CN110070113B - Training method and device for training set - Google Patents
Training method and device for training set Download PDFInfo
- Publication number
- CN110070113B CN110070113B CN201910252738.XA CN201910252738A CN110070113B CN 110070113 B CN110070113 B CN 110070113B CN 201910252738 A CN201910252738 A CN 201910252738A CN 110070113 B CN110070113 B CN 110070113B
- Authority
- CN
- China
- Prior art keywords
- training
- training set
- subset
- picture
- sample pictures
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the invention provides a training method and a device of a training set, wherein the method comprises the following steps: acquiring a training set for training a preset model; filtering the interference sample pictures from all sample pictures to construct the first training set, and reserving all first residual sample pictures except the interference sample pictures; filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture; acquiring the second target picture from all second residual sample pictures to construct the third training set; and respectively training the constructed first training set, the second training set and the third training set. The device performs the above method. The training method and the training device for the training set provided by the embodiment of the invention can improve the rationality of the construction of the training set, and further train the training set more reasonably.
Description
Technical Field
The embodiment of the invention relates to the technical field of picture processing, in particular to a training method and a training device for a training set.
Background
The capsule endoscopy has the advantages of no pain, no injury, large information amount of shot images and the like, and has wide application value.
The prior art adopts the manual mode discernment through the original picture of capsule scope shooting to classify original picture, in order to more accurately, high-efficiently discern original picture, need establish the model, but the model need train usually before using, need train the training set to the model can carry out the picture discernment more accurately, but, current training method to the training set, because the training set of establishing is reasonable inadequately, leads to the model accuracy after the training not high.
Therefore, it is an urgent problem to avoid the above-mentioned drawbacks, improve the rationality of training set construction, and train the training set more rationally.
Disclosure of Invention
Aiming at the problems in the prior art, the embodiment of the invention provides a training method and a training device for a training set.
The embodiment of the invention provides a training method of a training set, which comprises the following steps:
acquiring a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features;
filtering the interference sample pictures from all sample pictures to construct the first training set, and reserving all first residual sample pictures except the interference sample pictures;
filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture;
acquiring the second target picture from all second residual sample pictures to construct the third training set;
and respectively training the constructed first training set, the second training set and the third training set.
The embodiment of the invention provides a training device of a training set, which comprises:
the device comprises an acquisition unit, a processing unit and a control unit, wherein the acquisition unit is used for acquiring a training set used for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features;
a first constructing unit, configured to filter the interference sample pictures from all sample pictures to construct the first training set, and retain all first remaining sample pictures except the interference sample pictures;
a second constructing unit, configured to filter the first target picture from all first remaining sample pictures to construct the second training set, and retain all second remaining sample pictures except the interference sample picture and the first target picture;
a third constructing unit, configured to acquire the second target picture from all second remaining sample pictures to construct the third training set;
and the training unit is used for respectively training the constructed first training set, the second training set and the third training set.
An embodiment of the present invention provides an electronic device, including: a processor, a memory, and a bus, wherein,
the processor and the memory are communicated with each other through the bus;
the memory stores program instructions executable by the processor, the processor invoking the program instructions to perform a method comprising:
acquiring a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features;
filtering the interference sample pictures from all sample pictures to construct the first training set, and reserving all first residual sample pictures except the interference sample pictures;
filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture;
acquiring the second target picture from all second residual sample pictures to construct the third training set;
and respectively training the constructed first training set, the second training set and the third training set.
An embodiment of the present invention provides a non-transitory computer-readable storage medium, including:
the non-transitory computer readable storage medium stores computer instructions that cause the computer to perform a method comprising:
acquiring a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features;
filtering the interference sample pictures from all sample pictures to construct the first training set, and reserving all first residual sample pictures except the interference sample pictures;
filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture;
acquiring the second target picture from all second residual sample pictures to construct the third training set;
and respectively training the constructed first training set, the second training set and the third training set.
According to the training method and device for the training set, provided by the embodiment of the invention, the interference sample picture, the first target picture and the second target picture are sequentially filtered step by step, the corresponding training sets are constructed, and then the training sets are respectively trained, so that the rationality of the construction of the training sets can be improved, and the training sets are trained more reasonably.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a flowchart of an embodiment of a training method for a training set according to the present invention;
FIGS. 2(a) to 2(h) are all screenshots of full exposure pictures taken according to the embodiment of the present invention;
3(a) to 3(h) are all screenshots of a first target picture which is shot by the embodiment of the invention and has a shape change locally;
fig. 4(a) to 4(h) are all screenshots of a second target picture with raised erosion, which is taken according to an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of an embodiment of a training apparatus of the training set according to the present invention;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a flowchart of an embodiment of a training method of a training set according to the present invention, and as shown in fig. 1, the training method of the training set according to the embodiment of the present invention includes the following steps:
s101: acquiring a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the anomalous features include raised features and/or designated color features.
Specifically, the device acquires a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the anomalous features include raised features and/or designated color features. It should be noted that: the interference sample picture, the first target picture and the second target picture belong to the category of sample pictures, the sample pictures are selected from original pictures and can be used as training samples, the original pictures are shot through a capsule endoscope, and the working process of the capsule endoscope is described as follows:
the capsule endoscope enters the digestive tract from the oral cavity and is naturally discharged from the anus.
The battery of capsule endoscopy has limited endurance, and the effective working space is a part of the mouth, esophagus, stomach, duodenum, small intestine and large intestine.
Each activity of the capsule endoscope produces an in-field exam picture and an out-of-field exam picture.
The intra-field examination picture is a result of taking a certain section of the digestive tract.
The out-of-field inspection picture is a picture taken by the capsule endoscope in addition to the in-field inspection picture.
All pictures can be automatically identified without any human intervention (including image pre-processing).
After the images are identified, the images taken by the capsule endoscopy are divided into six major categories (125 minor categories) and automatically saved in 125 image folders, wherein the six major categories can be:
the first major category: one class of out-of-domain category labels (10 classes).
The second major category: class two out-of-domain category labels (13 categories).
The third major category: the tags (14 classes) are classified based on the first target picture of the local structural features.
The fourth major category: hole-structured first target picture classification tags (8 classes).
The fifth main category: the tags (24 classes) are classified based on the first target picture of the global structural features.
The sixth major class: the second target picture category label (56 categories).
It is possible to automatically recognize different parts of the digestive tract such as the oral cavity, the esophagus, the stomach, the duodenum, the small intestine, and the large intestine.
The number of the original pictures which can be shot by each capsule endoscope at each time can be 2000-3000, namely the number of the pictures which are acquired by the capsule endoscopes and concentrated.
Raw pictures taken of the capsule endoscopy (JPG format) can be derived from the hospital information system without any processing. The interference sample pictures can be understood as sample pictures which cannot be used for picture identification, and after the pictures are identified, the pictures need to be removed as early as possible, so that the operation amount in the process of training the preset model is reduced. It should be noted that: the interference sample picture may include a full exposure picture, fig. 2(a) to fig. 2(h) are all screenshots of the full exposure picture taken by the embodiment of the present invention, and the pictures are independent from each other and are respectively the representation forms of the full exposure picture. The anomalous features may include raised features and/or designated color features, and the raised features may include swollen, particulate raised features. The designated color characteristics may include red and white, and are not particularly limited. When the preset model output result comprises the second target picture, a special mark aiming at the abnormal feature can be generated, for example, the abnormal feature is selected by a box frame to indicate related personnel to carefully examine the box frame, namely, the abnormal feature can be used as an intermediate reference feature in some disease diagnosis processes, and the abnormal feature is not enough to diagnose the disease only by relying on the abnormal feature. The first target picture may include a first target picture with shape changes locally, and the content of the specific shape changes may include folds, cracks, interlaces, and the like, without specific limitation. Fig. 3(a) to 3(h) are all screenshots of the first target picture with the local shape change taken by the embodiment of the invention, and the screenshots are independent from each other and are respectively the representation forms of the first target picture with the local shape change. The second target pictures can include redness, swelling, erosion, ulcer and the like, and fig. 4(a) to 4(h) are all screenshots of the second target pictures with raised erosion taken by the embodiment of the present invention, and the pictures are independent from each other and are respectively representations of the second target pictures with raised erosion locally.
S102: and filtering the interference sample pictures from all the sample pictures to construct the first training set, and reserving all the first residual sample pictures except the interference sample pictures.
Specifically, the device filters the interference sample pictures from all sample pictures to construct the first training set, and retains all first remaining sample pictures except the interference sample pictures. Examples are as follows: all sample pictures including { A, B, C }, A, B, C are interference sample pictures, first target pictures, second target pictures, respectively, it is understood that each class in A, B, C is a set of pictures, this step filters a from the sample pictures, i.e., constructs a first training set containing a, and retains B, C as all first remaining sample pictures.
S103: and filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture.
Specifically, the device filters the first target picture from all the first remaining sample pictures to construct the second training set, and retains all the second remaining sample pictures except the interference sample picture and the first target picture. With reference to the above examples, the following are specifically described: all first remaining sample pictures include { B, C }, which filters out B from all first remaining sample pictures, i.e., constructs a second training set containing B, retains C, and serves as all second remaining sample pictures.
S104: and acquiring the second target picture from all the second residual sample pictures to construct the third training set.
Specifically, the device acquires the second target picture from all the second remaining sample pictures to construct the third training set. With reference to the above examples, the following are specifically described: all second remaining sample pictures include { C }, which is taken from all second remaining sample pictures, i.e., a third training set containing C is constructed. It should be noted that due to the characteristics of some pictures, such as confusability, etc., the second remaining sample picture may further include other types of pictures besides C, and by this step, the constructed third training set may include all the second target pictures as much as possible, but not include other types of pictures except C.
S105: and respectively training the constructed first training set, the second training set and the third training set.
Specifically, the device trains the constructed first training set, second training set and third training set respectively. The method for training the constructed training set is a mature technology in the field and is not described in detail.
According to the training method of the training set provided by the embodiment of the invention, the interference sample picture, the first target picture and the second target picture are sequentially filtered step by step, the corresponding training sets are constructed, and then the training sets are respectively trained, so that the rationality of the construction of the training sets can be improved, and the training sets are trained more reasonably.
On the basis of the above embodiment, the first training set includes a first training subset and a second training subset; correspondingly, the filtering the interference sample picture from all the sample pictures to construct the first training set includes:
splitting the first training set into the first training subset and the second training subset; the first training subset is a training subset corresponding to a class of out-of-domain classification labels, and the class of out-of-domain classification labels are determined based on shooting defects of an original picture and shooting parts irrelevant to a target part to be detected; the second training subset is a training subset corresponding to class II out-of-domain classification labels determined based on the original picture without medical judgment value, the original picture with the attached covering object and the original picture containing the digestion residues.
Specifically, the device splits the first training set into the first training subset and the second training subset; the first training subset is a training subset corresponding to a class of out-of-domain classification labels, and the class of out-of-domain classification labels are determined based on shooting defects of an original picture and shooting parts irrelevant to a target part to be detected; the second training subset is a training subset corresponding to class II out-of-domain classification labels determined based on the original picture without medical judgment value, the original picture with the attached covering object and the original picture containing the digestion residues. The first training subset and the second training subset respectively correspond to the first large class and the second large class, and the shooting defects may include full exposure pictures, full black pictures, half exposure pictures, local exposure pictures, structure-blurred pictures and detail-blurred pictures. The target site to be detected may be a stomach, and the photographing site may include a picture taken before an entrance of the capsule endoscope, a picture taken in an esophagus, an oral cavity picture, and an intestinal tract picture. The original picture without medical judgment value may include:
homogeneous whole images, namely the surface of the shot object is flat and smooth, no obvious texture exists, the color is uniform, although the shooting quality is high, the medical judgment value is lost due to the fact that the content is too single (the position, the angle, the organ carrier, the anatomical feature and the like of the shot object cannot be judged). The number of pictures is about 5.8% which is very high. Such pictures, although apparently not spam pictures, are not distinguished from "spam pictures" because they lose medical value. The subsequent treatment process can be completely omitted.
Waterline picture: namely the boundary line of air and water appearing in the picture, and the image structure is clear and simple. The part exposed in the air has the content similar to that of the homogenized image and has no medical value; the part submerged under the water surface is covered by the water film, valuable information is not exposed, so that the whole picture has no medical value and can be regarded as a 'garbage picture', and the number of the pictures is about 3.8%.
The covering in the original picture with the covering attached may comprise lumps of suspension, bubbles, mucous bodies, etc., which, because the photographic object is completely covered by the covering, results in such a picture being of no medical value.
Original picture of digestion residue: there is no clear food residue in the digestive tract, which is possible in the stomach and intestines, and the number of pictures is about 1%. In most cases, the coverage of the digestion residues is large, and the digestion residues occupy more than 50% of the area of the image, but as long as there is a place which is not covered, no abnormal feature is required to be ensured, so that the images guided by the category are all images without the abnormal feature, and the images can be classified into 'garbage images' and do not participate in subsequent processing.
And respectively filtering out interference sample pictures respectively corresponding to the first training subset and the second training subset from all sample pictures to respectively construct the first training subset and the second training subset.
Specifically, the device respectively filters interference sample pictures respectively corresponding to the first training subset and the second training subset from all sample pictures to respectively construct the first training subset and the second training subset. Referring to the above example, a is split into a1 and a2, corresponding to the first large class and the second large class, respectively, and the first training subset is constructed to correspond to the first large class a1 and the second training subset is constructed to correspond to the second large class a 2.
According to the training method of the training set provided by the embodiment of the invention, the first training subset and the second training subset are respectively constructed, so that the rationality of the construction of the training set can be further improved, and the training set can be trained more reasonably.
On the basis of the above embodiment, the second training set includes a third training subset, a fourth training subset, and a fifth training subset; correspondingly, the filtering out the first target picture from all the first remaining sample pictures to construct the second training set includes:
splitting the second training set into the third training subset, the fourth training subset, and the fifth training subset; the third training subset is a training subset corresponding to a first target picture classification label based on local structural features; the fourth training subset is a training subset corresponding to the first target picture classification label of the hole-shaped structure; the fifth training subset is a training subset corresponding to a first target picture classification label based on global structural features.
Specifically, the device splits the second training set into the third training subset, the fourth training subset, and the fifth training subset; the third training subset is a training subset corresponding to a first target picture classification label based on local structural features; the fourth training subset is a training subset corresponding to the first target picture classification label of the hole-shaped structure; the fifth training subset is a training subset corresponding to a first target picture classification label based on global structural features. The first target picture based on the local structural feature may include:
the first target picture of the edge defect, namely the picture of the edge defect, is usually the result of taking a picture of the side surface of the hole-shaped structure, most picture frames have simple structures, and only the edge part has a defective hole-shaped or semi-radial structure. Since many abnormal features also appear in this area, this category of pictures has a strong contrast effect.
Simple linear structure: that is, the picture only contains 1 to 2 short prismatic structures, the picture has a small amount of shaded areas, and the rest parts are smooth and have no obvious texture. Many pictures containing abnormal features also have similar background structures, so this category of pictures can be compared well with the abnormal feature pictures.
The first target picture of the hole-shaped structure, namely the shot object, is of the hole-shaped structure and can be divided into a big hole structure and a small hole structure according to the size of the hole.
The first target picture based on the global structural feature may include:
the stomach corner structure, i.e. the stomach corner is the structure formed by the lesser curvature of the stomach at the lowest end, generally presenting a 90 ° corner, which is the boundary between the stomach body and the pyloric part at the lesser curvature of the stomach. The surface of the stomach corner often has thread-like folds as the stomach wall contracts, and the folds disappear as the stomach wall relaxes. Therefore, the picture in the special position is classified into a category.
Dense texture structures, i.e. structures of the stomach wall taken in a long-range view, which are usually taken when the inner wall is contracted, present densely arranged curved textures, and some structural information is also superimposed on the textures, so that the background of the whole picture becomes very complicated, and it becomes extremely difficult to find and identify abnormal features on the background of such pictures. The number of pictures is about 4.5%.
Respectively filtering out first target pictures respectively corresponding to the third training subset, the fourth training subset and the fifth training subset from all first remaining sample pictures to respectively construct the third training subset, the fourth training subset and the fifth training subset.
Specifically, the apparatus respectively filters out first target pictures respectively corresponding to the third training subset, the fourth training subset, and the fifth training subset from all first remaining sample pictures to respectively construct the third training subset, the fourth training subset, and the fifth training subset. Referring to the above example, B is split into B1, B2, and B3, which correspond to the third to fifth broad classes, respectively, the constructed third training subset corresponds to the third broad class B1, the constructed fourth training subset corresponds to the fourth broad class B2, and the constructed fifth training subset corresponds to the fifth broad class B3.
According to the training method of the training set provided by the embodiment of the invention, the third training subset to the fifth training subset are respectively constructed, so that the rationality of the construction of the training set can be further improved, and the training set can be trained more reasonably.
On the basis of the above embodiment, the training of the constructed first training set, second training set, and third training set respectively includes:
training the constructed first, second, third, fourth, fifth and third training subsets, respectively.
Specifically, the apparatus trains the first training subset, the second training subset, the third training subset, the fourth training subset, the fifth training subset, and the third training set that have been constructed, respectively. The method for training the constructed training set is a mature technology in the field and is not described in detail.
According to the training method of the training set provided by the embodiment of the invention, the rationality of the construction of the training set can be further improved by respectively training the first training subset to the fifth training subset and the third training set, so that the training set is trained more reasonably.
On the basis of the above embodiment, the method further includes:
acquiring training durations respectively corresponding to the first training subset, the second training subset, the third training subset, the fourth training subset, the fifth training subset and the third training set.
Specifically, the device obtains training durations respectively corresponding to the first training subset, the second training subset, the third training subset, the fourth training subset, the fifth training subset, and the third training set. The training duration may be understood as the duration from the training start time to the training completion time.
And splitting the target training set corresponding to the target training time length reaching the preset time length if judging and knowing that at least one training time length reaches the preset time length, so that the training time length corresponding to the split target training set is smaller than the preset time length.
Specifically, if the device judges that at least one training time length reaches the preset time length, splitting a target training set corresponding to the target training time length reaching the preset time length so that the training time length corresponding to the split target training set is smaller than the preset time length. The preset duration can be set independently according to actual conditions. For the case that there is a training duration reaching the preset duration, for example: only when the training duration T1 of the first training subset reaches the preset duration, the target training set is the first training subset, and then the first training subset is split, and the specific splitting manner is not specifically limited. For the case that a plurality of training durations reach the preset duration, for example, two training durations are used, that is, only the training duration T1 of the first training subset and the training duration T2 of the second training subset reach the preset duration, the target training set is the first training subset and the second training subset, and then the first training subset and the second training subset are respectively split.
According to the training method of the training set provided by the embodiment of the invention, the training set with too long training time is split, so that the reasonability of the construction of the training set can be further improved, and the training set is trained more reasonably.
Fig. 5 is a schematic structural diagram of an embodiment of a training apparatus of a training set according to the present invention, and as shown in fig. 5, an embodiment of the present invention provides a training apparatus of a training set, which includes an obtaining unit 501, a first constructing unit 502, a second constructing unit 503, a third constructing unit 504, and a training unit 505, where:
the obtaining unit 501 is configured to obtain a training set used for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features; the first constructing unit 502 is configured to filter the interference sample pictures from all sample pictures to construct the first training set, and retain all first remaining sample pictures except the interference sample pictures; the second constructing unit 503 is configured to filter the first target picture from all the first remaining sample pictures to construct the second training set, and retain all the second remaining sample pictures except the interference sample picture and the first target picture; the third constructing unit 504 is configured to obtain the second target picture from all the second remaining sample pictures to construct the third training set; the training unit 505 is configured to train the constructed first training set, the second training set, and the third training set, respectively.
Specifically, the obtaining unit 501 is configured to obtain a training set used for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features; the first constructing unit 502 is configured to filter the interference sample pictures from all sample pictures to construct the first training set, and retain all first remaining sample pictures except the interference sample pictures; the second constructing unit 503 is configured to filter the first target picture from all the first remaining sample pictures to construct the second training set, and retain all the second remaining sample pictures except the interference sample picture and the first target picture; the third constructing unit 504 is configured to obtain the second target picture from all the second remaining sample pictures to construct the third training set; the training unit 505 is configured to train the constructed first training set, the second training set, and the third training set, respectively.
According to the training device for the training set, provided by the embodiment of the invention, the interference sample picture, the first target picture and the second target picture are sequentially filtered step by step, the corresponding training sets are constructed, and then the training sets are respectively trained, so that the rationality of construction of the training sets can be improved, and the training sets are trained more reasonably.
The training device of the training set provided in the embodiment of the present invention may be specifically configured to execute the processing flows of the above method embodiments, and its functions are not described herein again, and refer to the detailed description of the above method embodiments.
Fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, and as shown in fig. 6, the electronic device includes: a processor (processor)601, a memory (memory)602, and a bus 603;
the processor 601 and the memory 602 complete mutual communication through a bus 603;
the processor 601 is configured to call program instructions in the memory 602 to perform the methods provided by the above-mentioned method embodiments, for example, including: acquiring a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features; filtering the interference sample pictures from all sample pictures to construct the first training set, and reserving all first residual sample pictures except the interference sample pictures; filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture; acquiring the second target picture from all second residual sample pictures to construct the third training set; and respectively training the constructed first training set, the second training set and the third training set.
The present embodiment discloses a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, enable the computer to perform the method provided by the above-mentioned method embodiments, for example, comprising: acquiring a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features; filtering the interference sample pictures from all sample pictures to construct the first training set, and reserving all first residual sample pictures except the interference sample pictures; filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture; acquiring the second target picture from all second residual sample pictures to construct the third training set; and respectively training the constructed first training set, the second training set and the third training set.
The present embodiments provide a non-transitory computer-readable storage medium storing computer instructions that cause the computer to perform the methods provided by the above method embodiments, for example, including: acquiring a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features; filtering the interference sample pictures from all sample pictures to construct the first training set, and reserving all first residual sample pictures except the interference sample pictures; filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture; acquiring the second target picture from all second residual sample pictures to construct the third training set; and respectively training the constructed first training set, the second training set and the third training set.
Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present 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 solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (7)
1. A training method of a training set is used for a classification and identification process of a picture taken by a capsule endoscope, and is characterized by comprising the following steps:
acquiring a training set for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features;
filtering the interference sample pictures from all sample pictures to construct the first training set, and reserving all first residual sample pictures except the interference sample pictures;
filtering the first target picture from all the first residual sample pictures to construct the second training set, and reserving all the second residual sample pictures except the interference sample picture and the first target picture;
acquiring the second target picture from all second residual sample pictures to construct the third training set;
respectively training the constructed first training set, the second training set and the third training set;
the first training set comprises a first training subset and a second training subset, the second training set comprises a third training subset, a fourth training subset, and a fifth training subset; acquiring training durations respectively corresponding to the first training subset, the second training subset, the third training subset, the fourth training subset, the fifth training subset and the third training set;
and splitting the target training set corresponding to the target training time length reaching the preset time length if judging and knowing that at least one training time length reaches the preset time length, so that the training time length corresponding to the split target training set is smaller than the preset time length.
2. The method of claim 1, wherein the filtering out the interference sample picture from all sample pictures to construct the first training set comprises:
splitting the first training set into the first training subset and the second training subset; the first training subset is a training subset corresponding to a class of out-of-domain classification labels, and the class of out-of-domain classification labels are determined based on shooting defects of an original picture and shooting parts irrelevant to a target part to be detected; the second training subset is a training subset corresponding to a class II out-of-domain classification label, and the class II out-of-domain classification label is determined based on an original picture without medical judgment value, an original picture attached with a covering object and an original picture containing a digestion residue object;
and respectively filtering out interference sample pictures respectively corresponding to the first training subset and the second training subset from all sample pictures to respectively construct the first training subset and the second training subset.
3. The method of claim 2, wherein the filtering out the first target picture from all first remaining sample pictures to construct the second training set comprises:
splitting the second training set into the third training subset, the fourth training subset, and the fifth training subset; the third training subset is a training subset corresponding to a first target picture classification label based on local structural features; the fourth training subset is a training subset corresponding to the first target picture classification label of the hole-shaped structure; the fifth training subset is a training subset corresponding to a first target picture classification label based on global structural features;
respectively filtering out first target pictures respectively corresponding to the third training subset, the fourth training subset and the fifth training subset from all first remaining sample pictures to respectively construct the third training subset, the fourth training subset and the fifth training subset.
4. The method of claim 3, wherein the training the constructed first training set, second training set, and third training set respectively comprises:
training the constructed first, second, third, fourth, fifth and third training subsets, respectively.
5. The utility model provides a training set's trainer for to the categorised recognition process of the picture of taking of capsule scope, its characterized in that includes:
the device comprises an acquisition unit, a processing unit and a control unit, wherein the acquisition unit is used for acquiring a training set used for training a preset model; the training sets comprise a first training set containing an interference sample picture, a second training set corresponding to a first target picture of which the outer surface of the shot does not contain abnormal features, and a third training set corresponding to a second target picture of which the outer surface of the shot contains the abnormal features; the abnormal features comprise raised features and/or designated color features;
a first constructing unit, configured to filter the interference sample pictures from all sample pictures to construct the first training set, and retain all first remaining sample pictures except the interference sample pictures;
a second constructing unit, configured to filter the first target picture from all first remaining sample pictures to construct the second training set, and retain all second remaining sample pictures except the interference sample picture and the first target picture;
a third constructing unit, configured to acquire the second target picture from all second remaining sample pictures to construct the third training set;
the training unit is used for respectively training the constructed first training set, the second training set and the third training set;
the first training set comprises a first training subset and a second training subset, the second training set comprises a third training subset, a fourth training subset, and a fifth training subset; acquiring training durations respectively corresponding to the first training subset, the second training subset, the third training subset, the fourth training subset, the fifth training subset and the third training set;
and splitting the target training set corresponding to the target training time length reaching the preset time length if judging and knowing that at least one training time length reaches the preset time length, so that the training time length corresponding to the split target training set is smaller than the preset time length.
6. An electronic device, comprising: a processor, a memory, and a bus, wherein,
the processor and the memory are communicated with each other through the bus;
the memory stores program instructions executable by the processor, the processor invoking the program instructions to perform the method of any of claims 1 to 4.
7. A non-transitory computer-readable storage medium storing computer instructions that cause a computer to perform the method of any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910252738.XA CN110070113B (en) | 2019-03-29 | 2019-03-29 | Training method and device for training set |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910252738.XA CN110070113B (en) | 2019-03-29 | 2019-03-29 | Training method and device for training set |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110070113A CN110070113A (en) | 2019-07-30 |
| CN110070113B true CN110070113B (en) | 2021-03-30 |
Family
ID=67366818
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910252738.XA Active CN110070113B (en) | 2019-03-29 | 2019-03-29 | Training method and device for training set |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110070113B (en) |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104809465A (en) * | 2014-01-23 | 2015-07-29 | 北京三星通信技术研究有限公司 | Classifier training method, target detection, segmentation or classification method and target detection, segmentation or classification device |
| CN105488453A (en) * | 2015-11-30 | 2016-04-13 | 杭州全实鹰科技有限公司 | Detection identification method of no-seat-belt-fastening behavior of driver based on image processing |
| CN106097340A (en) * | 2016-06-12 | 2016-11-09 | 山东大学 | A kind of method automatically detecting and delineating Lung neoplasm position based on convolution grader |
| CN106778583A (en) * | 2016-12-07 | 2017-05-31 | 北京理工大学 | Vehicle attribute recognition methods and device based on convolutional neural networks |
| CN107103187A (en) * | 2017-04-10 | 2017-08-29 | 四川省肿瘤医院 | The method and system of Lung neoplasm detection classification and management based on deep learning |
| CN107833219A (en) * | 2017-11-28 | 2018-03-23 | 腾讯科技（深圳）有限公司 | Image-recognizing method and device |
| CN107909572A (en) * | 2017-11-17 | 2018-04-13 | 合肥工业大学 | Pulmonary nodule detection method and system based on image enhancement |
| CN107945875A (en) * | 2017-11-17 | 2018-04-20 | 合肥工业大学 | Pulmonary nodule detection method and system based on data enhancing |
| CN107977963A (en) * | 2017-11-30 | 2018-05-01 | 北京青燕祥云科技有限公司 | Decision method, device and the realization device of Lung neoplasm |
| CN108133476A (en) * | 2017-12-26 | 2018-06-08 | 安徽科大讯飞医疗信息技术有限公司 | A kind of Lung neoplasm automatic testing method and system |
| CN108364006A (en) * | 2018-01-17 | 2018-08-03 | 超凡影像科技股份有限公司 | Medical Images Classification device and its construction method based on multi-mode deep learning |
| CN108664924A (en) * | 2018-05-10 | 2018-10-16 | 东南大学 | A kind of multi-tag object identification method based on convolutional neural networks |
| CN109086716A (en) * | 2018-08-01 | 2018-12-25 | 北京嘀嘀无限科技发展有限公司 | A kind of method and device of seatbelt wearing detection |
| CN109508741A (en) * | 2018-11-09 | 2019-03-22 | 哈尔滨工业大学 | Method based on deep learning screening training set |
-
2019
- 2019-03-29 CN CN201910252738.XA patent/CN110070113B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104809465A (en) * | 2014-01-23 | 2015-07-29 | 北京三星通信技术研究有限公司 | Classifier training method, target detection, segmentation or classification method and target detection, segmentation or classification device |
| CN105488453A (en) * | 2015-11-30 | 2016-04-13 | 杭州全实鹰科技有限公司 | Detection identification method of no-seat-belt-fastening behavior of driver based on image processing |
| CN106097340A (en) * | 2016-06-12 | 2016-11-09 | 山东大学 | A kind of method automatically detecting and delineating Lung neoplasm position based on convolution grader |
| CN106778583A (en) * | 2016-12-07 | 2017-05-31 | 北京理工大学 | Vehicle attribute recognition methods and device based on convolutional neural networks |
| CN107103187A (en) * | 2017-04-10 | 2017-08-29 | 四川省肿瘤医院 | The method and system of Lung neoplasm detection classification and management based on deep learning |
| CN107909572A (en) * | 2017-11-17 | 2018-04-13 | 合肥工业大学 | Pulmonary nodule detection method and system based on image enhancement |
| CN107945875A (en) * | 2017-11-17 | 2018-04-20 | 合肥工业大学 | Pulmonary nodule detection method and system based on data enhancing |
| CN107833219A (en) * | 2017-11-28 | 2018-03-23 | 腾讯科技（深圳）有限公司 | Image-recognizing method and device |
| CN107977963A (en) * | 2017-11-30 | 2018-05-01 | 北京青燕祥云科技有限公司 | Decision method, device and the realization device of Lung neoplasm |
| CN108133476A (en) * | 2017-12-26 | 2018-06-08 | 安徽科大讯飞医疗信息技术有限公司 | A kind of Lung neoplasm automatic testing method and system |
| CN108364006A (en) * | 2018-01-17 | 2018-08-03 | 超凡影像科技股份有限公司 | Medical Images Classification device and its construction method based on multi-mode deep learning |
| CN108664924A (en) * | 2018-05-10 | 2018-10-16 | 东南大学 | A kind of multi-tag object identification method based on convolutional neural networks |
| CN109086716A (en) * | 2018-08-01 | 2018-12-25 | 北京嘀嘀无限科技发展有限公司 | A kind of method and device of seatbelt wearing detection |
| CN109508741A (en) * | 2018-11-09 | 2019-03-22 | 哈尔滨工业大学 | Method based on deep learning screening training set |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110070113A (en) | 2019-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110232383B (en) | Focus image recognition method and focus image recognition system based on deep learning model | |
| CN113743384B (en) | Stomach picture identification method and device | |
| CN112184699A (en) | Aquatic product health detection method, terminal device and storage medium | |
| CN113129287A (en) | Automatic lesion mapping method for upper gastrointestinal endoscope image | |
| CN108615045B (en) | Method, device and equipment for screening images shot by capsule endoscopy | |
| CN112232977A (en) | Aquatic product cultivation evaluation method, terminal device and storage medium | |
| CN114842000A (en) | Endoscope image quality evaluation method and system | |
| CN113052843B (en) | Method, apparatus, system, storage medium and computing device for assisting endoscopy | |
| CN113762422B (en) | Image training set composition method and system | |
| CN111062929A (en) | Intelligent analysis and diagnosis system for livestock and poultry disease pictures through software design | |
| CN110110750B (en) | Original picture classification method and device | |
| CN110070113B (en) | Training method and device for training set | |
| CN116596927B (en) | Endoscope video processing method, system and device | |
| CN109886916B (en) | Capsule mirror image screening method and device | |
| CN110083727B (en) | Method and device for determining classification label | |
| CN110097082B (en) | Splitting method and device of training set | |
| CN110084276B (en) | Splitting method and device of training set | |
| CN110084277B (en) | Splitting method and device of training set | |
| CN115797729A (en) | Model training method and device, and motion artifact identification and prompting method and device | |
| CN110084278B (en) | Splitting method and device of training set | |
| CN110084280B (en) | Method and device for determining classification label | |
| CN109993226B (en) | Splitting method and device of training set | |
| CN114581402A (en) | Capsule endoscope quality inspection method, device and storage medium | |
| CN113920355A (en) | Part type identification method and inspection quality monitoring system | |
| CN110097081A (en) | A kind of construction method and device of training set |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |