CN117710285A - Cervical lesion cell mass detection method and system based on self-adaptive feature extraction - Google Patents
Cervical lesion cell mass detection method and system based on self-adaptive feature extraction Download PDFInfo
- Publication number
- CN117710285A CN117710285A CN202311370758.XA CN202311370758A CN117710285A CN 117710285 A CN117710285 A CN 117710285A CN 202311370758 A CN202311370758 A CN 202311370758A CN 117710285 A CN117710285 A CN 117710285A
- Authority
- CN
- China
- Prior art keywords
- adaptive
- network
- yolov7
- self
- cell mass
- 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.)
- Granted
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 46
- 230000003902 lesion Effects 0.000 title claims abstract description 46
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000000105 evaporative light scattering detection Methods 0.000 title claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 56
- 238000012360 testing method Methods 0.000 claims abstract description 29
- 238000001514 detection method Methods 0.000 claims abstract description 22
- 230000003044 adaptive effect Effects 0.000 claims description 34
- 238000010276 construction Methods 0.000 claims description 14
- 238000010586 diagram Methods 0.000 claims description 10
- 238000007781 pre-processing Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 7
- 238000004806 packaging method and process Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 230000003213 activating effect Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 27
- 206010008342 Cervix carcinoma Diseases 0.000 description 6
- 208000006105 Uterine Cervical Neoplasms Diseases 0.000 description 6
- 201000010881 cervical cancer Diseases 0.000 description 6
- 238000004590 computer program Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 5
- 238000012216 screening Methods 0.000 description 5
- 230000006978 adaptation Effects 0.000 description 3
- 230000000052 comparative effect Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000002679 ablation Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 239000008188 pellet Substances 0.000 description 2
- 206010028980 Neoplasm Diseases 0.000 description 1
- 208000017442 Retinal disease Diseases 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000002380 cytological effect Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000013399 early diagnosis Methods 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30024—Cell structures in vitro; Tissue sections in vitro
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of medical images, and particularly relates to a cervical lesion cell mass detection method and system based on self-adaptive feature extraction, wherein a test set and a training set of the cervical lesion cell mass are firstly obtained; then constructing a grouping self-adaptive feature extraction network and constructing a dynamic focusing loss function of which the angle influences the center distance; the network and the dynamic focus loss function are extracted based on the self-adaptive characteristics of the packets, and an improved yolov7 network is constructed; then, the training set is preprocessed and then is input into an improved yolov7 network, and a training result is output; then taking the test set as an input parameter of the improved yolov7 network model, outputting a test result, comparing the test result with a training result, and reserving the optimal improved yolov7 network model; and finally, detecting cervical lesion cell masses by using an optimal improved yolov7 network model, and reserving a detection result picture. The invention can solve the problems of insufficient detection precision and large hardware facility requirements of the existing cervical lesion cell mass.
Description
Technical Field
The invention belongs to the technical field of medical images, and particularly relates to a cervical lesion cell mass detection method and system based on self-adaptive feature extraction.
Background
At present, the main method for preventing and treating cervical cancer is early screening, early diagnosis and early treatment.
The early screening of cervical cancer aims to detect precancerous lesions of cervical cancer, treat the precancerous lesions of cervical cancer in time and reduce risks and mortality of cervical cancer lesions in middle and late stages. Liquid-based cytology examination (TCT) is the primary screening method in the early screening method of cervical cancer. The method requires cytologists to slice and dye the dropped cervical cells, and finally, the primary diagnosis result is obtained by observing the sample under a microscope according to the slight difference between the cervical lesion cells and the normal cells in morphology and structure. However, manual interpretation of thousands of cells detected in a billion-level pixelated cytological image slide is often time consuming and error prone.
With the important breakthrough of deep learning in the image field, the method has excellent performance in various fields. It is also embodied in medicine, including classification of skin malignant tumors, image diagnosis of tumors, detection and classification of retinal diseases, and the like. Aiming at a series of problems existing in manual interpretation, the use of deep learning to assist in interpreting cervical cell medical images has great potential. The method can lighten the workload of doctors, reduce the influence of subjective emotion, improve the working efficiency and improve the reliability of screening results. However, deep learning is still in an exploration stage in cervical image research at present, the detection accuracy of cervical lesion cells needs to be improved, the requirements on hardware facilities are larger, and meanwhile, the problem that the appearance difference is larger due to consistent lesion types needs to be solved.
Disclosure of Invention
The invention aims to provide a cervical lesion cell mass detection method and a cervical lesion cell mass detection system based on self-adaptive feature extraction, so as to solve the problems of insufficient detection precision and large hardware facility requirements of the existing cervical lesion cell mass detection.
The basic scheme provided by the invention is as follows: the cervical lesion cell mass detection method based on the self-adaptive feature extraction comprises the following steps:
s1: obtaining a data set of cervical lesion cell mass, the data set comprising a test set and a training set;
s2: constructing a grouping self-adaptive feature extraction network and constructing a dynamic focusing loss function of which the angle influences the center distance; the network and the dynamic focus loss function are extracted based on the self-adaptive characteristics of the packets, and an improved yolov7 network is constructed;
s3: preprocessing a training set, inputting the preprocessed training set into an improved yolov7 network, and outputting a training result;
s4: taking the test set as an input parameter of the improved yolov7 network, outputting a test result, comparing the test result with a training result, and reserving an optimal improved yolov7 network model;
s5: and detecting cervical lesion cell mass by using an optimal improved yolov7 network model, and reserving a detection result picture.
Further, in the step S2, constructing a packet adaptive feature extraction network specifically includes:
the input feature map (W×H×C) 1 ) Press channel (C) 1 ) Is divided into two groups to generate two groups of characteristic diagrams (W×H×C 1 /2)；
Two sets of feature maps (W X H X C) 1 2) obtaining adaptive feature maps (W x H x C) at two scales by deformable convolution with a hole of 6 and a hole of 12, respectively 2 /2)；
The two adaptive feature maps are spliced to obtain a locally adaptive feature map (W×H×C) 2 )；
Will input a feature map (W x H x C) 1 ) By adaptive averaging, the feature map of each channel is compressed to 1×1 to obtain a global feature (1×1×c 1 ) Then the global features (1×1×c) obtained by a 1×1 convolution pair 1 ) Further extraction to obtain new global features (1×1×C 2 ) And obtaining final global features (1×1×C) by SiLU activation function 2 )；
The obtained local adaptive feature map (W×H×C 2 ) And global features (1 x C) 2 ) Using Add operation connections, a final global adaptive feature map (w×h×c 2 )。
Further, the dynamic focusing loss function of the construction angle influencing the center distance is specifically:
the angle component is used for constructing an influence factor of the angle, the included angle between the central line and the x axis is alpha, if the angle alpha is less than or equal to pi/4, the central line is deviated to the x axis, otherwise pi/2-alpha is adopted, the central line is deviated to the y axis, and the angle component can be expressed as follows:
human = 1-2sin 2 (arcsin(x)-π/4)
Packaging the angle component by an angle factor θ, which can be expressed as:
θ＝Λ+1
bringing an angle factor into WIOU v3 The TIOU loss function can be expressed as:
Δ＝((x-x gt ) 2 +(y-y gt ) 2 )/(W 2 +H 2 )
wherein IOU represents the cross-over ratio between the predicted and real frames,representing the mean of the IOU; delta and epsilon are constants, delta represents the distance between the center points, (x, y) represents the coordinates of the center points of the prediction frame, (x) gt ，y gt ) Representing the coordinates of the center point of the real frame; w and H represent the width and height of the minimum closure.
Further, the preprocessing in S3 includes preprocessing the training set using the method of mosaics data enhancement and multi-scale training.
A cervical lesion cell mass detection system based on adaptive feature extraction, comprising:
the data acquisition module is used for acquiring a data set of cervical lesion cell masses, wherein the data set comprises a test set and a training set;
the model construction module is used for constructing a grouping self-adaptive feature extraction network and constructing a dynamic focusing loss function of which the angle influences the center distance; extracting a network and a dynamic focus loss function based on the self-adaptive characteristics of the packet, and constructing an improved yolov7 network model;
the model training module is used for inputting the training set after pretreatment to the improved yolov7 network and outputting a training result;
the model optimization module is used for taking the test set as an input parameter of the improved yolov7 network, outputting a test result, comparing the test result with a training result, and reserving an optimal improved yolov7 network model;
and the image detection module is used for detecting the cervical lesion cell mass image by using the optimal improved yolov7 network model, and retaining a detection result picture.
The model building module comprises a local grouping self-adaptive module and a global bias module, wherein the local self-adaptive module is used for grouping the feature images, then the features on different scales are obtained through deformable convolution with a cavity of 6 and a cavity of 12 respectively, and finally the features on different scales are spliced to obtain local self-adaptive features; the global bias module is used for carrying out self-adaptive average pooling on the input feature map to obtain a global feature, extracting further features of the global feature through a convolution of 1×1, and then activating the global feature by using an activation function.
Further, the dynamic focusing loss function of the construction angle influencing the center distance is specifically:
determining the included angle between the central line and the x axis as alpha, if the angle alpha is less than or equal to pi/4, taking alpha to indicate that the central line is biased towards the x axis, otherwise taking pi/2-alpha to indicate that the central line is biased towards the y axis, and the angle component can be expressed as follows:
Λ＝1-2sin 2 (arcsin(x)-π/4)
packaging the angle component by an angle factor θ, which can be expressed as:
θ＝Λ+1
bringing an angle factor into WIOU v3 The TIOU loss function can be expressed as:
Δ＝((x-x gt ) 2 +(y-y gt ) 2 )/(W 2 +H 2 )
wherein IOU represents the cross-over ratio between the predicted and real frames,representing the mean of the IOU; delta and epsilon are constants, delta represents the distance between the center points, (x, y) represents the coordinates of the center points of the prediction frame, (x) gt ，y gt ) Representing the coordinates of the center point of the real frame; w and H represent the width and height of the minimum closure.
Furthermore, the model training module pre-processes the training set, specifically, the training set is pre-processed by using a method of Mosaic data enhancement and multi-scale training.
The principle and the advantages of the invention are as follows: in the application, when image data of cervical lesion cell masses are detected, the detection is carried out by relying on an improved yolov7 network, a grouping self-adaptive feature extraction network and a dynamic focusing loss function are constructed in the improved yolov7 network, and local features and global features can be effectively and adaptively extracted in the grouping self-adaptive feature extraction network, so that the average precision of a model is improved while model parameters are reduced; in the dynamic focusing loss function, the aspect ratio is not used under the condition of considering the coincidence ratio of the prediction frame and the real frame, the center point distance and the angle relation, the generalization capability of the model is improved, and the average precision of the model is improved.
Meanwhile, in the construction process of the grouping self-adaptive feature extraction network, a model scaling module with residual errors is integrated into the feature extraction network, namely, the number of channels is reduced after stacking connection is changed into residual error connection, and model parameters are reduced under the condition that average model accuracy is not reduced, so that hardware requirements are reduced.
Therefore, the technical scheme of the application can effectively improve the precision and accuracy of detection and reduce the requirements of hardware facilities when being used for image detection of cervical lesion cell masses.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention;
FIG. 2 is a flow chart of a network model according to an embodiment of the present invention;
FIG. 3 is a block diagram of a packet-adaptive feature extraction network in accordance with an embodiment of the present invention;
FIG. 4 is an analysis diagram of a prediction block and a real block according to an embodiment of the present invention;
FIG. 5 is a diagram of a model structure of an improved yolov7 network in accordance with an embodiment of the present invention;
FIG. 6 is a graph comparing features extracted by a packet-adaptive feature extraction network according to an embodiment of the present invention;
FIG. 7 is a graph of the effect of using a modified yolov7 network in accordance with an embodiment of the present invention;
FIG. 8 is a functional block diagram of a system according to an embodiment of the present invention;
fig. 9 is a block diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The following is a further detailed description of the embodiments:
the labels in the drawings of this specification include: an electronic device 400, a processor 401, a memory 402, an input means 403, an output means 404.
An example is substantially as shown in figures 1 and 2: the cervical lesion cell mass detection method based on the self-adaptive feature extraction comprises the following steps:
s1: obtaining a data set of cervical lesion cell mass, the data set comprising a test set and a training set; in the application, the cell mass comprises cells and masses, the acquired test set and training set are used for constructing and training a model, after an optimal model is generated, the actual operation is performed, and the image data set to be detected of the cervical lesion cell mass is also acquired.
S2: constructing a grouping self-adaptive feature extraction network and constructing a dynamic focusing loss function of which the angle influences the center distance; the network and the dynamic focus loss function are extracted based on the self-adaptive characteristics of the packets, and an improved yolov7 network is constructed;
in the application, a grouping self-adaptive feature extraction network is constructed, and a model scaling idea is combined, and a residual structure originally used for feature fusion is used for feature extraction, wherein the specific construction process is as follows:
the input feature map (W×H×C) 1 ) Press channel (C) 1 ) Is divided into two groups to generate two groups of characteristic diagrams (W×H×C 1 /2)；
Two sets of feature maps (W X H X C) 1 2) obtaining adaptive feature maps (W x H x C) at two scales by deformable convolution with a hole of 6 and a hole of 12, respectively 2 /2)；
The two adaptive feature maps are spliced to obtain a locally adaptive feature map (W×H×C) 2 )；
Will input a feature map (W x H x C) 1 ) By adaptive averaging, the feature map of each channel is compressed to 1×1 to obtain a global feature (1×1×c 1 ) Then the global features (1×1×c) obtained by a 1×1 convolution pair 1 ) Further extraction to obtain new global features (1×1×C 2 ) And obtaining final global features (1×1×C) by SiLU activation function 2 )；
The obtained local adaptive feature map (W×H×C 2 ) And global features (1 x C) 2 ) Using Add operation connections, a final global adaptive feature map (w×h×c 2 )。
In table 1 below, the present application devised a set of comparative experiments to verify the performance of model scaling modules with residuals in feature extraction, specifically:
table 1 model scaling with residuals results for feature extraction
In the experimental results of table 1, model scaling using residuals was used for feature extraction reduced by 4.1m compared to the original model parameters, with an increase of 0.2% for map, while model scaling using residuals was used for feature extraction reduced by 4.1m compared to the original model parameters, with an increase of 0.4% for map after using transfer learning.
Based on the above performance, the process of constructing the packet adaptive feature extraction network can be expressed as:
firstly, combining the ideas of grouping convolution, deformable convolution and cavity convolution, providing a local grouping self-adaptive module, and then introducing a global bias module for acquiring global features; finally, the local grouping self-adaptive module and the global bias module are combined to form a complete grouping self-adaptive module, and the complete grouping self-adaptive module is shown in fig. 3.
The local self-adaptive module groups the feature graphs firstly, the feature graphs are divided into two groups according to the channel in the application, and then the features on different scales are obtained through deformable convolution with a cavity of 6 and a cavity of 12 respectively. And then splicing the two features to obtain the local self-adaptive feature.
The global bias module carries out self-adaptive average pooling on the input feature map to obtain a global feature, extracts the further feature of the global feature through a convolution of 1 multiplied by 1, then uses an activation function to activate, and increases the nonlinear expression capacity of the global feature.
In table 2, we designed a set of ablation experiments to analyze the positional impact of the packet adaptation module in a model scaling network with residuals, with specific positions as shown in fig. 5. Experimental results show that the parameter amount is minimum 30.68M and mAP is 63.8% when the packet adaptation module is placed at the third convolution. The parameter number of the grouping self-adaptive module is 30.72M when the grouping self-adaptive module is placed at the 6 th convolution, the parameter number is slightly higher than the minimum parameter number, mAP is 64.4 percent and is 0.6 percent higher than the third position, so that the analysis shows that the grouping self-adaptive module achieves the best effect when the grouping self-adaptive module is placed at the 6 th convolution.
Table 2 results of position ablation experiments of packet adaptation modules in model scaling networks with residuals
In addition, to more intuitively feel the effectiveness of the grouping adaptive module, we visualize the extracted feature map, as shown in fig. 6, where a in fig. 6 represents the original label picture of the cervical cell mass, b represents the model thermodynamic diagram of the baseline yolov7, and c represents the model thermodynamic diagram of the modified yolov 7.
After the construction of the packet adaptive feature extraction network is completed, a dynamic focusing loss function of which the angle influences the center distance is then constructed, wherein the dynamic focusing loss function is used for considering the influence on the result of the target detection algorithm, as shown in fig. 4, specifically:
the angle component is used for constructing an influence factor of the angle, the included angle between the central line and the x axis is alpha, if the angle alpha is less than or equal to pi/4, the central line is deviated to the x axis, otherwise pi/2-alpha is adopted, the central line is deviated to the y axis, and the angle component can be expressed as follows:
Λ＝1-2sin 2 (arcsin(x)-π/4)
packaging the angle component by an angle factor θ, which can be expressed as:
θ＝A+1
bringing an angle factor into WIOU v3 The TIOU loss function can be expressed as:
Δ＝((x-x gt ) 2 +(y-y gt ) 2 )/(W 2 +H 2 )
wherein IOU represents the cross-over ratio between the predicted and real frames,representing the mean of the IOU; delta and epsilon are constants, delta represents the distance between the center points, (x, y) represents the coordinates of the center points of the prediction frame, (x) gt ，y gt ) Representing the coordinates of the center point of the real frame; w and H represent the width and height of the minimum closure; the minimum closure refers to the smallest rectangle that can enclose the predicted and real frames.
In Table 3, we designed a comparative set of experiments to analyze the effectiveness of the improved TIOU with the CIOU and popular EIOU, SIOU and MIOU of the original model. Experimental results show that the high and wide losses of the EIOU refinement prediction frame and the real frame have little influence on the experimental results, and even mAP is reduced by 0.2%; in the loss function SIOU reconstructed from angle, the resulting mAP is improved by 0.4%; under the condition of not using the high and wide losses of the prediction frame and the real frame, the WIOU is proposed again through focusing on the center distance and the focusing criterion, and the mAP is improved by 0.6%; the effect of adding SIOU to WIOU was that TIOU was proposed, resulting in an increase of 0.7% in mAP, indicating improved effectiveness.
Table 3 results of comparing TIOU with CIOU, EIOU, SIOU and WIOU
S3: preprocessing a training set, inputting the preprocessed training set into an improved yolov7 network, and outputting a training result;
s4: taking the test set as an input parameter of the improved yolov7 network, outputting a test result, comparing the test result with a training result, and reserving an optimal improved yolov7 network model;
in the application, the preprocessing adopts a method of Mosaic data enhancement and multi-scale training, then the pictures in the training set are input into an improved yolov7 network model, the structure of the improved yolov7 model is shown in fig. 5, the pre-training weight of yolov7 on an ImageNet is used for initialization, the training result is checked, and the model weight is saved.
In table 4, we designed a set of comparative experiments to compare the effectiveness of the improved yolov7 versus the baseline yolov7 model. Experimental results showed that the modified yolov7 was reduced in reference amount by about 12% and the maps were increased by 2.3% over the baseline yolov 7.
TABLE 4 baseline yolov7 and final modified yolov7 experimental results
S5: the detection of the cervical lesion cell mass is carried out by using the optimal improved yolov7 network model, the detection result picture is reserved, the effect picture generated by adopting the detection method is shown in fig. 7, a in fig. 7 represents an original label picture of the cervical cell mass, and b represents the detection result picture of the cervical cell mass by using the improved yolov 7.
As shown in fig. 8, in another embodiment of the present embodiment, a cervical lesion cell pellet detection system based on adaptive feature extraction is further included, including a data acquisition module for acquiring a data set of cervical lesion cell pellets, the data set including a test set training set;
the model construction module is used for constructing a grouping self-adaptive feature extraction network and constructing a dynamic focusing loss function of which the angle influences the center distance; extracting a network and a dynamic focus loss function based on the self-adaptive characteristics of the packet, and constructing an improved yolov7 network model; the model construction module comprises a local grouping self-adaptive module and a global bias module, wherein the local self-adaptive module is used for grouping the feature images, then the features on different scales are obtained through deformable convolution with a cavity of 6 and a cavity of 12 respectively, and finally the features on different scales are spliced to obtain local self-adaptive features; the global bias module is used for carrying out self-adaptive average pooling on the input feature map to obtain a global feature, extracting further features of the global feature through a convolution of 1×1, and then activating the global feature by using an activation function.
The dynamic focusing loss function of the construction angle influencing the center distance is specifically as follows:
determining the included angle between the central line and the x axis as alpha, if the angle alpha is less than or equal to pi/4, taking alpha to indicate that the central line is biased towards the x axis, otherwise taking pi/2-alpha to indicate that the central line is biased towards the y axis, and the angle component can be expressed as follows:
Λ＝1-2sin 2 (arcsin(x)-π/4)
packaging the angle component by an angle factor θ, which can be expressed as:
θ＝Λ+1
bringing an angle factor into WIOU v3 The TIOU loss function can be expressed as:
Δ＝((x-x gt ) 2 +(y-y gt ) 2 )/(W 2 +H 2 )
wherein IOU represents the cross-over ratio between the predicted and real frames,representing the mean of the IOU; delta and epsilon are constants, delta represents the distance between the center points, (x, y) represents the coordinates of the center points of the prediction frame, (x) gt ，y gt ) Representing the coordinates of the center point of the real frame; w and H represent the width and height of the minimum closure.
The model training module is used for inputting the training set after pretreatment to the improved yolov7 network and outputting a training result; the model training module pre-processes the training set, specifically, the training set is pre-processed by using a method of Mosaic data enhancement and multi-scale training.
The model optimization module is used for taking the test set as an input parameter of the improved yolov7 network, outputting a test result, comparing the test result with a training result, and reserving an optimal improved yolov7 network model;
and the image detection module is used for detecting the cervical lesion cell mass image by using the optimal improved yolov7 network model, and retaining a detection result picture.
Further, an electronic device is included, as shown in fig. 9, the electronic device 400 including one or more processors 401 and a memory 402.
The processor 401 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities and may control other components in the electronic device 400 to perform desired functions.
Memory 402 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, random Access Memory (RAM) and/or cache memory (cache), and the like. The non-volatile memory may include, for example, read Only Memory (ROM), hard disk, flash memory, and the like. One or more computer program instructions may be stored on the computer readable storage medium that may be executed by the processor 401 to implement the cervical lesion cell mass detection method based on adaptive feature extraction and/or other desired functions of any of the embodiments of the invention described above. Various content such as initial arguments, thresholds, etc. may also be stored in the computer readable storage medium.
In one example, the electronic device 400 may further include: an input device 403 and an output device 404, which are interconnected by a bus system and/or other forms of connection mechanisms (not shown). The input device 403 may include, for example, a keyboard, a mouse, and the like. The output device 404 may output various information to the outside, including early warning prompt information, braking force, etc. The output device 404 may include, for example, a display, speakers, a printer, and a communication network and remote output devices connected thereto, etc.
Of course, only some of the components of the electronic device 400 that are relevant to the present invention are shown in fig. 9 for simplicity, components such as buses, input/output interfaces, etc. are omitted. In addition, electronic device 400 may include any other suitable components depending on the particular application.
In addition to the methods and apparatus described above, embodiments of the invention may also be a computer program product comprising computer program instructions which, when executed by a processor, cause the processor to perform the steps of the cervical lesion cell mass detection method based on adaptive feature extraction provided by any of the embodiments of the invention.
The computer program product may write program code for performing operations of embodiments of the present invention in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server.
Furthermore, embodiments of the present invention may also be a computer-readable storage medium, having stored thereon computer program instructions, which when executed by a processor, cause the processor to perform the steps of the cervical lesion cell mass detection method based on adaptive feature extraction provided by any embodiment of the present invention.
The computer readable storage medium may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a hard disk, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
The foregoing is merely exemplary of the present invention, and the specific structures and features well known in the art are not described in any way herein, so that those skilled in the art will be able to ascertain all prior art in the field, and will not be able to ascertain any prior art to which this invention pertains, without the general knowledge of the skilled person in the field, before the application date or the priority date, to practice the present invention, with the ability of these skilled persons to perfect and practice this invention, with the help of the teachings of this application, with some typical known structures or methods not being the obstacle to the practice of this application by those skilled in the art. It should be noted that modifications and improvements can be made by those skilled in the art without departing from the structure of the present invention, and these should also be considered as the scope of the present invention, which does not affect the effect of the implementation of the present invention and the utility of the patent. The protection scope of the present application shall be subject to the content of the claims, and the description of the specific embodiments and the like in the specification can be used for explaining the content of the claims.
Claims (8)
1. The cervical lesion cell mass detection method based on self-adaptive feature extraction is characterized by comprising the following steps of: comprising the following steps:
s1: obtaining a data set of cervical lesion cell mass, the data set comprising a test set and a training set;
s2: constructing a grouping self-adaptive feature extraction network and constructing a dynamic focusing loss function of which the angle influences the center distance; the network and the dynamic focus loss function are extracted based on the self-adaptive characteristics of the packets, and an improved yolov7 network is constructed;
s3: preprocessing a training set, inputting the preprocessed training set into an improved yolov7 network, and outputting a training result;
s4: taking the test set as an input parameter of the improved yolov7 network, outputting a test result, comparing the test result with a training result, and reserving an optimal improved yolov7 network model;
s5: and detecting cervical lesion cell mass by using an optimal improved yolov7 network model, and reserving a detection result picture.
2. The cervical lesion cell mass detection method based on adaptive feature extraction according to claim 1, wherein: in the step S2, the construction of the packet adaptive feature extraction network specifically includes:
the input feature map (W×H×C) 1 ) Press channel (C) 1 ) Is divided into two groups to generate two groups of characteristic diagrams (W×H×C 1 /2)；
Two sets of feature maps (W X H X C) 1 Respectively/2) byA deformable convolution with a hole of 6 and a hole of 12, yielding an adaptive feature map (W H C) at two scales 2 /2)；
The two adaptive feature maps are spliced to obtain a locally adaptive feature map (W×H×C) 2 )；
Will input a feature map (W x H x C) 1 ) By adaptive averaging, the feature map of each channel is compressed to 1×1 to obtain a global feature (1×1×c 1 ) Then the global features (1×1×c) obtained by a 1×1 convolution pair 1 ) Further extraction to obtain new global features (1×1×C 2 ) And obtaining final global features (1×1×C) by SiLU activation function 2 )；
The obtained local adaptive feature map (W×H×C 2 ) And global features (1 x C) 2 ) Using Add operation connections, a final global adaptive feature map (w×h×c 2 )。
3. The cervical lesion cell mass detection method based on adaptive feature extraction according to claim 2, wherein: the dynamic focusing loss function of the construction angle influencing the center distance is specifically as follows:
the angle component is used for constructing an influence factor of the angle, the included angle between the central line and the x axis is alpha, if the angle alpha is less than or equal to pi/4, the central line is deviated to the x axis, otherwise pi/2-alpha is adopted, the central line is deviated to the y axis, and the angle component can be expressed as follows:
Λ＝1-2sin 2 (arcsin(x)-π/4)
packaging the angle component by an angle factor θ, which can be expressed as:
θ＝Λ+1
bringing an angle factor into WIOU v3 The TIOU loss function can be expressed as:
Δ＝((x-x gt ) 2 +(y-y gt ) 2 )/(W 2 +H 2 )
wherein IOU represents the cross-over ratio between the predicted and real frames,representing the mean of the IOU; delta and epsilon are constants, delta represents the distance between the center points, (x, y) represents the coordinates of the center points of the prediction frame, (x) gt ,y gt ) Representing the coordinates of the center point of the real frame; w and H represent the width and height of the minimum closure.
4. The cervical lesion cell mass detection method based on adaptive feature extraction according to claim 3, wherein: the preprocessing described in S3 includes preprocessing the training set using the method of mosaics data enhancement and multi-scale training.
5. Cervical lesion cell mass detecting system based on self-adaptation characteristic extraction, its characterized in that: comprising the following steps:
the data acquisition module is used for acquiring a data set of cervical lesion cell masses, wherein the data set comprises a test set and a training set;
the model construction module is used for constructing a grouping self-adaptive feature extraction network and constructing a dynamic focusing loss function of which the angle influences the center distance; extracting a network and a dynamic focus loss function based on the self-adaptive characteristics of the packet, and constructing an improved yolov7 network model;
the model training module is used for inputting the training set after pretreatment to the improved yolov7 network and outputting a training result;
the model optimization module is used for taking the test set as an input parameter of the improved yolov7 network, outputting a test result, comparing the test result with a training result, and reserving an optimal improved yolov7 network model;
and the image detection module is used for detecting the cervical lesion cell mass image by using the optimal improved yolov7 network model, and retaining a detection result picture.
6. The cervical lesion cell mass detection system based on adaptive feature extraction according to claim 5, wherein: the model construction module comprises a local grouping self-adaptive module and a global bias module, wherein the local self-adaptive module is used for grouping the feature images, then the features on different scales are obtained through deformable convolution with a cavity of 6 and a cavity of 12 respectively, and finally the features on different scales are spliced to obtain local self-adaptive features; the global bias module is used for carrying out self-adaptive average pooling on the input feature map to obtain a global feature, extracting further features of the global feature through a convolution of 1×1, and then activating the global feature by using an activation function.
7. The cervical lesion cell mass detection system based on adaptive feature extraction according to claim 6, wherein: the dynamic focusing loss function of the construction angle influencing the center distance is specifically as follows:
determining the included angle between the central line and the x axis as alpha, if the angle alpha is less than or equal to pi/4, taking alpha to indicate that the central line is biased towards the x axis, otherwise taking pi/2-alpha to indicate that the central line is biased towards the y axis, and the angle component can be expressed as follows:
Λ＝1-2sin 2 (arcsin(x)-π/4)
packaging the angle component by an angle factor θ, which can be expressed as:
θ＝Λ+1
bringing an angle factor into WIOU v3 The TIOU loss function can be expressed as:
Δ＝((x-x gt ) 2 +(y-y gt ) 2 )/(W 2 +H 2 )
wherein IOU represents the cross-over ratio between the predicted and real frames,representing the mean of the IOU; delta and epsilon are constants, delta represents the distance between the center points, (x, y) represents the coordinates of the center points of the prediction frame, (x) gt ,y gt ) Representing the coordinates of the center point of the real frame; w and H represent the width and height of the minimum closure.
8. The cervical lesion cell mass detection system based on adaptive feature extraction according to claim 7, wherein: the model training module pre-processes the training set, specifically, the training set is pre-processed by using a method of Mosaic data enhancement and multi-scale training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311370758.XA CN117710285B (en) | 2023-10-20 | 2023-10-20 | Cervical lesion cell mass detection method and system based on self-adaptive feature extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311370758.XA CN117710285B (en) | 2023-10-20 | 2023-10-20 | Cervical lesion cell mass detection method and system based on self-adaptive feature extraction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117710285A true CN117710285A (en) | 2024-03-15 |
| CN117710285B CN117710285B (en) | 2024-07-16 |
Family
ID=90146797
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311370758.XA Active CN117710285B (en) | 2023-10-20 | 2023-10-20 | Cervical lesion cell mass detection method and system based on self-adaptive feature extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117710285B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220031227A1 (en) * | 2018-10-02 | 2022-02-03 | Industry Academic Cooperation Foundation, Hallym University | Device and method for diagnosing gastric lesion through deep learning of gastroendoscopic images |
| CN114972254A (en) * | 2022-05-25 | 2022-08-30 | 宁夏理工学院 | Cervical cell image segmentation method based on convolutional neural network |
| CN115546614A (en) * | 2022-12-02 | 2022-12-30 | 天津城建大学 | Safety helmet wearing detection method based on improved YOLOV5 model |
-
2023
- 2023-10-20 CN CN202311370758.XA patent/CN117710285B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220031227A1 (en) * | 2018-10-02 | 2022-02-03 | Industry Academic Cooperation Foundation, Hallym University | Device and method for diagnosing gastric lesion through deep learning of gastroendoscopic images |
| CN114972254A (en) * | 2022-05-25 | 2022-08-30 | 宁夏理工学院 | Cervical cell image segmentation method based on convolutional neural network |
| CN115546614A (en) * | 2022-12-02 | 2022-12-30 | 天津城建大学 | Safety helmet wearing detection method based on improved YOLOV5 model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117710285B (en) | 2024-07-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US12051199B2 (en) | Image processing method and apparatus, server, medical image processing device and storage medium | |
| CN111160269A (en) | Face key point detection method and device | |
| CN111695463B (en) | Training method of face impurity detection model and face impurity detection method | |
| CN111368672A (en) | Construction method and device for genetic disease facial recognition model | |
| CN111860169B (en) | Skin analysis method, device, storage medium and electronic equipment | |
| Shaik et al. | Lesion-aware attention with neural support vector machine for retinopathy diagnosis | |
| CN111611851A (en) | Model generation method, iris detection method and device | |
| US11967125B2 (en) | Image processing method and system | |
| CN116912597A (en) | Intellectual property intelligent management system and method thereof | |
| Wang et al. | Multiscale feature fusion for skin lesion classification | |
| CN114419087A (en) | Focus image generation method and device, electronic equipment and storage medium | |
| Liu et al. | Lightweight target detection algorithm based on YOLOv4 | |
| CN111881803B (en) | Face recognition method based on improved YOLOv3 | |
| CN117596058A (en) | Network information security protection system and method | |
| CN117710285B (en) | Cervical lesion cell mass detection method and system based on self-adaptive feature extraction | |
| CN116130088A (en) | Multi-mode face diagnosis method, device and related equipment | |
| CN112699907A (en) | Data fusion method, device and equipment | |
| CN113222989B (en) | Image grading method and device, storage medium and electronic equipment | |
| Mohan et al. | Enhancing skin disease classification leveraging transformer-based deep learning architectures and explainable ai | |
| CN115641541A (en) | Based on patient's body surface mud surface state data processing system | |
| Wang et al. | A fixed-point rotation-based feature selection method for micro-expression recognition | |
| CN114649092A (en) | Auxiliary diagnosis method and device based on semi-supervised learning and multi-scale feature fusion | |
| Milczarski et al. | Blue-white veil classification of dermoscopy images using convolutional neural networks and invariant dataset augmentation | |
| Chen et al. | LMSA‐Net: A lightweight multi‐scale aware network for retinal vessel segmentation | |
| Chen et al. | Caf-yolo: A robust framework for multi-scale lesion detection in biomedical imagery |
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 |