A biological image identification system based on telecentric camera matrix and its identification method

Abstract

The present invention discloses a biological image identification system based on a telecentric camera matrix, the system comprising: a camera matrix for collecting a plurality of high-precision original images of a local field of view of a detected organism; an image stitching module for registering and stitching a plurality of high-precision original images of a local field of view into a stitched image of a high-precision global field of view; a genus and species analysis module for analyzing the stitched images and obtaining a conclusion on the classification prediction of the genus and species of the organism. The present invention also discloses a biological image identification method based on a telecentric camera matrix. Compared with the prior art, the system provided by the present invention can take into account the accuracy and imaging range of the imaging of the detected organism, greatly reducing the equipment cost and the difficulty of its equipment assembly, and facilitating operation. The method provided by the present invention realizes automated identification by identifying the genus and species of the detected organism through the steps of image acquisition-registration stitching-feature extraction, analysis and prediction, and has the beneficial effects of high identification efficiency, high accuracy, and saving labor costs.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing, and in particular relates to a method for identifying biological species through images. Background Art

[0002] Using the appearance, morphology, physiological characteristics and other shapes of organisms to determine the genus and species to which the organism belongs is a common method in biological taxonomy. Identifying the biological classification of the inspected organism is of great significance in the fields of customs supervision, scientific investigation, ecological assessment, etc. For a long time, the identification of biological species has faced difficulties such as large identification workload, high professional requirements, and high misjudgment rate. In recent years, with the rapid development of technology in various fields, in order to improve the efficiency of biological species classification and identification, more and more identification personnel are considering combining image processing technology with machine learning methods to realize the automated identification of the inspected biological species.

[0003] When implementing the method of automated identification of biological species that combines image processing technology and machine learning, it is usually necessary to pre-train a suitable and available deep learning model, and then collect image data of the inspected organism. The collected image data is further input into the deep learning model, and the model automatically analyzes it, predicts its classification results, and draws a conclusion on the genus and species to which the inspected organism belongs.

[0004] The above method is widely used in the existing field of biological genus and species identification. Taking the Chinese patent application document with patent publication number "CN111930988A" as an example, the method of quickly identifying spider species using computer image recognition technology disclosed in the application document involves the following steps: (1) establishing a database of suborders, families, genera, and species of known spider groups based on the spider list; (2) taking photos of important features of the spider specimen to be identified; (3) inputting the photos into the computer and comparing and identifying with the established spider image database; (4) identifying whether the spider individual is mature or not, keeping the mature ones and eliminating the immature ones; (5) identifying the gender of the mature spider individual; (6) identifying the suborder to which the mature spider individual belongs; (7) identifying the family to which the mature spider individual belongs; (8) identifying the genus to which the mature spider individual belongs; (9) identifying the species to which the mature spider individual belongs; (10) outputting detailed identification information, and marking the species that cannot be identified as a new species. The above patent application document clearly records the method steps of modeling-photographing-identification, which is a typical application of the combination of image processing technology and machine learning.

[0005] When the method of combining the above-mentioned image processing technology and machine learning to realize automatic identification of biological species is applied in practice, it will face the contradiction between image resolution, image recording range and image processing method: on the one hand, in order to obtain the biological characteristics of the inspected organism as much as possible, in detail and clearly, a sufficiently high image resolution should be guaranteed as much as possible; but on the other hand, the identification of biological species often requires a global image of the inspected organism to comprehensively observe the appearance of the inspected organism, so a sufficiently large image recording range should be guaranteed as much as possible, and the lens field of view should record the entire outline of the inspected organism as much as possible; however, since high precision and wide range cannot be achieved at the same time, if it is necessary to forcibly take into account both a sufficiently high image resolution and a sufficiently large image recording range, not only will it place strict requirements on the shooting equipment, but it will also have high requirements on the data processing capabilities of the image processor, which greatly increases the equipment cost and its implementation difficulty. Summary of the invention

[0006] To solve the above problems, the purpose of the present invention is to provide a biological image identification system based on a telecentric camera matrix. The system constructs a camera matrix and sets an image stitching module. On the one hand, it can take into account the accuracy and imaging range of the image of the inspected organism, thereby improving the accuracy of the image of the inspected organism and obtaining a larger field of view. On the other hand, by combining the camera matrix, the image stitching module and the genus and species analysis module, not only the equipment cost and the difficulty of equipment assembly are greatly reduced, but its simple architecture is also very convenient to operate, and it can very conveniently identify the inspected organism and obtain its corresponding genus and species conclusion.

[0007] Another object of the present invention is to provide a biological image identification method based on a telecentric camera matrix. The method is based on the image data of the inspected organism, with image acquisition-registration splicing-feature extraction, analysis and prediction as the core, to identify the genus and species of the inspected organism, and has the beneficial effects of high identification efficiency, high accuracy, saving labor costs, and easy implementation.

[0008] To achieve the above object, the technical solution of the present invention is as follows:

[0009] A biological image identification system based on a telecentric camera matrix, the system comprising: a camera matrix for collecting a plurality of high-precision original images of local fields of view of a detected organism; an image stitching module for registering and stitching a plurality of high-precision original images of local fields of view into a stitched image of a high-precision global field of view; and a genus and species analysis module for analyzing the stitched images and obtaining a conclusion on the prediction of the genus and species classification of the organism.

[0010] Specifically, the camera matrix includes: a stage for placing the inspected organism; and several telecentric lens cameras for independently shooting high-precision, original images of corresponding parts of the inspected organism; the telecentric lens camera is mounted above the stage facing the stage, and several telecentric lens cameras are mounted in sequence, with the adjacent telecentric lens cameras maintaining partial overlap of the field of view. Due to the telecentric lens, each camera in the camera matrix can record high-resolution, ultra-wide depth of field, and ultra-low distortion images of the inspected organism in its corresponding field of view within a certain object distance range. Such images are more conducive to subsequent image registration, image stitching, and other processing, and can bring more reliable original images to the system, reduce system errors, and provide a strong guarantee for further improving the identification accuracy of the system in the future.

[0011] The stage includes a stage body and a receiving groove. The receiving groove is adapted to the external dimensions of the organism to be examined and is provided on the upper surface of the stage body. The organism to be examined or the specimen slide prepared as required is embedded in the receiving groove and placed firmly on the stage.

[0012] Furthermore, the image stitching module includes:

[0013] An ORB algorithm unit for preliminarily extracting feature point descriptors from two input original images;

[0014] A SURF algorithm unit used to screen and recheck the initially extracted feature point descriptors to obtain the final reliable feature point descriptors;

[0015] A KNN algorithm unit for processing feature point descriptors and matching the feature point descriptors of two original images;

[0016] The RANSAC algorithm unit used to solve and refine the matching process and calculate the overlapping area between the two original images;

[0017] And, a fusion unit for fusing the overlapping areas of the two original images to form a new stitched image.

[0018] Furthermore, the genus and species analysis module includes:

[0019] Pre-training has been completed in advance to store a standard feature database of standard image features of each organism;

[0020] A comparison and analysis unit is used to extract features from spliced ​​images and compare them with a standard feature database to obtain a prediction conclusion on biological genus and species classification.

[0021] The present application also discloses a biological image identification method based on a telecentric camera matrix, which is:

[0022] S1: Image acquisition: The camera matrix takes several original images;

[0023] S2: Image registration and stitching: The image stitching module stitches several original images into a stitched image;

[0024] S3: Image analysis prediction: The spliced ​​image is input into the genus and species analysis module to obtain the genus and species classification prediction conclusion of the inspected organism.

[0025] Furthermore, S1 is specifically:

[0026] S11: placing the organism to be examined on the stage;

[0027] S12: Start the camera matrix, and each telecentric lens camera collects the original image in its corresponding field of view.

[0028] Furthermore, S2 is specifically:

[0029] S21: Image registration: The original images collected by two adjacent telecentric lens cameras are input into the image stitching module, and the overlapping area between the two original images is obtained after being processed in sequence by the ORB algorithm unit, the SURF algorithm unit, the KNN algorithm unit and the RANSAC algorithm unit.

[0030] S22: Image fusion and stitching: After the overlapping areas of the two original images are fused, the two original images are mapped to a new blank image through affine transformation to complete the fusion and stitching of the two original images;

[0031] S23: Repeat S21-S23 until all original images are fused and stitched into a complete large stitched image.

[0032] Furthermore, S21 is specifically:

[0033] S211: Input the original images collected by two adjacent telecentric lens cameras into the ORB algorithm unit to preliminarily extract feature point descriptors; after the original images are input into the ORB algorithm unit, the ORB algorithm unit first uses the FAST operator to detect feature points, and after finding the feature points, the ORB algorithm unit uses the BRIEF operator to calculate the descriptor of each feature point to describe the properties of these feature points. The FAST operator specifically used in the ORB algorithm unit has a very fast speed and good real-time performance, and can quickly complete the calculation of the input original image to obtain the feature point descriptor.

[0034] S212: The feature point descriptors initially extracted from the two original images are screened and re-examined by the SURF algorithm unit to obtain the final reliable feature point descriptors; the input original image is processed by the SURF algorithm, and its processing process will have very excellent rotation robustness, fuzzy robustness and scale invariant robustness. It is applied to the scene of biological species image identification with many uncontrollable factors interfering with the image, which will be able to significantly reduce the impact of accidental errors on the prediction conclusion and improve the accuracy of the system. However, due to its own limitations, it will take a long time to align the images if the SURF algorithm is fully used. In the method provided by the present application, the ORB algorithm unit is first used to process the original image to obtain the feature point descriptor of the image, and then the SURF algorithm unit is used to review the above feature point descriptors. The feature point descriptors screened and reviewed will be more reliable. In the whole feature point descriptor extraction process, the system can simultaneously obtain the superior processing speed and good real-time performance during the execution of the ORB algorithm unit, and can also harvest the excellent robustness during the execution of the SURF algorithm, and obtain fast and good image processing results.

[0035] S213: The KNN algorithm unit constructs a transformation matrix between the image sequences for the matching point pairs, and matches the reliable feature point descriptors extracted from the two original images;

[0036] S214: The RANSAC algorithm unit solves and refines the transformation matrix between the image sequences, and calculates the overlapping area between the two original images.

[0037] Furthermore, S3 is specifically:

[0038] S31: A feature database is pre-established in the genus and species analysis module;

[0039] S32: inputting standard images of various organisms into the feature database for pre-training, extracting corresponding biological features from the standard images of various organisms and storing them as standard image features, and updating the feature database to a standard feature database;

[0040] S33: inputting the spliced ​​image obtained in S2 into a comparison and analysis unit, and the comparison and analysis unit extracts the corresponding biological image features from the spliced ​​image;

[0041] S34: The comparison and analysis unit compares the extracted biological image features with the standard feature database to obtain the final biological genus and species classification prediction conclusion. It should be emphasized that before the system is put into use, it is realized to establish a feature database in the genus and species analysis module and put it into the training atlas to complete pre-training, and then automatically identify biological genus and species after it is put into use to obtain the genus and species classification prediction result. Such operation is a prior art and is not the core of protection of this application. This application intends to request protection for its combination with S1 image acquisition and S2: image registration and splicing to process the spliced ​​image. As for the specific form in which the feature database is established, the way in which the biological image features in the spliced ​​image are extracted, and the way in which the comparison is drawn, etc., those skilled in the art can refer to the existing technical information and choose the appropriate method according to the specific application situation to achieve it, which will not be repeated here.

[0042] This application has the following distinguishing technical effects:

[0043] Good image acquisition effect: In the system provided in the present application, by combining multiple telecentric lens cameras to form a matrix, the high precision of a single camera and the wide field of view of the camera group are taken into account, and good imaging effects can be achieved.

[0044] Good image processing effect: The image registration and stitching method provided in this application combines the ORB algorithm and the SURF algorithm, and achieves excellent robustness while achieving superior feature point descriptor extraction speed.

[0045] The system has good identification effect: on the one hand, the high-precision and wide-field-of-view original images can collect a large amount of detailed and rich image information; on the other hand, the fast and robust image registration and stitching method will accurately extract reliable feature point descriptors from the original images. Finally, the comprehensive, detailed and standardized biometric database can automatically calculate the feature point descriptors and automatically derive classification predictions. The whole process is easy to operate, highly automated and has good conclusion reliability, which greatly reduces the labor cost and achieves good biological identification effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a system block diagram of a biological image identification system based on a telecentric camera matrix implemented in a specific implementation manner.

[0047] Figure 2 It is a schematic diagram of the specific structure of the camera matrix in the biological image identification system based on the telecentric camera matrix implemented in the specific implementation manner.

[0048] Figure 3 It is an overall workflow diagram of the biological image identification method based on the telecentric camera matrix implemented in the specific implementation manner.

[0049] Figure 4 It is a specific flow chart of step S21: image registration in the biological image identification method based on the telecentric camera matrix implemented in the specific implementation manner. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0051] To achieve the above object, the technical solution of the present invention is as follows:

[0052] See also Figure 1-4 .

[0053] In this specific embodiment, a biological image identification system based on a telecentric camera matrix is ​​provided, and the system includes: a camera matrix 1 for collecting a plurality of high-precision original images of the local field of view of the inspected organism; an image stitching module 2 for aligning and stitching a plurality of high-precision original images of the local field of view into a stitched image of the high-precision global field of view; and a genus and species analysis module 3 for analyzing the stitched images and obtaining a conclusion on the prediction of the biological genus and species classification.

[0054] Specifically, the camera matrix 1 includes: a stage 11 for placing the inspected organism; and a plurality of telecentric lens cameras 12 for independently taking high-precision original images of corresponding parts of the inspected organism; the telecentric lens cameras 12 are mounted above the stage 11 facing the stage 11, and a plurality of telecentric lens cameras 12 are mounted in sequence, with adjacent telecentric lens cameras 12 maintaining partial overlap of the field of view.

[0055] In this specific embodiment, the stage 11 includes a stage body 111 and a receiving groove 112; the receiving groove 112 is adapted to the external dimensions of the organism to be examined, and the receiving groove 112 is provided on the upper surface of the stage body 111. The organism to be examined itself or the specimen slide prepared in accordance with the regulations is embedded in the receiving groove 112 and stably placed on the stage 11.

[0056] In this specific implementation, the image stitching module 2 includes:

[0057] An ORB algorithm unit 21 for preliminarily extracting feature point descriptors from two input original images;

[0058] A SURF algorithm unit 22 for screening and rechecking the initially extracted feature point descriptors to obtain final reliable feature point descriptors;

[0059] A KNN algorithm unit 23 for processing feature point descriptors and matching the feature point descriptors of two original images;

[0060] A RANSAC algorithm unit 24 for solving and refining the matching process and calculating the overlapping area between the two original images;

[0061] And, a fusion unit 25 for fusing the overlapping areas of the two original images to form a new stitched image.

[0062] In this specific implementation, the genus and species analysis module 3 includes:

[0063] Pre-training has been completed in advance to store a standard feature database 31 of standard image features of each organism;

[0064] The comparison and analysis unit 32 is used for extracting features from the spliced ​​image and comparing it with a standard feature database to obtain a prediction conclusion of biological genus and species classification.

[0065] In this specific embodiment, a biological image identification method based on a telecentric camera matrix is ​​also disclosed, and the method is:

[0066] S1: Image acquisition: Camera matrix 1 takes several original images;

[0067] S2: Image registration and stitching: The image stitching module 2 stitches several original images into a stitched image;

[0068] S3: Image analysis prediction: The spliced ​​image is input into the genus and species analysis module 3 to obtain the genus and species classification prediction conclusion of the inspected organism.

[0069] In this specific implementation, further, S1 is specifically:

[0070] S11: placing the organism to be examined on the stage 11;

[0071] S12: Start the camera matrix 1, and each telecentric lens camera 12 collects the original image in its corresponding field of view.

[0072] In this specific implementation, further, S2 is specifically:

[0073] S21: Image registration: The original images collected by two adjacent telecentric lens cameras are input into the image stitching module, and the overlapping area between the two original images is obtained after being processed in sequence by the ORB algorithm unit, the SURF algorithm unit, the KNN algorithm unit and the RANSAC algorithm unit.

[0074] S22: Image fusion and stitching: After the overlapping areas of the two original images are fused, the two original images are mapped to a new blank image through affine transformation to complete the fusion and stitching of the two original images;

[0075] S23: Repeat S21-S23 until all original images are fused and stitched into a complete large stitched image.

[0076] In this specific implementation, further, S21 is specifically:

[0077] S211: Input the original images captured by two adjacent telecentric lens cameras 12 into the ORB algorithm unit 21 to preliminarily extract feature point descriptors; after the original images are input into the ORB algorithm unit 21, the ORB algorithm unit 21 first uses the FAST operator to detect feature points, and after finding the feature points, in order to describe the attributes of these feature points, the ORB algorithm unit 21 calculates the descriptor of each feature point using the BRIEF operator.

[0078] S212: The SURF algorithm unit 22 screens and rechecks the feature point descriptors initially extracted from the two original images to obtain final reliable feature point descriptors;

[0079] S213: The KNN algorithm unit 23 constructs a transformation matrix between the image sequences for the matching point pairs, and matches the reliable feature point descriptors extracted from the two original images;

[0080] S214: The RANSAC algorithm unit 24 solves and refines the transformation matrix between the image sequences, and calculates the overlapping area between the two original images.

[0081] In this specific implementation, further, S3 is specifically:

[0082] S31: A feature database is pre-established in the genus and species analysis module 3;

[0083] S32: inputting standard images of various organisms into the feature database for pre-training, extracting corresponding biological features from the standard images of various organisms and storing them as standard image features, and updating the feature database to the standard feature database 31;

[0084] S33: inputting the spliced ​​image obtained in S2 into the comparison and analysis unit 32, and the comparison and analysis unit 32 extracts the corresponding biological image features from the spliced ​​image;

[0085] S34: The comparison and analysis unit 32 compares the extracted biological image features with the standard feature database 31 to obtain the final biological genus and species classification prediction conclusion.

[0086] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A biological image identification system based on a telecentric camera matrix, characterized in that: The system includes: A camera matrix for acquiring a plurality of high-precision original images of a local field of view of the inspected organism; An image stitching module for registering and stitching a number of high-precision local field-of-view original images into a stitched image of a high-precision global field of view; Genus and species analysis module for analyzing spliced ​​images and drawing conclusions on biological genus and species classification prediction; The camera matrix includes: A stage for placing the organisms to be examined; Several telecentric lens cameras for independently taking high-precision original images of corresponding parts of the inspected organism; The telecentric lens camera is mounted above the stage facing the stage, and a plurality of the telecentric lens cameras are mounted in sequence, with the fields of view of adjacent telecentric lens cameras partially overlapping; The stage comprises a stage body and a receiving groove; the receiving groove is adapted to the external dimensions of the organism to be inspected, and the receiving groove is arranged on the upper surface of the stage body; The image stitching module includes: An ORB algorithm unit for preliminarily extracting feature point descriptors from two input original images; A SURF algorithm unit used to screen and recheck the initially extracted feature point descriptors to obtain the final feature point descriptors; A KNN algorithm unit for processing feature point descriptors and matching the final feature point descriptors of two original images; The RANSAC algorithm unit used to solve and refine the matching process and calculate the overlapping area between the two original images; And, a fusion unit for fusing the overlapping areas of the two original images to form a new stitched image.

2. The biological image identification system based on a telecentric camera matrix as claimed in claim 1, characterized in that: The genus and species analysis module includes: Pre-training has been completed in advance to store a standard feature database of standard image features of each organism; A comparison and analysis unit is used to extract features from spliced ​​images and compare them with a standard feature database to obtain a prediction conclusion on biological genus and species classification.

3. A biological image identification method based on a telecentric camera matrix, characterized in that: The method is applied in a biological image identification system based on a telecentric camera matrix as described in claim 1, and the method is: S1: Image acquisition: the camera matrix takes several original images; S2: Image registration and stitching: the image stitching module stitches several original images into a stitched image; S3: Image analysis prediction: The spliced ​​image is input into the genus and species analysis module to obtain the genus and species classification prediction conclusion of the inspected organism.

4. The biological image identification method based on a telecentric camera matrix as claimed in claim 3, characterized in that: The S1 is specifically: S11: placing the organism to be tested on the stage; S12: Start the camera matrix, and each of the telecentric lens cameras collects an original image in its corresponding field of view.

5. The biological image identification method based on a telecentric camera matrix as claimed in claim 4, characterized in that: The S2 is specifically: S21: Image registration: inputting the original images collected by the two adjacent telecentric lens cameras into the image stitching module, and obtaining the overlapping area between the two original images after being processed in sequence by the ORB algorithm unit, the SURF algorithm unit, the KNN algorithm unit and the RANSAC algorithm unit; S22: Image fusion and stitching: After the overlapping areas of the two original images are fused, the two original images are mapped to a new blank image through affine transformation to complete the fusion and stitching of the two original images; S23: Repeat S21-S23 until all original images are fused and stitched into a complete large stitched image.

6. The biological image identification method based on a telecentric camera matrix as claimed in claim 5, characterized in that: The S21 is specifically: S211: inputting the original images collected by two adjacent telecentric lens cameras into the ORB algorithm unit to preliminarily extract feature point descriptors; S212: The SURF algorithm unit screens and rechecks the feature point descriptors initially extracted from the two original images to obtain final feature point descriptors; S213: The KNN algorithm unit constructs a transformation matrix between the image sequences for the matching point pairs, and matches the final feature point descriptors extracted from the two original images; S214: The RANSAC algorithm unit solves and refines the transformation matrix between the image sequences, and calculates the overlapping area between the two original images.

7. The biological image identification method based on a telecentric camera matrix as claimed in claim 6, characterized in that: The S3 is specifically: S31: A feature database is pre-established in the genus and species analysis module; S32: inputting standard images of various organisms into the feature database for pre-training, extracting corresponding biological features from the standard images of various organisms and storing them as standard image features, and updating the feature database to a standard feature database; S33: inputting the spliced ​​image obtained in S2 into a comparison and analysis unit, wherein the comparison and analysis unit extracts the corresponding biological image features from the spliced ​​image; S34: The comparison and analysis unit compares the extracted biological image features with the standard feature database to obtain the final biological genus and species classification prediction conclusion.

Images