Automatic cleaning method for target infrared original gray feature image data

Abstract

The invention discloses an automatic cleaning method for target infrared original gray feature image data. The method sequentially comprises the steps of extraction of image data and related parameter data, image mapping, image target detection and recognition, target feature extraction and differentiation judgment, image data association and output and storage in a specified data format. Aiming at large-scale and high-capacity images generated when photoelectric equipment performs infrared characteristic measurement on a target, the problem of difficulty in automatic cleaning of large-scale and high-capacity target infrared characteristic data is solved.

Description

Technical Field

[0001] The invention relates to the field of image processing, and in particular to an automatic cleaning method for screening and extracting high-value data of target infrared original grayscale feature video stream image data collected by a photoelectric device. Background Art

[0002] When measuring the infrared features of a target, a large amount of infrared original grayscale feature image data will be continuously collected. Only images that contain the target and can characterize the feature differences have a higher retention and analysis value.

[0003] Therefore, in the process of collecting target infrared feature data, the original infrared grayscale feature image needs to go through a data cleaning process of viewing and screening. The viewing and screening of large-scale and large-capacity images will increase the workload of equipment operators, cause fatigue, and affect work quality.

[0004] Therefore, it is necessary to rely on technologies such as artificial intelligence to realize the automatic cleaning function of data. Through highly intelligent, automated and precise software algorithm tools, powerful technical means are provided for collection tasks. Summary of the invention

[0005] In view of the large-scale and large-volume images generated when optoelectronic equipment measures the infrared characteristics of the target, in order to solve the data cleaning problem that it is difficult for equipment operators to view, screen and retain high-value analysis data for a long time and efficiently, the present invention provides an automatic cleaning method for the original grayscale characteristic image data of the target infrared, which realizes a highly intelligent, automated and precise automatic cleaning function of the target infrared characteristic data, and provides technical support for realizing efficient data collection.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for automatically cleaning target infrared original grayscale feature image data, comprising the following steps:

[0007] S1, extraction of image data and related parameter data: according to the infrared image Cameralink interface transmission protocol, extract the 14-bit target infrared original grayscale feature image data I representing the grayscale img14 , extract the acquisition parameter information according to the Aurora high-speed serial communication transmission protocol, and extract the latest infrared calibration data packet according to the infrared calibration data Cameralink interface transmission protocol;

[0008] S2, image mapping: the original grayscale feature image data of the target infrared image I img14 Perform grayscale transformation and map each pixel in the image to 8-bit space to obtain an 8-bit image I img8 ;

[0009] S3, image target detection and recognition: Through the pre-trained infrared surface ship target intelligent detection and recognition model based on Yolov5-m, the 8-bit image I mapped to the image img8 Perform model inference to determine whether there is a surface ship target in the image. If there is no surface ship target, extract a new image for the next round of processing;

[0010] S4, target feature extraction and differential contrast discrimination: For the current image with the target, from the 8-bit image I img8 I img8 Extract the target area sub-image I sub ,Use SIFT operator to extract the target area sub-image I sub feature points, and obtain the feature point set M sub-SIFT ; For the previous image retained after cleaning, from its corresponding 8-bit image I′ img8 Extract the corresponding target area sub-image I′ sub , use SIFT operator to extract the corresponding target area sub-image I′ sub feature points, and obtain the corresponding feature point set N sub-SIFT ; for the set M sub-SIFT and N sub-SIFT The feature points are encoded and matched to obtain the number of matched feature points s; when s is relative to the set M sub-SIFT When the total proportion of ≥80%, it means that the current image is not significantly different from the previous cleaned image, and the current image is not retained;

[0011] S5, image data association and output: for the image with significant difference characteristics completed by differential comparison, associate the parameter information of the acquired image with the current original grayscale characteristic image; associate the latest infrared calibration data with the current original grayscale characteristic image; set the field and attribute labels, and save it in the specified data format.

[0012] Furthermore, the collection parameters in step S1 include a transmission delimiter and a main field type.

[0013] Furthermore, when the 14-bit representation is performed in step S2, the maximum pixel value is P max , minimum pixel value P min , original grayscale feature image data I img14 Each pixel P 14 The mapping formula for (x,y) is (x, y) is the image coordinate value of the pixel point, Indicates rounding down.

[0014] Furthermore, the parameter information of the image collected in step S5 includes time, space, ambient temperature, target type and image coordinates.

[0015] The beneficial effects of the present invention are as follows: the present invention solves the difficulty of automatic cleaning of large-scale and large-capacity target infrared characteristic data for large-scale and large-capacity images generated when photoelectric equipment measures infrared characteristics of targets. Compared with relying on equipment operators to view, screen and retain high-value analysis data, the workload of personnel operating equipment is greatly reduced. Through highly intelligent, automated and precise automatic cleaning means, efficient collection of target infrared characteristics is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The figure is a flow chart of the cleaning method of the present invention. DETAILED DESCRIPTION

[0017] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0018] Reference Figure 1 As shown in FIG. , the present invention discloses an automatic cleaning method for target infrared original grayscale feature image data, which includes the following steps.

[0019] S1, extraction of image data and related parameter data.

[0020] According to the infrared image Cameralink interface transmission protocol, the 14-bit target infrared original grayscale feature image data representing the grayscale is extracted. img14 , according to the Aurora high-speed serial communication transmission protocol, such as the transmission delimiter, the main field type, etc., the acquisition parameter information is extracted, and according to the infrared calibration data Cameralink interface transmission protocol, the latest infrared calibration data packet is extracted.

[0021] S2, image mapping.

[0022] The target infrared original grayscale feature image data I img14 Perform grayscale transformation and map each pixel in the image to 8-bit space to obtain an 8-bit image I img8 .

[0023] For 14-bit representation, the maximum pixel value is P max , minimum pixel value P min , each pixel P in the original grayscale feature image data 14 The mapping formula for (x,y) is (x, y) is the image coordinate value of the pixel point, Indicates rounding down.

[0024] S3, image object detection and recognition.

[0025] Through the pre-trained infrared surface ship target intelligent detection and recognition model based on Yolov5-m, the 8-bit image I img8 Perform model inference to determine whether there is a target in the image. If there is no target, extract a new image for the next round of processing.

[0026] S4, target feature extraction and differential contrast discrimination.

[0027] For the current image with the target, from the 8-bit image I img8 Extract the target area sub-image I sub ,Use SIFT operator to extract the target area sub-image I sub The feature points get the feature point set M sub-SIFT ; For the previous image retained after cleaning, from its corresponding 8-bit image I′ img8 Extract the corresponding target area sub-image I′ sub , use SIFT operator to extract the corresponding target area sub-image I′ sub The feature points get the corresponding feature point set N sub-SIFT ; for the set M sub-SIFT and N sub-SIFT The feature points are encoded and matched to obtain the number of matched feature points s; when s is relative to the set M sub-SIFT When the total proportion of ≥80%, it means that the difference between the current image and the previous image retained after cleaning is not significant, and the current image is not retained.

[0028] S5, image data association and output.

[0029] For images with significant difference characteristics completed by differential comparison, the image acquisition parameter information, such as time, space, ambient temperature, target type, image coordinates, etc., is associated with the current original grayscale feature image; the latest infrared calibration data is associated with the current original grayscale feature image; the fields and attribute labels are set according to the format requirements and saved in the specified data format.

[0030] The above embodiments are only illustrative of the principles and effects of the present invention, as well as some embodiments of its application. For those skilled in the art, several modifications and improvements may be made without departing from the creative concept of the present invention, and all of these belong to the protection scope of the present invention.

Claims

1. A method for automatically cleaning target infrared raw grayscale feature image data, characterized in that: The following steps are included S1, according to the infrared image Cameralink interface transmission protocol, extract the 14-bit target infrared original grayscale feature image data I img14 , extract the acquisition parameter information according to the Aurora high-speed serial communication transmission protocol, and extract the latest infrared calibration data packet according to the infrared calibration data Cameralink interface transmission protocol; S2, the target infrared original grayscale feature image data I img14 Perform grayscale transformation and map each pixel in the image to 8-bit space to obtain an 8-bit image I img8 ; S3, through the pre-trained infrared surface ship target intelligent detection and recognition model based on Yolov5-m, the 8-bit image I img8 Perform model inference to determine whether there is a target in the image. If there is no target, extract a new image for the next round of processing; S4, for the current image with the target, from the 8-bit image I img8 Extract the target area sub-image I sub , use SIFT operator to extract target area sub-image I sub The feature points get the feature point set M sub-SIFT ; For the previous image retained after cleaning, from its corresponding 8-bit image I′ img8 Extract the corresponding target area sub-image I′ sub , use SIFT operator to extract the corresponding target area sub-image I′ sub The feature points get the corresponding feature point set N sub-SIFT ; for the set M sub-SIFT and N sub-SIFT The feature points are encoded and matched to obtain the number of matched feature points s; when s is relative to the set M sub-SIFT When the total number of ≥80%, it means that the difference between the current image and the previous image after cleaning is not significant, and only images with significant difference are retained; S5, for the feature image with significant difference, associating the image acquisition parameter information with the current original grayscale feature image; associating the latest infrared calibration data with the current original grayscale feature image; Set field and attribute labels and save in the specified data format.

2. The automatic cleaning method of target infrared raw grayscale feature image data according to claim 1 is characterized in that: The acquisition parameters in step S1 include a transmission delimiter and a main field type.

3. The automatic cleaning method of target infrared raw grayscale feature image data according to claim 2 is characterized in that: When the 14-bit representation is performed in step S2, the maximum pixel value is P max , minimum pixel value P min , original grayscale feature image data I img14 Each pixel P 14 The mapping formula for (x, y) is (x, y) is the image coordinate value of the pixel point, Indicates rounding down.

4. The automatic cleaning method of target infrared raw grayscale feature image data according to claim 3 is characterized in that: The parameter information of the image collected in step S5 includes time, space, ambient temperature, target type and image coordinates.

Images