KR20050062004A - Method for providing the classification of multispectral imagery using a network - Google Patents

Abstract

The present invention relates to a method for automatically classifying a multispectral imagery using a network, comprising the steps of: dividing a multispectral image through an image training unit; and a feature for a classification class defined in an object; And setting up as training data, performing an accuracy assessment on the classification result of the training data, and providing database-specific feature information to the user for cover classification through the network. And performing a function setting process for the image classifying unit at a user side. According to the present invention, since the use of the database through the network in the surface coating classification using the multispectral image, and the separate training data setting work is not required on the user side, high classification accuracy can be obtained with little effort of the user.

Description

Automatic classification of multispectral images using a network {METHOD FOR PROVIDING THE CLASSIFICATION OF MULTISPECTRAL IMAGERY USING A NETWORK}

The present invention relates to a technique for performing surface coating classification using an image taken from the ground, and in particular, a method for automatically classifying a multispectral image using a network capable of easily performing land cover classification for a multispectral image. It is about.

Previously, image classification techniques using multispectral images used pixel-based classification, which is the basic unit of image formation, but recently, object-based classification that classifies objects based on image segmentation. The theory of -based classification or object-oriented classification has been raised. Unlike the pixel-based classification results, object-based classification techniques with higher classification accuracy are now mainstream because there is no effect like "Salt and pepper" in object-based classification results.

In addition, the classification techniques commonly used in terms of algorithms include the unsupervised classification technique and the supervised classification technique, among which the supervised classification technique is widely used.

1 illustrates a process of classifying an image based on pixel-based supervised classification as a conventional technique.

First, in step S100, a classification class is set, which can be arbitrarily set by the user.

After setting the classification class, as shown in step S102, training data is extracted according to the type of class. Since this is the standard data when performing image classification, it is important to extract appropriate training data that can represent the class.

Subsequently, in step S104 and step S106, statistics of the extracted training data are calculated, and the statistics calculated based on one pixel are classified.

Finally, in step S108, a precision check is performed on the results classified in step S106.

As such, the supervised classification technique has the advantage of higher precision than the unsupervised classification.

However, the supervised classification technique has a disadvantage in that the user has to set up training data, which can be a burden for both experienced users and novice users because it is repetitive and has a great effect on precision.

That is, novice users may not be familiar with how to extract what, and even experienced users may be less efficient due to repetitive work characteristics.

In conclusion, the image classification technology to date has various problems in terms of precision and convenience.

The present invention has been implemented to solve the above-mentioned problems of the prior art, by providing a database of features for a predetermined classification class while providing object-based classification and supervised classification techniques for multispectral images. The purpose of the present invention is to provide a method for automatically classifying multiple spectroscopic images using a network, which enables users to easily implement high-precision land cover classification without providing various task execution programs for image classification.

According to a preferred embodiment of the present invention for achieving this object, in the automatic classification method of a multispectral image connected to the image training unit and the image classifier via a network, the step of segmenting the image through the image training unit, and determined in the object Databaseing the feature of the classification class and setting it as training data, performing an accuracy assessment on the classification result of the set training data, and database-based feature information. The present invention provides a method for automatically classifying a multispectral image using a network, the method comprising providing a multispectral image covering classification through a network, and performing a function setting process for the image classifying unit.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the present invention.

Prior to the description, the core technical gist of the present invention provides an object-based classification technique based on image segmentation, and a supervised classification technique for image classification, in order to provide a user with an environment for facilitating surface coating classification and the like. In order to eliminate the inconvenience of setting up training materials for users, it is possible to easily achieve the object of the present invention from this technical idea by constructing a database for a predetermined classification class in advance and providing the user through a network. Could be.

2 is a block diagram of the entire system for performing the automatic classification method of the multispectral image according to the present invention.

As shown in FIG. 2, the system is divided into two parts, a training application 200 and a classification & change detection application 222, and the training unit 200 and the classification unit 222. Commonly included in the configuration, that is, display module (208), feature extraction module (Feature extraction module) 210, segmentation module (Segmentation module) 212, Classification module (214) The multispectral image DB 216, the feature image DB 218, and the reference image DB 220 exist.

The training unit 200 and the classification unit 222 are interconnected through a network not shown.

After performing the image segmentation through the segmentation module 214, the training unit 200 is responsible for the database of the necessary features (Features) as the training data through the training module 202 in the feature image DB 218.

Also, an accuracy assessment of the classification result output from the classification unit 222 is performed through the precision inspection module 204.

The training data setting process and the precision test performing process will be described in more detail with reference to FIGS. 3 and 4 to be described later.

The classification unit 222 performs a function of classifying and editing an image through the classification module 214 and the class modification module 226 using the constructed feature image as training material.

Furthermore, the classifying unit 222 performs a change detection function of finding an area where the characteristics of the surface have changed using at least two or more multispectral images observed at different times through the change detection module 224. do. The detailed description is described in detail in FIG. 5 below.

Meanwhile, the training unit 200 including the common components 208, 210, 212, 214, 216, and 218 and 220 of FIG. 2 attempts to build a feature image DB through training on feature information. The classifier 222, which may be provided with a database to a data provider and includes a common configuration 208, 210, 212, 214, 216, 218, 220, may be used for user classification. It may be provided to.

That is, since the features of the various classes of the images acquired through the training unit 200 are databased, the user may receive the characteristics of the class through the network with only the classification unit 222 having a common configuration.

Therefore, the user side does not need repetitive tasks such as training data setting, and because the database data constructed by experts can be utilized, more accurate classification results can be obtained.

3 is a view for explaining the process of setting the training data of the two functions of the above-described training unit 200.

First, the training unit 200 inputs image information provided from the reference image DB 220 and the multispectral image DB 216 as in steps S300 and S302. In this case, the information on the input multispectral image may be displayed to the outside through the display module 208.

In step S304, the multispectral image is segmented using an image segmentation technique, and the average, standard deviation, and various statistical information of each band are extracted from the segmented image information.

Subsequently, in step S306, the multispectral image information and the reference image information are trained, and the extracted feature is stored in the feature image DB 218.

The information stored in the feature image DB 218 may be provided to the user through a network, and thus the user may classify the image using a built database without setting training data.

4 is a view for explaining a process of checking the precision of the two functions of the training unit 200.

First, as in step S400, image information is input from the reference image DB 220, and as in step S406, a multispectral image is input from the multispectral image DB 216.

The input multispectral image is divided and classified through steps S408 and S410, and step S412, and then proceeds to steps S414 and S402 and reference is made to the result. ) Compare and analyze the data and check the classification.

Based on this check, the database data constructed in the feature DB 218 can be corrected and updated (S404).

5 illustrates a function setting process for the classification unit 222 used by an actual user.

First, when image information provided from the multispectral image DB 216 and the reference image DB 220 is input, as in steps S500 and S514, the user may use a function of segmenting and changing the image. There is (S504) (S516).

When the image segmentation provided by the common part is completed, the feature is extracted (S506).

In this case, the extracted feature is assigned a class by comparing the values from the database constructed in the feature image DB 218, it is not necessary to set a separate training material.

In addition, as in step S508, the user can view and determine the final classification result to modify the class.

In addition, if there are at least two images or if there is classification criteria data, change detection function is provided to indicate the area of significant change between the two data.

According to the present invention described above, when performing the classification for the multispectral image, it is possible to facilitate the parallel training data setting work, and the problem of purchasing expensive image processing software and the prerequisite to deal with it It is considerably solved by the present invention. For the purpose of surface coating classification, only software implemented by the present invention may be performed without any additional software. In addition, repetitive work for each class is not required to set up training data, which saves time and labor. Users can have an efficient working environment by using pre-built database data, and the result of classification can also be obtained with high accuracy by using data built by experts.

As mentioned above, although this invention was concretely demonstrated based on the Example, this invention is not limited to this Example, Of course, various changes are possible within the range which does not deviate from the summary.

1 is a flowchart illustrating a general pixel-based supervised classification technique according to the prior art;

2 is an overall configuration diagram for performing a multispectral image automatic classification method according to the present invention,

3 is a view for explaining a training material setting process of the training unit of FIG.

4 is a view for explaining a precision test process of the training unit of FIG.

5 is a view for explaining the function of the classification unit of FIG.

Claims (5)

In the multispectral image automatic classification method in which the image training unit and the image classification unit are connected through a network, Segmenting the multispectral image through an image training unit; Setting up a database of features for a class defined in the object as training data; Performing an accuracy assessment on the classification result of the set training data; Providing the databased feature information to a user for the purpose of classifying a multispectral image through the network; Performing a function setting process for the image classification unit at the user side; Multispectral image automatic classification method using a network comprising a. The method of claim 1, The training data setting step, Dividing the multispectral image and extracting an average, standard deviation, and various other statistical information of each band from the divided image information; After the training process for the multispectral image information and the reference image information, database the extracted feature to the feature image Multispectral image automatic classification method using a network comprising a. The method of claim 1, The precision check step, Performing division and classification on the input image information, and comparing and analyzing the result with reference data to perform a test to determine the degree of classification; Modifying and updating the database data constructed in the feature DB based on the inspection Multispectral image automatic classification method using a network comprising a. The method of claim 1, The performing of a function setting process for the image classifying unit may include: Dividing and changing detection of image information provided through the image training unit; Dividing an image provided from a common part of the image training unit and the image classification unit and extracting a feature when image segmentation is completed; Assigning a class by comparing the extracted feature with the database of feature image data; Editing the class for the final classification result at the user side Multispectral image automatic classification method using a network comprising a. The method of claim 1, The method is a method for automatically classifying multiple spectroscopic images using a network, which comprises object-based classification and supervised classification.