CN116486230A - Image detection method based on semi-recursion characteristic pyramid structure and storage medium - Google Patents
Image detection method based on semi-recursion characteristic pyramid structure and storage medium Download PDFInfo
- Publication number
- CN116486230A CN116486230A CN202310433523.4A CN202310433523A CN116486230A CN 116486230 A CN116486230 A CN 116486230A CN 202310433523 A CN202310433523 A CN 202310433523A CN 116486230 A CN116486230 A CN 116486230A
- Authority
- CN
- China
- Prior art keywords
- feature
- semantic
- low
- features
- image
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000004364 calculation method Methods 0.000 claims abstract description 15
- 230000000007 visual effect Effects 0.000 claims abstract description 12
- 230000004927 fusion Effects 0.000 claims description 18
- 238000000605 extraction Methods 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 12
- 230000003044 adaptive effect Effects 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 7
- 238000012805 post-processing Methods 0.000 claims description 6
- 230000001965 increasing effect Effects 0.000 claims description 5
- 230000001629 suppression Effects 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- 239000000523 sample Substances 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 25
- 230000006870 function Effects 0.000 description 14
- 230000008569 process Effects 0.000 description 14
- 238000012545 processing Methods 0.000 description 9
- 230000005540 biological transmission Effects 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000011897 real-time detection Methods 0.000 description 3
- 230000002829 reductive effect Effects 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 239000010949 copper Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Classifications
-
- 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
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/048—Activation functions
-
- 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
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the application discloses an image detection method and a storage medium based on a semi-recursion characteristic pyramid structure, and relates to the technical field of deep learning image detection, wherein the method comprises the following steps: acquiring a real-time image through an image acquisition device; inputting the real-time image into a low semantic layer of a semi-recursion feature pyramid to generate first low semantic features; performing feedback feature selection operation on the first low-semantic features to generate feedback features; inputting the feedback characteristics and the real-time image into the low semantic layer again to carry out recursive calculation to obtain second low semantic characteristics; respectively inputting the first low-semantic features and the second low-semantic features into the high-semantic layer of the semi-recursive feature pyramid for downsampling to obtain two high-semantic features; and respectively fusing the first low-semantic features, the second low-semantic features and the two high-semantic features to generate features for prediction, performing multi-stage prediction by using the self-adaptive detection head to obtain a prediction result and performing visual display.
Description
Technical Field
The application relates to the technical field of deep learning image detection, in particular to an image detection method based on a semi-recursion characteristic pyramid structure and a storage medium.
Background
With the development of image recognition technology, image detection algorithms based on deep learning are dominant in the field of natural images. The existing image detection method has the problems of low recognition accuracy, complex recognition flow, high calculation complexity, difficulty in meeting real-time performance and the like on images with high noise and low resolution and more overlapping targets. Particularly in medical clinical diagnosis, the ultrasonic technology is used as a medical imaging technology with wide application, and has the advantages of low cost, safety and no radiation. However, the ultrasonic image has the defects of serious noise, lower resolution and the like, and the ultrasonic image has the characteristic of real-time performance, and the defects increase the recognition difficulty of clinical lesions. And the method brings great challenges in the field of medical image identification.
Therefore, the traditional image detection method has poor detection effect on images with high noise and low resolution and more overlapped targets, and is difficult to meet the real-time requirement.
Disclosure of Invention
An object of the embodiments of the present application is to provide an image detection method and a storage medium based on a semi-recursive feature pyramid structure, so as to solve the problems of low image detection precision and poor real-time performance in the prior art.
To achieve the above object, an embodiment of the present application provides an image detection method based on a semi-recursive feature pyramid structure, including: acquiring a real-time image through an image acquisition device;
inputting the real-time image to a low semantic layer of a semi-recursion feature pyramid to generate first low semantic features;
performing feedback feature selection operation on the first low-semantic features to generate feedback features, and reducing local errors caused by noise points;
inputting the feedback characteristics and the real-time image into the low semantic layer again for recursive calculation to obtain second low semantic characteristics;
respectively inputting the first low-semantic features and the second low-semantic features into the high-semantic layers of the semi-recursive feature pyramid for downsampling, and expanding the feature level to obtain two high-semantic features;
and respectively fusing the first low-semantic features, the second low-semantic features and the two high-semantic features to generate features for prediction, performing multi-stage prediction by using the self-adaptive detection head to obtain a prediction result and performing visual display.
Optionally, the acquiring, by the image capturing device, a real-time image includes:
acquiring a real-time image stream by using an image acquisition probe;
and extracting the image stream frame by frame and filling the image stream in a fixed scale, and detecting the image frame by frame.
Optionally, the semi-recursive feature pyramid comprises a low semantic layer and a high semantic layer;
the low semantic layer comprises a backbone network, a low-level feature pyramid network and a feedback feature selection module, wherein the backbone network adopts an increased depth convolution kernel and an inverted bottleneck structure to effectively enlarge a receptive field and enhance global feature extraction capacity, the low-level feature pyramid network realizes low-level feature fusion from top to bottom, and the feedback feature selection module adopts a channel and spatial attention to realize important feature enhancement and invalid feature suppression;
the high semantic layer comprises a high-level feature pyramid network, the high-level recursive computation is canceled by utilizing the image downsampling operation, the feature level is expanded, and the operation is accelerated.
Optionally, the performing a feedback feature selection operation on the first low semantic feature includes:
performing point convolution aiming at channel dimensions on the first low-semantic features through parallel convolution branches, outputting feature images under different receptive fields through hole convolution with different expansion rates and image pooling operation, and splicing the feature images in the channel dimensions to obtain multi-scale features;
and calculating channel feature attention and space feature attention of the multi-scale features, generating channel and space feature selection weights through a Sigmoid function, and sequentially multiplying the channel and space feature selection weights, so that the quality of feature information is improved, and feedback feature selection is realized.
Optionally, the recursively calculating the feedback feature and the real-time image input to the low semantic layer again includes:
channel transformation is carried out on the feedback features through point convolution, if the feedback features are not null, the feedback features are accumulated with the feature images after downsampling of the low semantic layer, and if the feedback features are null, the feedback features are set to be 0 and the feature images after downsampling of the low semantic layer are accumulated;
and carrying out second feature extraction on the accumulated features in the low semantic layer, and realizing recursive calculation to obtain second low semantic features.
Optionally, the step of downsampling the expansion feature level by the high semantic layer to obtain two high semantic features includes:
inputting the highest-level feature map of the first low-semantic features into a high-semantic layer, and performing two downsampling operations on the feature map by the high-semantic layer to generate 2 layers of feature levels which do not participate in feature fusion, so as to obtain first high-semantic features;
and carrying out the same operation as the first operation on the highest-level feature map of the second low-semantic features to obtain second high-semantic features.
Optionally, the fusing the first low-semantic feature, the second low-semantic feature and the two high-semantic features in corresponding layers respectively to generate features for prediction includes:
performing point convolution on the second low semantic features and the second high semantic features, and generating a selection weight w by using a Sigmoid function;
multiplying the first low-semantic feature, the second low-semantic feature and the two high-semantic features with 1-w and w respectively, and accumulating the first low-semantic feature, the second low-semantic feature and the two high-semantic features with different weights to generate prediction features.
Optionally, the adaptive detection head comprises a multi-level feature mapping block and a single detection head;
the multi-level feature mapping block is mapped through point convolution after residual connection by the depth convolution with the convolution kernel of 7, and the weights of the multi-level feature mapping block are learnable and not shared so as to enhance the self-adaptive capacity of multi-level features;
the single detection head adopts a decoupling mode, uses two parallel convolution branches to conduct category and regression frame prediction, and simultaneously utilizes regression branch prediction centrality, and the weight of the single detection head is shared.
Optionally, the visually displaying the prediction result includes:
post-processing is carried out on the prediction result, and a prediction frame with low confidence coefficient and a prediction frame deviating from the center of the target are restrained, so that the coordinates and the category of the prediction frame to be visualized are obtained;
and drawing the prediction frame coordinates and the categories of the real-time image, generating a visual image stream and outputting the visual image stream.
To achieve the above object, the present application also provides a computer storage medium having stored thereon a computer program which, when executed by a machine, implements the steps of the method as described above.
The embodiment of the application has the following advantages:
the embodiment of the application provides an image detection method based on a semi-recursion characteristic pyramid structure, which comprises the following steps: acquiring a real-time image through an image acquisition device; inputting the real-time image to a low semantic layer of a semi-recursion feature pyramid to generate first low semantic features; performing feedback feature selection operation on the first low-semantic features to generate feedback features, and reducing local errors caused by noise points; inputting the feedback characteristics and the real-time image into the low semantic layer again for recursive calculation to obtain second low semantic characteristics; respectively inputting the first low-semantic features and the second low-semantic features into the high-semantic layers of the semi-recursive feature pyramid for downsampling, and expanding the feature level to obtain two high-semantic features; and respectively fusing the first low-semantic features, the second low-semantic features and the two high-semantic features to generate features for prediction, performing multi-stage prediction by using the self-adaptive detection head to obtain a prediction result and performing visual display.
Based on the method, acquiring a real-time image; at a low semantic layer, feedback feature selection and recursion fusion calculation are carried out on the image, so that local fine feature recursion extraction is realized; at a high semantic layer, downsampling the low semantic layer image to expand the feature level, so as to realize multi-level detection; fusing the two features before and after recursion to generate prediction features; and carrying out multistage prediction on the fusion characteristics by using the self-adaptive detection head to obtain a detection result and visualizing the detection result. The real-time detection of the image is realized through the semi-recursion feature pyramid structure, the recursion computation complexity and time consumption in the image detection can be greatly reduced, the multi-level feature self-adaption capability of an algorithm is enhanced, the problems that the computation complexity is high, the flow is complex, the real-time performance cannot be met, the noise resistance is poor, the influence of the resolution is large in the image detection at the present stage are solved, and the image detection effect is improved.
Drawings
In order to more clearly illustrate the embodiments of the present application 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. It will be apparent to those skilled in the art from this disclosure that the drawings described below are merely exemplary and that other embodiments may be derived from the drawings provided without undue effort.
Fig. 1 is an application environment diagram of an image detection method based on a semi-recursive feature pyramid structure according to an embodiment of the present application;
fig. 2 is a schematic flow chart of an image detection method based on a semi-recursive feature pyramid structure according to an embodiment of the present application;
fig. 3 is a block diagram of a semi-recursive feature pyramid network of an image detection method based on a semi-recursive feature pyramid structure according to an embodiment of the present application;
fig. 4 is a schematic diagram of a feedback feature selection module of an image detection method based on a semi-recursive feature pyramid structure according to an embodiment of the present application;
fig. 5 is a schematic diagram of a single stage of a backbone network of an image detection method based on a semi-recursive feature pyramid structure according to an embodiment of the present application;
fig. 6 is a block diagram of an adaptive detection head of an image detection method based on a semi-recursive feature pyramid structure according to an embodiment of the present application, where (a) is a schematic diagram of a detection head calculation flow, (b) is a schematic diagram of a hybrid convolution block, and (c) is a schematic diagram of a weight sharing detection head;
fig. 7 is an image prediction result visualization diagram of an image detection method based on a semi-recursive feature pyramid structure according to an embodiment of the present application, where (a) is a real result schematic diagram, (b) is a prediction result schematic diagram corresponding to (a), (c) is another real result schematic diagram, and (d) is a prediction result schematic diagram corresponding to (c).
Detailed Description
Other advantages and advantages of the present application will become apparent to those skilled in the art from the following description of specific embodiments, which is to be read in light of the present disclosure, wherein the present embodiments are described in some, but not all, of the several embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
In addition, the technical features described below in the different embodiments of the present application may be combined with each other as long as they do not collide with each other.
The image detection method based on the semi-recursion characteristic pyramid structure provided by the embodiment of the application can be applied to an application environment shown in fig. 1. As shown in FIG. 1, the application environment includes a computer device 110. Computer device 110 may acquire real-time images; computer device 110 may input the real-time image to the low semantic layer of the semi-recursive feature pyramid to generate first time low semantic features; the computer device 110 can perform feedback feature selection operation on the first low-semantic features, reduce local errors caused by noise points, and generate feedback features; the computer device 110 may re-input the feedback features and the real-time image to the low semantic layer for recursive computation to generate second low semantic features; the computer device 110 may input the first low semantic features and the second low semantic features to the high semantic layer of the semi-recursive feature pyramid to downsample the expanded feature levels to obtain two high semantic features; the computer device 110 may fuse the two high semantic features to generate features for prediction, and perform multi-level prediction using the adaptive detection head, so as to obtain a prediction result and visualize the prediction result. The computer device 110 may be, but is not limited to, various personal computers, notebook computers, smart phones, robots, unmanned aerial vehicles, tablet computers, and the like.
An embodiment of the present application provides an image detection method based on a semi-recursive feature pyramid structure, referring to fig. 2, fig. 2 is a flowchart of an image detection method based on a semi-recursive feature pyramid structure provided in an embodiment of the present application, and it should be understood that the method may further include additional blocks not shown and/or blocks shown may be omitted, and the scope of the present application is not limited in this respect.
In this embodiment, referring to fig. 3, the method includes performing fixed scale filling on a real-time image, then performing feedback feature selection and recursive fusion calculation on the image at a low semantic layer to implement local fine feature recursive extraction, performing downsampling and expanding feature level on the image at the low semantic layer at a high semantic layer to implement multi-level detection, fusing two features before and after recursion to generate a prediction feature, and performing multi-level prediction on the fusion feature by using an adaptive detection head to obtain a detection result and visualizing the detection result. The method has the advantages that the real-time detection of the image is realized through the semi-recursion feature pyramid structure, so that the recursion computation complexity and time consumption in the image detection are greatly reduced, the multi-level feature self-adaption capability of an algorithm is enhanced, the problems that the computation complexity is high, the flow is complex, the real-time performance cannot be met, the noise resistance is poor, the influence of resolution is large in the image detection at the present stage are solved, and the image detection effect is improved.
The method specifically comprises the following steps:
at step 201, a real-time image is acquired by an image acquisition device.
Specifically, the computer device may obtain a real-time image stream, extract and fill the image stream in a fixed scale frame by frame, and detect the image frame by frame.
At step 202, the real-time image is input to the low semantic layer of the semi-recursive feature pyramid to generate first low semantic features.
At step 203, the first low semantic feature is subjected to feedback feature selection operation, so as to generate feedback features, and reduce local errors caused by noise points.
At step 204, the feedback features and the real-time image are input again to the low semantic layer for recursive computation to obtain second low semantic features.
Specifically, at a low semantic layer, feedback feature selection and recursive fusion calculation are performed on the image, so that local fine feature recursive extraction is realized.
The computer device may utilize a backbone network, a low-level feature pyramid network, and a feedback feature selection module to perform low-level semantic recursive computation, where the backbone network effectively expands receptive fields by increasing a deep convolution kernel and inverting a bottleneck structure, and enhances global feature extraction capability, and in some embodiments, may employ ConvNeXt based on the deep convolution kernel, where each stage includes ConvNeXt blocks of 3,3,9,3, respectively. The low-level feature pyramid network adopts low-level feature fusion from top to bottom. The feedback feature selection module adopts a channel and spatial attention to realize important feature enhancement and invalid feature suppression.
At step 205, the first low-semantic features and the second low-semantic features are respectively input to the high-semantic layer of the semi-recursive feature pyramid for downsampling, and feature levels are expanded to obtain two high-semantic features.
Specifically, at a high semantic layer, the expansion feature level is sampled downwards for the low semantic layer image, so that multi-level detection is realized.
The computer equipment can input the highest-level feature map P5 of the first low-semantic features into a high-semantic layer, and the high-semantic layer performs two downsampling operations on the feature map to generate 2 layers of feature levels P6 and P7 which do not participate in feature fusion, so as to obtain the first high-semantic features; and carrying out the same operation as the first operation on the highest level characteristic map P5 of the second low-semantic characteristics to obtain the second high-semantic characteristics.
At step 206, the first low-semantic feature, the second low-semantic feature and the second high-semantic feature are respectively fused in corresponding layers to generate features for prediction, and the adaptive detection head is used for performing multi-level prediction to obtain a prediction result and performing visual display.
Specifically, corresponding layer fusion is performed on two low-semantic and high-semantic features (a first low-semantic and high-semantic feature and a second low-semantic and high-semantic feature) before and after recursion to generate prediction features.
The computer equipment can carry out point convolution on the low-semantic features and the high-semantic features of the second time, generate a selection weight w by using a Sigmoid function, multiply the low-semantic features and the high-semantic features of the second time with 1-w and w respectively, and accumulate the low-semantic features and the high-semantic features of the second time with different weights to generate prediction features.
And carrying out multistage prediction on the fusion characteristics by using the self-adaptive detection head to obtain a detection result and visualizing the detection result.
The computer equipment can map the multi-level feature map, the depth convolution with the convolution kernel of 7 is mapped through point convolution after residual connection, mapping weights are learnable and not shared, so that the self-adaptive capacity of the multi-level features is enhanced, a single detection head adopts a decoupling mode, two parallel convolution branches are used for carrying out category and regression frame prediction, meanwhile, the regression branch prediction centrality is utilized for carrying out post-processing on a prediction result, a prediction frame with low confidence coefficient and a prediction frame deviating from a target center are restrained, the coordinate and the category of the prediction frame to be visualized are obtained, the drawing of the coordinate and the category of the prediction frame is carried out, and a visual image stream is generated and output.
In some embodiments, the acquiring, by the image acquisition device, the real-time image includes:
acquiring a real-time image stream by using an image acquisition probe;
and extracting the image stream frame by frame and filling the image stream in a fixed scale, and detecting the image frame by frame.
Specifically, the image detection method based on the semi-recursion characteristic pyramid structure can further comprise a process of acquiring real-time images through the image acquisition equipment, and the specific process comprises the following steps: and acquiring a real-time image stream by using an image acquisition probe, extracting the image stream frame by frame, filling the image stream in a fixed scale, and detecting the image frame by frame.
The computer device may populate the real-time image with images having a width of 1024 and a height of 800 and perform frame-by-frame detection of the real-time image.
In some embodiments, the semi-recursive feature pyramid includes a low semantic layer and a high semantic layer;
the low semantic layer comprises a backbone network, a low-level feature pyramid network and a feedback feature selection module, wherein the backbone network adopts an increased depth convolution kernel and an inverted bottleneck structure to effectively enlarge a receptive field and enhance global feature extraction capacity, the low-level feature pyramid network realizes low-level feature fusion from top to bottom, and the feedback feature selection module adopts a channel and spatial attention to realize important feature enhancement and invalid feature suppression;
the high semantic layer comprises a high-level feature pyramid network, the high-level recursive computation is canceled by utilizing the image downsampling operation, the feature level is expanded, and the operation is accelerated.
In some embodiments, the performing the feedback feature selection operation on the first low-semantic features includes:
performing point convolution aiming at channel dimensions on the first low-semantic features through parallel convolution branches, outputting feature images under different receptive fields through hole convolution with different expansion rates and image pooling operation, and splicing the feature images in the channel dimensions to obtain multi-scale features;
and calculating channel feature attention and space feature attention of the multi-scale features, generating channel and space feature selection weights through a Sigmoid function, and sequentially multiplying the channel and space feature selection weights, so that the quality of feature information is improved, and feedback feature selection is realized.
Specifically, the image detection method based on the semi-recursion feature pyramid structure can further comprise the step that the low semantic layer utilizes a feedback feature selection module to realize feature selection, and the specific process comprises the following steps: the feedback feature selection module adopts a channel and spatial attention to realize important feature enhancement and invalid feature suppression.
As shown in fig. 4, the computer device performs a point convolution with a convolution kernel of 1 on the first low semantic feature through 4 parallel branches, outputs 4 feature graphs with the number of original channels of 1/4 by using a cavity convolution with a expansion rate of 3 and 6 and an image pooling operation, splices the feature graphs to obtain multi-scale features, then performs calculation of channel feature attention and space feature attention, generates channel and space feature selection weights through a Sigmoid function, and sequentially performs multiplication to realize feedback feature selection.
In some embodiments, said recursively computing the feedback features and the real-time image again into the low semantic layer comprises:
channel transformation is carried out on the feedback features through point convolution, if the feedback features are not null, the feedback features are accumulated with the feature images after downsampling of the low semantic layer, and if the feedback features are null, the feedback features are set to be 0 and the feature images after downsampling of the low semantic layer are accumulated;
and carrying out second feature extraction on the accumulated features in the low semantic layer, and realizing recursive calculation to obtain second low semantic features.
Specifically, the image detection method based on the semi-recursion feature pyramid structure can further comprise a process of inputting feedback features and real-time images into the low semantic layer again to carry out recursion calculation, and the specific process comprises the following steps: and carrying out channel transformation on the feedback characteristics through point convolution, accumulating the feedback characteristics with the characteristic diagram after the downsampling of the low semantic layer if the feedback characteristics are not null, and accumulating the feedback characteristics as 0 and the characteristic diagram after the downsampling of the low semantic layer if the feedback characteristics are null.
As shown in fig. 5, the computer device adds the feedback feature to the backbone network computation of the second time by adding an input branch of the backbone network, and the ConvNeXt enlarges the convolution receptive field and the network width by increasing the design of the deep convolution kernel and the inverted bottleneck block, thereby enhancing the global feature extraction capability. ConvNeXt comprises 4 stages, C2-C5 is generated respectively, when a feedback characteristic R is input to a backbone network, R branches are added in the later 3 stages to adapt to recursive computation, the feedback characteristic is accumulated after the channel number of the feedback characteristic is the same as that of a downsampling characteristic through point convolution, N ConvNeXt block extraction characteristics are utilized to generate an input characteristic of the next stage, and the number N of blocks in the later 3 stages is 3,9,3 respectively.
In some embodiments, the high semantic layer downsampling the expanded feature levels to obtain two high semantic features, including:
inputting the highest-level feature map of the first low-semantic features into a high-semantic layer, and performing two downsampling operations on the feature map by the high-semantic layer to generate 2 layers of feature levels which do not participate in feature fusion, so as to obtain first high-semantic features;
and carrying out the same operation as the first operation on the highest-level feature map of the second low-semantic features to obtain second high-semantic features.
Specifically, the image detection method based on the semi-recursion feature pyramid structure can further comprise a high semantic layer downsampling and expanding feature level process, and the specific process comprises the following steps: inputting the highest level feature map of the first low-semantic features into a high-semantic layer, performing two downsampling operations on the feature map by the high-semantic layer to generate 2 layers of feature levels which do not participate in feature fusion, obtaining the first high-semantic features, and performing the same operations on the highest level feature map of the second low-semantic features to obtain the second high-semantic features.
In some embodiments, the fusing the first low-semantic feature, the second low-semantic feature, and the two high-semantic features in corresponding layers respectively, to generate a feature for prediction includes:
performing point convolution on the second low semantic features and the second high semantic features, and generating a selection weight w by using a Sigmoid function;
multiplying the first low-semantic feature, the second low-semantic feature and the two high-semantic features with 1-w and w respectively, and accumulating the first low-semantic feature, the second low-semantic feature and the two high-semantic features with different weights to generate prediction features.
Specifically, the image detection method based on the semi-recursion feature pyramid structure can further comprise a process of generating predicted features by fusing low-semantic features and high-semantic features twice, and the specific process comprises the following steps: and carrying out point convolution with convolution kernel of 1 on the second low-semantic and high-semantic features, generating a selection weight w by using a Sigmoid function, multiplying the two low-semantic and high-semantic features with 1-w and w respectively, and accumulating the two low-semantic and high-semantic features with different weights to generate a prediction feature.
In some embodiments, the adaptive detection head comprises a multi-level feature mapping block and a single detection head;
the multi-level feature mapping block is mapped through point convolution after residual connection by the depth convolution with the convolution kernel of 7, and the weights of the multi-level feature mapping block are learnable and not shared so as to enhance the self-adaptive capacity of multi-level features;
the single detection head adopts a decoupling mode, uses two parallel convolution branches to conduct category and regression frame prediction, and simultaneously utilizes regression branch prediction centrality, and the weight of the single detection head is shared.
Specifically, the image detection method based on the semi-recursion characteristic pyramid structure can further comprise a multi-stage prediction process performed by the self-adaptive detection head, and the specific process comprises the following steps: the multi-stage feature mapping block performs feature mapping to enhance the self-adaption capability of the multi-stage features, and the single detection head detects the multi-stage features to generate a prediction result.
As shown in fig. 6, the computer device maps through point convolution after residual connection by means of depth convolution with convolution kernel of 7, and the weights of the blocks are not shared, single detection head weights are shared, and adopts a decoupling mode to predict a class and a regression frame by using two parallel convolution branches, and simultaneously predicts the centrality by using the regression branches to obtain a predicted regression frame, class and centrality.
In some embodiments, the visually presenting the prediction results includes:
post-processing is carried out on the prediction result, and a prediction frame with low confidence coefficient and a prediction frame deviating from the center of the target are restrained, so that the coordinates and the category of the prediction frame to be visualized are obtained;
and drawing the prediction frame coordinates and the categories of the real-time image, generating a visual image stream and outputting the visual image stream.
Specifically, the image detection method based on the semi-recursion characteristic pyramid structure can further comprise a prediction result visualization process, and the specific process comprises the following steps: and (3) carrying out post-processing on the prediction result, inhibiting a prediction frame with low confidence and a prediction frame deviating from the target center, obtaining the coordinates and the category of the prediction frame to be visualized, generating a visualized image stream and outputting the visualized image stream.
As shown in fig. 7, the computer device performs post-processing on the multi-level prediction result, suppresses the prediction frame with low confidence and the prediction frame deviating from the target center, obtains effective prediction frame coordinates and categories, and performs prediction image drawing by using cv2.Rectangle, thereby generating a prediction result visualized image.
Based on the method, acquiring a real-time image; at a low semantic layer, feedback feature selection and recursion fusion calculation are carried out on the image, so that local fine feature recursion extraction is realized; at a high semantic layer, downsampling the low semantic layer image to expand the feature level, so as to realize multi-level detection; fusing the two features before and after recursion to generate prediction features; and carrying out multistage prediction on the fusion characteristics by using the self-adaptive detection head to obtain a detection result and visualizing the detection result. The real-time detection of the image is realized through the semi-recursion feature pyramid structure, the recursion computation complexity and time consumption in the image detection can be greatly reduced, the multi-level feature self-adaption capability of an algorithm is enhanced, the problems that the computation complexity is high, the flow is complex, the real-time performance cannot be met, the noise resistance is poor, the influence of the resolution is large in the image detection at the present stage are solved, and the image detection effect is improved.
The present application may be a method, apparatus, system, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for performing the various aspects of the present application.
The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, punch cards or in-groove structures such as punch cards or grooves having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media, as used herein, are not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses through fiber optic cables), or electrical signals transmitted through wires.
The computer readable program instructions described herein may be downloaded from a computer readable storage medium to a respective computing/processing device or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. The network interface card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in the respective computing/processing device.
Computer program instructions for performing the operations of the present application may be assembly instructions, instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, c++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, aspects of the present application are implemented by personalizing electronic circuitry, such as programmable logic circuitry, field Programmable Gate Arrays (FPGAs), or Programmable Logic Arrays (PLAs), with state information for computer readable program instructions, which may execute the computer readable program instructions.
Various aspects of the present application are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.
These computer readable program instructions may be provided to a processing unit of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable medium having the instructions stored therein includes an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.
The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
Note that all features disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise. Thus, unless expressly stated otherwise, each feature disclosed is one example only of a generic set of equivalent or similar features. Where used, further, preferably, still further and preferably, the brief description of the other embodiment is provided on the basis of the foregoing embodiment, and further, preferably, further or more preferably, the combination of the contents of the rear band with the foregoing embodiment is provided as a complete construct of the other embodiment. A further embodiment is composed of several further, preferably, still further or preferably arrangements of the strips after the same embodiment, which may be combined arbitrarily.
While the application has been described in detail with respect to the general description and specific embodiments thereof, it will be apparent to those skilled in the art that certain modifications and improvements may be made thereto based upon the application. Accordingly, such modifications or improvements may be made without departing from the spirit of the application and are intended to be within the scope of the invention as claimed.
Claims (10)
1. An image detection method based on a semi-recursive feature pyramid structure is characterized by comprising the following steps:
acquiring a real-time image through an image acquisition device;
inputting the real-time image to a low semantic layer of a semi-recursion feature pyramid to generate first low semantic features;
performing feedback feature selection operation on the first low-semantic features to generate feedback features, and reducing local errors caused by noise points;
inputting the feedback characteristics and the real-time image into the low semantic layer again for recursive calculation to obtain second low semantic characteristics;
respectively inputting the first low-semantic features and the second low-semantic features into the high-semantic layers of the semi-recursive feature pyramid for downsampling, and expanding the feature level to obtain two high-semantic features;
and respectively fusing the first low-semantic features, the second low-semantic features and the two high-semantic features to generate features for prediction, performing multi-stage prediction by using the self-adaptive detection head to obtain a prediction result and performing visual display.
2. The image detection method based on semi-recursive feature pyramid structure according to claim 1, the acquiring real-time images by an image acquisition device, comprising:
acquiring a real-time image stream by using an image acquisition probe;
and extracting the image stream frame by frame and filling the image stream in a fixed scale, and detecting the image frame by frame.
3. The image detection method based on semi-recursive feature pyramid structure of claim 1, wherein the semi-recursive feature pyramid comprises a low semantic layer and a high semantic layer;
the low semantic layer comprises a backbone network, a low-level feature pyramid network and a feedback feature selection module, wherein the backbone network adopts an increased depth convolution kernel and an inverted bottleneck structure to effectively enlarge a receptive field and enhance global feature extraction capacity, the low-level feature pyramid network realizes low-level feature fusion from top to bottom, and the feedback feature selection module adopts a channel and spatial attention to realize important feature enhancement and invalid feature suppression;
the high semantic layer comprises a high-level feature pyramid network, the high-level recursive computation is canceled by utilizing the image downsampling operation, the feature level is expanded, and the operation is accelerated.
4. The image detection method based on the semi-recursive feature pyramid structure according to claim 1, wherein the performing a feedback feature selection operation on the first low-semantic features includes:
performing point convolution aiming at channel dimensions on the first low-semantic features through parallel convolution branches, outputting feature images under different receptive fields through hole convolution with different expansion rates and image pooling operation, and splicing the feature images in the channel dimensions to obtain multi-scale features;
and calculating channel feature attention and space feature attention of the multi-scale features, generating channel and space feature selection weights through a Sigmoid function, and sequentially multiplying the channel and space feature selection weights, so that the quality of feature information is improved, and feedback feature selection is realized.
5. The image detection method based on semi-recursive feature pyramid structure according to claim 1, wherein said recursively calculating the feedback features and the real-time image again input to a low semantic layer, comprises:
channel transformation is carried out on the feedback features through point convolution, if the feedback features are not null, the feedback features are accumulated with the feature images after downsampling of the low semantic layer, and if the feedback features are null, the feedback features are set to be 0 and the feature images after downsampling of the low semantic layer are accumulated;
and carrying out second feature extraction on the accumulated features in the low semantic layer, and realizing recursive calculation to obtain second low semantic features.
6. The image detection method based on the semi-recursive feature pyramid structure according to claim 1, wherein the high semantic layer downsampling expands feature levels to obtain two high semantic features, comprising:
inputting the highest-level feature map of the first low-semantic features into a high-semantic layer, and performing two downsampling operations on the feature map by the high-semantic layer to generate 2 layers of feature levels which do not participate in feature fusion, so as to obtain first high-semantic features;
and carrying out the same operation as the first operation on the highest-level feature map of the second low-semantic features to obtain second high-semantic features.
7. The image detection method based on the semi-recursive feature pyramid structure according to claim 6, wherein the fusing the first low-semantic feature, the second low-semantic feature and the two high-semantic features in the corresponding layers respectively, to generate features for prediction, includes:
performing point convolution on the second low semantic features and the second high semantic features, and generating a selection weight w by using a Sigmoid function;
multiplying the first low-semantic feature, the second low-semantic feature and the two high-semantic features with 1-w and w respectively, and accumulating the first low-semantic feature, the second low-semantic feature and the two high-semantic features with different weights to generate prediction features.
8. The image detection method based on semi-recursive feature pyramid structure of claim 1, wherein the adaptive detection head comprises a multi-level feature mapping block and a single detection head;
the multi-level feature mapping block is mapped through point convolution after residual connection by the depth convolution with the convolution kernel of 7, and the weights of the multi-level feature mapping block are learnable and not shared so as to enhance the self-adaptive capacity of multi-level features;
the single detection head adopts a decoupling mode, uses two parallel convolution branches to conduct category and regression frame prediction, and simultaneously utilizes regression branch prediction centrality, and the weight of the single detection head is shared.
9. The method for detecting an image based on a semi-recursive feature pyramid structure according to claim 1, wherein the visually displaying the prediction result comprises:
post-processing is carried out on the prediction result, and a prediction frame with low confidence coefficient and a prediction frame deviating from the center of the target are restrained, so that the coordinates and the category of the prediction frame to be visualized are obtained;
and drawing the prediction frame coordinates and the categories of the real-time image, generating a visual image stream and outputting the visual image stream.
10. A computer storage medium having stored thereon a computer program, which when executed by a machine performs the steps of the method according to any of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310433523.4A CN116486230B (en) | 2023-04-21 | 2023-04-21 | Image detection method based on semi-recursion characteristic pyramid structure and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310433523.4A CN116486230B (en) | 2023-04-21 | 2023-04-21 | Image detection method based on semi-recursion characteristic pyramid structure and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116486230A true CN116486230A (en) | 2023-07-25 |
| CN116486230B CN116486230B (en) | 2024-02-02 |
Family
ID=87218928
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310433523.4A Active CN116486230B (en) | 2023-04-21 | 2023-04-21 | Image detection method based on semi-recursion characteristic pyramid structure and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116486230B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111104962A (en) * | 2019-11-05 | 2020-05-05 | 北京航空航天大学青岛研究院 | Semantic segmentation method and device for image, electronic equipment and readable storage medium |
| CN113255699A (en) * | 2021-06-10 | 2021-08-13 | 浙江华睿科技有限公司 | Small target object image detection method and device, electronic equipment and storage medium |
| CN114549555A (en) * | 2022-02-25 | 2022-05-27 | 北京科技大学 | Human ear image planning and division method based on semantic division network |
| CN114783072A (en) * | 2022-03-17 | 2022-07-22 | 哈尔滨工业大学（威海） | Image identification method based on remote domain transfer learning |
| CN114841972A (en) * | 2022-05-09 | 2022-08-02 | 浙江大学 | Power transmission line defect identification method based on saliency map and semantic embedded feature pyramid |
| CN115439470A (en) * | 2022-10-14 | 2022-12-06 | 深圳职业技术学院 | Polyp image segmentation method, computer-readable storage medium, and computer device |
| US20230102467A1 (en) * | 2021-09-29 | 2023-03-30 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method of detecting image, electronic device, and storage medium |
-
2023
- 2023-04-21 CN CN202310433523.4A patent/CN116486230B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111104962A (en) * | 2019-11-05 | 2020-05-05 | 北京航空航天大学青岛研究院 | Semantic segmentation method and device for image, electronic equipment and readable storage medium |
| CN113255699A (en) * | 2021-06-10 | 2021-08-13 | 浙江华睿科技有限公司 | Small target object image detection method and device, electronic equipment and storage medium |
| US20230102467A1 (en) * | 2021-09-29 | 2023-03-30 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method of detecting image, electronic device, and storage medium |
| CN114549555A (en) * | 2022-02-25 | 2022-05-27 | 北京科技大学 | Human ear image planning and division method based on semantic division network |
| CN114783072A (en) * | 2022-03-17 | 2022-07-22 | 哈尔滨工业大学（威海） | Image identification method based on remote domain transfer learning |
| CN114841972A (en) * | 2022-05-09 | 2022-08-02 | 浙江大学 | Power transmission line defect identification method based on saliency map and semantic embedded feature pyramid |
| CN115439470A (en) * | 2022-10-14 | 2022-12-06 | 深圳职业技术学院 | Polyp image segmentation method, computer-readable storage medium, and computer device |
Non-Patent Citations (3)
| Title |
|---|
| J DAI .ETAL: "deformable convolutional networks", IEEE * |
| 杨建华: "基于FPN的高分辨率建筑物图像的语义分割", 遥感信息, vol. 36, no. 04, pages 133 - 141 * |
| 郭启帆 等: "基于特征金字塔的多尺度特征融合网络", 工程数学学报, no. 05 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116486230B (en) | 2024-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6745328B2 (en) | Method and apparatus for recovering point cloud data | |
| US20210158533A1 (en) | Image processing method and apparatus, and storage medium | |
| US11200424B2 (en) | Space-time memory network for locating target object in video content | |
| CN111104962B (en) | Semantic segmentation method and device for image, electronic equipment and readable storage medium | |
| US10984556B2 (en) | Method and apparatus for calibrating relative parameters of collector, device and storage medium | |
| CN108710885B (en) | Target object detection method and device | |
| US9213899B2 (en) | Context-aware tracking of a video object using a sparse representation framework | |
| JP6471448B2 (en) | Noise identification method and noise identification apparatus for parallax depth image | |
| US11100646B2 (en) | Future semantic segmentation prediction using 3D structure | |
| US20190301861A1 (en) | Method and apparatus for binocular ranging | |
| CN112488999B (en) | Small target detection method, small target detection system, storage medium and terminal | |
| CN113538235B (en) | Training method and device for image processing model, electronic equipment and storage medium | |
| CN111860398A (en) | Remote sensing image target detection method and system and terminal equipment | |
| CN109035303B (en) | SLAM system camera tracking method and device, and computer readable storage medium | |
| CN111368668B (en) | Three-dimensional hand recognition method and device, electronic equipment and storage medium | |
| CN113111684B (en) | Training method and device for neural network model and image processing system | |
| CN112884702A (en) | Polyp identification system and method based on endoscope image | |
| CN116486230B (en) | Image detection method based on semi-recursion characteristic pyramid structure and storage medium | |
| CN113610856B (en) | Method and device for training image segmentation model and image segmentation | |
| CN113554550B (en) | Training method and device for image processing model, electronic equipment and storage medium | |
| CN112995433B (en) | Time sequence video generation method and device, computing equipment and storage medium | |
| CN114332509A (en) | Image processing method, model training method, electronic device and automatic driving vehicle | |
| CN113642510A (en) | Target detection method, device, equipment and computer readable medium | |
| CN110610185A (en) | Method, device and equipment for detecting salient object of image | |
| US20230298326A1 (en) | Image augmentation method, electronic device and readable storage medium |
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 |