S750: When the feature loss value does not reach the preset threshold, the discriminant result is obtained as being the same.

[0060] Specifically, the picture classification cluster and the matching standard feature set are respectively input into the standard recognition sub-model and the sample recognition sub-model of the twin network discriminant model, the sample recognition sub-model is the same model as the standard recognition sub-model sharing weights, feature recognition analysis is performed on the matching standard feature set by the standard recognition sub-model in the twin network discriminant model, the Kai wood features in the matching standard feature set are extracted, a feature vector is generated as standard feature analysis data, and similarly, feature recognition analysis is performed on each picture in the picture classification cluster by the sample recognition sub-model in the twin network discriminant model, another feature vector is generated as sample feature analysis data.

[0061] Further, the standard feature analysis data and the sample feature analysis data are input into a loss function for calculation to obtain a feature loss value, the loss function is a function for reflecting the difference between the input pictures by calculating the "distance" between the two feature vectors, the difference between the pictures in the picture classification cluster and the matching standard feature set is the feature loss value, the greater the feature loss value, the greater the "distance" between the two feature vectors, and the greater the difference between the two pictures. When the feature loss value reaches a preset threshold, it means that the difference between the two pictures exceeds the preset maximum difference, and the picture similarity is low, so the discriminant result is different, that is, the picture does not meet the characteristics of Kai wood and cannot be stored in the Kai wood resource library. On the contrary, when the feature loss value does not reach the preset threshold, it means that the difference between the two pictures is less than the maximum difference, so the discriminant result is the same, and the picture can be stored in the Kai wood resource library.

[0062] S800: When the discriminant result meets the archiving requirement, the corresponding picture is stored in the Kai wood resource library corresponding to the category resource set.

[0063] Specifically, when the determination result meets the archiving requirement, that is, the feature loss value obtained after the automatically crawled picture from the Internet is analyzed by the feature recognition of the twin network determination model is less than the preset feature loss threshold, and the picture feature meets the characteristics of Kai wood, the corresponding picture is stored in the Kai wood resource library corresponding to the resource set of the category, so as to achieve the purpose of expanding the data amount of the Kai wood resource library.

[0064] Further, the embodiment of the present application further includes step S900, and step S900 further includes:

[0065] S910: determining a classification confusion feature according to the picture classification cluster;

[0066] S920: collecting abnormal identification cases based on the classification confusion feature to obtain an abnormal case set;

[0067] S930: determining an abnormal feature set according to the abnormal case set;

[0068] S940: constructing a twin network abnormal determination model based on the network shared weight;

[0069] S950: inputting the picture classification cluster and the abnormal feature set into the twin network abnormal determination model to obtain an abnormal determination result;

[0070] S960: setting an abnormal determination screening weight based on the picture classification cluster;

[0071] S970: screening the determination result according to the abnormal determination screening weight and the abnormal determination result, and archiving the screened picture in the Kai wood resource library.

[0072] Specifically, according to the category features of the picture classification cluster, a plurality of classification confusion features are determined, such as color, texture and other features that are easy to be confused with other wood, a plurality of abnormal identification cases are collected based on the classification confusion features as an abnormal case set, a plurality of abnormal features are extracted from the abnormal case set, such as color identification error, texture identification error features, and an abnormal feature set is constructed.

[0073] Further, based on the network sharing weight, a twin network anomaly discrimination model is constructed, the twin network anomaly discrimination model comprising two same anomaly recognition models sharing the weight, the anomaly recognition model being obtained by supervised training of a neural network with the anomaly feature set as training data. The picture classification cluster and the anomaly feature set are input into the twin network anomaly discrimination model to obtain standard anomaly feature analysis data and sample anomaly feature analysis data. Feature loss value calculation is performed using the standard anomaly feature analysis data and the sample anomaly feature analysis data. Anomaly determination is performed according to the feature loss value calculation result to obtain an anomaly discrimination result.

[0074] Further, based on the picture classification cluster, an anomaly discrimination screening weight of each picture classification cluster is set. Then, based on the anomaly discrimination screening weight, the discrimination result is screened through the anomaly discrimination result to screen out and remove pictures matching the anomaly discrimination result in the discrimination result, that is, to remove easily confused false calligraphy pictures in the discrimination result. The screened pictures are archived in the calligraphy resource library to improve the accuracy of the obtained calligraphy pictures and further improve the picture quality of the calligraphy resource library.

[0075] To sum up, the embodiments of the present application have at least the following technical effects:

[0076] The present application sets a crawling factor to automatically crawl Internet pictures and classify them to obtain a picture classification cluster. A calligraphy standard feature set is constructed and used as training data to obtain a standard recognition sub-model. A network sharing weight is determined to construct a twin network discrimination model. A matching standard feature set is determined. The picture classification cluster and the matching standard feature set are input into the twin network discrimination model to obtain a discrimination result. When the discrimination result meets the archiving requirements, the corresponding pictures are stored in the corresponding category resource set of the calligraphy resource library.

[0077] The technical effect of obtaining calligraphy picture resources with high accuracy through Internet information crawling and screening and improving the resource quality of the calligraphy resource library is achieved.

[0078] Embodiment Two

[0079] Based on the same inventive concept as the calligraphy carving resource library construction method based on Internet information crawling in the foregoing embodiments, as shown in Figure 4 The present application provides a calligraphy carving resource library construction system based on Internet information crawling. The system and method embodiments in the present application are based on the same inventive concept. The system comprises:

[0080] A picture data set construction module 11 is configured to set a crawling factor, automatically crawl Internet pictures based on the crawling factor, and construct a picture data set.

[0081] The picture classification cluster obtaining module 12 is configured to classify the picture data set according to a preset classification feature, and obtain a picture classification cluster, wherein the preset classification feature is a standard feature set of a calligraphy brush.

[0082] The calligraphy brush standard feature set construction module 13 is configured to construct a calligraphy brush standard feature set corresponding to the preset classification feature.

[0083] The standard recognition sub-model construction module 14 is configured to obtain a standard recognition sub-model according to the calligraphy brush standard feature set as training data, and determine network shared weights based on the standard recognition sub-model.

[0084] The Siamese network discriminant model construction module 15 is configured to construct a Siamese network discriminant model based on the standard recognition sub-model and the network shared weights.

[0085] The matching standard feature set determination module 16 is configured to determine a matching standard feature set by performing feature matching between the picture classification cluster and the calligraphy brush standard feature set.

[0086] The discriminant result acquisition module 17 is configured to input the picture classification cluster and the matching standard feature set into the Siamese network discriminant model, and obtain a discriminant result.

[0087] The picture archiving module 18 is configured to store a corresponding picture in a corresponding category resource set of a calligraphy brush resource library when the discriminant result meets archiving requirements.

[0088] Further, the picture classification cluster obtaining module 12 is further configured to perform the following steps:

[0089] Inputting pictures in the picture data set into a frame recognition model to obtain picture frame recognition results.

[0090] S220: Clustering based on the picture frame recognition results and the preset classification feature, and determining a classification result of the pictures according to a clustering result.

[0091] S230: Storing all pictures in corresponding classification picture sets to construct the picture classification cluster.

[0092] Further, the picture classification cluster obtaining module 12 is further configured to perform the following steps:

[0093] Preprocessing the picture data set and setting a starting point.

[0094] S212: setting a distance constraint threshold according to a gray value distribution of the picture data set;

[0095] S213: performing a liquid surface rising fitting based on the starting point, when the liquid surfaces corresponding to any starting point intersect, generating a watershed line at a corresponding position, and stopping the liquid surface fitting when the liquid surface rises to a maximum gray value of the image;

[0096] S214: performing image segmentation using the watershed line to obtain the picture frame recognition result.

[0097] Further, the standard recognition sub-model construction module 14 is further configured to perform the following steps:

[0098] determining a classification cluster core area according to the picture classification cluster;

[0099] S420: determining a core discrimination area according to the picture frame recognition result and the classification cluster core area, and marking the picture by area.

[0100] Further, the standard recognition sub-model construction module 14 is further configured to perform the following steps:

[0101] when the classification cluster core area is a multi-area, collecting a case library based on the picture classification cluster and the multi-area recognition area;

[0102] S422: calculating a weight of each core area based on the classification cluster case library to determine the weight of each area;

[0103] S423: setting a recognition area level according to the weight of each area, and generating an area mark based on the recognition area level.

[0104] Further, the discrimination result acquisition module 17 is further configured to perform the following steps:

[0105] performing feature recognition analysis on the matching standard feature set by a standard recognition sub-model in the twin network discrimination model to obtain standard feature analysis data;

[0106] S720: performing feature recognition analysis on each picture in the picture classification cluster by a sample recognition sub-model in the twin network discrimination model to obtain sample feature analysis data;

[0107] S730: calculating a feature loss value based on the standard feature analysis data and the sample feature analysis data by a loss function;

[0108] S740: when the feature loss value reaches a preset threshold, obtaining a discrimination result that is not the same;

[0109] S750: When the feature loss value does not reach the preset threshold, obtaining the discrimination result as discriminating the same.

[0110] Further, the system further comprises:

[0111] A classification confusion feature determination module is configured to determine a classification confusion feature according to the picture classification cluster.

[0112] An abnormal case set obtaining module is configured to collect abnormal recognition cases based on the classification confusion feature, and obtain an abnormal case set.

[0113] An abnormal feature set determination module is configured to determine an abnormal feature set according to the abnormal case set.

[0114] A Siamese network abnormal discrimination model construction module is configured to construct a Siamese network abnormal discrimination model based on the network shared weight.

[0115] An abnormal discrimination result obtaining module is configured to input the picture classification cluster and the abnormal feature set into the Siamese network abnormal discrimination model, and obtain an abnormal discrimination result.

[0116] An abnormal discrimination screening weight setting module is configured to set an abnormal discrimination screening weight based on the picture classification cluster.

[0117] An abnormal discrimination result screening module is configured to screen the discrimination result according to the abnormal discrimination screening weight and the abnormal discrimination result, and archive the screened pictures in the wood resource library.

[0118] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above describes a specific embodiment of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be executed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0119] The above only describes the preferred embodiments of the present application and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0120] The specification and drawings are, of course, subject to various interpretations and should not be viewed in any limiting sense. It will be understood that various modifications and changes can be made to the application disclosed without departing from the scope of the application. It is therefore intended that the application be limited only by the scope of the appended claims.

Claims

1. A method for constructing a Chinese privet resource database based on Internet information crawling, characterized in that, The method includes: Set crawling factors, and automatically crawl Internet images based on the crawling factors to construct an image dataset; The image dataset is classified according to a preset classification feature to obtain an image classification cluster. The preset classification feature is the morphological feature of the Chinese tallow tree. Construct a standard feature set for Chinese mahogany, which corresponds to the preset classification features; Using the standard feature set of the Chinese privet as training data, a standard recognition sub-model is obtained, and based on the standard recognition sub-model, the network shared weights are determined. Based on the aforementioned standard identification sub-model and network shared weights, a twin network discrimination model is constructed. The matching standard feature set is determined by performing feature matching between the image classification cluster and the Kaimu standard feature set; The image classification cluster and matching standard feature set are input into the Siamese network discriminant model to obtain the discriminant result; When the discrimination result meets the archiving requirements, the corresponding image will be stored in the corresponding category resource set of the Kaimu Resource Library; The image classification cluster and matching standard feature set are input into the Siamese network discriminant model to obtain the discrimination result, including: The matching standard feature set is analyzed by the standard recognition sub-model in the twin network discriminant model to obtain standard feature analysis data; The sample identification sub-model in the Siamese network discriminant model is used to perform feature identification and analysis on each image in the image classification cluster to obtain sample feature analysis data; Based on the standard feature parsing data and sample feature parsing data, the feature loss value is obtained by calculating the loss function. When the feature loss value reaches a preset threshold, the discrimination result is obtained as "discriminated as different". When the feature loss value does not reach the preset threshold, the discrimination result is obtained as "discriminated as the same". The method further includes: Based on the image classification clusters, determine the classification confusion features; Based on the aforementioned classification confusion features, anomaly identification cases are collected to obtain an anomaly case set; Based on the set of abnormal cases, determine the set of abnormal features; Based on the network shared weights, a twin network anomaly detection model is constructed; The image classification cluster and the abnormal feature set are input into the Siamese network anomaly detection model to obtain the anomaly detection result; Based on the image classification cluster, anomaly detection filtering weights are set; Based on the anomaly detection filtering weight and the anomaly detection result, the detection result is filtered, and the filtered images are archived in the Kaimu resource library. The step of obtaining a standard recognition sub-model based on the standard feature set of *Kaloula spp.* as training data, and determining the network shared weights based on the standard recognition sub-model, further includes: Based on the image classification clusters, determine the core area of ​​each cluster; Based on the image border recognition results and the core region of the classification cluster, the core discrimination region is determined and the image is marked for feature extraction.

2. The method as described in claim 1, characterized in that, The step of classifying the image dataset according to preset classification features to obtain image classification clusters includes: Input the images from the image dataset into the border recognition model to obtain the image border recognition results; Clustering is performed based on the image border recognition results and preset classification features, and the classification result of the image is determined based on the clustering results; All images are stored in their corresponding category image sets to construct the image category cluster.

3. The method as described in claim 2, characterized in that, The step of inputting images from the image dataset into the border recognition model to obtain image border recognition results includes: The image dataset is preprocessed, and a starting point is set; Set a distance constraint threshold based on the grayscale value distribution of the image dataset; Based on the starting point, the liquid level rise is fitted. When the liquid levels corresponding to any starting point intersect, a watershed line is generated at the corresponding position. When the liquid level rises to the maximum gray value of the image, the liquid level fitting is stopped. Image segmentation is performed using watershed lines to obtain the image border recognition results.

4. The method as described in claim 1, characterized in that, The method further includes: When the core region of the classification cluster is multiple regions, a case library is collected based on the image classification cluster and the multiple region recognition region. The weights of each core region are calculated based on the classification cluster case library to determine the weights of each region. Based on the weights of each region, a recognition region level is set, and a region label is generated based on the recognition region level.

5. A system for constructing a Chinese privet resource database based on internet information crawling, characterized in that, The system is used to perform the method according to any one of claims 1-4, comprising: The image dataset construction module is used to set crawling factors and automatically crawl Internet images based on the crawling factors to construct an image dataset. The image classification cluster acquisition module is used to classify the image dataset according to a preset classification feature to obtain an image classification cluster. The preset classification feature is the morphological feature of the Chinese mahogany tree. A standard feature set construction module for *Kaloula chinensis* is used to construct a standard feature set for *Kaloula chinensis*, which corresponds to the preset classification features. A standard recognition sub-model construction module is used to obtain a standard recognition sub-model based on the standard feature set of the Chinese privet as training data, and to determine the network shared weights based on the standard recognition sub-model. A twin network discriminant model construction module is used to construct a twin network discriminant model based on the standard identification sub-model and network shared weights. A matching standard feature set determination module is used to determine the matching standard feature set by performing feature matching between the image classification cluster and the standard feature set. The discrimination result acquisition module is used to input the image classification cluster and the matching standard feature set into the Siamese network discrimination model to obtain the discrimination result; The image archiving module is used to store the corresponding image into the corresponding category resource set of the Kaimu resource library when the discrimination result meets the archiving requirements.

Images