CN109978962B

CN109978962B - Low-contrast indicating value image intelligent identification method for darkroom illuminometer calibration

Info

Publication number: CN109978962B
Application number: CN201910280450.3A
Authority: CN
Inventors: 南瑞亭
Original assignee: Guangzhou Communications Senior Technical School (guangzhou Communications Technician Institute)
Current assignee: Guangzhou Communications Senior Technical School (guangzhou Communications Technician Institute)
Priority date: 2019-04-09
Filing date: 2019-04-09
Publication date: 2022-05-17
Anticipated expiration: 2039-04-09
Also published as: CN109978962A

Abstract

The invention discloses a darkroom illuminometer calibration-oriented intelligent identification method for low-contrast indicating value images, which comprises the following steps: respectively performing data enhancement on the high-contrast image and the low-contrast image to complete model training of indication value image prediction of the CNN target detection frame under different contrasts; calculating the optimal recognition contrast according to the prediction probability and the contrast of the correct prediction sample in the training process; in the actual prediction process, the image contrast is enhanced to the optimal recognition contrast, and the recognition accuracy is further improved. The method realizes the identification of the illuminance indicating value in the low-contrast environment and has good applicability.

Description

Low-contrast indicating value image intelligent identification method for darkroom illuminometer calibration

Technical Field

The invention relates to the field of machine vision, in particular to an intelligent identification method for illuminance meter indication values in a low-illumination environment.

Background

In the illuminometer calibration industry, since the illuminometer calibration needs to be carried out in a darkroom and under the condition of no stray light interference, the difficulty of calibrating personnel in identifying the indication value of the illuminometer in a low-light environment is greatly increased, and meanwhile, fatigue work is easily caused to influence the identification accuracy; on the other hand, working in a darkroom with severely limited light can easily cause health and safety problems for the certified personnel. The widely used optical character recognition method at present needs to have good illumination conditions to accurately recognize characters in an image, which conflicts with the environment of illuminometer verification, so that the method for efficiently, accurately and intelligently recognizing the illuminometer indication value in the verification process has important practical significance.

Disclosure of Invention

In order to solve the existing problems, the invention adopts a Convolutional Neural Network (CNN) target detection framework, trains the CNN target detection framework by constructing a data set and a data enhancement algorithm, searches for the optimal prediction contrast and realizes the accurate identification of the illuminometer readings in the low-contrast environment.

The purpose of the invention is realized by the following technical scheme:

a darkroom illuminometer calibration-oriented intelligent low-contrast indicating image identification method comprises the following steps:

A. making a training and testing data set of a convolutional neural network target detection framework, wherein the training data set comprises a high-contrast nixie tube image set I_HAnd low contrast nixie tube image set I_LAnd calculate I_HMinimum value of contrast C for collective image_{H_min}、I_LMaximum value of contrast C of collective image_{L_max}；

B. Calculating image contrast C in CNN target detection frame at each input during training and testing_trJudging the set to which the image belongs and changing the contrast of the input image;

C. the CNN target detection framework generates a prediction target frame and a probability P output for an input image, and after training is finished, the optimal recognition contrast C is calculated according to a P-C curve_best；

D. Calculating low contrast illuminometer representative image contrast C in darkroom predicted by CNN target detection framework_realAnd performing enhancement.

According to the method, data enhancement is respectively carried out on the high-contrast image and the low-contrast image, model training of the CNN target detection frame in value image prediction under different contrasts is completed, and the optimal recognition contrast is calculated according to the prediction probability and the contrast of a correct prediction sample in the training process. In the actual prediction process, the image contrast is enhanced to the optimal recognition contrast, and the recognition accuracy is further improved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:

FIG. 1 is a flow chart of the structure of the method for intelligently identifying a low-contrast indicating image for darkroom illuminometer calibration according to the present invention.

Detailed Description

According to the technical scheme of the invention, a plurality of structural modes and manufacturing methods of the invention can be provided by a person with ordinary skill in the art without changing the essential spirit of the invention. Therefore, the following detailed description and the accompanying drawings are only specific illustrations of the technical solutions of the present invention, and should not be construed as all of the present invention or as limitations or limitations on the technical solutions of the present invention.

The invention is described in further detail below with reference to examples and figures:

as shown in fig. 1, a darkroom illuminometer calibration-oriented intelligent identification method for low-contrast indicating value images comprises the following specific steps:

step 10, making a training and testing data set of a convolutional neural network target detection framework, wherein the training data set comprises a high-contrast nixie tube image set I_HAnd low contrast nixie tube image set I_LAnd calculate I_HMinimum value of contrast C for collective image_{H_min}、I_LMaximum value of contrast C of collective image_{L_max}；

Step 20 calculating image contrast C in the CNN target detection frame for each input during training and testing_trJudging the set to which the image belongs and changing the contrast of the input image by using a corresponding data enhancement algorithm;

step 30, the CNN target detection framework generates a prediction target frame and probability P output for the input image, and after training is finished, the optimal recognition contrast C is calculated according to the P-C curve_best；

Step 40 calculating the actual low contrast illuminometer representative image contrast C in the darkroom predicted using the CNN target detection framework_realAnd performing enhancement;

the step 10 specifically includes: i is_HImage dataset is { F_H,1,F_H,2,F_H,3,……,F_{H,Num_H}}，I_LImage dataset is { F_L,1,F_L,2,F_L,3,……,F_{L,Num_L}Wherein Num _ H and Num _ L are I respectively_H、I_LNumber of middle images, F ═ F_ijk)_m×n×3As a RGB color space mapLike a matrix. Converting RGB image into grayscale image F in calculating image contrast^gray＝(f_ij)_m×nThen, the method for calculating the contrast C of the image is as follows:

i can be calculated according to the formula_HAnd I_LSet of contrasts { C over all images in training dataset_H,1,C_H,2,C_H,2,……,C_{H,Num_H}And { C }_L,1,C_L,2,C_L,2,……,C_{L,Num_L}Get C_{H_min}And C_{L_max}And by adjusting the set I_HAnd I_LGuarantee C_{H_min}>C_{L_max}。

The step 20 specifically includes: calculating a training image F according to an image contrast calculation formula_trContrast ratio C of_trThen F is_trThe method for judging the belonged set comprises the following steps:

F_trthe data enhancement method comprises the following steps:

wherein s is²For contrast enhancement factor, the function random (x, y) acts to randomly generate a bit in the interval [ x, y ]]Of (4).

The step 30 specifically includes: selecting a part for correctly predicting the indicating value of the nixie tube image in the whole training process, wherein the contrast of each training image is { C_right,1,C_right,2,C_rhght,3,……,C_{right,Num_right}With a corresponding prediction probability of { P }_right,1,P_right,2,P_right,3,……,P_{right,Num_right}Num _ right is to make the correct predictionMeasuring the number of fractions, then C_bestThe calculation method comprises the following steps:

the step 40 specifically includes: contrast C from images captured by industrial cameras before actual recognition of light level indications in low light environments_realContrast enhancement is performed with a contrast enhancement factor of

Although the embodiments of the present invention have been described above, the above descriptions are only for the convenience of understanding the present invention, and are not intended to limit the present invention. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims

1. A darkroom illuminometer calibration-oriented intelligent identification method for low-contrast indicating value images is characterized by comprising the following specific steps:

A. making a training and testing data set of a convolutional neural network target detection framework, wherein the training data set comprises a high-contrast nixie tube image set I_HAnd low contrast nixie tube image set I_LAnd calculate I_HMinimum value of set image contrast C_{H_min}、I_LMaximum value of contrast C of collective image_{L_max}；

B. Calculating image contrast C in CNN target detection frame at each input during training and testing_trJudging the set to which the image belongs and changing the contrast of the input image by using a corresponding data enhancement algorithm;

C. the CNN target detection framework generates a prediction target frame and probability P output for an input image, and after training is finished, the probability is predicted according to the imageP and the contrast C of the image, and calculating the optimal recognition contrast C_best；

D. Calculating low contrast illuminometer representative image contrast C in darkroom predicted by CNN target detection framework_realAnd performing enhancement;

in the step C, a part for correctly predicting the numerical tube image indicating value in the whole training process is selected, and the contrast of each training image is { C_right,1,C_right,2,C_rhght,3,……,C_{right,Num_right}Corresponding to a prediction probability of { P }_right,1,P_right,2,P_right,3,……,P_{right,Num_right}Num _ right is the number of parts to make the correct prediction, then C_bestThe calculation method comprises the following steps:

2. the darkroom illuminometer-certification-oriented intelligent identification method for low-contrast indicating images according to claim 1, wherein in the step A, I_HImage dataset is { F_H,1,F_H,2,F_H,3,……,F_{H,Num_H}}，I_LImage dataset is { F_L,1,F_L,2,F_L,3,……,F_{L,Num_L}H, Num _ H and Num _ L are I_H、I_LNumber of middle images, F ═ F_ijk)_m×n×3Is an RGB color space image matrix; converting RGB image into grayscale image F in calculating image contrast^gray＝(f_ij)_m×nThen, the method for calculating the contrast C of the image is as follows:

i can be calculated according to a formula_HAnd I_LSet of contrasts { C over all images in training dataset_H,1,C_H,2,C_H,2,……,C_{H,Num_H}And { C }_L,1,C_L,2,C_L,2,……,C_{L,Num_L}Get C_{H_min}And C_{L_max}And by adjusting the set I_HAnd I_LGuarantee C_{H_min}>C_{L_max}。

3. The darkroom illuminometer-certification-oriented intelligent identification method for low-contrast indicating-value images according to claim 1, wherein in the step B, the training image F is calculated according to an image contrast calculation formula_trContrast ratio C of_trThen F is_trThe method for judging the belonged set comprises the following steps:

4. the darkroom illuminometer-certification-oriented intelligent identification method for low-contrast indicating images according to claim 3, wherein F is_trThe data enhancement algorithm is as follows:

wherein the function random (x, y) acts to randomly generate a number in the interval x, y.

5. The method for intelligently identifying low-contrast-indicated-image for darkroom illuminometer calibration according to claim 1, wherein in the step D, before actually identifying the illuminometer indication in the low-illumination environment, the contrast C of the low-contrast-indicated-image in the darkroom is predicted according to the actually used CNN target detection framework_realContrast enhancement is carried out with a contrast enhancement factor of

C_bestFor low contrast in darkroomThe illuminance representation image optimally identifies the contrast.