What Is Information Extraction?

In surveying and mapping, information extraction refers to the extraction of remote sensing image information. Remote sensing information extraction refers to extracting a large amount of useful information (such as ground features, vegetation, temperature, etc.) from users of massive, incomplete, noisy, fuzzy, random practical remote sensing image data. And put it (forming structured data) into the database or provide it to the user for query and use in other forms.

Remote sensing images are a reflection of the collection of ground objects.Each object has a close internal connection with the surrounding environment.For example, trees and buildings will have shadows, and residential areas will have houses, street networks, vegetation, and so on. There is also a certain hierarchical relationship.For example, the general vegetation has green characteristics, and grass and trees have their own characteristics in addition to the green commonality of vegetation, such as different height distributions and unique images.

At present, the classification and extraction of most remote sensing information is mainly a combination of mathematical statistics and manual interpretation. This method not only has relatively low accuracy, low efficiency, high labor intensity, but also relies on people involved in interpretation and analysis, which is largely non-repetitive. Especially for the multi-phase, multi-sensor, multi-platform, multi-spectral band remote sensing data processing, the problem is more prominent. Therefore, research on the intelligent extraction method of remote sensing information is of great significance to improve the accuracy and efficiency of remote sensing information extraction ^[2]

Information extraction based on ground texture knowledge

When the composition of the features is complex and the size is greater than the spatial resolution of the sensor, the structure and composition of the features may be felt remotely. The image has obvious texture features. When there are texture features different from the background features When it is difficult to completely solve the problem of information extraction based only on the extraction of knowledge based on spectral features, it is necessary to use the spectral knowledge of the features together with the texture knowledge to extract the information. Texture refers to the spatial change of gray values. It It is a pattern composed of some texture primitives according to different spatial configuration forms. The spatial configuration of texture primitives can be random, deterministic, probability, and functional. Textures can be divided into structured textures and unstructured textures. Unstructured texture is also called random texture. In visual interpretation, texture is generally described and expressed in terms of thickness, smoothness, graininess, randomness, directionality, linearity, periodicity, and repeatability. When identifying textures with textures, it is necessary to compare and analyze the texture features of a topic with the textures of surrounding features. The methods of finding texture knowledge include the co-occurrence matrix method and the semi-variogram function method. , Fractal fractal dimension method, Markov random field and wavelet transform method, extreme value texture method, structural element method, etc. Among them, the most commonly used are co-occurrence matrix texture method and semi-variogram function method.

Information extraction based on ground shape knowledge

Sometimes, features and backgrounds are not only the same or similar in spectral features, but also similar in texture features. In this case, you must perform a deep extraction based on the shape knowledge of the features. For example, Li Xia et al. Used the differences in the shape of ponds and rivers to distinguish the two, and realized the automatic computer extraction of information. The description of shapes has regular geometric shapes, such as rectangles, squares, rhombuses, circles, ovals, and triangles. , Pentagons, spindles, etc., with a few irregular geometric figures. The methods to discover the shape knowledge of ground features include the method based on perimeter and area, the method based on area, and the method based on area and area length.

Feature-based feature extraction

Different features have different edge shapes. The description of the edge shapes of the features includes straight lines, jagged shapes, wavy shapes, etc. The line fractal dimension can be used to find the edge feature knowledge of the features, and the edge feature knowledge can be used for the features. When used for qualitative localization and identification, first, the edge information is extracted through edge enhancement, and then the feature of the feature is used to qualitatively identify the feature. When used for qualitative extraction, the extracted Information further affirms its attributes.

Feature-based process knowledge extraction

When we have multiple periods of remote sensing images, we can establish remote sensing information extraction based on the ground feature process. This is because different ground features have different change cycles and have unique changes in their change cycles. For example, deciduous trees generally shed leaves in winter, and evergreen plants do not shed leaves throughout the year. According to this feature, they can be extracted. For another example, based on their process spectral characteristics, extraction based on their process knowledge can be established model.

Information extraction based on image spatial relationship knowledge

Image spatial relationship knowledge refers to the spatial configuration relationship between features.It can generally be divided into adjacent, contained, and contained.According to its possible, it can be divided into deterministic spatial relationship and probability spatial relationship.At the same time, it can also be divided into positive Sexual spatial relationship and negative spatial relationship. Deterministic spatial relationship means that the spatial relationship between two features is certain. As long as the feature A exists (or does not exist), there must be a feature B (or Non-existent.) Probabilistic spatial relationship means that the spatial relationship between two features exists (or does not exist) with a certain probability. The existence (or absence) of feature A can indicate the existence or non-existence of The probability of the object B can be expressed with a certain probability. This spatial relationship knowledge, for example, there are trees on both sides of the road, so that we can identify the road by identifying the trees on both sides of the road. In the application of class knowledge, the feature A can be extracted from the image, and then the existence of B is confirmed or excluded according to the correlation between A and B ^[3] .

Utilization of multi-source data and knowledge in GIS

In information extraction, in addition to using remote sensing data, a large amount of related data is generally used. Most of these data are graphic data and non-graphic data from GIS. Graphic data refers to existing maps. Non-graphic data Generally refers to demographic, social, economic and other statistical data. In the use of graphic data, there are two steps. The first step requires mining knowledge; the second step is to use this knowledge to link the graphic data with remote sensing images. To support the extraction of information. These knowledge are some positively related knowledge and anti-relevant knowledge. For these two types of knowledge, they can be further divided into deterministic and probabilistic knowledge. For graphical data, according to their relationship with us The relationship of the information to be extracted can be further divided into four cases, one is the qualitative and positioning description of the past or future state of the information; the two is the past or future state of the higher-level information containing the information The qualitative positioning description of the condition information that makes the topic exist, the qualitative positioning description of the condition information that makes the topic exist, including proximity conditions and overlapping conditions; the fourth, the condition information that restricts the existence of the topic Qualitative and positioning descriptions, which include proximity constraints and overlay constraints ^[4] .