CN116152162A - Digitizing method and device for appearance quality residual injury index of cured tobacco leaves - Google Patents
Digitizing method and device for appearance quality residual injury index of cured tobacco leaves Download PDFInfo
- Publication number
- CN116152162A CN116152162A CN202211539266.4A CN202211539266A CN116152162A CN 116152162 A CN116152162 A CN 116152162A CN 202211539266 A CN202211539266 A CN 202211539266A CN 116152162 A CN116152162 A CN 116152162A
- Authority
- CN
- China
- Prior art keywords
- tobacco leaf
- edge
- pixel
- picture
- residual injury
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 241000208125 Nicotiana Species 0.000 title claims abstract description 296
- 235000002637 Nicotiana tabacum Nutrition 0.000 title claims abstract description 296
- 230000006378 damage Effects 0.000 title claims abstract description 178
- 208000027418 Wounds and injury Diseases 0.000 title claims abstract description 174
- 208000014674 injury Diseases 0.000 title claims abstract description 174
- 238000000034 method Methods 0.000 title claims abstract description 50
- 230000011218 segmentation Effects 0.000 claims abstract description 60
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 29
- 238000012545 processing Methods 0.000 claims abstract description 27
- 238000004891 communication Methods 0.000 claims abstract description 23
- 238000002372 labelling Methods 0.000 claims abstract description 18
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000013461 design Methods 0.000 claims abstract description 10
- 238000004364 calculation method Methods 0.000 claims description 38
- 239000011159 matrix material Substances 0.000 claims description 37
- 238000010586 diagram Methods 0.000 claims description 25
- 230000004913 activation Effects 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 11
- 101100129590 Schizosaccharomyces pombe (strain 972 / ATCC 24843) mcp5 gene Proteins 0.000 claims description 10
- 230000001174 ascending effect Effects 0.000 claims description 10
- 238000003708 edge detection Methods 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 206010061245 Internal injury Diseases 0.000 claims description 2
- 239000012535 impurity Substances 0.000 claims description 2
- 210000003462 vein Anatomy 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 3
- 238000013139 quantization Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000011002 quantification Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/457—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by analysing connectivity, e.g. edge linking, connected component analysis or slices
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the specification provides a method, a device and electronic equipment for digitizing a residual injury index of appearance quality of cured tobacco leaves, wherein the method comprises the following steps: labeling the cured tobacco leaf picture, training the cured tobacco leaf picture based on a deep self-learning network to obtain a semantic segmentation model, inputting the cured tobacco leaf picture into the semantic segmentation model, and processing tobacco leaf background and other sundries on a tobacco leaf placing platform according to the output of the semantic segmentation model to obtain the processed cured tobacco leaf picture; calculating and analyzing the communication area of the pretreated cured tobacco leaf picture, and obtaining the residual injury area inside the tobacco leaf; carrying out edge extraction on the pretreated flue-cured tobacco leaf picture, and solving a residual injury area on the edge according to an extracted edge characteristic design algorithm; and calculating the internal residual injury ratio and the edge residual injury ratio according to the extracted tobacco leaf internal and edge residual injury area information, and taking the internal residual injury ratio and the edge residual injury ratio as a digital result of the appearance quality residual injury index of the flue-cured tobacco leaf.
Description
Technical Field
The document relates to the technical field of computers, in particular to a method and a device for digitizing residual injury indexes of appearance quality of cured tobacco leaves and electronic equipment.
Background
With the development of the tobacco industry, tobacco grading is increasingly important, and in the aspect of grading prediction of tobacco leaves through appearance quality indexes, the current common practice in the industry is manual grading, and great waste is caused to manpower, material resources and the like. The reason for this is that in the national standard for flue-cured tobacco of the people's republic of China, the regulations for 7 indexes related to the appearance quality are qualitative regulations. For example, the chromaticity index is divided into a middle and a weak, while the residual injury index is expressed as a percentage, but the percentage is a maximum value of a corresponding grade, and how the percentage is defined and calculated is not mentioned, and the percentage is considered to be a multiple of 5% and is estimated manually according to the experience of a grading expert.
In conclusion, the tobacco residual injury is an important influencing factor of the appearance quality of tobacco, but the quality specification of the index is qualitative specification by the national standard of the flue-cured tobacco at present, and the method has strong subjectivity and uncertainty and has negative influence on classification, subsequent processing and the like of the tobacco.
Disclosure of Invention
The invention aims to provide a method and a device for digitizing a residual injury index of appearance quality of cured tobacco leaves and electronic equipment, and aims to solve the problems in the prior art.
The invention provides a digitizing method of a residual injury index of appearance quality of cured tobacco leaves, which comprises the following steps:
labeling the cured tobacco leaf picture, training the cured tobacco leaf picture based on a deep self-learning network to obtain a semantic segmentation model, inputting the cured tobacco leaf picture into the semantic segmentation model, and processing tobacco leaf background and other sundries on a tobacco leaf placing platform according to the output of the semantic segmentation model to obtain the processed cured tobacco leaf picture;
calculating and analyzing the communication area of the pretreated cured tobacco leaf picture, and obtaining the residual injury area inside the tobacco leaf; carrying out edge extraction on the pretreated flue-cured tobacco leaf picture, and solving a residual injury area on the edge according to an extracted edge characteristic design algorithm;
and calculating the internal residual injury ratio and the edge residual injury ratio according to the extracted tobacco leaf internal and edge residual injury area information, and taking the internal residual injury ratio and the edge residual injury ratio as a digital result of the appearance quality residual injury index of the flue-cured tobacco leaf.
The invention provides a digitizing device for residual injury indexes of appearance quality of cured tobacco leaves, which comprises:
The semantic segmentation model module is used for marking the cured tobacco leaf picture, training the cured tobacco leaf picture based on the marked tobacco leaf picture based on the deep self-learning network to obtain a semantic segmentation model, inputting the cured tobacco leaf picture into the semantic segmentation model, and processing tobacco leaf background and other sundries on a tobacco leaf placing platform according to the output of the semantic segmentation model to obtain the processed cured tobacco leaf picture;
the analysis and extraction module is used for carrying out calculation and analysis on the communication area of the pretreated cured tobacco leaf picture and solving the residual injury area inside the tobacco leaf; carrying out edge extraction on the pretreated flue-cured tobacco leaf picture, and solving a residual injury area on the edge according to an extracted edge characteristic design algorithm;
the computing module is used for computing the internal residual injury ratio and the edge residual injury ratio according to the extracted tobacco leaf internal and edge residual injury area information and taking the internal residual injury ratio and the edge residual injury ratio as a digital result of the appearance quality residual injury index of the flue-cured tobacco leaf.
The embodiment of the invention also provides digital electronic equipment for the appearance quality residual injury index of the cured tobacco leaves, which comprises the following components: the method comprises the steps of a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the computer program realizes the digitizing method of the appearance quality residual injury index of the cured tobacco leaves when being executed by the processor.
The embodiment of the invention also provides a computer readable storage medium, wherein the computer readable storage medium is stored with an implementation program of information transmission, and the program is executed by a processor to realize the steps of the digitizing method of the appearance quality residual injury index of the cured tobacco leaves.
By adopting the embodiment of the invention, the problems of strong subjectivity, low detection efficiency, high resource consumption and the like caused by manual grading and qualitative indexes are solved. The embodiment of the invention utilizes deep learning semantic segmentation to preprocess the tobacco leaf picture, extracts the related characteristic information of the inside and the edge of the tobacco leaf, provides an algorithm for quantizing the residual injury index, uses part of tobacco leaves with known grades for verification, confirms the effectiveness of the quantization method and can realize the digital quantization of the appearance residual injury index. Compared with the original qualitative analysis, the method has more scientificity, and is favorable for the research of the aspects of intelligent classification, related index prediction and the like of follow-up tobacco leaves. The hardware equipment required by the embodiment of the invention is simple, and the high-resolution camera only needs to shoot the tobacco leaf pictures, and has the advantages of low cost and high speed compared with other complex hardware devices.
Drawings
For a clearer description of one or more embodiments of the present description or of the solutions of the prior art, the drawings that are necessary for the description of the embodiments or of the prior art will be briefly described, it being apparent that the drawings in the description that follow are only some of the embodiments described in the description, from which, for a person skilled in the art, other drawings can be obtained without inventive faculty.
FIG. 1 is a flowchart of a method for digitizing an index of the appearance quality residue of cured tobacco leaves in accordance with an embodiment of the present invention;
FIG. 2 is a detailed step flow chart of a method for digitizing an index of the appearance quality residue of cured tobacco leaves according to an embodiment of the present invention;
FIG. 3 is a flowchart of a method for calculating the internal injury ratio of tobacco leaves according to an embodiment of the invention;
FIG. 4 is a flowchart of a method for calculating the residual injury ratio of the edges of tobacco leaves according to an embodiment of the invention;
fig. 5 is a schematic diagram of labeling of tobacco leaf picture semantic segmentation labels according to an embodiment of the present invention;
FIG. 6 is a schematic view of an internal residual injury area of tobacco leaves according to an embodiment of the invention;
FIG. 7 is a schematic view of a tobacco leaf edge residual injury area according to an embodiment of the invention;
FIG. 8 is a waveform of tobacco leaf edge characteristics of an embodiment of the present invention;
FIG. 9 is a differential waveform of tobacco leaf edges according to an embodiment of the present invention;
FIG. 10 is a schematic diagram of a digitizing apparatus for appearance quality residual indicators of cured tobacco leaves in accordance with an embodiment of the present invention;
fig. 11 is a schematic diagram of a digital electronic device for the appearance quality residual injury index of the cured tobacco leaves according to the embodiment of the invention.
Detailed Description
The invention provides a digital quantification method of a residual injury index of appearance quality of cured tobacco leaves, which mainly aims at designing a quantification algorithm for the residual injury in the appearance quality index in national standard of flue-cured tobacco of the people's republic of China, and solves the problems of strong subjectivity, low detection efficiency, more resource consumption and the like caused by manual grading of the qualitative index. Firstly, preprocessing a photographed tobacco leaf picture by deep learning semantic segmentation and connected region screening to eliminate the interference of sundries such as background, broken leaves and the like; secondly, defining tobacco leaf areas through a cutting frame, sequencing by utilizing the communicated areas to obtain residual injury characteristic areas inside the tobacco leaves, and calculating the internal residual injury ratio; then, constructing a tobacco leaf edge characteristic oscillogram by extracting the tobacco leaf edge, carrying out waveform analysis and processing to obtain a tobacco leaf edge residual injury characteristic region and calculating an edge residual injury ratio; and finally, comparing the calculated residual injury ratio with the maximum residual injury percentage of the national standard, and verifying the rationality and the effectiveness of the algorithm.
In order to enable a person skilled in the art to better understand the technical solutions in one or more embodiments of the present specification, the technical solutions in one or more embodiments of the present specification will be clearly and completely described below with reference to the drawings in one or more embodiments of the present specification, and it is obvious that the described embodiments are only some embodiments of the present specification, not all embodiments. All other embodiments, which can be made by one or more embodiments of the present disclosure without inventive faculty, are intended to be within the scope of the present disclosure.
Method embodiment
According to an embodiment of the present invention, there is provided a method for digitizing an index of appearance quality and damage of cured tobacco leaves, and fig. 1 is a flowchart of the method for digitizing an index of appearance quality and damage of cured tobacco leaves according to an embodiment of the present invention, as shown in fig. 1, the method for digitizing an index of appearance quality and damage of cured tobacco leaves according to an embodiment of the present invention specifically includes:
step 101, labeling the cured tobacco leaf picture, training the cured tobacco leaf picture based on the labeled based on a deep self-learning network to obtain a semantic segmentation model, inputting the cured tobacco leaf picture into the semantic segmentation model, and processing tobacco leaf background and other sundries on a tobacco leaf placing platform according to the output of the semantic segmentation model to obtain the processed cured tobacco leaf picture;
102, calculating and analyzing a communication area of the pretreated cured tobacco leaf picture, and obtaining a residual injury area inside the tobacco leaf; carrying out edge extraction on the pretreated flue-cured tobacco leaf picture, and solving a residual injury area on the edge according to an extracted edge characteristic design algorithm;
and 103, calculating the internal residual injury ratio and the edge residual injury ratio according to the extracted tobacco leaf internal and edge residual injury area information, and taking the internal residual injury ratio and the edge residual injury ratio as a digital result of the appearance quality residual injury index of the flue-cured tobacco leaf.
The step 101 specifically includes:
pixel-level labeling is carried out on the picture characteristic region and the background of the cured tobacco leaf picture, and a self-defined polygonal curve is used for labeling a tobacco leaf part and a background part on each cured tobacco leaf picture to be trained, so that a pixel-level label with pixel position information of the tobacco leaf region is generated;
training according to the baked tobacco leaf picture marked with the label by adopting a deep Lab v3+ semantic segmentation network based on the mobiletv 2, segmenting tobacco leaf parts in the baked tobacco leaf picture, wherein 186 layers are formed in the deep Lab v3+ semantic segmentation network based on the mobiletv 2, carrying out batch normalization according to a formula 1 after carrying out convolution operation through a convolution layer each time, inputting an activation layer, using a truncated rectification linear unit as an activation function according to a formula 2, restoring an output result of an encoder into the size of an input feature map by using transposed convolution, carrying out center cutting on an original map, outputting a segmentation result with the number of channels being the number of labels after the softmax activation function shown in the formula 3, and inputting output into a next convolution layer. The final semantic segmentation result is a category matrix with a category of a label category, namely the elements of the matrix are label names of different categories or label identification character strings of labels, namely labels 'StopSeg' of tobacco leaf parts and labels 'back' of background parts respectively;
Wherein x is (k) And y (k) Original input data and output data, mu (k) Sum sigma (k) Respectively the mean value and standard deviation, beta of input data (k) And gamma (k) The upper mark k represents the kth dimension of the data, epsilon is a small amount preventing denominator from being 0;
wherein, ceilling is a set threshold;
wherein z is i The output value of the ith node is C, which is the number of label types;
extracting element positions with elements of back from a category matrix output by a semantic segmentation model aiming at the background, setting 0 as a pixel value at the corresponding position in an original image pixel matrix, performing preliminary processing, converting a primarily processed flue-cured tobacco leaf picture into a single-channel gray level image according to a formula 4, converting the single-channel gray level image into a binary image to obtain a connected region, determining a proper pixel threshold value to binarize the pixel when the gray level image is converted into the binary image, searching the pixel threshold value by adopting a method for calculating the variance between foreground and background categories according to a formula 5, traversing the pixel value in the gray level image when calculation, taking a part higher than the current pixel value as a foreground part and the rest as a background part, sequentially calculating the pixel value corresponding to the maximum variance as a threshold value, converting the gray level image into the binary image, obtaining the connected region in the binary image, selecting 8 connected modes, namely 8 pixel points with the elements of 1, namely 8 directions are all the pixel values in the upper, lower left and right directions and 8 directions, regarding the region as one connected region, selecting the largest connected region, obtaining all the rest connected regions, taking the rest connected regions as the foreground region, taking the rest pixel values, setting the rest pixel values as the rest pixel values, and setting the rest pixel values as the rest of the pixel values, and standing on the background region, standing the rest as the pixel values, and standing on the background image with no influence platform, and standing on the pixel value and the rest being 0;
I=0.30r+0.59g+0.11b equation 4;
wherein R, G, B are three channel matrixes of the color picture, and I is a combined single channel matrix;
std＝num1×(ave1-ave) 2 +num2×(ave2-ave) 2 equation 5;
where std is the inter-class variance, num1 is the number of pixels in the foreground portion, num2 is the number of pixels in the background portion, ave1 is the average gray value in the foreground portion, ave2 is the average gray value in the background portion, and ave is the total average gray value.
the preprocessed flue-cured tobacco leaf picture is cut by using a minimum rectangular frame, namely, the picture pixel matrix is traversed row by row or column by column along the up-down, left-right and up-down directions respectively, the up-down direction is traversed row by row, the left-right direction is traversed column by column, and the first row or column which is not 0 is searched as the boundary of the minimum rectangular frame.
And inverting the obtained binary image, then solving the connected areas of the 8 connected modes, comparing the number of elements in each connected area and arranging the elements in a descending order, selecting the first four areas, wherein the four areas correspond to the background formed by the four corners of the rectangular frame and the edges of the tobacco leaves, and removing the four areas to obtain the left connected areas which are small connected areas, namely the residual injury characteristic areas in the tobacco leaves.
Filling the residual injury area in the tobacco leaves, and converting the residual injury area into a binary image, wherein the foreground and the background in the binary image respectively represent the integral part of the tobacco leaves and the background of the image; the filling mode is as follows: setting the R channel value of the residual injury part to 255 according to the residual injury area in the tobacco, namely filling all the R channel values into red, and obtaining a complete tobacco area after binarization;
Extracting tobacco leaf edges in a binary image by using a canny edge detection algorithm, wherein a canny operator smoothes the image by Gaussian filtering, differentially calculates the amplitude and direction of an image gradient, suppresses non-maximum values, and finally screens the binary image by using a double threshold to extract edges, wherein the gradient amplitude and direction of a certain pixel point C in the image are determined according to a formula 6 and a formula 7, and an edge point set is obtained by 8-neighborhood traversal according to the calculated gradient information of each pixel point:
wherein G is c The amplitude value of the pixel point C, the theta is the direction angle G x ,G y Gradient values of the point C in the X and Y directions calculated by using a first-order edge detection operator and other methods respectively;
taking a point inside the tobacco leaf and determining as a reference point O r ，O r Selecting the center of the tobacco leaf picture, and respectively calculating all points and O in the edge point set r Euclidean distance of (2) and O r The inclination angles of the straight lines are arranged in ascending order according to the inclination angle, and an edge point set R= { (x) is arranged 1 ,y 1 ),(x 2 ,y 2 ),...,(x i ,y i ) Determining a distance set L according to equation 8:
L＝{‖(x 1 ,y 1 )-O r ‖ 2 ,‖(x 2 ,y 2 )-O r ‖ 2 ,...,‖(x i ,y i )-O r ‖ 2 equation 8;
calculating the tilt angle A according to equations 9-11 requires the current point coordinates to be compared with the reference point O r Coordinates (x) r ,y r ) Comparing;
when x is i <x r When the calculation formula A is:
when x is i >x r And y is i >y r When the calculation formula A is:
When x is i >x r And y is i <y r When the calculation formula A is:
wherein, when x i ＝x r And y is i >y r When a=270°; when x is i ＝x r And y is i <y r When a=90°; when x is i <x r And y is i ＝y r When a=180°; when x is i >x r And y is i ＝y r When a=0°;
after calculating the corresponding inclination angles A of all points in the edge point set R, arranging the inclination angles in ascending order, taking the arranged inclination angle set as an abscissa, taking the distance in L corresponding to each A as an ordinate, and constructing an edge characteristic waveform diagram;
reconstructing an edge characteristic waveform chart in a fluctuation area where the leaf stalks are removed, wherein the fluctuation area is in a range of an inclination angle + -10 DEG, namely, the part where the inclination angle is smaller than 10 DEG and larger than 350 DEG;
the large fluctuation area in the processed oscillogram reflects the edge residual injury characteristic area, the first-order difference processing is carried out on the edge oscillogram when the positions of the fluctuation areas are extracted, the difference oscillogram after the difference processing is obtained, and the first-order difference is respectively carried out on the horizontal coordinates and the vertical coordinates of the edge oscillogram according to the formula 12:
Δx＝x t+1 -x t equation 12;
and 3 is taken as a threshold value, the amplitude value of the part smaller than the threshold value is set to be 0, and the part larger than the threshold value is kept unchanged, and at the moment, the part with the absolute value of the amplitude value not being 0 in the differential waveform diagram is the edge residual injury characteristic part.
Step 103 specifically includes:
the formula for calculating the internal residual injury ratio of tobacco leaves according to formula 13:
Wherein Rate_in is the residual injury ratio in tobacco leaf and num i For the number of elements of the ith connected region, N is the total number of the connected regions, block_all is the total pixel value occupied by the tobacco leaf part, and the calculation method of the total pixel value is as follows: and searching non-zero elements for the pixel matrix of the cut picture to form a set, and solving the total number of elements of the set.
The pixel point set corresponding to the coordinate with the amplitude of not 0 in the waveform difference diagram is B= { B 1 ,b 2 ,...,b n If bound_all is the total number of elements of the edge point set, then the tobacco edge residual ratio rate_bound is calculated according to equation 14:
after step 103 is performed, the method may further include:
and shooting a picture of the tobacco leaves to be detected, recording a minimum residual injury value according to the national standard according to the real grade of each tobacco leaf, and comparing whether the residual injury result obtained by the algorithm meets the minimum residual injury requirement of the national standard.
The above technical solutions of the embodiments of the present invention are described in detail below with reference to the accompanying drawings.
Fig. 2 is a flowchart of the detailed steps of the method for digitizing the index of the appearance quality residual injury of the cured tobacco leaves according to the embodiment of the invention, as shown in fig. 2, specifically including the following steps:
the method comprises the following steps of (1) preprocessing pictures based on deep learning semantic segmentation: as shown in fig. 5, the flue-cured tobacco leaf picture is marked first, and a depth network is selected for training to obtain a semantic segmentation model. Then reading in tobacco pictures by using the trained model, and processing tobacco background and other sundries on a tobacco placement platform according to model output;
(1) Tobacco leaf picture feature labeling
Before the picture is subjected to semantic segmentation, pixel-level labeling is required to be carried out on the picture characteristic region and the background, a self-defined polygonal curve is used for labeling a tobacco leaf part and a background part on each picture to be trained, and a pixel-level label with the pixel position information of the tobacco leaf region is generated.
(2) Tobacco leaf picture tobacco leaf partial segmentation
According to the embodiment of the invention, a deep Lab v3+ semantic segmentation network based on mobiletv 2 is adopted, and the marked label in the step (1) is used for training, so that tobacco leaf parts in pictures are segmented. The whole deep Lab v3+ network architecture is an encoder-decoder architecture, and a mobiletv 2 network is adopted as a basic network in the invention as a main body part of the encoder. The network has 186 layers, and each time, the convolution operation is carried out through the convolution layers, then batch normalization is carried out, the activation layer is input, and then the output is input into the next convolution layer. The batch normalization formula used was:
x (k) and y (k) Original input data and output data, mu (k) Sum sigma (k) Respectively the mean value and standard deviation, beta of input data (k) And gamma (k) The upper-label k represents the kth dimension of the data, epsilon is a small amount that prevents the denominator from being 0, for the learnable translation parameter and the scaling parameter, respectively. The activation layer uses a truncated rectifying linear unit as an activation function, and the formula is as follows:
The ceiling is a set threshold, here set to 6. The decoder uses transpose convolution to restore the output result of the encoder into the size of an input feature diagram, and performs center clipping on the original diagram, and outputs a segmentation result with the number of channels being the number of labels after a softmax activation function, wherein the softmax activation function formula is as follows:
z i for the output value of the ith node, C is the label typeNumber of the pieces. The final semantic segmentation result is a category matrix with a category of label category, namely, the elements of the matrix are label names of different categories or label identification character strings of labels, namely, label "StopSeg" of tobacco leaf parts and label "back" of background parts.
(3) Treatment of tobacco picture background and broken tobacco and other sundries
Firstly, extracting element positions with the element of back by using a category matrix output by a semantic segmentation model for the background, corresponding to the element positions in an original image pixel matrix, setting the pixel value of the corresponding position to 0, and performing primary processing. And converting the primarily processed tobacco leaf picture into a single-channel gray level image, and converting the single-channel gray level image into a binary image to obtain a connected region. The formula for converting the tobacco leaf picture into the single-channel gray picture is as follows:
I＝0.30R+0.59G+0.11B⑷
r, G, B are three channel matrices of the color picture, I is the merged single channel matrix. When converting a gray level image into a binary image, a proper pixel threshold value needs to be found to binarize the pixel, and the pixel threshold value is found by adopting a method for calculating the variance between foreground and background classes. The calculation formula of the inter-class variance is:
std＝num1×(ave1-ave) 2 +num2×(ave2-ave) 2 ⑸
std is the inter-class variance, num1 is the number of pixels in the foreground portion, num2 is the number of pixels in the background portion, ave1 is the average gray value in the foreground portion, ave2 is the average gray value in the background portion, and ave is the total average gray value. And traversing pixel values in the gray level image during calculation, taking a part higher than the current pixel value as a foreground part and the rest as a background part, sequentially calculating the inter-class variance, and then taking the pixel value corresponding to the maximum variance as a threshold value to convert the gray level image into a binary image.
And (3) obtaining a connected region in the binary image, and selecting an 8-connected mode, namely, 8 directions of pixel values of up, down, left, right and oblique directions of a pixel point with an element of 1 are 1, wherein the region is considered to be a connected region. After all the connected areas are obtained, the largest connected area is selected, the pixel value of the other areas is set to 0, the pixel value of the background part of the picture after twice treatment is set to 0, and the influence of impurities such as broken leaves and the like remained on a workbench is avoided.
Step (2) extracting a residual injury characteristic region in tobacco leaves, as shown in fig. 3: calculating and analyzing the communication area of the pretreated tobacco leaf picture, and solving the residual injury area in the tobacco leaf;
the residual injury in the tobacco leaves is mostly represented by abnormal areas such as holes, focal spots and the like, and the color characteristics are obvious compared with the difference of the intact parts of the tobacco leaves. Firstly, cutting the picture preprocessed in the step (1) by using a minimum rectangular frame, wherein the cutting method is to traverse the picture pixel matrix row by row or column by column along four directions, namely, traversing the picture pixel matrix row by row in the up-down direction and traversing the picture pixel matrix column by column in the left-right direction, and searching a first row or column which is not 0 as a boundary of the minimum rectangular frame. Most of background areas are removed through clipping, the size of the picture is reduced, and the calculation efficiency of a subsequent algorithm is improved. And (2) inverting the binary image obtained in the specific step (3) in the step (1), obtaining a communication region of an 8 communication mode, comparing the number of elements in each communication region, arranging the elements in a descending order, selecting the first four regions, wherein the four regions correspond to the background formed by four corners of a rectangular frame and edges of tobacco leaves, and removing the four regions to obtain a small communication region, namely a residual injury characteristic region in the tobacco leaves. Regarding the influence of main veins of tobacco leaves, when the pictures of the tobacco leaves are shot according to industry specifications, the front sides of the veins are shot, the color difference of the rear sides of the veins is larger, compared with other residual injury areas, the front sides of the veins are very small, the color of the veins is highly similar to that of the black background, and the veins are classified as the background when the communicated areas are obtained, so that the color difference of the veins is not greatly influenced when the residual injury of the tobacco leaves is calculated.
Step (3) extracting the residual injury characteristic region of the tobacco leaf edge, as shown in fig. 4: carrying out edge extraction on the preprocessed tobacco leaf picture, and solving a residual injury area on the edge according to an extracted edge feature design algorithm;
(1) tobacco leaf edge extraction
The residual injury of the tobacco leaf edge is mainly represented by edge defect, mechanical crease, and the like. Firstly, extracting edges of tobacco leaf parts in a picture, wherein the edge extraction is generally carried out in a binary picture, filling the residual injury area in the tobacco leaf in the step (2), and then converting the residual injury area into the binary picture, wherein the foreground and the background in the binary picture respectively represent the integral part of the tobacco leaf and the background of the picture. The filling mode is that the R channel value of the residual injury part is set to 255 according to the residual injury area in the tobacco leaf obtained in the step (1), namely, all the residual injury part is filled with red, and the whole tobacco leaf area can be obtained after binarization is carried out at the moment.
And extracting tobacco leaf edges in the binary image by using a canny edge detection algorithm, smoothing the image by using a canny operator through Gaussian filtering, calculating the amplitude and the direction of the image gradient in a differential mode, inhibiting a non-maximum value, and finally screening and extracting edges from the binary image by using a double threshold value. The gradient amplitude and direction calculation formula of a certain pixel point C in the image is as follows:
G c The amplitude value of the pixel point C, the theta is the direction angle G x ,G y The gradient values in the X and Y directions of the point C calculated using a first order edge detection operator or the like, respectively. And traversing through 8 neighborhood according to the calculated gradient information of each pixel point to obtain an edge point set.
(2) Constructing a tobacco leaf edge characteristic waveform chart as shown in fig. 8:
taking a point inside the tobacco leaf and determining as a reference point O r ，O r Selecting the center of the tobacco leaf picture, and respectively calculating all points and O in the edge point set obtained in the step (1) r Euclidean distance of (2) and O r The inclination angles of the connected straight lines are arranged in ascending order according to the inclination angle. Let the edge point set R = { (x) 1 ,y 1 ),(x 2 ,y 2 ),...,(x i ,y i ) The distance set L is calculated as:
L＝{‖(x 1 ,y 1 )-O r ‖ 2 ,‖(x 2 ,y 2 )-O r ‖ 2 ,...,‖(x i ,y i )-O r ‖ 2 } ⑻
calculating the tilt angle A requires the coordinates of the current point and the reference point O r Coordinates (x) r ,y r ) Comparing;
when x is i <x r When the calculation formula A is:
when x is i >x r And y is i >y r When the calculation formula A is:
when x is i >x r And y is i <y r When the calculation formula A is:
when x is i ＝x r And y is i >y r When a=270°;
when x is i ＝x r And y is i <y r When a=90°;
when x is i <x r And y is i ＝y r When a=180°;
when x is i >x r And y is i ＝y r When a=0°;
and after calculating the inclination angles A corresponding to all the points in the edge point set R, arranging the points in an ascending order, taking the point serial numbers corresponding to the arranged inclination angles as the abscissa, taking the distance in L corresponding to each A as the ordinate, and constructing the edge characteristic waveform diagram.
(3) Edge characteristic waveform diagram analysis for extracting edge residual injury characteristic region
The overall edge waveform is shown in FIG. 9, which is shaped like a sine-cosine function, but the peduncles are due to the fact that they are aligned with the reference point O r The method is very close and has very large distance jump, and severe fluctuation is presented in the oscillogram, so that the subsequent analysis is not facilitated.
The method specifically comprises the steps of carrying out first-order difference on the horizontal and vertical coordinates of the edge waveform graph respectively, wherein the first-order difference formula is as follows:
Δx＝x t+1 -x t ⑿
i.e. adjacent two of the sets are differenced. In the edge differential waveform diagram of tobacco leaves, the absolute value of the amplitude of the complete normal part of the edge is smaller than 3, so that the amplitude of the part smaller than the threshold value is set as 0 and the part larger than the threshold value is kept unchanged by using 3 as the threshold value for processing the original differential waveform signal. At this time, the part with the absolute value of the amplitude value not being 0 in the differential waveform diagram is the edge residual injury characteristic part.
Step (4) designing a residual injury index quantization algorithm: calculating an internal residual injury ratio and an edge residual injury ratio by using the tobacco leaf internal and edge residual injury area information extracted in the steps (2) and (3) through a design algorithm, as shown in figures 6 and 7;
(1) calculation of residual injury ratio in tobacco leaf
The total number of elements in the connected area is the sum of pixel points in the area, so that the formula for calculating the residual injury ratio in the tobacco leaf is as follows:
rate_in is the residual injury ratio in tobacco leaf and num i The number of elements of the ith connected region, N is the total number of connected regionsThe block_all is the total pixel value occupied by the tobacco leaf part, and the calculation method is to find non-zero elements to form a set for the pixel matrix of the picture cut in the step (2), and calculate the total number of elements of the set.
(2) Calculation of tobacco leaf edge residual injury ratio
The pixel point set corresponding to the coordinate with the amplitude of not 0 in the waveform difference diagram obtained in the step (3) is B= { B 1 ,b 2 ,...,b n And if bound_all is the total number of elements of the edge point set, the calculation formula of the tobacco leaf edge residual ratio rate_bound is as follows:
and (5) verifying the result obtained by the algorithm: carrying out residual injury calculation on tobacco pictures with known grades, and verifying the effectiveness of an algorithm;
(1) collecting verification sample data
And shooting a picture of the tobacco leaves to be detected by using a high-resolution industrial camera, and recording the maximum residual injury value according to the national standard according to the real grade of each tobacco leaf. If the maximum residual injury is 25%, it means that less than 25% of the calculated residual injury is considered to be the grade.
(2) Checking algorithm calculation result validity
And comparing whether the residual injury result obtained by the algorithm meets the minimum residual injury requirement of the national standard.
Table 1 shows the verification effect of 10 tobacco pictures:
table 1 comparison of tobacco leaf image residual injury calculation
It can be seen that only one of the 10 groups of samples has a calculated residual injury slightly exceeding the national standard minimum of 2%, confirming the effectiveness of the algorithm. While the present invention has been described in detail with reference to the drawings, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.
Device embodiment 1
According to an embodiment of the present invention, there is provided a digitizing apparatus for post-baking tobacco leaf appearance quality residual injury index, fig. 10 is a schematic diagram of the digitizing apparatus for post-baking tobacco leaf appearance quality residual injury index according to the embodiment of the present invention, as shown in fig. 10, the digitizing apparatus for post-baking tobacco leaf appearance quality residual injury index according to the embodiment of the present invention specifically includes:
the semantic segmentation model module 100 is used for labeling the cured tobacco leaf picture, training the cured tobacco leaf picture based on the labeled based on a deep self-learning network to obtain a semantic segmentation model, inputting the cured tobacco leaf picture into the semantic segmentation model, and processing tobacco leaf background and other sundries on a tobacco leaf placing platform according to the output of the semantic segmentation model to obtain the processed cured tobacco leaf picture;
The analysis and extraction module 102 is used for carrying out calculation and analysis on the communication area of the pretreated cured tobacco leaf picture and solving the residual injury area inside the tobacco leaf; carrying out edge extraction on the pretreated flue-cured tobacco leaf picture, and solving a residual injury area on the edge according to an extracted edge characteristic design algorithm;
the calculating module 104 is configured to calculate an internal residual injury ratio and an edge residual injury ratio according to the extracted tobacco leaf internal and edge residual injury area information, and use the internal residual injury ratio and the edge residual injury ratio as a digitalized result of the appearance quality residual injury index of the flue-cured tobacco leaf.
The semantic segmentation model module 100 is specifically configured to:
pixel-level labeling is carried out on the picture characteristic region and the background of the cured tobacco leaf picture, and a self-defined polygonal curve is used for labeling a tobacco leaf part and a background part on each cured tobacco leaf picture to be trained, so that a pixel-level label with pixel position information of the tobacco leaf region is generated;
training according to the baked tobacco leaf picture marked with the label by adopting a deep Lab v3+ semantic segmentation network based on the mobiletv 2, segmenting tobacco leaf parts in the baked tobacco leaf picture, wherein 186 layers are formed in the deep Lab v3+ semantic segmentation network based on the mobiletv 2, carrying out batch normalization according to a formula 1 after carrying out convolution operation through a convolution layer each time, inputting an activation layer, using a truncated rectification linear unit as an activation function according to a formula 2, restoring an output result of an encoder into the size of an input feature map by using transposed convolution, carrying out center cutting on an original map, outputting a segmentation result with the number of channels being the number of labels after the softmax activation function shown in the formula 3, and inputting output into a next convolution layer. The final semantic segmentation result is a category matrix with a category of a label category, namely the elements of the matrix are label names of different categories or label identification character strings of labels, namely labels 'StopSeg' of tobacco leaf parts and labels 'back' of background parts respectively;
Wherein x is (k) And y (k) Original input data and output data, mu (k) Sum sigma (k) Respectively the mean value and standard deviation, beta of input data (k) And gamma (k) The upper mark k represents the kth dimension of the data, epsilon is a small amount preventing denominator from being 0;
wherein, ceilling is a set threshold;
wherein z is i The output value of the ith node is C, which is the number of label types;
extracting element positions with elements of back from a category matrix output by a semantic segmentation model aiming at the background, setting 0 as a pixel value at the corresponding position in an original image pixel matrix, performing preliminary processing, converting a primarily processed flue-cured tobacco leaf picture into a single-channel gray level image according to a formula 4, converting the single-channel gray level image into a binary image to obtain a connected region, determining a proper pixel threshold value to binarize the pixel when the gray level image is converted into the binary image, searching the pixel threshold value by adopting a method for calculating the variance between foreground and background categories according to a formula 5, traversing the pixel value in the gray level image when calculation, taking a part higher than the current pixel value as a foreground part and the rest as a background part, sequentially calculating the pixel value corresponding to the maximum variance as a threshold value, converting the gray level image into the binary image, obtaining the connected region in the binary image, selecting 8 connected modes, namely 8 pixel points with the elements of 1, namely 8 directions are all the pixel values in the upper, lower left and right directions and 8 directions, regarding the region as one connected region, selecting the largest connected region, obtaining all the rest connected regions, taking the rest connected regions as the foreground region, taking the rest pixel values, setting the rest pixel values as the rest pixel values, and setting the rest pixel values as the rest of the pixel values, and standing on the background region, standing the rest as the pixel values, and standing on the background image with no influence platform, and standing on the pixel value and the rest being 0;
I=0.30r+0.59g+0.11b equation 4;
wherein R, G, B are three channel matrixes of the color picture, and I is a combined single channel matrix;
std＝num1×(ave1-ave) 2 +num2×(ave2-ave) 2 equation 5;
wherein std is the inter-class variance, num1 is the number of pixels of the foreground part, num2 is the number of pixels of the background part, ave1 is the average gray value of the foreground part, ave2 is the average gray value of the background part, and ave is the total average gray value;
the analysis extraction module 102 is specifically configured to:
the preprocessed flue-cured tobacco leaf picture is cut by using a minimum rectangular frame, namely, the picture pixel matrix is traversed row by row or column by column along the up-down, left-right and up-down directions respectively, the up-down direction is traversed row by row, the left-right direction is traversed column by column, and the first row or column which is not 0 is searched as the boundary of the minimum rectangular frame.
The obtained binary pictures are inverted, then the communication areas of an 8 communication mode are obtained, the number of elements in each communication area is compared and arranged in a descending order, the first four areas are selected, the four areas correspond to the background formed by four corners of a rectangular frame and edges of tobacco leaves, and the left communication areas after the four areas are removed are small communication areas, namely residual injury characteristic areas in the tobacco leaves;
filling the residual injury area in the tobacco leaves, and converting the residual injury area into a binary image, wherein the foreground and the background in the binary image respectively represent the integral part of the tobacco leaves and the background of the image; the filling mode is as follows: setting the R channel value of the residual injury part to 255 according to the residual injury area in the tobacco, namely filling all the R channel values into red, and obtaining a complete tobacco area after binarization;
Extracting tobacco leaf edges in a binary image by using a canny edge detection algorithm, wherein a canny operator smoothes the image by Gaussian filtering, differentially calculates the amplitude and direction of an image gradient, suppresses non-maximum values, and finally screens the binary image by using a double threshold to extract edges, wherein the gradient amplitude and direction of a certain pixel point C in the image are determined according to a formula 6 and a formula 7, and an edge point set is obtained by 8-neighborhood traversal according to the calculated gradient information of each pixel point:
wherein G is c The amplitude value of the pixel point C, the theta is the direction angle G x ,G y Gradient values of the point C in the X and Y directions calculated by using a first-order edge detection operator and other methods respectively;
taking a point inside the tobacco leaf and determining as a reference point O r ，O r Selecting the center of the tobacco leaf picture, and respectively calculating all points and O in the edge point set r Euclidean distance of (2) and O r The inclination angles of the straight lines are arranged in ascending order according to the inclination angle, and an edge point set R= { (x) is arranged 1 ,y 1 ),(x 2 ,y 2 ),...,(x i ,y i ) According to }, according toEquation 8 determines the distance set L:
L＝{‖(x 1 ,y 1 )-O r ‖ 2 ,‖(x 2 ,y 2 )-O r ‖ 2 ,...,‖(x i ,y i )-O r ‖ 2 equation 8;
calculating the tilt angle A according to equations 9-11 requires the current point coordinates to be compared with the reference point O r Coordinates (x) r , r ) Comparing;
when x is i <x r When the calculation formula A is:
when x is i >x r And y is i >y r When the calculation formula A is:
When x is i >x r And y is i <y r When the calculation formula A is:
wherein, when x i ＝x r And y is i >y r When a=270°; when x is i ＝x r And y is i <y r When a=90°; when x is i <x r And y is i ＝y t When a=180°; when x is i >x r And y is i ＝y r When a=0°;
after calculating the corresponding inclination angles A of all points in the edge point set R, arranging the inclination angles in ascending order, taking the arranged inclination angle set as an abscissa, taking the distance in L corresponding to each A as an ordinate, and constructing an edge characteristic waveform diagram;
reconstructing an edge characteristic waveform chart in a fluctuation area where the leaf stalks are removed, wherein the fluctuation area is in a range of an inclination angle + -10 DEG, namely, the part where the inclination angle is smaller than 10 DEG and larger than 350 DEG;
the large fluctuation area in the processed oscillogram reflects the edge residual injury characteristic area, the first-order difference processing is carried out on the edge oscillogram when the positions of the fluctuation areas are extracted, the difference oscillogram after the difference processing is obtained, and the first-order difference is respectively carried out on the horizontal coordinates and the vertical coordinates of the edge oscillogram according to the formula 12:
Δx＝x t+1 -x t equation 12;
processing the original differential waveform signal by taking 3 as a threshold value, wherein the amplitude value of the part smaller than the threshold value is set to 0, and the part larger than the threshold value is kept unchanged, and at the moment, the part with the absolute value of the amplitude value not being 0 in the differential waveform diagram is the edge residual injury characteristic part;
the computing module 104 is specifically configured to:
The formula for calculating the internal residual injury ratio of tobacco leaves according to formula 13:
wherein Rate_in is the residual injury ratio in tobacco leaf and num i For the number of elements of the ith connected region, N is the total number of the connected regions, block_all is the total pixel value occupied by the tobacco leaf part, and the calculation method of the total pixel value is as follows: searching non-zero elements for a pixel matrix of the cut picture to form a set, and solving the total number of elements of the set;
the pixel point set corresponding to the coordinate with the amplitude of not 0 in the waveform difference diagram is B= { B 1 ,b 2 ,...,b n If bound_all is the total number of elements of the edge point set, then the tobacco edge residual ratio rate_bound is calculated according to equation 14:
the device further comprises:
the verification module is used for shooting pictures of the tobacco leaves to be detected, recording the minimum residual injury value according to the national standard according to the real grade of each tobacco leaf, and comparing whether the residual injury result obtained by the algorithm meets the minimum residual injury requirement of the national standard.
The embodiment of the present invention is an embodiment of a device corresponding to the embodiment of the method, and specific operations of each module may be understood by referring to descriptions of the embodiment of the method, which are not repeated herein.
Device example two
The embodiment of the invention provides digital electronic equipment for the appearance quality residual injury index of cured tobacco leaves, as shown in fig. 11, comprising: memory 110, processor 112, and a computer program stored on the memory 110 and executable on the processor 112, which when executed by the processor 112, performs the steps as described in the method embodiments.
Device example two
Embodiments of the present invention provide a computer-readable storage medium having stored thereon a program for carrying out information transmission, which when executed by the processor 112, carries out the steps described in the method embodiments.
The computer readable storage medium of the present embodiment includes, but is not limited to: ROM, RAM, magnetic or optical disks, etc.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.
Claims (10)
1. The method for digitizing the index of the appearance quality residual injury of the flue-cured tobacco leaves is characterized by comprising the following steps:
labeling the cured tobacco leaf picture, training the cured tobacco leaf picture based on a deep self-learning network to obtain a semantic segmentation model, inputting the cured tobacco leaf picture into the semantic segmentation model, and processing tobacco leaf background and other sundries on a tobacco leaf placing platform according to the output of the semantic segmentation model to obtain the processed cured tobacco leaf picture;
Calculating and analyzing the communication area of the pretreated cured tobacco leaf picture, and obtaining the residual injury area inside the tobacco leaf; carrying out edge extraction on the pretreated flue-cured tobacco leaf picture, and solving a residual injury area on the edge according to an extracted edge characteristic design algorithm;
and calculating the internal residual injury ratio and the edge residual injury ratio according to the extracted tobacco leaf internal and edge residual injury area information, and taking the internal residual injury ratio and the edge residual injury ratio as a digital result of the appearance quality residual injury index of the flue-cured tobacco leaf.
2. The method of claim 1, wherein labeling the cured tobacco leaf picture, training the cured tobacco leaf picture based on the labeled based on a deep self-learning network to obtain a semantic segmentation model, inputting the cured tobacco leaf picture into the semantic segmentation model, processing tobacco leaf background and other impurities on a tobacco leaf placement platform according to the output of the semantic segmentation model, and obtaining the processed cured tobacco leaf picture specifically comprises:
pixel-level labeling is carried out on the picture characteristic region and the background of the cured tobacco leaf picture, and a self-defined polygonal curve is used for labeling a tobacco leaf part and a background part on each cured tobacco leaf picture to be trained, so that a pixel-level label with pixel position information of the tobacco leaf region is generated;
Training according to the baked tobacco leaf picture marked with the label by adopting a deep Lab v3+ semantic segmentation network based on the mobiletv 2, segmenting tobacco leaf parts in the baked tobacco leaf picture, wherein 186 layers are added to the deep Lab v3+ semantic segmentation network based on the mobiletv 2, carrying out batch normalization according to a formula 1 after carrying out convolution operation through a convolution layer each time, inputting the activation layer, using a truncated rectification linear unit as an activation function according to a formula 2, restoring an output result of an encoder into the size of an input feature map by using transposed convolution, carrying out center cutting on an original map, outputting a segmentation result with the number of channels being the number of labels after the softmax activation function shown in the formula 3, and inputting the output into a next convolution layer. The final semantic segmentation result is a category matrix with a category of a label category, namely the elements of the matrix are label names of different categories or label identification character strings of labels, namely labels 'StopSeg' of tobacco leaf parts and labels 'back' of background parts respectively;
wherein x is (k) And y (k) Original input data and output data, mu (k) Sum sigma (k) Respectively the mean value and standard deviation, beta of input data (k) And gamma (k) The upper mark k represents the kth dimension of the data, epsilon is a small amount preventing denominator from being 0;
wherein, ceilling is a set threshold;
wherein z is i The output value of the ith node is C, which is the number of label types;
extracting element positions with elements of back from a category matrix output by a semantic segmentation model aiming at the background, setting 0 as a pixel value at the corresponding position in an original image pixel matrix, performing preliminary processing, converting a primarily processed flue-cured tobacco leaf picture into a single-channel gray level image according to a formula 4, converting the single-channel gray level image into a binary image to obtain a connected region, determining a proper pixel threshold value to binarize the pixel when the gray level image is converted into the binary image, searching the pixel threshold value by adopting a method for calculating the variance between foreground and background categories according to a formula 5, traversing the pixel value in the gray level image when calculation, taking a part higher than the current pixel value as a foreground part and the rest as a background part, sequentially calculating the pixel value corresponding to the maximum variance as a threshold value, converting the gray level image into the binary image, obtaining the connected region in the binary image, selecting 8 connected modes, namely 8 pixel points with the elements of 1, namely 8 directions are all the pixel values in the upper, lower left and right directions and 8 directions, regarding the region as one connected region, selecting the largest connected region, obtaining all the rest connected regions, taking the rest connected regions as the foreground region, taking the rest pixel values, setting the rest pixel values as the rest pixel values, and setting the rest pixel values as the rest of the pixel values, and standing on the background region, standing the rest as the pixel values, and standing on the background image with no influence platform, and standing on the pixel value and the rest being 0;
I=0.30r+0.59g+0.11b equation 4;
wherein R, G, B are three channel matrixes of the color picture, and I is a combined single channel matrix;
std＝num1×(ave1-ave) 2 +num2×(ave2-ave) 2 equation 5;
where std is the inter-class variance, num1 is the number of pixels in the foreground portion, num2 is the number of pixels in the background portion, ave1 is the average gray value in the foreground portion, ave2 is the average gray value in the background portion, and ave is the total average gray value.
3. The method of claim 2, wherein the calculating and analyzing the communication area of the pretreated cured tobacco leaf picture to find the residual injury area inside the tobacco leaf specifically comprises:
the preprocessed flue-cured tobacco leaf picture is cut by using a minimum rectangular frame, namely, the picture pixel matrix is traversed row by row or column by column along the up-down, left-right and up-down directions respectively, the up-down direction is traversed row by row, the left-right direction is traversed column by column, and the first row or column which is not 0 is searched as the boundary of the minimum rectangular frame.
And inverting the obtained binary image, then solving the connected areas of the 8 connected modes, comparing the number of elements in each connected area and arranging the elements in a descending order, selecting the first four areas, wherein the four areas correspond to the background formed by the four corners of the rectangular frame and the edges of the tobacco leaves, and removing the four areas to obtain the left connected areas which are small connected areas, namely the residual injury characteristic areas in the tobacco leaves.
4. A method according to claim 3, wherein the edge extraction of the pre-processed cured tobacco leaf picture and the calculation of the residual injury area on the edge by the extracted edge feature design algorithm specifically comprises:
filling the residual injury area in the tobacco leaves, and converting the residual injury area into a binary image, wherein the foreground and the background in the binary image respectively represent the integral part of the tobacco leaves and the background of the image; the filling mode is as follows: setting the R channel value of the residual injury part to 255 according to the residual injury area in the tobacco, namely filling all the R channel values into red, and obtaining a complete tobacco area after binarization;
extracting tobacco leaf edges in a binary image by using a canny edge detection algorithm, wherein a canny operator smoothes the image by Gaussian filtering, differentially calculates the amplitude and direction of an image gradient, suppresses non-maximum values, and finally screens the binary image by using a double threshold to extract edges, wherein the gradient amplitude and direction of a certain pixel point C in the image are determined according to a formula 6 and a formula 7, and an edge point set is obtained by 8-neighborhood traversal according to the calculated gradient information of each pixel point:
wherein G is c The amplitude value of the pixel point C, the theta is the direction angle G x ，G y Gradient values of the point C in the X and Y directions calculated by using a first-order edge detection operator and other methods respectively;
taking a point inside the tobacco leaf and determining as a reference point O r ，O r Selecting the center of the tobacco leaf picture, and respectively calculating all points and O in the edge point set r Euclidean distance of (2) and O r The inclination angles of the straight lines are arranged in ascending order according to the inclination angle, and an edge point set R= { (x) is arranged 1 ，y 1 )，(x 2 ，y 2 )，...，(x i ，y i ) Determining a distance set L according to equation 8:
L＝{||(x 1 ，y 1 )-O r || 2 ，||(x 2 ，y 2 )-O r || 2 ，...，||(x i ，y i )-O r || 2 equation 8;
calculating the tilt angle A according to equations 9-11 requires the current point coordinates to be compared with the reference point O r Coordinates (x) r ，y r ) Comparing;
when x is i ＜x r When the calculation formula A is:
when x is i ＞x r And y is i ＞y r When the calculation formula A is:
when x is i ＞x r And y is i ＜y r When the calculation formula A is:
wherein, when x i ＝x r And y is i ＞y r When a=270°; when x is i ＝x r And y is i ＜y r When a=90°; when x is i ＜x r And y is i ＝y r When a=180°; when x is i ＞x r And y is i ＝y r When a=0°;
after calculating the corresponding inclination angles A of all points in the edge point set R, arranging the inclination angles in ascending order, taking the arranged inclination angle set as an abscissa, taking the distance in L corresponding to each A as an ordinate, and constructing an edge characteristic waveform diagram;
reconstructing an edge characteristic waveform chart in a fluctuation area where the leaf stalks are removed, wherein the fluctuation area is in a range of an inclination angle + -10 DEG, namely, the part where the inclination angle is smaller than 10 DEG and larger than 350 DEG;
The large fluctuation area in the processed waveform diagram reflects the edge residual injury characteristic area, and the first-order difference processing is carried out on the edge waveform diagram when the positions of the fluctuation areas are extracted to obtain a difference wave after the difference processing
The graph is characterized in that the first order difference is respectively carried out on the abscissa and the ordinate of the edge waveform graph according to the formula 12:
Δx＝x t+1 -x t equation 12;
and 3 is taken as a threshold value, the amplitude value of the part smaller than the threshold value is set to be 0, and the part larger than the threshold value is kept unchanged, and at the moment, the part with the absolute value of the amplitude value not being 0 in the differential waveform diagram is the edge residual injury characteristic part.
5. The method according to claim 1, wherein calculating the internal injury ratio and the edge injury ratio based on the extracted tobacco leaf internal and edge injury area information, as the digitized result of the post-cured tobacco leaf appearance quality injury index, specifically comprises:
the formula for calculating the internal residual injury ratio of tobacco leaves according to formula 13:
wherein Rate_in is the residual injury ratio in tobacco leaf and num i For the number of elements of the ith connected region, N is the total number of the connected regions, block_all is the total pixel value occupied by the tobacco leaf part, and the calculation method of the total pixel value is as follows: and searching non-zero elements for the pixel matrix of the cut picture to form a set, and solving the total number of elements of the set.
The amplitude in the waveform difference diagram is notThe pixel point set corresponding to the 0 coordinate is B= { B 1 ,b 2 ,...,b n If bound_all is the total number of elements of the edge point set, then the tobacco edge residual ratio rate_bound is calculated according to equation 14:
6. the method according to claim 1, wherein the method further comprises:
and shooting a picture of the tobacco leaves to be detected, recording a minimum residual injury value according to the national standard according to the real grade of each tobacco leaf, and comparing whether the residual injury result obtained by the algorithm meets the minimum residual injury requirement of the national standard.
7. The utility model provides a digitizing device of incomplete damage index of flue-cured tobacco leaf appearance quality which characterized in that includes:
the semantic segmentation model module is used for marking the cured tobacco leaf picture, training the cured tobacco leaf picture based on the marked tobacco leaf picture based on the deep self-learning network to obtain a semantic segmentation model, inputting the cured tobacco leaf picture into the semantic segmentation model, and processing tobacco leaf background and other sundries on a tobacco leaf placing platform according to the output of the semantic segmentation model to obtain the processed cured tobacco leaf picture;
the analysis and extraction module is used for carrying out calculation and analysis on the communication area of the pretreated cured tobacco leaf picture and solving the residual injury area inside the tobacco leaf; carrying out edge extraction on the pretreated flue-cured tobacco leaf picture, and solving a residual injury area on the edge according to an extracted edge characteristic design algorithm;
The computing module is used for computing the internal residual injury ratio and the edge residual injury ratio according to the extracted tobacco leaf internal and edge residual injury area information and taking the internal residual injury ratio and the edge residual injury ratio as a digital result of the appearance quality residual injury index of the flue-cured tobacco leaf.
8. The apparatus of claim 7, wherein the device comprises a plurality of sensors,
the semantic segmentation model module is specifically configured to:
pixel-level labeling is carried out on the picture characteristic region and the background of the cured tobacco leaf picture, and a self-defined polygonal curve is used for labeling a tobacco leaf part and a background part on each cured tobacco leaf picture to be trained, so that a pixel-level label with pixel position information of the tobacco leaf region is generated;
training according to the baked tobacco leaf picture marked with the label by adopting a deep Lab v3+ semantic segmentation network based on the mobiletv 2, segmenting tobacco leaf parts in the baked tobacco leaf picture, wherein 186 layers are formed in the deep Lab v3+ semantic segmentation network based on the mobiletv 2, carrying out batch normalization according to a formula 1 after carrying out convolution operation through a convolution layer each time, inputting an activation layer, using a truncated rectification linear unit as an activation function according to a formula 2, restoring an output result of an encoder into the size of an input feature map by using transposed convolution, carrying out center cutting on an original map, outputting a segmentation result with the number of channels being the number of labels after the softmax activation function shown in the formula 3, and inputting output into a next convolution layer. The final semantic segmentation result is a category matrix with a category of a label category, namely the elements of the matrix are label names of different categories or label identification character strings of labels, namely labels 'StopSeg' of tobacco leaf parts and labels 'back' of background parts respectively;
Wherein x is (k) And y (k) Original input data and output data, mu (k) Sum sigma (k) Respectively the mean value and standard deviation, beta of input data (k) And gamma (k) The upper mark k represents the kth dimension of the data, epsilon is a small amount preventing denominator from being 0;
wherein, ceilling is a set threshold;
wherein z is i The output value of the ith node is C, which is the number of label types;
extracting element positions with elements of back from a category matrix output by a semantic segmentation model aiming at the background, setting 0 as a pixel value at the corresponding position in an original image pixel matrix, performing preliminary processing, converting a primarily processed flue-cured tobacco leaf picture into a single-channel gray level image according to a formula 4, converting the single-channel gray level image into a binary image to obtain a connected region, determining a proper pixel threshold value to binarize the pixel when the gray level image is converted into the binary image, searching the pixel threshold value by adopting a method for calculating the variance between foreground and background categories according to a formula 5, traversing the pixel value in the gray level image when calculation, taking a part higher than the current pixel value as a foreground part and the rest as a background part, sequentially calculating the pixel value corresponding to the maximum variance as a threshold value, converting the gray level image into the binary image, obtaining the connected region in the binary image, selecting 8 connected modes, namely 8 pixel points with the elements of 1, namely 8 directions are all the pixel values in the upper, lower left and right directions and 8 directions, regarding the region as one connected region, selecting the largest connected region, obtaining all the rest connected regions, taking the rest connected regions as the foreground region, taking the rest pixel values, setting the rest pixel values as the rest pixel values, and setting the rest pixel values as the rest of the pixel values, and standing on the background region, standing the rest as the pixel values, and standing on the background image with no influence platform, and standing on the pixel value and the rest being 0;
I=0.30r+0.59g+0.11b equation 4;
wherein R, G, B are three channel matrixes of the color picture, and I is a combined single channel matrix;
std＝num1×(ave1-ave) 2 +num2×(ave2-ave) 2 equation 5;
wherein std is the inter-class variance, num1 is the number of pixels of the foreground part, num2 is the number of pixels of the background part, ave1 is the average gray value of the foreground part, ave2 is the average gray value of the background part, and ave is the total average gray value;
the analysis and extraction module is specifically used for:
the preprocessed flue-cured tobacco leaf picture is cut by using a minimum rectangular frame, namely, the picture pixel matrix is traversed row by row or column by column along the up-down, left-right and up-down directions respectively, the up-down direction is traversed row by row, the left-right direction is traversed column by column, and the first row or column which is not 0 is searched as the boundary of the minimum rectangular frame.
The obtained binary pictures are inverted, then the communication areas of an 8 communication mode are obtained, the number of elements in each communication area is compared and arranged in a descending order, the first four areas are selected, the four areas correspond to the background formed by four corners of a rectangular frame and edges of tobacco leaves, and the left communication areas after the four areas are removed are small communication areas, namely residual injury characteristic areas in the tobacco leaves;
filling the residual injury area in the tobacco leaves, and converting the residual injury area into a binary image, wherein the foreground and the background in the binary image respectively represent the integral part of the tobacco leaves and the background of the image; the filling mode is as follows: setting the R channel value of the residual injury part to 255 according to the residual injury area in the tobacco, namely filling all the R channel values into red, and obtaining a complete tobacco area after binarization;
Extracting tobacco leaf edges in a binary image by using a canny edge detection algorithm, wherein a canny operator smoothes the image by Gaussian filtering, differentially calculates the amplitude and direction of an image gradient, suppresses non-maximum values, and finally screens the binary image by using a double threshold to extract edges, wherein the gradient amplitude and direction of a certain pixel point C in the image are determined according to a formula 6 and a formula 7, and an edge point set is obtained by 8-neighborhood traversal according to the calculated gradient information of each pixel point:
wherein G is c The amplitude value of the pixel point C, the theta is the direction angle G x ,G y Gradient values of the point C in the X and Y directions calculated by using a first-order edge detection operator and other methods respectively;
taking a point inside the tobacco leaf and determining as a reference point O r ，O r Selecting the center of the tobacco leaf picture, and respectively calculating all points and O in the edge point set r Euclidean distance of (2) and O r The inclination angles of the straight lines are arranged in ascending order according to the inclination angle, and an edge point set R= { (x) is arranged 1 ,y 1 ),(x 2 ,y 2 ),...,(x i ,y i ) Determining a distance set L according to equation 8:
L＝{‖(x 1 ,y 1 )-O r ‖ 2 ,‖(x 2 ,y 2 )-O r ‖ 2 ,...,‖(x i ,y i )-O r ‖ 2 equation 8;
calculating the tilt angle A according to equations 9-11 requires the current point coordinates to be compared with the reference point O r Coordinates (x) r ,y r ) Comparing;
when x is i <x r When the calculation formula A is:
when x is i >x r And y is i >y r When the calculation formula A is:
When x is i >x r And y is i <y r When the calculation formula A is:
wherein, when x i ＝x r And y is i >y r When a=270°; when x is i ＝x r And y is i <y r When a=90°; when x is i <xx r And y is i ＝y r When a=180°; when x is i >x r And y is i ＝y r When a=0°;
after calculating the corresponding inclination angles A of all points in the edge point set R, arranging the inclination angles in ascending order, taking the arranged inclination angle set as an abscissa, taking the distance in L corresponding to each A as an ordinate, and constructing an edge characteristic waveform diagram;
reconstructing an edge characteristic waveform chart in a fluctuation area where the leaf stalks are removed, wherein the fluctuation area is in a range of an inclination angle + -10 DEG, namely, the part where the inclination angle is smaller than 10 DEG and larger than 350 DEG;
the large fluctuation area in the processed oscillogram reflects the edge residual injury characteristic area, the first-order difference processing is carried out on the edge oscillogram when the positions of the fluctuation areas are extracted, the difference oscillogram after the difference processing is obtained, and the first-order difference is respectively carried out on the horizontal coordinates and the vertical coordinates of the edge oscillogram according to the formula 12:
Δx＝x t+1 -x t equation 12;
processing the original differential waveform signal by taking 3 as a threshold value, wherein the amplitude value of the part smaller than the threshold value is set to 0, and the part larger than the threshold value is kept unchanged, and at the moment, the part with the absolute value of the amplitude value not being 0 in the differential waveform diagram is the edge residual injury characteristic part;
the computing module is specifically configured to:
The formula for calculating the internal residual injury ratio of tobacco leaves according to formula 13:
wherein Rate_in is the residual injury ratio in tobacco leaf and num i For the number of elements of the ith connected region, N is the total number of the connected regions, block_all is the total pixel value occupied by the tobacco leaf part, and the calculation method of the total pixel value is as follows: for the cut graphSearching non-zero elements to form a set by the pixel matrix of the slice, and solving the total number of elements of the set;
the pixel point set corresponding to the coordinate with the amplitude of not 0 in the waveform difference diagram is B= { B 1 ,b 2 ,...,b n If bound_all is the total number of elements of the edge point set, then the tobacco edge residual ratio rate_bound is calculated according to equation 14:
the device further comprises:
the verification module is used for shooting pictures of the tobacco leaves to be detected, recording the minimum residual injury value according to the national standard according to the real grade of each tobacco leaf, and comparing whether the residual injury result obtained by the algorithm meets the minimum residual injury requirement of the national standard.
9. A digital electronic device for a post-baking tobacco leaf appearance quality residual injury index, comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed by the processor, performs the steps of the method for digitizing indicators of post-cured tobacco leaf appearance quality residue as defined in any one of claims 1 to 6.
10. A computer-readable storage medium, wherein a program for realizing information transmission is stored on the computer-readable storage medium, and the program when executed by a processor realizes the steps of the digitizing method for the appearance quality residual index of the cured tobacco leaves according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211539266.4A CN116152162A (en) | 2022-12-01 | 2022-12-01 | Digitizing method and device for appearance quality residual injury index of cured tobacco leaves |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211539266.4A CN116152162A (en) | 2022-12-01 | 2022-12-01 | Digitizing method and device for appearance quality residual injury index of cured tobacco leaves |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116152162A true CN116152162A (en) | 2023-05-23 |
Family
ID=86372609
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211539266.4A Pending CN116152162A (en) | 2022-12-01 | 2022-12-01 | Digitizing method and device for appearance quality residual injury index of cured tobacco leaves |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116152162A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118411368A (en) * | 2024-07-04 | 2024-07-30 | 中国农业科学院烟草研究所(中国烟草总公司青州烟草研究所) | Intelligent detection method, medium and system for baked tobacco leaves |
-
2022
- 2022-12-01 CN CN202211539266.4A patent/CN116152162A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118411368A (en) * | 2024-07-04 | 2024-07-30 | 中国农业科学院烟草研究所(中国烟草总公司青州烟草研究所) | Intelligent detection method, medium and system for baked tobacco leaves |
| CN118411368B (en) * | 2024-07-04 | 2024-09-13 | 中国农业科学院烟草研究所(中国烟草总公司青州烟草研究所) | Intelligent detection method, medium and system for baked tobacco leaves |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108416377B (en) | Information extraction method and device in histogram | |
| CN109978890B (en) | Target extraction method and device based on image processing and terminal equipment | |
| CN107665492B (en) | Colorectal panoramic digital pathological image tissue segmentation method based on depth network | |
| CN111257341B (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN109409355B (en) | Novel transformer nameplate identification method and device | |
| CN103577838A (en) | Face recognition method and device | |
| CN110766708B (en) | Image comparison method based on contour similarity | |
| CN110728640B (en) | Fine rain removing method for double-channel single image | |
| CN107240084A (en) | A kind of removing rain based on single image method and device | |
| CN113435407B (en) | Small target identification method and device for power transmission system | |
| CN109064419A (en) | A kind of removing rain based on single image method based on WLS filtering and multiple dimensioned sparse expression | |
| CN105740844A (en) | Insulator cracking fault detection method based on image identification technology | |
| CN108615401B (en) | Deep learning-based indoor non-uniform light parking space condition identification method | |
| CN111222564A (en) | Image identification system, method and device based on image channel correlation | |
| CN113158969A (en) | Apple appearance defect identification system and method | |
| CN113887472A (en) | Remote sensing image cloud detection method based on cascade color and texture feature attention | |
| CN115457327A (en) | Method and device for classifying flue-cured tobacco leaf parts based on vision and electronic equipment | |
| CN112509026A (en) | Insulator crack length identification method | |
| CN110991434B (en) | Self-service terminal certificate identification method and device | |
| CN116152162A (en) | Digitizing method and device for appearance quality residual injury index of cured tobacco leaves | |
| CN115272826A (en) | Image identification method, device and system based on convolutional neural network | |
| CN111046838A (en) | Method and device for identifying wetland remote sensing information | |
| CN113052234A (en) | Jade classification method based on image features and deep learning technology | |
| CN116524725B (en) | Intelligent driving traffic sign image data identification system | |
| CN117218672A (en) | Deep learning-based medical records text recognition method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |