An artificial intelligence-based target identification method in a complex environment

Abstract

The application discloses a target recognition method and device based on artificial intelligence in a complex environment, and the target recognition method comprises the following steps: extracting image materials corresponding to an environment in a database based on a detection image of a detection environment; performing feature extraction on the image materials to obtain a plurality of image feature contours in the corresponding environment; performing recognition and removal on image feature data of a complete contour in the detection image to obtain a defective image; extracting a plurality of defective feature contours in the defective image, repairing the defective feature contours to obtain completed image contour data; filling the plurality of defective feature contours after the repair in the defective image, verifying the completed image contour data; correcting the completed image contour data based on a verification result, obtaining a plurality of feature contours in the detection image, and obtaining a target feature contour. Through the extraction of the feature contours in the detection image and the repair and completion of the defective feature contours, the accuracy of target detection in the detection image is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and particularly relates to a target recognition method based on artificial intelligence in a complex environment. BACKGROUND

[0002] Target recognition extraction is one of main methods of image processing. For a photographed picture in a complex scene, since multiple interference objects exist in the scene and the multiple interference objects cause occlusion to a target unit, the difficulty of recognizing and extracting the target unit in the scene is increased. An existing recognition method generally sets multiple candidate boxes, filters a region where the target unit appears, and thus improves the accuracy of target recognition. Since there are many feature outlines of interference objects in the complex scene, the target detection in the candidate region is interfered, and thus the accuracy of the candidate box selection mode is reduced. SUMMARY

[0003] The present application aims at overcoming the deficiencies of the prior art, and provides a target recognition method based on artificial intelligence in a complex environment and a device thereof. The method includes extracting feature outlines in a detection image, repairing and completing the incomplete feature outlines, and improving the accuracy of target detection in the detection image.

[0004] The present application provides a target recognition method based on artificial intelligence in a complex environment, which includes the following steps.

[0005] Confirming a detection environment according to a detection image, extracting image materials of a corresponding environment in a database based on the detection environment;

[0006] Extracting features from the image materials, and obtaining a plurality of image feature outlines in the corresponding environment;

[0007] Recognizing outlines of the detection image, removing image feature data of complete outlines, and obtaining incomplete images;

[0008] Extracting a plurality of incomplete feature outlines from the incomplete images, repairing the incomplete feature outlines according to the plurality of image feature outlines, and obtaining complete image outline data;

[0009] Filling the plurality of repaired incomplete feature outlines in the incomplete images, and verifying the complete image outline data;

[0010] Based on the verification result, correcting the complete image outline data, obtaining a plurality of feature outlines in the detection image, and obtaining a target feature outline in the plurality of feature outlines.

[0011] Further, the step of confirming the detection environment according to the detection image and extracting the image materials of the corresponding environment in the database based on the detection environment includes the following steps.

[0012] inputting the detection image into a recognition system, obtaining a detection environment according to a shooting environment of the detection image;

[0013] extracting image data of the same environment as the detection environment from a database according to the detection environment, and extracting the image data as image materials.

[0014] Further, the feature extraction of the image materials and the acquisition of a plurality of image feature contours in the corresponding environment include:

[0015] The contour extraction of the image materials by the Canny algorithm extracts different feature contours in the image materials, thereby constructing a feature contour database of the detection environment.

[0016] Further, the contour extraction of the image materials by the Canny algorithm extracts different feature contours in the image materials, thereby constructing a feature contour database of the detection environment includes:

[0017] The acquired image materials are subjected to gray value processing, and the gray value image of the acquired image materials is subjected to Gaussian filtering processing.

[0018] The image after Laplace transform is further sharpened by non-maximum suppression method to obtain an accurate feature contour image.

[0019] Further, the contour recognition of the detection image, the recognition and removal of the image feature data of the complete contour, and the acquisition of the incomplete image include:

[0020] The detection image is preprocessed to obtain a preprocessed image.

[0021] The preprocessed image is subjected to contour recognition, and the complete feature contour is marked and extracted.

[0022] The complete feature contour part in the preprocessed image is subjected to undercolor filling to obtain an incomplete image.

[0023] Further, the extraction of a plurality of incomplete feature contours in the incomplete image, the repair of the incomplete feature contour according to the plurality of image feature contours, and the acquisition of the complete image contour data include:

[0024] The incomplete feature contour is extracted, and the approximate image feature contour is acquired in the plurality of image feature contours according to the incomplete feature contour.

[0025] The deflection angle of the approximate image feature is adjusted based on the incomplete feature contour, and the incomplete feature contour is completed based on the approximate image feature contour.

[0026] Further, the extracting the incomplete feature contour, and obtaining the approximate image feature contour according to the comparison of the incomplete feature contour in the plurality of image feature contours comprises:

[0027] The incomplete feature contour is obtained around the feature contour in the plurality of hollow states in the incomplete image, and the incomplete feature contour is connected with the feature contour in the hollow state.

[0028] Further, the filling the plurality of incomplete feature contours after repair in the incomplete image, and verifying the complete image contour data comprises:

[0029] The complete image contour is filled in the incomplete image, the complete image contour is verified based on the hollow area in the incomplete image, and whether the complete part of the complete image contour is completely contained in the hollow area of the incomplete image is detected;

[0030] If yes, the complete image contour is marked as a repaired image contour, if not, the complete image contour is marked as an error contour, and the complete image contour is removed.

[0031] Further, the correcting the complete image contour data based on the verification result, obtaining the plurality of feature contours in the detection image, and obtaining the target feature contour in the plurality of feature contours comprises:

[0032] The plurality of feature contours of the corrected detection image are extracted, and the feature contour corresponding to the target feature contour comparison data in the plurality of feature contours is extracted and marked as a target unit;

[0033] The complete image contour data and the incomplete incomplete feature contour data of the corrected detection image are obtained, and the first type of target image data and the second type of target image data are obtained after comparison through the target feature contour comparison data.

[0034] The application also provides a target recognition device based on artificial intelligence in a complex environment, the recognition device comprising:

[0035] An image material obtaining module is used for confirming a detection environment according to a detection image, and extracting image materials of a corresponding environment in a database based on the detection environment;

[0036] A feature extraction module is used for extracting features of the image materials, and obtaining a plurality of image feature contours in the corresponding environment;

[0037] A feature removal module is used for recognizing contours of the detection image, removing image feature data of complete contours, and obtaining an incomplete image;

[0038] The feature repairing module is used for extracting a plurality of incomplete feature contours in the incomplete image, repairing the incomplete feature contours according to a plurality of image feature contours, and obtaining complete image contour data.

[0039] The verifying module is used for verifying the plurality of repaired incomplete feature contours filled in the incomplete image, and verifying the complete image contour data.

[0040] The correcting module is used for correcting the complete image contour data based on the verification result, obtaining a plurality of feature contours in the detection image, and obtaining a target feature contour in the plurality of feature contours.

[0041] The present application provides a target recognition method and device based on artificial intelligence in a complex environment, according to the scene of the detection image, a plurality of image materials are extracted, the complete feature contours in the detection image are removed to obtain an incomplete image, and the incomplete feature contours in the incomplete image are completed based on the feature contours of the plurality of image materials, which improves the convenience of extracting feature contours in the detection image and improves the accuracy of recognizing target units in the detection image. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0043] Figure 1 is a flow chart of the target recognition method based on artificial intelligence in a complex environment in the embodiment of the present application;

[0044] Figure 2 is a schematic diagram of the target recognition device based on artificial intelligence in a complex environment in the embodiment of the present application. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0046] Embodiment one:

[0047] The embodiment of the application discloses a target recognition method based on artificial intelligence in a complex environment, which comprises the following steps:

[0048] The image material is subjected to feature extraction to obtain a plurality of image feature contours in the corresponding environment, and the feature contours of the detection image are filtered according to the image material, thereby improving the reliability of target detection.

[0049] The detection image is subjected to contour recognition, and the image feature data of the complete contour is removed to obtain a defective image, a plurality of defective feature contours are extracted in the defective image, the defective feature contours are repaired according to the plurality of image feature contours, and complete image contour data is obtained; by recognizing the feature contours of the detection image and removing the complete feature contours, the defective feature contours are repaired in combination with the image feature contours of the image material, a plurality of feature contour information of the detection image can be obtained by repairing, so as to detect the corresponding target contour feature in the detection image.

[0050] The plurality of defective feature contours after repairing are filled in the defective image, the complete image contour data is verified, the complete image contour data is corrected based on the verification result, a plurality of feature contours in the detection image are obtained, and a target feature contour is obtained in the plurality of feature contours; in each time of repairing the defective feature contour, the defective image formed based on the detection image is detected and corrected, so that the repair of the defective image can meet the repair requirements of the detection image, and the contour recognition reliability and accuracy of the detection image are improved.

[0051] Figure 1 The embodiment of the application shows a flowchart of a target recognition method based on artificial intelligence in a complex environment, and the recognition method comprises the following steps:

[0052] S11: Extract the detection environment.

[0053] Obtain the detection image, and obtain the image data of the complex environment through a visual component, wherein the visual component can be a camera, a camera, a mobile device, or other electronic devices with a shooting function.

[0054] Input the detection image into a recognition system, obtain the detection environment according to the shooting environment of the detection image, extract the image data of the same environment as the detection environment from the database according to the detection environment, and extract the image data as image material.

[0055] Specifically, according to the detection shooting features of the detection image, the detection environment is extracted, and based on keyword semantic retrieval, image materials corresponding to the detection environment are obtained in the database. The selection of the image materials needs to meet the following rules: the image features of a single image material are less, and the image features of each image material are different from each other. By selecting the image materials with less image features, the image feature extraction step of the image materials can be simplified, and the convenience of feature contour extraction of the image materials can be improved. By setting a plurality of image materials with different image features, the plurality of image materials can include a plurality of different feature contours, so as to meet the accuracy and reliability of the feature contour recognition of the detection image.

[0056] Further, the plurality of image materials can include image data corresponding to different shooting angles of the detection environment, thereby expanding the diversity of the feature contours in the detection environment and improving the reliability of target detection and recognition.

[0057] S12: Extracting a feature contour.

[0058] The Canny algorithm is used for contour extraction of the image materials, different feature contours are extracted in the image materials, and a feature contour database of the detection environment is constructed, so as to compare and identify the contour features of the detection image, to identify each contour feature of the detection image, and to improve the connection reliability of the target feature contour obtained in the detection image.

[0059] The Canny algorithm is a standard algorithm for edge detection, which aims to find an optimal edge detection solution or find the position with the strongest gray intensity change in an image, so as to accurately obtain the image feature contour.

[0060] The obtained image materials are subjected to gray value processing, and the gray value image of the obtained image materials is subjected to Gaussian filtering processing. A Gaussian kernel is used for filtering. Each cell on the gray value image is provided with a pixel value. In this embodiment, a Gaussian kernel with a size of 3*3 is set, so that the Gaussian kernel covers the gray value image. The pixel value of the center cell after Gaussian convolution of the Gaussian kernel is calculated. The pixel value of each cell of the gray value image is calculated by convolution through the Gaussian kernel. The gray value image is filtered by Gaussian kernel convolution, which can remove the noise points of the gray value image and make the image of the gray value image smooth.

[0061] The gray value image after Gaussian filtering processing is subjected to partial derivative processing. The position where the mutation value of the contour edge of the image occurs is confirmed by the partial derivative, so as to improve the accuracy of contour extraction.

[0062] The Laplacian operator is used to perform partial derivation on the gray value image. The Laplacian operator is an isotropic edge extraction operator, which can sharpen lines in any direction in the image and achieve good smoothing effect.

[0063] The calculation formula of the Laplacian operator is as follows:

[0064] ;

[0065] The Laplacian operator is used to perform partial derivation on the gray value image. The Laplacian operator is an isotropic edge extraction operator, which can sharpen lines in any direction in the image and achieve good smoothing effect. is a Laplace transform value, is a coordinate value of the image pixel value.

[0066] The image after Laplace transformation is further sharpened by the non-maximum suppression method. According to the gray value data of the image edge, a plurality of candidate boxes are set, and the plurality of candidate boxes are set at the positions of the image edge. A local maximum value in the image edge gray value is searched, and the pixel values of the candidate box corresponding to the local maximum value and the adjacent candidate box are calculated. The overlap rate of the adjacent candidate box and the maximum value candidate box is calculated, and the overlap rate is compared with a preset threshold. If the overlap rate is greater than the preset threshold, the corresponding candidate box is deleted, and only the maximum value candidate box is retained. Thus, the position of the most prominent change of the image edge is obtained, the image edge is sharpened, and an accurate feature contour image is obtained.

[0067] Further, a plurality of feature contour images are extracted from the image material, and it is detected whether there is a target feature image contour in the plurality of feature contour images. If yes, the corresponding target feature image contour is copied and arranged into target feature contour comparison data. The target feature contour comparison data includes a plurality of target contour feature image data.

[0068] S13: Extracting a complete image feature contour.

[0069] The image feature data of the complete contour is identified and removed, and a defective image is obtained. By obtaining the image contour involved in the detection image, the reliability of the contour recognition of the detection image is improved.

[0070] Specifically, the detection image is preprocessed to obtain a preprocessed image.

[0071] The detection image is processed to obtain a gray value image of the detection image, so as to identify the feature contour of the detection image.

[0072] The preprocessed image is subjected to contour recognition, and the detection image is subjected to contour feature extraction by Gaussian filtering, Laplace transformation and non-maximum suppression method.

[0073] The complete contour image is extracted in the contour image of the detection image by contour extraction, and the complete contour image is filled with the undercolor, so that the complete feature contour in the detection image is in a hollow state, that is, the detection image is formed into a defective image.

[0074] The defective feature contour is obtained around the feature contour in the hollow state in the defective image, and the defective feature contour is connected with the feature contour in the hollow state, that is, the defective feature contour formed by the partially occluded feature contour is obtained, and the defective feature contour is repaired by comparing the plurality of feature contours extracted from the image material, so that the accurate feature contour in the detection image is obtained.

[0075] S14: complete the defective feature contour.

[0076] The plurality of defective feature contours are extracted in the defective image, and the defective feature contour is repaired according to the plurality of image feature contours, so as to obtain the complete image contour data. The plurality of image feature contours are extracted in the image material, and the image feature contour matched successfully is repaired for the corresponding defective image feature contour by comparing and matching the image feature contour with the plurality of defective feature contours in the defective image, so as to repair the defective image and obtain the feature contour image in the defective image.

[0077] The defective feature contour is extracted, and the approximate image feature contour is obtained by comparing the defective feature contour in the plurality of image feature contours. The defective feature contour of the edge of the hollow area of the defective image is extracted, and the plurality of image feature contours of the image material are compared with the defective feature contour. When the defective feature contour matches one or more of the plurality of image feature contours, the image feature contour matched successfully is arranged as the approximate image feature contour.

[0078] The deflection angle of the approximate image feature contour is adjusted based on the defective feature contour, the defective feature contour is completed based on the approximate image feature contour, the deflection angle of the approximate image feature contour is adjusted, so that the approximate image feature contour completely overlaps the defective feature contour, thereby realizing the completion of the defective feature contour.

[0079] Based on the image feature contour matched successfully with the defective feature contour, a plurality of complete image contour data are obtained by completing the defective feature contour.

[0080] S15: correct the feature contour.

[0081] Specifically, the completed image contour is filled in the incomplete image, the completed image contour is verified based on the hollow region in the incomplete image, it is detected whether the completed part of the completed image contour is completely contained in the hollow region of the incomplete image, if yes, the completed image contour is marked as a repaired image contour, if not, the completed image contour is marked as an error contour, and the completed image contour is rejected.

[0082] If the completed image contour is filled in the incomplete image, and the completed image contour can simultaneously satisfy the matching requirements of two incomplete characteristic contours, that is, the completed image contour forms two incomplete characteristic contours based on the hollow region, the completed image contour is marked as a repaired image contour, and the remaining completed image contours of the corresponding two incomplete characteristic contours are rejected.

[0083] Further, the completed image contour is detected and corrected based on the hollow region of the incomplete image, and a repaired image contour is obtained which can satisfy the repair requirements of the incomplete image.

[0084] Further, after the incomplete characteristic contour corresponding to the hollow region in the incomplete image is repaired and completed, the repaired incomplete characteristic contour is filled with a base color, thereby expanding the hollow region of the incomplete image, obtaining a new incomplete image, and performing incomplete characteristic contour repair and completion on the new incomplete image until all the completed image characteristic contours of the detection image are obtained.

[0085] Specifically, the repaired image contour is matched with the target characteristic contour data, if the repaired image contour matches the target characteristic contour data, the corresponding repaired image contour data is marked as target characteristic contour data.

[0086] S16: Obtain a target characteristic contour.

[0087] Specifically, a plurality of image characteristic contours are obtained in the detection image, the image contour features of the image material extraction are combined to repair and complete the image contour features in the detection image, and all the completed image characteristic contours and incomplete characteristic contours in the detection image are obtained.

[0088] A plurality of characteristic contours of the corrected detection image are extracted, and a plurality of characteristic contours corresponding to the target characteristic contour data are extracted and matched, and are marked as target units.

[0089] The completed image characteristic contours in the detection image are compared one by one with the target characteristic contour data, and the image data similar to the target characteristic contour data in the completed image characteristic contours is obtained and marked as first-class target pending image data.

[0090] The incomplete and defective feature contour is matched with the target feature contour matching data, if there is a matched defective feature contour data, the defective feature contour data is marked as the second type of target pending image data.

[0091] According to the first type of target pending image data and the second type of target pending image data, the corresponding position of the contour feature image in the detection image is marked, and the target feature contour is output.

[0092] The embodiment of the application provides a target recognition method based on artificial intelligence in a complex environment, which recognizes the detection environment of the detection image, obtains image materials corresponding to the detection environment, identifies and extracts the contour features in the detection image in combination with the image materials, repairs and completes the defective feature contour after obtaining the defective image, and obtains a plurality of contour image features of the detection image, improves the accuracy of detection and identification of the contour features in the detection image, and improves the accuracy and convenience of recognizing and outputting the target contour in the detection image.

[0093] Embodiment two:

[0094] Figure 2 The embodiment of the application shows a schematic diagram of a target recognition device based on artificial intelligence in a complex environment, the target recognition device comprises:

[0095] The image material acquisition module 10 is used for confirming the detection environment according to the detection image, and extracting image materials corresponding to the detection environment in the database based on the detection environment.

[0096] The detection image is input into the recognition system, the detection environment is obtained according to the shooting environment of the detection image, the image data of the same environment as the detection environment is extracted in the database according to the detection environment, and the image data is extracted as image materials.

[0097] Specifically, according to the detection and shooting features of the detection image, the detection environment is extracted, a plurality of image materials corresponding to the detection environment are obtained in the database based on keyword semantic retrieval, and the selection of the image materials needs to meet the following rules: the image features of a single image material are less, and the image features of each image material are different from each other. By selecting the image material with less image features, the image feature extraction step of the image material can be simplified, and the convenience of feature contour extraction of the image material is improved; by setting a plurality of image materials with different image features, a plurality of different feature contours can be covered by the plurality of image materials, so as to meet the accuracy and reliability of the image material to the contour feature recognition of the detection image.

[0098] The feature extraction module 20 is used for feature extraction of the image material, and a plurality of image feature contours in the corresponding environment are obtained.

[0099] The Canny algorithm is used to extract the contour of the image material, different feature contours are extracted in the image material, a feature contour database of the detection environment is constructed, the contour features of the detection image are compared and identified, each contour feature of the detection image is identified, and the connection reliability of the target feature contour obtained in the detection image is improved.

[0100] The feature removal module 30 is used for contour identification of the detection image, identification and removal of the image feature data of the complete contour, and acquisition of the incomplete image.

[0101] The contour of the detection image is identified, the image feature data of the complete contour is identified and removed, and the incomplete image is acquired, the image contour involved in the detection image is acquired, and the reliability of the contour identification of the detection image is improved.

[0102] Specifically, the detection image is preprocessed to obtain a preprocessed image;

[0103] The detection image is subjected to gray value processing to obtain a gray value image of the detection image, so as to identify the feature contour of the detection image.

[0104] The preprocessed image is subjected to contour identification, and the detection image is subjected to contour feature extraction through Gaussian filtering, Laplace transformation and non-maximum suppression.

[0105] The complete contour image is extracted in the contour image of the detection image through contour extraction, and the complete contour image is subjected to bottom color filling, so that the complete feature contour in the detection image is in a hollow state, that is, the detection image is formed into an incomplete image.

[0106] The incomplete feature contour is acquired around the feature contour in the hollow state in the incomplete image, and the incomplete contour feature is connected with the feature contour in the hollow state, that is, the incomplete feature contour formed by the partially occluded feature contour is acquired, the incomplete feature contour is repaired according to the comparison of the plurality of feature contours extracted from the image material, so as to acquire the accurate feature contour in the detection image.

[0107] The feature repair module 40 is used for extracting a plurality of incomplete feature contours in the incomplete image, repairing the incomplete feature contour according to the plurality of image feature contours, and obtaining complete image contour data.

[0108] The several image feature contours are extracted from the image material, and the approximate image feature contour is obtained by comparing the several image feature contours with the several incomplete feature contours in the incomplete image. The incomplete feature contour is extracted, and the approximate image feature contour is obtained by comparing the incomplete feature contour with the several image feature contours. The approximate image feature contour is adjusted based on the incomplete feature contour, the incomplete feature contour is completed based on the approximate image feature contour, the deflection angle of the approximate image feature contour is adjusted, the approximate image feature contour is completely overlapped with the incomplete feature contour, and the completion of the incomplete feature contour is realized.

[0109] The several image feature contours are extracted from the image material, and the approximate image feature contour is obtained by comparing the several image feature contours with the several incomplete feature contours in the incomplete image. The incomplete feature contour is extracted, and the approximate image feature contour is obtained by comparing the incomplete feature contour with the several image feature contours. The approximate image feature contour is adjusted based on the incomplete feature contour, the incomplete feature contour is completed based on the approximate image feature contour, the deflection angle of the approximate image feature contour is adjusted, the approximate image feature contour is completely overlapped with the incomplete feature contour, and the completion of the incomplete feature contour is realized.

[0110] The several image feature contours are extracted from the image material, and the approximate image feature contour is obtained by comparing the several image feature contours with the several incomplete feature contours in the incomplete image. The incomplete feature contour is extracted, and the approximate image feature contour is obtained by comparing the incomplete feature contour with the several image feature contours. The approximate image feature contour is adjusted based on the incomplete feature contour, the incomplete feature contour is completed based on the approximate image feature contour, the deflection angle of the approximate image feature contour is adjusted, the approximate image feature contour is completely overlapped with the incomplete feature contour, and the completion of the incomplete feature contour is realized.

[0111] The several image feature contours are extracted from the image material, and the approximate image feature contour is obtained by comparing the several image feature contours with the several incomplete feature contours in the incomplete image. The incomplete feature contour is extracted, and the approximate image feature contour is obtained by comparing the incomplete feature contour with the several image feature contours. The approximate image feature contour is adjusted based on the incomplete feature contour, the incomplete feature contour is completed based on the approximate image feature contour, the deflection angle of the approximate image feature contour is adjusted, the approximate image feature contour is completely overlapped with the incomplete feature contour, and the completion of the incomplete feature contour is realized.

[0112] The verification module 50 is used for verifying the several incomplete feature contours filled in the incomplete image after repair, and verifying the completed image contour data.

[0113] Specifically, the completed image contour is filled in the incomplete image, the completed image contour is verified based on the hollow area in the incomplete image, it is detected whether the completed part of the completed image contour is completely contained in the hollow area of the incomplete image, if yes, the completed image contour is marked as a repaired image contour, if not, the completed image contour is marked as an error contour, and the completed image contour is removed.

[0114] If the completed image contour is filled in the incomplete image, and the completed image contour can simultaneously meet the matching requirements of two incomplete feature contours, that is, the completed image contour forms two incomplete feature contours based on the hollow area, the completed image contour is marked as a repaired image contour, and the remaining completed image contours of the corresponding two incomplete feature contours are removed.

[0115] The correction module 60 is configured to correct the complete image contour data based on the verification result, acquire a plurality of feature contours in the detection image, and acquire a target feature contour from the plurality of feature contours.

[0116] After the incomplete feature contour corresponding to the hollow region in the incomplete image is repaired and filled with the base color, the hollow region of the incomplete image is expanded, a new incomplete image is obtained, and the incomplete feature contour of the new incomplete image is repaired and filled, until all the complete image feature contours of the detection image are acquired.

[0117] Specifically, the repaired image contour is matched with the target feature contour reference data, and if the repaired image contour matches the target feature contour reference data, the corresponding repaired image contour data is marked as target feature contour data.

[0118] The plurality of feature contours of the corrected detection image are extracted, and the feature contour corresponding to the target feature contour reference data is extracted from the plurality of feature contours and marked as a target unit.

[0119] Specifically, a plurality of image feature contours are acquired in the detection image, the image contour features in the detection image are repaired and completed in combination with the image contour features extracted from the image material, and all the complete image feature contours and incomplete feature contours in the detection image are acquired.

[0120] The complete image feature contours in the detection image are compared with the target feature contour reference data one by one, and the image data similar to the target feature contour reference data in the complete image feature contours is acquired and marked as first-class target pending image data.

[0121] The incomplete feature contours are matched with the target feature contour reference data, and if there is matched incomplete feature contour data, the incomplete feature contour data is marked as second-class target pending image data.

[0122] The contour feature image at the corresponding position in the detection image is marked according to the first-class target pending image data and the second-class target pending image data, and the target feature contour is output.

[0123] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiments can be completed by programs instructing related hardware, and the programs can be stored in a computer readable storage medium, which can include read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

[0124] In addition, the above detailed description of the embodiments of the present application is made, and the principles and implementation manners of the present application are described by using specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will be changed, and the above description should not be understood as a limitation of the present application.

Claims

1.A method for target recognition in a complex environment based on artificial intelligence, characterized in that, The target recognition method comprises: According to the detection image, the detection environment is confirmed, and the image material corresponding to the environment is extracted from the database based on the detection environment; Feature extraction is performed on the image material to obtain a plurality of image feature contours in the corresponding environment; Contour recognition is performed on the detection image, and the image feature data of the complete contour is recognized and removed to obtain a residual image; A plurality of residual feature contours are extracted from the residual image, the residual feature contours are repaired according to the plurality of image feature contours, and complete image contour data is obtained; The plurality of residual feature contours after repair are filled in the residual image, and the complete image contour data is verified through the residual image after filling; Based on the verification result, the complete image contour data is corrected, a plurality of feature contours in the detection image are obtained, and a target feature contour is obtained in the plurality of feature contours; The plurality of feature contours of the corrected detection image are extracted, and the feature contours matched with the target feature contour comparison data in the plurality of feature contours are extracted and marked as target units; The complete image contour data and the incomplete residual feature contour data of the corrected detection image are obtained, and first and second target pending image data are obtained after comparison through the target feature contour comparison data; According to the first and second target pending image data, the contour feature image of the corresponding position in the detection image is marked, and the target feature contour is output. According to the detection image, the detection environment is confirmed, and the image material corresponding to the environment is extracted from the database based on the detection environment; 2.The method of identifying a target in a complex environment based on artificial intelligence according to claim 1, wherein, The detection image is input into the recognition system, the detection environment is obtained according to the shooting environment of the detection image, and the detection environment is obtained according to the shooting environment of the detection image; The image data of the same environment as the detection environment is extracted from the database according to the detection environment, and the image data is extracted as image material. The feature extraction is performed on the image material to obtain a plurality of image feature contours in the corresponding environment; 3.The method of claim 1, wherein the AI-based target recognition method in a complex environment is characterized by, The Canny algorithm is used for contour extraction of the image material, different feature contours are extracted in the image material, and a feature contour database of the detection environment is constructed. The Canny algorithm is used for contour extraction of the image material, different feature contours are extracted in the image material, and a feature contour database of the detection environment is constructed. 4.The method of claim 3, wherein the AI-based target recognition method in a complex environment is characterized by, The Canny algorithm is used for contour extraction of the image material, different feature contours are extracted in the image material, and a feature contour database of the detection environment is constructed. The detection image is preprocessed to obtain a preprocessed image; The preprocessed image is subjected to contour recognition, and the complete feature contour is marked and extracted; 5.The method of identifying a target in a complex environment based on artificial intelligence according to claim 1, wherein, ​ ​ ​ The complete feature contour part is filled with a bottom color in the preprocessed image to obtain a defective image. 6.The method of identifying a target in a complex environment based on artificial intelligence according to claim 1, wherein, The several defective feature contours are extracted in the defective image, and the defective feature contours are repaired according to the several image feature contours to obtain the complete image contour data. The defective feature contour is extracted, and an approximate image feature contour is obtained by comparison of the defective feature contour in the several image feature contours; The deflection angle of the approximate image feature is adjusted based on the defective feature contour, and the defective feature contour is completed based on the approximate image feature contour. 7.The method of identifying a target in a complex environment based on artificial intelligence according to claim 6, wherein, The defective feature contour is extracted, and an approximate image feature contour is obtained by comparison of the defective feature contour in the several image feature contours; The defective feature contour is obtained around the feature contour in the several hollow states in the defective image, and the defective feature contour is connected with the feature contour in the hollow state. 8.The method of identifying a target in a complex environment based on artificial intelligence according to claim 1, wherein, The several defective feature contours after repair are filled in the defective image, and the complete image contour data is verified through the filled defective image. The complete image contour is filled in the defective image, and the complete image contour is verified based on the hollow area in the defective image to detect whether the complete part of the complete image contour is completely contained in the hollow area of the defective image; If yes, the complete image contour is marked as a repaired image contour, and if no, the complete image contour is marked as an error contour, and the complete image contour is removed. 9.A device for target recognition in a complex environment based on artificial intelligence, characterized in that, The recognition device comprises: An image material acquisition module is configured to confirm a detection environment according to a detection image, and extract image materials corresponding to the environment in a database based on the detection environment; A feature extraction module is configured to extract features of the image materials to obtain several image feature contours in the corresponding environment; A feature removal module is configured to recognize contours of the detection image, remove image feature data of complete contours, and obtain a defective image; A feature repair module is configured to extract several defective feature contours in the defective image, repair the defective feature contours according to the several image feature contours, and obtain complete image contour data; A verification module is configured to fill the several defective feature contours after repair in the defective image, and verify the complete image contour data through the filled defective image; A correction module is configured to correct the complete image contour data based on the verification result, obtain several feature contours in the detection image, and obtain a target feature contour in the several feature contours; Several feature contours of the corrected detection image are extracted, and a feature contour corresponding to the target feature contour in the target feature contour comparison data is extracted from the several feature contours and marked as a target unit; Complete image contour data and incomplete defective feature contour data of the corrected detection image are obtained, and first-class target pending image data and second-class target pending image data are obtained by comparison of the target feature contour comparison data; The contour feature image of the corresponding position is marked in the detection image according to the first-class target pending image data and the second-class target pending image data, and the target feature contour is output.

Images