Summary of the invention
The technical problem to be solved by the embodiment of the invention is that providing a kind of skin pressure damage graphic images knowledge
Method for distinguishing, system and mobile phone are analyzed image and temperature index by using machine learning method to complete pressure
The building of infrared imaging ideograph is damaged, realizes the identification to pressure injury graphic images.
In order to solve the above-mentioned technical problem, the embodiment of the invention provides a kind of skin pressure damage graphic images to know
Method for distinguishing the described method comprises the following steps:
Step S1, multiple original skin pressure damage graphic images are obtained;
Step S2, marking to each original skin pressure damage graphic images has label, and further right
The original skin pressure damage graphic images of each existing label carry out image dividing processing, have been partitioned into each
There is the target area image of the original skin pressure damage graphic images of label;
Step S3, image characteristics extraction is carried out to each target area image being partitioned into, obtains each target area
Characteristic after image characteristics extraction, and further carry out the characteristic after each target area image feature extraction
It pre-processes and establishes data set;
Step S4, it is based on preset engineering method, the pretreated characteristic in the data set is trained, structure
Build out skin pressure damage graphic images identification model;
Step S5, a skin graphic images to be measured are obtained and import constructed skin pressure out and damage graphic images
It is detected in identification model, determines whether the skin graphic images to be measured are skin pressure damage thermograph
Picture.
Wherein, the step S2 is specifically included:
If of a certain original skin pressure damage graphic images and preset skin pressure damage graphic images
When being greater than preset threshold with degree, then marking label is 1;Conversely, then marking label is 0;
After the completion of all original skin pressure damage graphic images label labels, to the original skin of each existing label
Skin pressure injury graphic images carry out true color enhancing processing;Wherein, the true color enhancing processing is image color
It remains unchanged, but brightness of image enhances;
Based on the color image segmentation method of preset visual color cluster, to each true color enhancing treated original skin
Skin pressure injury graphic images carry out image dividing processing, are partitioned into the original skin pressure of each existing label
Damage the target area image of graphic images.
Wherein, the original skin pressure damage graphic images to each existing label carry out true color increasing
Strength reason specific steps include:
By R, G in the original skin pressure damage graphic images of each existing label, the equal corresponding conversion of B component be H,
I, S component indicates;
Using grey linear transformation method, enhance in the original skin pressure damage graphic images of each existing label
The I component of conversion;
It, will be each in original skin pressure damage graphic images after each grey linear transformation after I component enhancing
H, I, S component reverse of original skin pressure damage graphic images after opening grey linear transformation are changed to R, G, B component comes
It indicates, obtains the enhancing of each true color treated original skin pressure damage graphic images.
Wherein, the color image segmentation method based on preset visual color cluster, enhances each true color
Treated, and original skin pressure damage graphic images carry out image dividing processing, are partitioned into each existing label
The specific steps of the target area image of original skin pressure damage graphic images include:
Determine gray threshold;
By the enhancing of each true color, treated that original skin pressure damage graphic images carry out gray processing processing, obtains
The gray value of each pixel in each gray processing treated original skin pressure damage graphic images, and further every
In one gray processing treated original skin pressure damage graphic images, gray value and the gray threshold phase are filtered out
Deng pixel and remain, obtain each gray processing treated in original skin pressure damage graphic images by institute
Retain the gray level image that pixel is formed;
Obtained each gray level image is reversed after being changed to corresponding RGB image, is further converted into H, I, S representation in components
Image;
It is H, I, S in H, I, S expression image to each greyscale image transitions using the color cluster algorithm of view-based access control model consistency
Component carries out cluster calculation, and further progress region merging technique and delete operation respectively, obtains each mesh with H, I, S representation in components
Mark area image;
Obtained each target area image using H, I, S representation in components is reversed respectively be changed to corresponding RGB image as point
The target area image output of the original skin pressure damage graphic images of each existing label after cutting.
Wherein, " image characteristics extraction is carried out to each target area image for being partitioned into " in the step S3
Specific steps include:
The approximate entropy and Sample Entropy of each target area image being partitioned into are extracted and as main feature, and into
One step uses Color Statistical feature extracting method, algorithm of co-matrix and local binarization method, every to what is be partitioned into
The color histogram of one target area image, color moment, energy, contrast, texture entropy, texture correlation, image local two-value
Change feature to extract and as complementary features.
Wherein, " the preset engineering method " in the step S4 is that algorithm of support vector machine or BP neural network are calculated
Method.
The embodiment of the invention also provides a kind of systems of skin pressure damage graphic images identification, including obtain single
Member, image segmentation unit, image characteristics extraction unit, identification model construction unit and result judgement unit;Wherein,
The acquiring unit, for obtaining multiple original skin pressure damage graphic images;
Described image cutting unit has mark for marking to each original skin pressure damage graphic images
Label, and image dividing processing is further carried out to the original skin pressure damage graphic images of each existing label,
It is partitioned into the target area image of the original skin pressure damage graphic images of each existing label;
Described image feature extraction unit, for carrying out image characteristics extraction to each target area image being partitioned into,
Characteristic after obtaining each target area image feature extraction, and further propose each target area image feature
Characteristic after taking is pre-processed and establishes data set;
The identification model construction unit, for being based on preset engineering method, to pretreated in the data set
Characteristic is trained, and constructs skin pressure damage graphic images identification model;
The result judgement unit, for obtaining a skin graphic images to be measured and importing constructed skin pressure out
It is detected in damage graphic images identification model, determines whether the skin graphic images to be measured are skin pressure
Damage graphic images.
Wherein, described image cutting unit includes:
Image tag mark module, if for a certain original skin pressure damage graphic images and preset skin pressure
Property damage graphic images when match degree is greater than the preset threshold, then marking label is 1;Conversely, then marking label is 0;
Image color processing module is used for after the completion of all original skin pressures damage graphic images labels labels, right
The original skin pressure damage graphic images of each existing label carry out true color enhancing processing;Wherein, described true
Colour enhancing processing remains unchanged for image color, then brightness of image enhances;
Image segmentation module, the color image segmentation method for being clustered based on preset visual color, to each true color
Treated that original skin pressure damage graphic images carry out image dividing processing for enhancing, is partitioned into each and has mark
The target area image of the original skin pressure damage graphic images of label.
The embodiment of the present invention provides a kind of mobile phone again, for skin pressure damage graphic images identification, including obtains
Take unit, image segmentation unit, image characteristics extraction unit, identification model construction unit and result judgement unit;Wherein,
The acquiring unit, for obtaining multiple original skin pressure damage graphic images;
Described image cutting unit has mark for marking to each original skin pressure damage graphic images
Label, and image dividing processing is further carried out to the original skin pressure damage graphic images of each existing label,
It is partitioned into the target area image of the original skin pressure damage graphic images of each existing label;
Described image feature extraction unit, for carrying out image characteristics extraction to each target area image being partitioned into,
Characteristic after obtaining each target area image feature extraction, and further propose each target area image feature
Characteristic after taking is pre-processed and establishes data set;
The identification model construction unit, for being based on preset engineering method, to pretreated in the data set
Characteristic is trained, and constructs skin pressure damage graphic images identification model;
The result judgement unit, for obtaining a skin graphic images to be measured and importing constructed skin pressure out
It is detected in damage graphic images identification model, determines whether the skin graphic images to be measured are skin pressure
Damage graphic images.
Wherein, described image cutting unit includes:
Image tag mark module, if for a certain original skin pressure damage graphic images and preset skin pressure
Property damage graphic images when match degree is greater than the preset threshold, then marking label is 1;Conversely, then marking label is 0;
Image color processing module is used for after the completion of all original skin pressures damage graphic images labels labels, right
The original skin pressure damage graphic images of each existing label carry out true color enhancing processing;Wherein, described true
Colour enhancing processing remains unchanged for image color, then brightness of image enhances;
Image segmentation module, the color image segmentation method for being clustered based on preset visual color, to each true color
Treated that original skin pressure damage graphic images carry out image dividing processing for enhancing, is partitioned into each and has mark
The target area image of the original skin pressure damage graphic images of label.
The implementation of the embodiments of the present invention has the following beneficial effects:
The original skin pressure damage graphic images of present invention acquisition are gone forward side by side, and row label marks, image procossing and target area divide
After cutting, target area image characteristic is extracted by using machine learning method, image and temperature index are analyzed to come
The building of pressure injury infrared imaging ideograph is completed, realizes the identification to pressure injury graphic images, not only quickly
It is convenient, and accuracy is high, so that solving the prior art can not intuitively recognise that as pressure injury thermograph
The problem of picture.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, the present invention is made into one below in conjunction with attached drawing
Step ground detailed description.
As shown in Figure 1, in the embodiment of the present invention, a kind of side of skin pressure damage graphic images identification of proposition
Method the described method comprises the following steps:
Step S1, multiple original skin pressure damage graphic images are obtained;
Step S2, marking to each original skin pressure damage graphic images has label, and further right
The original skin pressure damage graphic images of each existing label carry out image dividing processing, have been partitioned into each
There is the target area image of the original skin pressure damage graphic images of label;
Step S3, image characteristics extraction is carried out to each target area image being partitioned into, obtains each target area
Characteristic after image characteristics extraction, and further carry out the characteristic after each target area image feature extraction
It pre-processes and establishes data set;
Step S4, it is based on preset engineering method, the pretreated characteristic in the data set is trained, structure
Build out skin pressure damage graphic images identification model;
Step S5, a skin graphic images to be measured are obtained and import constructed skin pressure out and damage graphic images
It is detected in identification model, determines whether the skin graphic images to be measured are skin pressure damage thermograph
Picture.
Detailed process is in step s 2, can be compared automatically by artificial or image data base come original to each
Skin pressure damages graphic images and marks label, to be measured to identify as the training classification of subsequent engineering method
Whether the ownership of skin graphic images is skin pressure damage graphic images.It should be noted that handmarking's mark
Label are after being determined by the existing experience of expert, and by computer record storage, and it is logical that image data base compares automatically
It crosses image similarity to determine automatically, and by computer record storage.
Step S2 specifically includes the following steps:
If step S21, a certain original skin pressure damage graphic images and preset skin pressure damage thermal imaging
When match degree is greater than the preset threshold for image (such as matching similarity is more than threshold value 90%), then marking label is 1;Conversely, then marking
Label is 0;
Step S22, after the completion of all original skin pressure damage graphic images label labels, to each existing label
Original skin pressure damage graphic images carry out true color enhancing processing;Wherein, true color enhancing processing is image
Color remains unchanged, then brightness of image enhances;
Step S23, the color image segmentation method based on preset visual color cluster, after handling each true color enhancing
Original skin pressure damage graphic images carry out image dividing processing, be partitioned into the original skin of each existing label
The target area image of skin pressure injury graphic images.
In step S22, the original skin pressure damage graphic images of each existing label are carried out very color
Color enhances the specific steps handled
(1) by R, G in the original skin pressure damage graphic images of each existing label, the equal corresponding conversion of B component
It is indicated for tone H, brightness I, saturation degree S component;
(2) grey linear transformation method is utilized, the original skin pressure for enhancing each existing label damages graphic images
In switched I component;
It (3), will be every in the original skin pressure damage graphic images after each grey linear transformation after I component enhancing
H, I, S component reverse of original skin pressure damage graphic images after one grey linear transformation are changed to R, G, B component
It indicates, obtains each true color enhancing treated original skin pressure damage graphic images.
In step S23, the image segmentation algorithm based on global threshold is used to remove image background first, retain heat at
It is poly- using view-based access control model color then in order to be partitioned into pressure injury target area as the part of the representative skin in image
The color image segmentation method of class, obtains target area image.
Therefore, the color image segmentation method based on preset visual color cluster, handles each true color enhancing
Original skin pressure damage graphic images afterwards carry out image dividing processing, are partitioned into the original of each existing label
The specific steps of target area image of skin pressure damage graphic images include:
(1) gray threshold is determined;Where it is assumed that its gray value value range between, then gray threshold, meets;
(2) by the enhancing of each true color, treated that original skin pressure damage graphic images carry out gray processing processing,
Each gray processing is obtained treated the gray value of each pixel in original skin pressure damage graphic images, and further
In each gray processing treated original skin pressure damage graphic images, gray value and the gray scale threshold are filtered out
It is worth equal pixel and remains, obtains each gray processing treated in original skin pressure damage graphic images
The gray level image formed by retained pixel;
(3) obtained each gray level image is reversed after being changed to corresponding RGB image, is further converted into H, I, S component
The image of expression;
(4) use view-based access control model consistency color cluster algorithm, to each greyscale image transitions be H, I, S indicate image in H,
I, S component carries out cluster calculation, and further progress region merging technique and delete operation respectively, obtains with H, I, S representation in components
Each target area image;
(5) obtained each target area image using H, I, S representation in components is reversed respectively be changed to corresponding RGB image as
The target area image output of the original skin pressure damage graphic images of each existing label after segmentation.
In step s3, the feature of every piece image, more classical method are extracted using the method that various features are extracted
Have: algorithm of co-matrix, local binary method and Hough transformation method etc., we will choose for graphic images and close
Suitable feature extracting method acquires representative feature.
Therefore, the specific steps that each target area image for being partitioned into carries out image characteristics extraction include:
The approximate entropy and Sample Entropy of each target area image being partitioned into are extracted and as main feature, and into
One step uses Color Statistical feature extracting method, algorithm of co-matrix and local binarization method, every to what is be partitioned into
The color histogram of one target area image, color moment, energy, contrast, texture entropy, texture correlation, image local two-value
Change feature to extract and as complementary features.
Skin graphic images in embodiments of the present invention, can successfully distinguish pressure injury region and image remaining
Partial feature.According to pertinent literature, the entropy feature of graphic images is main identification feature.
In step s 4, for collected feature, can choose a variety of machine learning methods, (such as support vector machines is calculated
Method, BP neural network algorithm etc.) image recognition model is established, performance and the optimal method of accuracy rate are selected more afterwards, are repeatedly tried
It tests, obtains stable and ideal model;Secondly, acquisition various features, are built using same machine learning algorithm (optimal algorithm)
Vertical image recognition model, after Feature Selection, the feature of selection most representative pressure damage image obtains after test
Model with higher resolution.
It the use of the training set that different machines learning algorithm is trained is " number × a skin for skin graphic images
The character matrix of the feature of graphic images ", wherein the feature of image is the feature that computer learns automatically in project.It answers
The training set is calculated with different machine learning methods, obtaining has the new skin pressure damage graphic images of identification
(can calculate new images whether be pressure injury image probability) model, reach the mesh of pressure injury Risk-warning
's.
In step s 5, a skin graphic images to be measured are obtained and import constructed skin pressure out and damage heat
It is detected in image identification model, if output classification is 1, it is determined that going out skin graphic images to be measured is skin pressure
Property damage graphic images;Conversely, output classification is 0, it is determined that going out skin graphic images to be measured is not skin pressure damage
Get mildewed in warm conditions image.It should be noted that the category attribute of output is determined by the label setting in step S2.
As shown in Fig. 2, in the embodiment of the present invention, a kind of skin pressure damage graphic images identification of proposition is
System, including acquiring unit 110, image segmentation unit 120, image characteristics extraction unit 130,140 and of identification model construction unit
Result judgement unit 150;Wherein,
The acquiring unit 110, for obtaining multiple original skin pressure damage graphic images;
Described image cutting unit 120 has for marking to each original skin pressure damage graphic images
Label, and further the original skin pressure damage graphic images of each existing label are carried out at image segmentation
Reason is partitioned into the target area image of the original skin pressure damage graphic images of each existing label;
Described image feature extraction unit 130 is mentioned for carrying out characteristics of image to each target area image being partitioned into
It takes, the characteristic after obtaining each target area image feature extraction, and further by each target area image spy
Characteristic after sign is extracted is pre-processed and establishes data set;
The identification model construction unit 140, for being based on preset engineering method, after the pretreatment in the data set
Characteristic be trained, construct skin pressure damage graphic images identification model;
The result judgement unit 150, for obtaining a skin graphic images to be measured and importing constructed skin pressure out
It is detected in power damage graphic images identification model, determines whether the skin graphic images to be measured are skin pressure
Power damages graphic images.
Wherein, described image cutting unit 120 includes:
Image tag mark module 1201, if for a certain original skin pressure damage graphic images and preset skin
For pressure injury graphic images when match degree is greater than the preset threshold, then marking label is 1;Conversely, then marking label is 0;
Image color processing module 1202, for being completed to all original skin pressure damage graphic images label labels
Afterwards, true color enhancing processing is carried out to the original skin pressure damage graphic images of each existing label;Wherein, institute
It states true color enhancing processing to remain unchanged for image color, then brightness of image enhances;
Image segmentation module 1203, the color image segmentation method for being clustered based on preset visual color, to each Zhang Zhen
Treated that original skin pressure damage graphic images carry out image dividing processing for colour enhancing, has been partitioned into each
There is the target area image of the original skin pressure damage graphic images of label.
As shown in Figure 3 and Figure 4, a kind of skin pressure to propose in the embodiment of the present invention damages graphic images identification
Systematic difference scene figure.
By a large amount of clinical care monitoring data, show that physical feeling is easiest to occur the position of pressure injury in sacrum tail
Portion, therefore choose sacrococcygeal region skin when measurement and carry out temperature measurement, after obtaining graphic images, by experienced person to it
Carry out interpretation, be each width image tagged, as shown in Figure 3.
By image processing algorithm, target area sacrococcygeal region is gone out by Computer Automatic Recognition, is removed unrelated in original image
Information.The following figure is ideal segmentation effect figure.Pressure occurs by 1 phase skin can be easily found in figure B, C and D after dividing
Property damage position (arrow as shown in Figure 4).
The feature of every piece image is completed and then extracts using the method that various features are extracted in image dividing processing,
Comparing classical method has: algorithm of co-matrix, local binary method and Hough transformation method etc., will be directed to thermal imaging
Image chooses suitable feature extracting method, acquires representative feature.
After the feature extraction step for completing image, the image data set of acquisition is subjected to arrangement and normalization operation.So
After select suitable classification algorithm training they, obtain the graphic images identification model with higher resolution.It is current extensive
The algorithm used has support vector machines, random forest and artificial neural network etc., and suitable algorithm is therefrom selected to be trained.?
When collecting enough amount of images, it will also attempt to establish model using the method for deep learning.Tentatively establish model it
Afterwards, the image data that persistent collection is new, is continued to optimize and test model, to increase the stability and accuracy of model.
Thermograph is damaged finally, obtaining a skin graphic images to be measured and importing constructed skin pressure out
As being detected in identification model, if output classification is 1, it is determined that going out skin graphic images to be measured is skin pressure damage
Graphic images;Conversely, output classification be 0, it is determined that go out skin graphic images to be measured be not skin pressure damage heat at
As image.
As shown in figure 5, in the embodiment of the present invention, a kind of mobile phone provided damages thermograph for skin pressure
As identification, including acquiring unit 210, image segmentation unit 220, image characteristics extraction unit 230, identification model construction unit
240 and result judgement unit 250;Wherein,
The acquiring unit 210, for obtaining multiple original skin pressure damage graphic images;
Described image cutting unit 220 has for marking to each original skin pressure damage graphic images
Label, and further the original skin pressure damage graphic images of each existing label are carried out at image segmentation
Reason is partitioned into the target area image of the original skin pressure damage graphic images of each existing label;
Described image feature extraction unit 230 is mentioned for carrying out characteristics of image to each target area image being partitioned into
It takes, the characteristic after obtaining each target area image feature extraction, and further by each target area image spy
Characteristic after sign is extracted is pre-processed and establishes data set;
The identification model construction unit 240, for being based on preset engineering method, after the pretreatment in the data set
Characteristic be trained, construct skin pressure damage graphic images identification model;
The result judgement unit 250, for obtaining a skin graphic images to be measured and importing constructed skin pressure out
It is detected in power damage graphic images identification model, determines whether the skin graphic images to be measured are skin pressure
Power damages graphic images.
Wherein, described image cutting unit 220 includes:
Image tag mark module 2201, if for a certain original skin pressure damage graphic images and preset skin
For pressure injury graphic images when match degree is greater than the preset threshold, then marking label is 1;Conversely, then marking label is 0;
Image color processing module 2202, for being completed to all original skin pressure damage graphic images label labels
Afterwards, true color enhancing processing is carried out to the original skin pressure damage graphic images of each existing label;Wherein, institute
It states true color enhancing processing to remain unchanged for image color, then brightness of image enhances;
Image segmentation module 2203, the color image segmentation method for being clustered based on preset visual color, to each Zhang Zhen
Treated that original skin pressure damage graphic images carry out image dividing processing for colour enhancing, has been partitioned into each
There is the target area image of the original skin pressure damage graphic images of label.
The implementation of the embodiments of the present invention has the following beneficial effects:
The original skin pressure damage graphic images of present invention acquisition are gone forward side by side, and row label marks, image procossing and target area divide
After cutting, target area image characteristic is extracted by using machine learning method, image and temperature index are analyzed to come
The building of pressure injury infrared imaging ideograph is completed, realizes the identification to pressure injury graphic images, not only quickly
It is convenient, and accuracy is high, so that solving the prior art can not intuitively recognise that as pressure injury thermograph
The problem of picture.
It is worth noting that, included each unit is only drawn according to function logic in the above system embodiment
Point, but be not limited to the above division, as long as corresponding functions can be realized;In addition, each functional unit is specific
Title is also only for convenience of distinguishing each other, the protection scope being not intended to restrict the invention.
Those of ordinary skill in the art will appreciate that implement the method for the above embodiments be can be with
Relevant hardware is instructed to complete by program, the program can be stored in a computer readable storage medium,
The storage medium, such as ROM/RAM, disk, CD.
Above disclosed is only a preferred embodiment of the present invention, cannot limit the power of the present invention with this certainly
Sharp range, therefore equivalent changes made in accordance with the claims of the present invention, are still within the scope of the present invention.