CN101114385A - A method for fully automatic generation of digital cities - Google Patents

Abstract

The invention discloses a method for fully automatic generation of a digital city, which is applied to a general-purpose computer system and includes the following steps: acquiring a remote sensing image of the ground in a predetermined area, and monitoring the shadows of all objects on the remote sensing image through a shadow detection algorithm length; vectorize the remote sensing image, obtain the shapes of different objects, and match the position of the shadow, and obtain the height of each object; identify different objects in the remote sensing image in the image library according to the characteristics of the image; according to the The type, base shape, top shape and height of objects in the area are combined with the model library to automatically generate a 3D model of the area object. The method of the present invention can automatically generate a large-scale digital city in real time, which meets the requirements of high timeliness urban applications; and only needs remote sensing images and DEM as basic data, and the cost is very low, making it widely used in government and business It becomes possible to apply in life, etc.

Description

A method for fully automatic generation of digital city

technical field

The invention relates to a geospatial simulation technology, in particular to a method for fully automatic generation of a digital city.

Background technique

The concepts involved in the present invention are first described below:

1. Digital city: a virtual and open city model that can realize comprehensive urban management and decision support;

2. Three-dimensional model (3D model): A three-dimensional polygonal representation of an object, usually displayed by a computer or other video equipment. The objects shown are real-world entities or fictional things, and can be as small as atoms or as large as gigantic dimensions. Anything that exists in physical nature can be represented by a three-dimensional model.

3. Texture: A texture is actually a bitmap. In this sense, when the word texture is used in computer graphics, it has a clear definition. Semantically, the term texture refers to both the pattern of color on an object and whether the surface of the object is rough or smooth.

4. Knowledge Base: a structured, easy-to-operate, easy-to-use, and comprehensively organized knowledge cluster in knowledge engineering is aimed at solving problems in a certain (or some) domains, using some (or several) Knowledge representation is a collection of interconnected pieces of knowledge stored, organized, managed and used in computer memory. These pieces of knowledge include domain-related theoretical knowledge, factual data, and heuristic knowledge obtained from expert experience, such as definitions, theorems, algorithms, and common sense knowledge related to a certain domain. The knowledge base makes the knowledge-based system (or expert system) intelligent, not all intelligent programs have a knowledge base, only the knowledge-based system has a knowledge base. Many applications today utilize knowledge, some at a very high level, but these applications may not be knowledge-based systems, nor do they have knowledge bases. The difference between a general application program and a knowledge-based system is that a general application program implicitly encodes problem-solving knowledge in the program, while a knowledge-based system explicitly expresses the problem-solving knowledge of the application domain , and form a relatively independent program entity separately.

The characteristics of the knowledge base are as follows:

1) The knowledge in the knowledge base is formed into an easy-to-use and structured organizational form according to their application field characteristics, background characteristics (background information at the time of acquisition), use characteristics, attribute characteristics, etc. Knowledge sheets are generally modular.

2) The knowledge in the knowledge base is hierarchical. The lowest level is "factual knowledge", the middle level is the knowledge used to control "facts" (usually represented by rules, processes, etc.); the highest level is "strategy", which takes the middle level knowledge as the control object. Policies are also often thought of as rules of rules. Therefore, the basic structure of the knowledge base is a hierarchical structure, which is determined by the characteristics of its knowledge itself. In the knowledge base, there is usually interdependence between knowledge slices. Rules are the most typical and commonly used piece of knowledge.

3) In the knowledge base, there may be a special form of knowledge that does not only belong to a certain level (or exists in any level)—credibility (or called trust degree, confidence measure, etc.). For a certain issue, relevant facts, rules and policies can be marked with credibility. In this way, an augmented knowledge base is formed. There is no uncertainty measure in the database. Because everything is "deterministic" in the processing of the database.

4) There may also be a special part of the knowledge base, usually referred to as the canonical method base. If the solution to certain problems is certain and inevitable, it can be directly stored in the typical method library as a part of the fairly certain solution to the problem. This macro storage will form another part of the knowledge base. When using this part, machine reasoning will be limited to a certain layer of the typical method library.

In addition, the knowledge base can also be implemented on a distributed network. In this way, it is necessary to build a distributed knowledge base. There are three advantages of building a distributed knowledge base:

(1) A larger knowledge base can be constructed at a lower price;

(2) The problem-solving tasks corresponding to knowledge bases at different levels or in different fields are relatively simple, so a more efficient system can be formed;

(3) It can be suitable for geographical distribution in a vast area.

The structure of the knowledge base must enable the knowledge in it to be effectively accessed and searched in the process of being used, the knowledge in the base can be easily modified and edited, and at the same time, the consistency and completeness of the knowledge in the base should be checked.

5. Image matching (image matching): refers to the process of spatially aligning two images recorded by two different sensors from the same scene to determine the relative translation between the two images, which can be widely used It is an extremely important technology in the field of modern information processing in terms of target tracking, resource analysis, and medical diagnosis.

6. Image semblance: refers to the probability of one image "spinning" into another image. At the same time, a simple image similarity algorithm is given. Through multiple experiments, this image similarity is effective and satisfactory for the recognition of complex patterns, and can be used for classification and retrieval of images. Similarity includes similarity in shape, structure, statistics, texture, environment, height and size.

7. Digital city image library: the library composed of images of all objects in the digital city is a digital city image library, which can be divided into building image sub-library, road image sub-library, bridge image sub-library, plant Image sub-library, animal image sub-library, water area image sub-library, atmospheric image sub-library, stratigraphic image sub-library.

8. Digital city model library: It is composed of three-dimensional models of all objects in the digital city, and its classification corresponds to the digital city image library. The model contains all the characteristic information of the object in three dimensions, including shape, color, texture and so on.

9. Digital elevation model (DEM), also known as digital terrain model (DTM), is a continuous representation of spatial fluctuations. Since DTM implies the meaning of terrain and landscape, DEM is commonly used to simply represent elevation. Free global elevation data with 30-meter accuracy can be downloaded from the web.

Digital city is a technical system that comprehensively uses GIS, remote sensing, telemetry, broadband network, multimedia and virtual simulation technologies to automatically collect information, dynamically monitor management and assist decision-making services for urban infrastructure and functional mechanisms; it has urban geography, Powerful functions such as digitization, networking, virtual simulation, optimization decision support and visualization of complex systems such as resources, ecological environment, population, economy, and society. The digital city provides an important supporting tool for the sustainable development of the city.

Visualization is the window and tool to realize the interaction between digital city and people. Without visualization technology, a bunch of numbers in the computer are meaningless. A notable feature of digital city is virtual reality technology. After the digital city is established, users can see the city emerge from the earth by wearing a display helmet or from a computer screen or a large-screen projection, and use the mouse or keyboard to zoom in on the digital image; with the continuous improvement of resolution, users You can see private houses, shops, trees and other natural and man-made landscapes. When users are interested in products, they can enter the store, appreciate the clothes in the mall, and construct a virtual scene of trying on clothes according to their own body shape.

Virtual reality technology provides an immersive feeling for human beings to observe nature, appreciate landscapes, and understand entities. In recent years, virtual reality technology has developed rapidly. Virtual Reality Modeling Language (VRML) is a Web-oriented, object-oriented 3D modeling language, and it is an interpreted language. It not only supports the three-dimensional representation of data and processes, but also enables users to walk into a virtual world with realistic audio-visual effects, so as to realize the representation of the digital earth and realize the research of various earth phenomena and people's daily applications through the digital earth. In fact, artificial virtual reality technology has long been a mature technology in photogrammetry. The development of digital photogrammetry in recent years has enabled the establishment of digital virtual technology for measurement on computers. Of course, the current technology is to take pictures of the same entity, generate parallax, and construct a three-dimensional model, usually when the model is processed. The further development is to seamlessly splice the entire earth, roam and zoom in at will, and construct a virtual three-dimensional from the three-dimensional data through the method of artificial parallax.

The existing technologies for building digital cities, as shown in Figure 1, Figure 2, and Figure 3, are three commonly used methods for building digital cities. They are similar to other solutions, except that the modeling and rendering tools are different. The common features of the above schemes are: according to the photos collected on site, manual 3D modeling is carried out, and the position of each object in the urban scene is manually marked, and then the 3D model of each object is manually added to the corresponding position in the urban scene.

In the various schemes of the existing technology, it is necessary to take pictures of all the objects in the city one by one with a camera and manually model them one by one. The workload is very heavy, and the built models need to be manually calibrated one by one and placed in the digital city scene appropriate location in . This process consumes a lot of manpower, including collecting photos, manual modeling, manual calibration and placement of the model, and at the same time consumes a lot of financial resources, for example, many cameras are needed for collecting photos, and many computers are needed for manual modeling, manual calibration and placement. Models, etc., will also consume a lot of time. For example, building a model sometimes takes 1 day. There are thousands of objects in a city that need to be modeled. For example, the digital city of Shenzhen needs at least 3 years with existing technology. time to complete.

However, with the rapid development of the city, officials want to direct emergency response, investigate and deal with violations, and residents want to travel without leaving home. These can only be done in digital cities. But if it takes a long time to build a digital city, for example, using existing technology, digital Shenzhen takes 3 years, then everything people see in the digital city is 3 years ago, which will give the city emergency and violations. Monitoring, etc. has disastrous consequences. In fact, the changes in Shenzhen are indeed changing with each passing day, and the appearance of the city will change every day, so unless the digital city is built within at least one day, the digital city drawn cannot truly represent and reflect the real city. Digital city can't really play a role in practical application, and it's impossible to achieve real-time.

Therefore, there are also defects in the prior art and need to be improved and developed.

Contents of the invention

The purpose of the present invention is to provide a method for fully automatic generation of digital cities, through the way of knowledge base, realize the automatic generation of digital cities, so that digital cities can be generated in real time, and provide services for urban emergency response, violation monitoring, traffic command, digital life, etc. Live support.

Technical scheme of the present invention comprises:

A method for fully automatic generation of a digital city, which is applied to a general computer system, comprising the following steps:

A. Acquire the remote sensing image of the ground in a predetermined area, and monitor the shadow length of all objects on the remote sensing image through the shadow detection algorithm;

B. Carry out vectorization on the remote sensing image, obtain the shapes of different objects, and match the position of the shadow, and obtain the height of each object;

C. Collect images and corresponding models of objects in the area to form a knowledge base;

D. Identify different objects in remote sensing images in the image library according to the characteristics of the images;

E. According to the type, base shape, top seat shape and height of the object in the area, combined with the model library, the 3D model of the area object is automatically generated;

F. According to the two-dimensional coordinate positions of the corresponding objects, insert the three-dimensional model of the regional object into the remote sensing image with elevation.

The method, wherein, the two-dimensional coordinate position in the step F is obtained according to the digital elevation model and the remote sensing image.

The method, wherein, the step D also includes:

D1. Extract representative images of various types of individuals from remote sensing images, and classify these representative images, extract their commonality, and form the first-level feature image;

D2. Divide subcategories at this level, and extract commonality from all images of each subcategory, and assign a feature image to each subcategory;

By analogy, until its division basically represents the various types that are representative of the individual.

The method described above, wherein the classification structure of the model library is consistent with the classification structure of the image library, and an image in the image library corresponds to a model in the model library.

The method described above, wherein the models in the model library are static models built using modeling tools.

The method described above, wherein the models in the model library are three-dimensional models described by using parameters and rendered in real time when needed.

Said method, wherein, when the digital city is automatically generated, the static model is used first, and then the previous static model is gradually replaced by the corrected dynamic model.

The method, wherein the mapping relationship between the image library and the model library includes the following steps:

D3. Extract and identify objects in remote sensing images according to the image library;

D4. Compare the similarity between the extracted object and the subclass of the corresponding category in the image library, and retrieve the image with the largest similarity with the object in the image library, and map it to the corresponding model in the model library;

D5. Automatically model the individuals in the remote sensing images through the knowledge base.

The method, wherein, the step D5 also includes:

D51. Scan the remote sensing images according to the first-level feature images in the image library to obtain a collection of objects of each category, and judge the similarity between the object and these feature images;

D52. Find the most accurate classification to which the object belongs from the image library.

The method, wherein, the step D52 includes:

D521. Comparing the image of the object with the feature image of the next level of the category to which it belongs, if the individual image has the highest similarity with the feature image of a certain category, determine that the individual image belongs to that category;

D522. Matching the individual image with each feature image of the next level of the class respectively, and calculating their similarity, finding the category to which the feature image with the largest similarity belongs, as the category to which the object image belongs;

And so on until the similarity reaches the expected requirement.

The method for fully automatic generation of a digital city provided by the present invention can fully automatically generate a large-scale digital city in real time, satisfying urban applications with high timeliness requirements; and only needs remote sensing images and DEM as basic data. However, the cost is very low, which satisfies the promotion of digital cities throughout the country and the world, and makes it possible to apply them in government, business, and life.

Description of drawings

FIG. 1 is a schematic diagram of a digital city generation technology in the prior art;

FIG. 2 is a schematic diagram of another digital city generation technology in the prior art;

FIG. 3 is a schematic diagram of another digital city generation technology in the prior art;

Fig. 4 is the schematic flow chart of building a database in the method for digital city automatic generation of the present invention;

Fig. 5 is the automatic generation digital city process schematic diagram of the inventive method;

Fig. 6 is the schematic diagram of the image storehouse of the method of the present invention;

Fig. 7 is a schematic diagram of a recognition rule library of the method of the present invention;

Fig. 8 is a schematic diagram of a model library of the method of the present invention;

Fig. 9 is a schematic diagram of the mapping relationship between the image library and the model library in the method of the present invention;

Fig. 10 is a remote sensing image diagram of an embodiment of the method of the present invention;

Fig. 11 is a schematic diagram of shadows processed according to Fig. 10 according to the method of the present invention;

Fig. 12 is a schematic diagram of the digital city drawn by the method of the present invention.

Detailed ways

Various preferred embodiments of the present invention will be described in more detail below in conjunction with the accompanying drawings.

The method for fully automatic generation of the digital city of the present invention utilizes remote sensing images, and specifically includes the following steps in a general-purpose computer:

Step 1, monitor the lengths of all shadows on the remote sensing image through the shadow monitoring algorithm, and the calculation of the shadow lengths is known in the prior art;

The second step is to vectorize the remote sensing image of the city to obtain the shapes of different urban objects; and match the position of the object with the position of the shadow to obtain the height of the object; the calculation process of vectorization is also the existing technology. Well known, therefore, no more details;

Step 3: Collect images of various buildings, vehicles, etc. in the city and their corresponding models, and put them into the knowledge base, as shown in Figure 4. First, analyze the urban data in the remote sensing image, and classify the individuals in the city, such as vehicles, Buildings, etc., and extract individual feature images from remote sensing images, and automatically add them to the image library. According to the individual feature images, through the recognition and judgment of the human eye and the 3D information of the individual collected on the spot, then model and add to the model library;

Step 4: Identify different objects in the remote sensing images according to the characteristics of the images of different objects in the image library, so as to obtain the types of different urban objects and their top shapes; The feature image matches the objects of this type in the remote sensing image, so as to identify which subcategory each object belongs to; and so on, knowing that the recognition effect meets the requirements, so as to obtain the specific types of all urban objects in the image;

Step 5, according to the different types of urban objects, base shape, top shape, and height combined with the model library, automatically generate a three-dimensional model of the urban object;

Step 6. Obtain the topography of the digital city and its ups and downs according to DEM and remote sensing images;

Step 7: Insert the 3D models of the aforementioned urban objects into the remote sensing images with elevation according to the positions of their 2D coordinates. By this step, a digital city has been automatically generated.

The whole process of above-mentioned automatic generation is as shown in Figure 5, introduces several key steps of the automatic generation digital city of the present invention in detail below:

Step 1: Automatically build a 3D model of urban objects;

Before the automatic generation of the digital city, the method of the present invention needs to first establish a recognition rule base, an image base, a model base, as shown in Figure 6, Figure 7 and Figure 8, the rules in the recognition rule base shown in Figure 7 are in accordance with Different aspects of matching are divided, such as: detection rules for shape similarity and difference, detection rules for structural similarity and difference, detection rules for statistical similarity and difference, detection rules for color similarity and difference, gray similarity and Difference detection rules, texture similarity and difference detection rules, environment similarity and difference detection rules, etc.

The images in the image library shown in Figure 6 are sampled from remote sensing images, specifically: extract representative images of various types of individuals from remote sensing images, classify these representative images, and extract commonality , the first-level feature image; then divide subcategories at this level, and extract commonality from all images of the subcategory, assign a feature image to the subcategory, and so on, until the division basically represents the individual Representative of various types so far.

In the method of the present invention, the classification structure of the model library as shown in Figure 8 is basically consistent with the classification structure of the image library, and an image in the image library basically corresponds to a model in the model library, but the model in the model library can be used Static models built by modeling tools can also be 3D models described using parameters that can be rendered in real time when needed. The dynamic model is easier to correct than the static model, and the static model is more real-time than the dynamic model, but the authenticity of the expression is not as good as the corrected dynamic model. Therefore, the method of the present invention can first use the static model when automatically generating the digital city, and then gradually replace the previous static model with the corrected dynamic model.

As shown in Figure 9, it is the mapping relationship between the image library and the model library of the method of the present invention. First, the objects in the remote sensing images are extracted and recognized according to the image library, and then the extracted objects are compared with the corresponding categories in the image library. The subclass performs similarity comparison, retrieves the image with the greatest similarity with the object in the image library, and maps it to the corresponding model in the model library. The process of automatically modeling individuals in remote sensing images through the knowledge base is as follows: Scan the remote sensing images according to the first-level feature images in the image base to obtain a collection of objects of each category. For example, the first-level classification includes characteristic images of buildings, bridges, squares, flowers and trees, water, and so on. Judge the similarity between the object and these feature images.

The similarity includes: similarity of shape, similarity of structure, similarity of statistics, similarity of color, similarity of gray scale, similarity of texture, similarity of environment and so on. It can be seen that the similarity has many components. The present invention decides which similarities to adopt through the initial analysis of the individual, and gives different weights to different types of similarities when judging the similarity between the individual and the image in the image library, and then A weighted method is used to judge the final similarity.

After the method of the present invention extracts a certain object from the remote sensing image, find out the most accurate classification (such as: building/high building/office building) to which the object belongs from the image library. Compared with the feature images of the next level of classification, if the individual image has the highest similarity with the feature image of class X, then it can be judged that the individual image belongs to class X; The images are matched separately, and their similarity is calculated, and the category of the feature image with the largest similarity (assumed to be Y) is found as the category of the object image, and then it can continue to compare with the next-level feature image of Y, and so on , until its similarity reaches the expected requirement.

As the method of the present invention stipulates according to needs: for buildings, the similarity can reach 80%. Then the feature image of the subclass that finally matches and has the greatest similarity is the twin image of the individual image, and the twin 3D model corresponding to the twin image can be obtained through the mapping rules between the image library and the model library. The 3D model contains a large amount of prior knowledge of people, as well as a large amount of fuzzy, inexpressible but actually existing knowledge between the natural world and the digital city. The mapping relationship between them is implicit.

Taking the example of the aforementioned building as an example, if the similarity between the individual and its twin image in the image library reaches 80%, the method of the present invention can also determine where their 20% difference is, based on the 20% difference, and Decompose the difference into differences in shape, structure, statistics, color, grayscale, texture, environment, and so on. The difference on these two-dimensional images will have a mapping rule with the difference on the three-dimensional model. According to this rule, the method of the present invention can correct the twin model of the individual, and finally obtain an ideal realistic model of the individual.

Step 2: Automatically generate the height of urban objects

The method of the invention calculates the length of each shadow according to the shadow in the remote sensing image, and then registers each shadow with the position of each object to obtain the height of each object. The remote sensing image is shown in Figure 10, and the shadow image detected by the method of the present invention is shown in Figure 11.

Step 3: Automatically embed objects into the remote sensing images (landforms) of the city

The process of reimplanting the object model in the remote sensing image is as follows: when the individual is extracted from the remote sensing image, the method of the present invention has already recorded the two-dimensional coordinates of the individual and the different edges of the individual in the program. orientation. According to the coordinates and orientation of the individual image in the remote sensing image, the method of the present invention can embed the realistic model automatically generated in the previous step into the remote sensing image with the correct orientation, angle, and position, thereby automatically reproducing the image in the city. various object images.

The fourth step: the method of the present invention covers the remote sensing image on the digital elevation model DEM, so that the remote sensing image fluctuates according to the data of the DEM, and at the same time, the three-dimensional models of all individuals on the remote sensing image also fluctuate accordingly, so that this is a simulation of a The actual geographic shape of the city.

Through the above several steps, the automatic generation of three-dimensional digital city from a single remote sensing image and corresponding elevation map has been fully realized. The digital city image generated by the method of the present invention through the remote sensing image 10 is shown in Figure 12. It can be seen that the method of the present invention can Fully automatic and real-time generation of a large-scale digital city, to meet the time-sensitive urban applications; and only need remote sensing images, DEM as the basic data, the cost is very low, to meet the promotion of digital cities in the country and even the world and Make it possible to apply it in government, business, life, etc.

The method of the present invention can serve the city emergency command system, integrate various remote sensing data to automatically generate a digital city in real time, based on real-time digital city layout, terrain, road and other information, and combine meteorological observation data (wind temperature data) to simulate the city Wind field, so that the development trend of emergencies such as air pollution diffusion can be dynamically simulated, and can be dynamically and realistically displayed to city commanders in real time to provide decision-making reference.

Utilizing the method of the present invention can serve for the monitoring of urban illegal buildings, and automatically generate a digital city in real time by fusing various remote sensing data, and by comparing the building and planning data in the digital city, different illegal buildings can be accurately found and displayed to the city planning manager.

The method of the present invention can also be used in many other aspects, for example, residents can enjoy the services of virtual shopping malls, virtual hospitals, virtual theaters and virtual tours without going out of their homes; Rescue routes and on-site conditions; the police can immediately locate the criminals without leaving the police station, monitor every move of the criminals, and immediately determine the best arrest route; the planning department can see all the criminals without on-site inspections. The most reasonable decision can be made based on the land and housing available; the traffic management department can see the traffic conditions of all roads without standing on the road, so as to make the most reasonable dispatch.

It should be noted that the method of the present invention can also use other methods to generate the height of the object, detect the type of the object, generate the elevation, etc., and besides the remote sensing image, it can also use photos, microwave remote sensing, wireless sensors, etc. to obtain data.

It should be understood that the above descriptions of the specific embodiments of the present invention are relatively specific, and should not be construed as limiting the scope of the patent protection of the present invention. The scope of patent protection of the present invention should be determined by the appended claims.

Claims (10)

1. A method for fully automatic generation of a digital city, which is applied to a general-purpose computer system, comprising the following steps: A. Acquire the remote sensing image of the ground in a predetermined area, and monitor the shadow length of all objects on the remote sensing image through the shadow detection algorithm; B. Carry out vectorization on the remote sensing image, obtain the shapes of different objects, and match the position of the shadow, and obtain the height of each object; C. Collect images and corresponding models of objects in the area to form a knowledge base; D. Identify different objects in remote sensing images in the image library according to the characteristics of the images; E. According to the type, base shape, top seat shape and height of the object in the area, combined with the model library, the 3D model of the area object is automatically generated; F. According to the two-dimensional coordinate positions of the corresponding objects, insert the three-dimensional model of the regional object into the remote sensing image with elevation. 2. The method according to claim 1, wherein the two-dimensional coordinate position in the step F is obtained according to a digital elevation model and a remote sensing image. 3. method according to claim 1, is characterized in that, described step D also comprises: D1. Extract representative images of various types of individuals from remote sensing images, and classify these representative images, extract their commonality, and form the first-level feature image; D2. Divide subcategories at this level, and extract commonality from all images of each subcategory, and assign a feature image to each subcategory; By analogy, until its division basically represents the various types that are representative of the individual. 4. The method according to claim 3, wherein the classification structure of the model library is consistent with the classification structure of the image library, and an image in the image library corresponds to a model in the model library. 5. The method according to claim 4, wherein the models in the model library are static models built using modeling tools. 6. The method according to claim 4, wherein the models in the model library are three-dimensional models described using parameters and rendered in real time when needed. 7. The method according to claim 5 or 6, characterized in that the static model is used first when the digital city is automatically generated, and then the previous static model is gradually replaced by the corrected dynamic model. 8. The method according to claim 1, wherein the mapping relationship between the image library and the model library comprises the following steps: D3. Extract and identify objects in remote sensing images according to the image library; D4. Compare the similarity between the extracted object and the subclass of the corresponding category in the image library, and retrieve the image with the largest similarity with the object in the image library, and map it to the corresponding model in the model library; D5. Automatic modeling of individuals in remote sensing images through the knowledge base. 9. method according to claim 8, is characterized in that, described step D5 also comprises: D51. Scanning the remote sensing images according to the first-level feature images in the image library to obtain a collection of objects of each category, and judging the similarity between the object and these feature images; D52. Find out the most accurate classification to which the object belongs from the image database. 10. The method according to claim 9, wherein said step D52 comprises: D521. Comparing the image of the object with the feature image of the next level of the category to which it belongs, if the individual image has the highest similarity with the feature image of a certain category, determine that the individual image belongs to that category; D522. Matching the individual image with each feature image of the next level of the class respectively, and calculating their similarity, finding the category to which the feature image with the largest similarity belongs, and taking it as the category to which the object image belongs; And so on until the similarity reaches the expected requirement.

Images