CN114202695A

CN114202695A - Remote sensing image automatic identification system based on artificial intelligence technology

Info

Publication number: CN114202695A
Application number: CN202111533659.XA
Authority: CN
Inventors: 梁吟君; 吴清涛
Original assignee: Individual
Current assignee: Individual
Priority date: 2021-12-15
Filing date: 2021-12-15
Publication date: 2022-03-18

Abstract

The invention relates to the technical field of remote sensing image recognition, in particular to an automatic remote sensing image recognition system based on an artificial intelligence technology. The system comprises an image acquisition unit, an image processing unit, a processing unit and a correction unit, wherein the image acquisition unit acquires image information of different types, the image processing unit is arranged on the image acquisition unit, the image information is processed to obtain spectrum semantic information, texture semantic information and shape semantic information, the three different types of semantic information are combined and compared to judge whether the single semantic information has distortion or not, and the single semantic information is quickly corrected when the single semantic information is judged to have distortion, so that the situation that the recognition accuracy of features in the semantic information is reduced due to the fact that the generated semantic information has distortion caused by the influence of weather or shooting angles during image acquisition is effectively avoided, the situation that the features of the ground objects to be recognized are wrongly judged is effectively avoided, and the recognition accuracy of the system of the ground object types in the region to be recognized is effectively improved.

Description

Remote sensing image automatic identification system based on artificial intelligence technology

Technical Field

The invention relates to the technical field of remote sensing image recognition, in particular to an automatic remote sensing image recognition system based on an artificial intelligence technology.

Background

The remote sensing image target automatic identification technology is an important remote sensing application technology, and aims to automatically identify an interested target from a mass of remote sensing images and acquire related information of the interested target. The remote sensing image target automatic identification has important application value, and in military aspect, the remote sensing image automatic identification technology is utilized, which is beneficial to realizing target monitoring of target countries and regions; in the civil aspect, the automatic identification technology using the remote sensing image is beneficial to realizing planning, dynamic monitoring and the like of urban and rural areas. Initially, the remote sensing image target identification can only be manually interpreted. Subsequently, a semi-automatic identification mode of human-computer interaction is gradually developed, but due to the fact that processing consumes much time and is long in period, the mode cannot bring the superiority of the remote sensing technology into play, especially in recent years, the scale of remote sensing image data is accumulated year by year, so that the remote sensing image data is brought into a big data era, and research and development of the automatic identification technology for large-scale remote sensing images gradually become a hotspot. By taking the mature technologies related to the field of computer vision and pattern recognition as a reference, the current methods for automatically recognizing remote sensing images are mainly divided into two types: the first is a top-down model driving method, and the second is a bottom-up data driving method. For large-scale remote sensing images, the latter has practical significance and feasibility based on calculation and time cost consideration, but the existing feature extraction method has poor robustness on complex changes of illumination, angle, scale, background and the like of the remote sensing images, and has an unsatisfactory application effect in large-scale remote sensing image target identification.

The method aims at the existing high-resolution, especially ultra-high-resolution remote sensing image data, the space information of the earth surface elements is highly detailed, and the corresponding typical earth surface elements are complex in type. However, the number of original spectral bands of the high-resolution remote sensing image is limited by the mutual restriction of the spectral resolution and the spatial resolution of the current sensor technology, and generally, the high-resolution remote sensing image only has four multispectral bands of red light, green light, blue light and near infrared, and the phenomenon of 'same object different spectrum' or 'same spectrum foreign matter' is more serious due to the influence of factors such as cloud layer shielding, ground object shadow, atmospheric reflection difference and the like. Meanwhile, the existing remote sensing identification technology still has many problems for building identification. For example, it is difficult to detect the edge of a building straight; the detection effect on buildings in urban areas, mountain areas and rural areas is greatly different; it is difficult to resist the interference of building shadows; due to the problem of the shooting angle of the remote sensing image, the side face of the inclined building is difficult to be correctly eliminated, and the identification accuracy is not high.

Disclosure of Invention

Therefore, the invention provides an automatic remote sensing image identification system based on an artificial intelligence technology, which is used for solving the problem of low identification precision caused by the fact that the prior art cannot quickly judge and correct semantic information distortion caused by the influence of weather and shooting angles.

In order to achieve the above object, the present invention provides an automatic remote sensing image recognition system based on artificial intelligence technology, comprising:

the image acquisition unit is used for acquiring image information of the area to be identified;

the image processing unit is connected with the image acquisition unit and used for receiving the image information acquired by the image acquisition unit and preprocessing the image information; the image processing unit generates a plurality of semantic information including spectrum semantic information, texture semantic information and shape semantic information after finishing preprocessing the image information;

the central control unit is connected with the image processing unit and used for receiving a plurality of semantic information output by the image processing unit and identifying the actual types and the actual shapes of all ground objects in the area to be identified according to all the semantic information; if the central control unit judges that the area to be recognized is a brand new area when receiving the semantic information transmitted by the image processing unit, the central control unit combines the semantic information transmitted by the image processing unit to complete recognition of the actual type and the actual shape of each feature in the area to be recognized;

the training unit is connected with the central control unit and used for determining the actual environment of the to-be-recognized area when the image acquisition unit acquires the image according to the known area and determining the ground object recognition standard according to the actual environment when the central control unit judges that the known area exists in the to-be-recognized area according to the semantic information;

the storage unit is respectively connected with the central control unit and the training unit and is used for storing the texture features of the corresponding ground objects, the spectrum semantic information of the corresponding ground objects in different environments and the shape features of the ground objects in the recognized regions; when the central control unit identifies the feature type and the feature shape in the semantic information, the central control unit selects the corresponding information from the storage unit and compares the corresponding information with the features in the corresponding semantic information to identify each feature in the semantic information.

Further, the central control unit selects shape semantic information when receiving the semantic information and compares the shape semantic information with the shape features of the ground objects in the recognized region stored in the training unit, and if the shape semantic information contains a plurality of shape features of adjacent ground objects which are the same as the shape features of the corresponding ground objects in the recognized region, the central control unit calculates the ratio S of the area of the shape features of the adjacent ground objects to the total area in the shape semantic information and judges whether a known region exists in the region to be recognized according to S; the central control unit is provided with a preset shape area ratio S0,

if S is larger than or equal to S0, the central control unit judges that a known region exists in the region to be recognized and transmits the received semantic information to the training unit, the training unit extracts the semantic information of the known region which is recognized from the storage unit and compares the semantic information with the characteristics of the known region in the semantic information transmitted by the central control unit to judge whether the semantic information obtained in the current recognition process has distortion, if not, the central control unit recognizes the feature type and the feature shape in the received semantic information, if so, the central control unit selects a corresponding correction mode according to the judgment result of the training unit to correct the corresponding semantic information and recognizes the feature type and the feature shape in the received semantic information after the correction is finished;

if S < S0, the central control unit preliminarily determines that no known region exists in the region to be recognized, and the central control unit detects whether the shape features of a plurality of adjacent ground objects are similar to the shape features of the corresponding ground objects in the region to be recognized so as to further determine whether the known region exists in the region to be recognized.

Further, when the central control unit detects whether a known area exists in the area to be recognized or not, if the similarity between the shape features of a plurality of adjacent ground objects and the shape features of the corresponding ground objects in the recognized area is greater than or equal to 85%, the central control unit determines that similar areas exist in the semantic information to be recognized, compares the overall contour of the adjacent ground objects with the contour of the shape of the similar corresponding ground object, counts the contour distance at the corresponding position of the two contours to obtain the maximum distance Dmax and the minimum distance Dmin, and calculates the average distance D between the corresponding positions of the two contours; the central control unit is also provided with a preset average distance D0 and a preset average distance difference Delta D0,

if D is less than D0, the central control unit calculates the difference value Delta Da between adjacent distances, sets the difference value Delta Da as D0-D, and if the difference value Delta Da is less than Delta D0, the central control unit calculates the difference value between D and Dmin and judges whether the shape semantic information of the area to be identified is distorted or whether the difference between the shape semantic information of the area to be identified and the shape characteristics of the corresponding ground object in the area to be identified is only in proportion; if the delta Da is more than or equal to the delta D0, the central control unit judges that the shape features of the plurality of adjacent ground objects and the shape features of the corresponding ground objects in the recognition completion area are not the same, judges the area to be recognized as a brand new area and combines a plurality of semantic information transmitted by the image processing unit to complete recognition of the actual type and the actual shape of each ground object in the area to be recognized;

if D is larger than or equal to D0, the central control unit calculates the difference value delta Db of the adjacent distance, the set delta Db is D-D0, and if the delta Db is smaller than delta D0, the central control unit calculates the difference value between Dmax and D and judges whether the shape semantic information of the area to be recognized is distorted or whether the difference between the shape semantic information of the area to be recognized and the shape characteristics of the corresponding ground object in the region for completing recognition is only in proportion or not according to the difference value; if the delta Db is not less than or equal to the delta D0, the central control unit judges that the shape features of the adjacent ground objects are not the same as the shape features of the corresponding ground objects in the recognition completion area, judges the area to be recognized as a brand new area and combines the semantic information transmitted by the image processing unit to complete the recognition of the actual type and the actual shape of each ground object in the area to be recognized.

Further, when D < D0 and Δ Da < Δ D0, if D-Dmin ≦ Δ D0, the central control unit determines that the shape features of the neighboring features are similar to those of the corresponding feature in the recognition-completed region and determines that the reason for the similarity between the two is less than 100% as a deviation of the shape ratio of the two due to the difference of the acquisition heights when the image of the region to be recognized is acquired, the central control unit adjusts the ratio of the shape semantic information of the region to be recognized to re-determine whether the shape features of the neighboring features and the shape features of the corresponding feature in the recognition-completed region are the same, if D-Dmin >. Δ D0, the central control unit judges that shape distortion occurs to shape semantic information of the area to be recognized, performs geometric correction to the shape semantic features, and judges whether the shape features of a plurality of adjacent ground objects and the shape features of corresponding ground objects in the recognized area are the same after correction;

when D is more than or equal to D0 and delta Db is less than delta D0, if Dmax-D is less than or equal to delta D0, the central control unit judges that the shape features of the plurality of adjacent ground objects are similar to the shape features of the corresponding ground objects in the recognition-finished region and judges that the reason that the similarity between the shape features and the corresponding ground objects is less than 100 percent is the deviation of the shape proportion of the ground objects and the corresponding ground objects caused by different acquisition heights when the images of the to-be-recognized region are acquired, the central control unit adjusts the proportion of the shape semantic information of the to-be-recognized region to judge whether the shape features of the plurality of adjacent ground objects and the shape features of the corresponding ground objects in the recognition-finished region are the same features again, if Dmax-D > -delta D0, the central control unit judges that shape distortion occurs to the shape semantic information of the region to be recognized, performs geometric correction on the shape semantic features, and judges whether the shape features of a plurality of adjacent ground objects and the shape features of the corresponding ground objects in the recognized region are the same after correction.

Further, when the central control unit judges that a known region exists in the region to be identified, the central control unit compares the spectral features of the corresponding ground objects in the identified region with the spectral features in the spectral semantic information to judge whether spectral distortion occurs in the spectral semantic features, and if the spectral distortion occurs, the central control unit eliminates distortion attached to radiation brightness in the spectral semantic information according to the spectral semantic information of the corresponding ground objects in the identified region and the spectral semantic information to finish radiation correction of the radiation semantic information;

when the central control unit judges that the area to be identified is a brand new area, the central control unit judges whether shape distortion and/or spectrum distortion exist or not according to the shape semantic information and the spectrum semantic information in sequence and performs targeted correction on the corresponding semantic information when the distortion exists;

and when the central control unit finishes the correction of the corresponding semantic information, the central control unit identifies the ground object type in the region to be identified according to the texture semantic information.

Further, when the central control unit judges that the area to be identified is a brand new area and spectral distortion exists in the area to be identified, the central control unit obtains an included angle theta between the irradiation angle of the sun and the ground through the texture semantic information so as to judge the cause of radiation distortion; the central control unit is internally provided with a first preset angle theta 1 and a second preset angle theta 2, wherein theta 1 is more than 0 and more than theta 2 and less than 90 degrees,

if theta is less than or equal to theta 1, the central control unit judges that the generation reason of radiation distortion is that the solar irradiation angle is too low;

if theta 1 is larger than theta and smaller than or equal to theta 1, the central control unit judges that the generation reason of the radiation distortion is atmospheric scattering or absorption in the to-be-identified area caused by fog or mist in the environment;

if theta 2 is larger than theta and smaller than 90 degrees, the central control unit judges that the instrument in the image acquisition unit has a problem, sends out a maintenance alarm of the image acquisition unit, controls the image acquisition unit to acquire the image information of the area to be identified again after the maintenance of the image acquisition unit is completed, and controls the image processing unit to generate corresponding semantic information again.

Further, a preset saturation C0 is arranged in the central control unit, when the central control unit judges that the generation cause of radiation distortion is too low because of the sun irradiation angle, the central control unit performs radiation correction on the spectrum semantic information and detects the saturation C of colors in the spectrum semantic information after completing the radiation correction, if C is less than C0, the central control unit judges that the atmosphere scattering and/or absorption condition exists in the area to be identified and selects a correction mode aiming at fog mist to perform secondary radiation correction on the corrected spectrum semantic information; and if C is larger than or equal to C0, the central control unit judges that the radiation correction is finished and identifies the ground objects in the area to be identified according to the texture semantic information.

Further, when the central control unit identifies the ground object in the area to be identified according to the texture semantic information, the central control unit extracts the pre-stored texture features from the storage unit and uses the texture features to identify the texture features in the texture semantic information in turn,

if the similarity between the texture features in the single piece of texture semantic information and the single pre-stored texture feature is more than 95%, the central control unit completes the identification of the types of the ground objects to which the texture features in the texture semantic information belong;

if the similarity between the texture features in the single piece of texture semantic information and the pre-stored texture features is less than 90%, the central control unit preliminarily judges that the definition of the texture features in the texture semantic information does not meet the standard and adjusts the definition of the texture features in the texture semantic information, if the definition of the texture features in the texture semantic information is adjusted, the similarity between the texture features in the texture semantic information and the pre-stored texture features is still less than 90%, and the central control unit preliminarily judges that the texture features in the texture semantic information are brand new texture features.

Further, when the central control unit preliminarily judges that the texture feature is a brand new texture feature, the central control unit detects whether the spectrum semantic information or the shape semantic information in the area to be identified is distorted,

if the shape semantic information has distortion, the central control unit readjusts the overall shape of the texture feature according to the shape correction process aiming at the distortion of the shape semantic information;

if the spectrum semantic information has distortion, the central control unit readjusts the tone and/or color of the texture feature according to a radiation correction process aiming at the distortion of the spectrum semantic information;

and if the texture feature is still judged to be a brand new texture feature by the central control unit after the adjustment is finished, the central control unit stores the texture feature into the storage unit after the manual identification.

Further, if the central control unit determines that the area to be recognized is a brand new area and the central control unit determines that the shape semantic information of the area to be recognized is distorted, the central control unit selects corresponding control points from the shape semantic information of the area to be recognized, establishes an overall mapping function according to the control points, rearranges the data of the shape features, and completes shape correction of the shape semantic information.

Compared with the prior art, the method has the advantages that the image processing unit is arranged to process the image information acquired by the image acquisition unit to obtain the spectrum semantic information, the texture semantic information and the shape semantic information, the three different types of semantic information are combined and compared to judge whether the single semantic information has distortion or not, and the single semantic information is quickly corrected when the single semantic information has distortion, so that the situation that the recognition accuracy of the features in the semantic information is reduced due to the fact that the generated semantic information is distorted due to the influence of weather or shooting angles when the image is acquired is effectively avoided. Meanwhile, by arranging the training unit, when the central control unit judges that the known area exists in the area to be recognized, whether the semantic information in the area to be recognized has distortion or not can be quickly determined by taking the actual condition of the prestored known area as a judgment reference, and the distorted semantic information is quickly corrected when the distortion is judged to exist, so that the condition of misjudgment of the feature of the ground object to be recognized is effectively avoided, and the recognition accuracy of the system for the type of the ground object in the area to be recognized is effectively improved.

Further, when the central control unit receives the semantic information, if the shape characteristics of a plurality of adjacent ground objects exist in the shape semantic information and the shape characteristics of corresponding ground objects in the recognition-completed region are the same, the central control unit calculates the ratio S of the area of the shape characteristics of the adjacent ground objects to the total area in the shape semantic information and judges whether a known region exists in the region to be recognized according to S, the invention can enable the central control unit to quickly finish the judgment of whether the known region exists in the shape semantic information by setting the preset shape-area ratio as the judgment reference of whether the known region exists, meanwhile, if the central control unit preliminarily judges that the known region does not exist in the region to be recognized, the central control unit detects whether the shape characteristics of the plurality of adjacent ground objects are similar to the shape characteristics of the corresponding ground objects in the recognition-completed region so as to further judge whether the known region exists in the region to be recognized, by detecting the shape characteristics of the similar ground objects, the occurrence of misjudgment caused by deviation between the acquisition shape of the ground objects and the century star shape of the ground objects due to weather, shooting height or shooting angle can be effectively avoided, so that the identification efficiency of the system is effectively improved.

Further, if the similarity between the shape features of a plurality of adjacent ground objects and the shape features of the corresponding ground objects in the recognition-completed region is greater than or equal to 85%, the central control unit judges that a similar region exists in the semantic information to be recognized, compares the overall contour of the adjacent ground objects with the contour of the similar corresponding ground object shape, counts the contour distance at the corresponding positions of the two contours to obtain the maximum distance Dmax and the minimum distance Dmin, and calculates the average distance D of the corresponding positions of the two contours, the present invention can quickly and accurately judge whether the ground objects described by the shape contour in the known region and the corresponding shape contour edge similar in the shape semantic information are the same ground objects by detecting that the distance between the shape contour in the known region and the corresponding shape contour edge similar in the shape semantic information is the same ground objects, thereby further improving the recognition efficiency of the system of the present invention, the identification precision of the system aiming at the types of the ground objects in the area to be identified is further improved.

Further, when the central control unit judges that a known region exists in the region to be identified, the central control unit compares the spectral features of the corresponding ground objects in the identified region with the spectral features in the spectral semantic information to judge whether spectral distortion occurs in the spectral semantic features, and if the spectral distortion occurs, the central control unit eliminates distortion attached to radiation brightness in the spectral semantic information according to the spectral semantic information of the corresponding ground objects in the identified region and the spectral semantic information to finish radiation correction of the radiation semantic information; when the central control unit judges that the area to be identified is a brand new area, the central control unit judges whether shape distortion and/or spectrum distortion exist or not according to the shape semantic information and the spectrum semantic information in sequence and performs targeted correction on the corresponding semantic information when the distortion exists; the invention respectively sets the corresponding correction flows aiming at the two conditions of the existence of the known area and the absence of the known area, can effectively avoid the condition that the distortion in the new area cannot be identified and corrected due to the absence of the known area in the area to be identified, and further improves the identification efficiency of the system of the invention and the identification precision of the system of the invention aiming at the type of the ground object in the area to be identified.

Further, when the central control unit judges that the area to be identified is a brand new area and the area to be identified has spectral distortion, the central control unit calculates an included angle theta between an irradiation angle of the sun and the ground through the texture semantic information to judge the cause of the radiation distortion.

Furthermore, the central control unit is provided with a preset saturation C0, when the central control unit judges that the generation reason of the radiation distortion is caused by the fact that the sun irradiation angle is too low, the central control unit carries out radiation correction on the spectral semantic information, detects the saturation C of colors in the spectral semantic information after the radiation correction is finished, compares the saturation C with the saturation C0 to further judge whether the atmosphere in the area to be identified has scattering and/or absorption or not, can carry out further judgment after the preliminary judgment of the reason of the radiation distortion in the area to be identified is finished by setting the corresponding color saturation, thereby effectively avoiding the situation that the distortion in the radiation semantic information can not be corrected by using a correction method aiming at a single situation when the sun irradiation angle and the atmospheric scattering and/or absorption exist simultaneously, therefore, the identification precision of the system aiming at the types of the ground objects in the area to be identified is further improved.

Furthermore, when the central control unit identifies the ground object in the area to be identified according to the texture semantic information, if the central control unit detects that the single texture feature is not similar to each pre-stored texture feature, the central control nano element preliminarily judges that the definition of the texture feature in the texture semantic information does not meet the standard, adjusts the definition of the texture feature in the texture semantic information and further judges the texture feature after adjustment.

Furthermore, when the central control unit preliminarily judges that the texture feature is a brand new texture feature, the central control unit detects whether the spectrum semantic information or the shape semantic information in the region to be identified is distorted or not and makes corresponding correction according to the actually distorted semantic information.

Drawings

Fig. 1 is a block diagram of the system for automatically identifying remote sensing images based on artificial intelligence technology according to the present invention.

Detailed Description

In order that the objects and advantages of the invention will be more clearly understood, the invention is further described below with reference to examples; it should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

Preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only for explaining the technical principle of the present invention, and do not limit the scope of the present invention.

It should be noted that in the description of the present invention, the terms of direction or positional relationship indicated by the terms "upper", "lower", "left", "right", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings, which are only for convenience of description, and do not indicate or imply that the device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and thus, should not be construed as limiting the present invention.

Furthermore, it should be noted that, in the description of the present invention, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.

Fig. 1 is a block diagram of an automatic remote sensing image recognition system based on artificial intelligence technology according to the present invention. The invention discloses an automatic remote sensing image identification system based on an artificial intelligence technology, which comprises:

The image processing unit is arranged to process the image information acquired by the image acquisition unit to obtain the spectrum semantic information, the texture semantic information and the shape semantic information, the three different types of semantic information are combined and compared to judge whether the single semantic information has distortion or not, and the single semantic information is quickly corrected when the single semantic information is judged to have distortion, so that the situation that the recognition accuracy of the features in the semantic information is reduced due to the fact that the generated semantic information is distorted due to the influence of weather or shooting angles when the image is acquired is effectively avoided. Meanwhile, by arranging the training unit, when the central control unit judges that the known area exists in the area to be recognized, whether the semantic information in the area to be recognized has distortion or not can be quickly determined by taking the actual condition of the prestored known area as a judgment reference, and the distorted semantic information is quickly corrected when the distortion is judged to exist, so that the condition of misjudgment of the feature of the ground object to be recognized is effectively avoided, and the recognition accuracy of the system for the type of the ground object in the area to be recognized is effectively improved.

Continuing to refer to fig. 1, when receiving the semantic information, the central control unit of the present invention selects shape semantic information and compares the shape semantic information with the shape features of the ground objects in the recognized region stored in the training unit, and if the shape semantic information includes a plurality of shape features of adjacent ground objects that are the same as the shape features of the corresponding ground objects in the recognized region, the central control unit calculates a ratio S between the area of the shape features of the adjacent ground objects and the total area in the shape semantic information and determines whether a known region exists in the region to be recognized according to S; the central control unit is provided with a preset shape area ratio S0,

The invention can enable the central control unit to quickly finish the judgment of whether the known area exists in the shape semantic information by setting the preset shape area ratio as the judgment reference of whether the known area exists, meanwhile, if the central control unit preliminarily judges that the known area does not exist in the area to be recognized, the central control unit detects whether the shape characteristics of a plurality of adjacent ground objects are similar to the shape characteristics of the corresponding ground objects in the area to be recognized so as to further judge whether the known area exists in the area to be recognized, and the occurrence of misjudgment caused by the deviation between the collected shape of the ground objects and the century star shape of the ground objects due to weather, shooting height or shooting angle can be effectively avoided by detecting the shape characteristics of the similar ground objects, so that the identification efficiency of the system is effectively improved.

Specifically, when the central control unit detects whether a known area exists in an area to be recognized or not, if the similarity between the shape features of a plurality of adjacent ground objects and the shape features of the corresponding ground objects in the recognized area is greater than or equal to 85%, the central control unit determines that a similar area exists in semantic information to be recognized, compares the overall contour of the adjacent ground objects with the contour of the shape of the similar corresponding ground object, counts the contour distance at the corresponding position of the two contours to obtain the maximum distance Dmax and the minimum distance Dmin, and calculates the average distance D between the corresponding positions of the two contours; the central control unit is also provided with a preset average distance D0 and a preset average distance difference Delta D0,

Specifically, when D < D0 and Δ Da < Δ D0, if D-Dmin ≦ Δ D0, the central control unit determines that the shape features of the neighboring features are similar to those of the corresponding feature in the recognition-completed region and determines that the reason for the similarity between the two is less than 100% as a deviation of the shape proportion of the two due to the difference of the acquisition heights when the image of the region to be recognized is acquired, the central control unit adjusts the proportion of the shape semantic information of the region to be recognized to re-determine whether the shape features of the neighboring features and the shape features of the corresponding feature in the recognition-completed region are the same, if D-Dmin >. Δ D0, the central control unit judges that shape distortion occurs to shape semantic information of the area to be recognized, performs geometric correction to the shape semantic features, and judges whether the shape features of a plurality of adjacent ground objects and the shape features of corresponding ground objects in the recognized area are the same after correction;

According to the method and the system, whether the shape outline in the known area and the corresponding shape outline edge similar to the shape semantic information are the same ground object or not is judged quickly by detecting the distance between the shape outline in the known area and the corresponding shape outline edge similar to the shape semantic information, and whether the ground object to which the corresponding shape feature in the shape semantic information belongs is the known ground object or not can be judged quickly and accurately, so that the identification efficiency of the system is further improved, and the identification precision of the system for the type of the ground object in the area to be identified is further improved.

As shown in fig. 1, when the central control unit determines that a known region exists in the region to be identified, the central control unit compares the spectral features of the corresponding ground object in the identified region with the spectral features in the spectral semantic information to determine whether spectral distortion occurs in the spectral semantic features, and if spectral distortion occurs, the central control unit eliminates distortion attached to the radiation brightness in the spectral semantic information according to the spectral semantic information of the corresponding ground object in the identified region and the spectral semantic information to complete radiation correction of the radiation semantic information;

The invention respectively sets the corresponding correction flows aiming at the two conditions of the existence of the known area and the absence of the known area, can effectively avoid the condition that the distortion in the new area cannot be identified and corrected due to the absence of the known area in the area to be identified, and further improves the identification efficiency of the system of the invention and the identification precision of the system of the invention aiming at the type of the ground object in the area to be identified.

Specifically, when the central control unit judges that the area to be identified is a brand new area and the area to be identified has spectral distortion, the central control unit calculates an included angle theta between an irradiation angle of the sun and the ground through the texture semantic information so as to judge the cause of radiation distortion; the central control unit is internally provided with a first preset angle theta 1 and a second preset angle theta 2, wherein theta 1 is more than 0 and more than theta 2 and less than 90 degrees,

According to the system and the method, the texture semantic information is used for calculating the irradiation angle of the sun when the image acquisition unit acquires the image of the area to be identified, and the reason for generating radiation distortion is determined according to the calculation result, so that the central control unit can effectively and quickly correct the distortion caused by different reasons, the identification efficiency of the system is further improved, and the identification precision of the system for the types of the ground objects in the area to be identified is further improved.

Specifically, the central control unit is provided with a preset saturation C0, when the central control unit judges that the generation cause of radiation distortion is too low because of the sun irradiation angle, the central control unit performs radiation correction on the spectrum semantic information and detects the saturation C of colors in the spectrum semantic information after completing the radiation correction, if C is less than C0, the central control unit judges that the atmosphere scattering and/or absorption condition exists in the area to be identified and selects a correction mode aiming at fog mist to perform secondary radiation correction on the corrected spectrum semantic information; and if C is larger than or equal to C0, the central control unit judges that the radiation correction is finished and identifies the ground objects in the area to be identified according to the texture semantic information.

According to the invention, through setting the corresponding color saturation, the preliminary judgment on the reason of the radiation distortion in the area to be identified can be completed, and then the further judgment is carried out, so that the situation that the distortion in the radiation semantic information cannot be corrected by using a correction method aiming at a single situation when the sun irradiation angle and the atmospheric scattering and/or absorption exist simultaneously is effectively avoided, and the identification precision of the system aiming at the types of the ground objects in the area to be identified is further improved.

Specifically, when the central control unit identifies the ground object in the area to be identified according to the texture semantic information, the central control unit extracts the pre-stored texture features from the storage unit and uses the texture features to identify the texture features in the texture semantic information in turn,

According to the method, when the measured single texture feature is not similar to each pre-stored texture feature, the definition of the texture feature is preliminarily judged to be not in accordance with the standard, and the definition of the texture feature is adjusted to be further identified, so that the condition of misjudgment aiming at the texture feature caused by unqualified definition can be effectively avoided, and the identification precision of the system aiming at the ground object type in the area to be identified is further improved.

Specifically, when the central control unit preliminarily judges the texture feature to be a brand new texture feature, the central control unit detects whether the spectral semantic information or the shape semantic information in the region to be identified is distorted,

By combining the texture features with other two semantic information, the invention can further improve the identification precision of the system of the invention for the ground object types in the area to be identified while further avoiding the occurrence of misjudgment aiming at the texture features caused by unqualified definition.

As shown in fig. 1, if the central control unit determines that the to-be-identified region is a brand-new region and the central control unit determines that the shape semantic information of the to-be-identified region is distorted, the central control unit selects a corresponding control point from the shape semantic information of the to-be-identified region, establishes an integral mapping function according to the control point, rearranges the data of the shape features, and completes the shape correction of the shape semantic information.

So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention; various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The utility model provides a remote sensing image automatic identification system based on artificial intelligence technique which characterized in that includes:

2. The system for automatically identifying remote sensing images based on the artificial intelligence technology as claimed in claim 1, wherein the central control unit selects shape semantic information when receiving the semantic information and compares the shape semantic information with the shape features of the ground objects in the recognized region stored in the training unit, if the shape semantic information has the shape features of a plurality of adjacent ground objects which are the same as the shape features of the corresponding ground objects in the recognized region, the central control unit calculates the ratio S of the area of the shape features of the adjacent ground objects to the total area in the shape semantic information and judges whether the known region exists in the region to be recognized according to S; the central control unit is provided with a preset shape area ratio S0,

3. The remote sensing image automatic identification system based on the artificial intelligence technology as claimed in claim 2, wherein when the central control unit detects whether there is a known area in the area to be identified, if there is a similarity between the shape features of a plurality of adjacent ground objects and the shape features of the corresponding ground objects in the area to be identified which is greater than or equal to 85%, the central control unit determines that there is a similar area in the semantic information to be identified, compares the overall contour of the adjacent ground objects with the contour of the similar corresponding ground object shape, counts the contour distance at the corresponding positions of the two contours to obtain the maximum distance Dmax and the minimum distance Dmin, and calculates the average distance D between the corresponding positions of the two contours; the central control unit is also provided with a preset average distance D0 and a preset average distance difference Delta D0,

4. The system of claim 2, wherein when D < D0 and Δ Da < Δ D0, if D-Dmin ≦ Δ D0, the central control unit determines that the shape features of the neighboring features are similar to those of the corresponding feature in the identification-completed region and determines that the similarity between the shape features and the corresponding feature is less than 100%, the central control unit adjusts the ratio of the shape semantic information of the region to be identified to determine whether the shape features of the neighboring features and the shape features of the corresponding feature in the identification-completed region are the same, and if D-Dmin > ΔD0, the central control unit determines that the shape semantic information of the region to be identified has shape distortion, geometrically corrects the shape semantic features, and after correction, re-determines that the shape features of the neighboring features and the shape features of the corresponding feature in the identification-completed region Whether the features are the same;

5. The system according to claim 3, wherein when the central control unit determines that a known region exists in the region to be identified, the central control unit compares the spectral features of the corresponding ground object in the identified region with the spectral features in the spectral semantic information to determine whether spectral distortion occurs in the spectral semantic features, and if spectral distortion occurs, the central control unit eliminates distortion attached to the radiation brightness in the spectral semantic information according to the spectral semantic information of the corresponding ground object in the identified region and the spectral semantic information to complete radiation correction of the radiation semantic information;

6. The system for automatically identifying remote sensing images based on the artificial intelligence technology as claimed in claim 5, wherein when the central control unit determines that the area to be identified is a brand new area and spectral distortion exists in the area to be identified, the central control unit determines an included angle theta between an irradiation angle of the sun and the ground through the texture semantic information so as to determine the cause of radiation distortion; the central control unit is internally provided with a first preset angle theta 1 and a second preset angle theta 2, wherein theta 1 is more than 0 and more than theta 2 and less than 90 degrees,

7. The remote sensing image automatic identification system based on artificial intelligence technology as claimed in claim 6, wherein the central control unit is provided with a preset saturation C0, when the central control unit determines that the generation cause of radiation distortion is too low because of the sun illumination angle, the central control unit performs radiation correction on the spectral semantic information and detects the saturation C of the color in the spectral semantic information after completing the radiation correction, if C < C0, the central control unit determines that the atmosphere scattering and/or absorption exists in the area to be identified and selects a correction mode aiming at fog to perform secondary radiation correction on the corrected spectral semantic information; and if C is larger than or equal to C0, the central control unit judges that the radiation correction is finished and identifies the ground objects in the area to be identified according to the texture semantic information.

8. The system of claim 7, wherein when the central control unit identifies the surface features in the area to be identified according to the texture semantic information, the central control unit extracts pre-stored texture features from the storage unit and uses the texture features to identify the texture features in the texture semantic information in sequence,

9. The remote sensing image automatic identification system based on artificial intelligence technology of claim 8, characterized in that, when the central control unit preliminarily judges the texture feature to be a brand new texture feature, the central control unit detects whether the spectral semantic information or the shape semantic information in the area to be identified is distorted,

10. The system of claim 5, wherein if the central control unit determines that the area to be identified is a brand new area and the central control unit determines that the shape semantic information of the area to be identified is distorted, the central control unit selects corresponding control points from the shape semantic information of the area to be identified, establishes an integral mapping function according to the control points, rearranges the data of the shape features, and completes shape correction of the shape semantic information.