CN113011415A - Improved target detection method and system based on Grid R-CNN model - Google Patents

Improved target detection method and system based on Grid R-CNN model Download PDF

Info

Publication number
CN113011415A
CN113011415A CN202011343605.2A CN202011343605A CN113011415A CN 113011415 A CN113011415 A CN 113011415A CN 202011343605 A CN202011343605 A CN 202011343605A CN 113011415 A CN113011415 A CN 113011415A
Authority
CN
China
Prior art keywords
anchor frame
grid
anchor
image
cnn model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011343605.2A
Other languages
Chinese (zh)
Inventor
刘嵩
周梓涵
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Qilu University of Technology
Original Assignee
Qilu University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Qilu University of Technology filed Critical Qilu University of Technology
Priority to CN202011343605.2A priority Critical patent/CN113011415A/en
Publication of CN113011415A publication Critical patent/CN113011415A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/24—Aligning, centring, orientation detection or correction of the image
    • G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a Grid R-CNN model-based improved target detection method and a system, comprising the following steps: acquiring a target image to be detected, performing image complexity processing on the image, and then distributing the image to different feature extraction networks to extract feature maps; for the extracted feature map, determining the position of an anchor frame by using grid-guided positioning; dynamically determining the shape of the anchor frame by combining shape prediction based on the position of the anchor frame, and determining the anchor frame with the anchor branch by the position and the shape; processing the anchor frame with the anchor branches and the FSAF branches in parallel to obtain a more accurate anchor frame; and based on the obtained anchor frame, classifying and positioning the target by utilizing a cascade detector. The invention has the beneficial effects that: according to the invention, the complexity of the image is processed, and the number of layers of the network is extracted according to the threshold range feature belonging to the complexity of the feature map, so that the accuracy of feature extraction can be improved. The position prediction is combined with the shape prediction, and the accuracy of the anchor frame is improved by continuously increasing the threshold of the detector.

Description

Improved target detection method and system based on Grid R-CNN model
Technical Field
The invention relates to the technical field of target detection, in particular to a Grid R-CNN (Grid area convolutional neural network) model-based improved target detection method and system.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
With the increase of the computing speed of computers, the artificial intelligence technology continuously obtains new achievements in a plurality of fields. Deep learning, one of the most important artificial intelligence techniques, is one of the latest trends of machine learning and artificial intelligence research.
The object detection is one of basic tasks to be solved in the field of computer vision, and the purpose of the object detection is to determine the position of a target object in an image or a video and the type of the target. With the research on the machine learning theory technology, the detection of the specific class of targets has been achieved well in recent ten years. However, the real environment is complex and various, and many problems are encountered when detecting the target. With the introduction of the deep learning concept, the research on the target detection is not limited to the detection of a certain class of targets, but extends to the detection of multiple classes of targets.
With the proposal of the deep neural network, the target detection is further developed, and the overall method is divided into two methods, namely an anchor frame-based target detection method and an anchor-free target detection method. The method based on the anchor frame is to preset a large number of anchor points for prediction on an image, further refine the anchor points and finally obtain an accurate detection result. The anchorless approach is to use a center point or region of the object to determine the location of the target.
With the increase of the complexity of the picture and the increase of the information contained in the picture, the target detection is more and more difficult, and the accurate positioning and classification of the target is the most important of the target detection.
The existing target detection method achieves end-to-end training and improves the performance of target detection. However, a large number of anchor frames are generally generated, and since the size of the anchor frame is fixed, but the size of the target object to be detected is not fixed, the problem that an object with an extreme size cannot be accurately detected is caused.
Grid R-CNN is a relatively excellent target detection algorithm at present. Although Grid R-CNN makes certain improvement in determination of the shape of the anchor frame, the anchor frame generated by the method cannot accurately mark the target object. And the Grid R-CNN uses a detector with a single IOU threshold value for detection, so that the problem that the anchor frame cannot be positioned more accurately exists.
Disclosure of Invention
In order to solve the problems, the invention provides a Grid R-CNN model-based improved target detection method and system. In the method, the shape prediction method in GA-RPN is used to replace the anchor frame prediction of Grid R-CNN, so that more accurate anchor frame shape can be obtained. Meanwhile, in order to solve the problem of selecting the overlapped anchor frame, the FSAF branch is adopted, and the anchor branch and the Grid R-CNN have anchor branches to perform maximum suppression in parallel, so that a proper anchor frame is selected.
In the implementation mode of the method, the following technical scheme is adopted:
a Grid R-CNN model-based improved target detection method comprises the following steps:
acquiring a target image to be detected, performing image complexity processing on the image, and then distributing the image to different feature extraction networks to extract feature maps;
for the extracted feature map, determining the position of an anchor frame by using grid-guided positioning;
dynamically determining the shape of the anchor frame by combining shape prediction based on the position of the anchor frame, and determining the anchor frame with the anchor branch by the position and the shape;
processing the anchor frame with the anchor branches and the FSAF branches in parallel to obtain a more accurate anchor frame;
and based on the obtained anchor frame, classifying and positioning the target by utilizing a cascade detector.
In other embodiments, the following technical solutions are adopted:
an improved target detection system based on a Grid R-CNN model comprises:
a module for obtaining a target image to be detected, processing the image complexity of the image, and then distributing the image to different feature extraction networks for extracting feature maps;
means for determining a location of an anchor box using grid-guided positioning for the extracted feature map;
means for dynamically determining an anchor frame with anchor branches in conjunction with shape prediction based on anchor frame position;
a module for processing the anchor frame with the anchor branch and the FSAF branch in parallel to obtain a more accurate anchor frame;
and a module for classifying and positioning the target by using the cascade detector based on the obtained anchor frame.
In other embodiments, the following technical solutions are adopted:
a terminal device comprising a processor and a computer-readable storage medium, the processor being configured to implement instructions; the computer-readable storage medium is configured to store a plurality of instructions adapted to be loaded by a processor and to perform the above-described method for object detection based on the Grid R-CNN model refinement.
In other embodiments, the following technical solutions are adopted:
a computer-readable storage medium, wherein a plurality of instructions are stored, the instructions being adapted to be loaded by a processor of a terminal device and to execute the above-mentioned Grid R-CNN model-based improved object detection method.
Compared with the prior art, the invention has the beneficial effects that:
(1) according to the invention, the complexity of the image is processed, and the number of layers of the network is extracted according to the threshold range feature belonging to the complexity of the feature map, so that the accuracy of feature extraction can be improved.
(2) The GA-RPN shape prediction method is used for replacing the Grid R-CNN anchor frame prediction, the GA-RPN anchor frame shape prediction method does not change the position of the anchor frame, and the accuracy of the position of the anchor frame can be ensured to the greatest extent; the anchor frame with the dynamic size can better predict an extremely high or wide object, so that a more accurate anchor frame shape can be obtained.
(3) The invention uses the anchor-free mode of FSAF branch to code the example, and uses the output prediction and the anchor branch to process in parallel, and the two branches work together to achieve the purpose of automatically selecting the best feature.
(4) The Cascade detection method of Cascade R-CNN is used, the output of the previous stage is used as the input of the next stage, and a better detection result is obtained through a continuously improved threshold value.
Additional features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
FIG. 1 is a flow chart of an improved target detection method based on a Grid R-CNN model in the embodiment of the present invention.
FIG. 2 is a diagram of a cascaded detector method in an embodiment of the invention.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
Example one
In one or more embodiments, an improved target detection method based on a Grid R-CNN model is disclosed, as shown in fig. 1, including the following steps:
(1) acquiring a target image to be detected, performing image complexity processing on the image, and then distributing the image to different feature extraction networks to extract feature maps;
specifically, firstly, the image complexity of the picture to be detected is calculated, and the feature map complexity calculation method is as follows:
Figure BDA0002799257850000051
where C represents the complexity of the image texture, z represents the gray scale, p (z)i) A histogram representing z, L representing the number of different grey levels, m representing the mean of z, is i representing the number of all individuals occupied.
The feature extraction is to extract all possible information that a target object may exist in an image on an input image. Due to the fact that the complexity of the images is different, the images with different complexity need feature extraction networks with different scales.
In this embodiment, the number of network layers for feature extraction is determined according to the threshold range to which the image complexity belongs. The pictures with higher complexity enter the processing network with more network layers, and the pictures with lower complexity enter the processing network with less network layers.
(2) The location of the anchor box is determined using grid-guided positioning based on the extracted feature map.
In this embodiment, it is necessary to mark the anchor frame as accurate as possible in the extracted candidate regions. The anchor frame is a bounding box which is generated by taking each pixel as a center and has different length-width ratios and sizes and is determined by combining the coordinate of one pixel point with the length and the width.
First, we use the Grid-guided positioning proposed in Grid R-CNN to determine the location of the anchor frame; the method specifically comprises the following steps:
Figure BDA0002799257850000052
Figure BDA0002799257850000061
wherein (H)x，Hy) Is the pixel coordinate on the heat map output by the trellis prediction branch, (I)x，Iy) Is the coordinate on the original image, (P)x，Py) Is the coordinate of the upper left corner of the input image, wpAnd hpIs the length and the sum of the candidate regions, w0And h0Is the length and width of the output in the heatmap.
(3) Dynamically determining the shape of the anchor frame by combining shape prediction based on the position of the anchor frame, and determining the anchor frame with the anchor branch by the position and the shape;
dynamically determining the shape of an anchor frame by using a GA-RPN shape prediction method; the shape prediction is not a predefined fixed-size anchor frame, but is dynamically variable, with a dynamic anchor frame at a determined position. Anchor boxes with dynamic sizes may be better predicted. The shape at each location is determined using the shape prediction method in the GA-RPN.
The GA-RPN anchor frame shape prediction method does not change the position of the anchor frame, and can ensure the accuracy of the position of the anchor frame to the maximum extent.
In the embodiment, the GA-RPN shape prediction method is used for replacing the Grid R-CNN anchor frame prediction, so that more accurate anchor frame shapes can be obtained.
Since the method of determining the anchor frame of Grid R-CNN is not sufficient to accurately locate the boundary of the target object, the present embodiment proposes to determine the shape of the anchor frame using shape prediction of GA-RPN. The shape prediction is not a predefined fixed-size anchor frame, but is dynamically variable, with a dynamic anchor frame at a determined position. An anchor frame with dynamic dimensions may better predict objects that are extremely high or wide.
The GA-RPN anchor frame shape prediction method does not change the position of an anchor frame, and can ensure the accuracy of the position of the anchor frame to the maximum extent, which is different from the traditional bounding box regression. The shape prediction is to predict the length h and the width w of the anchor frame, and since the numerical range of h and w is large, it is difficult to directly predict the two numerical values, so the embodiment converts h and w. The calculation formula is as follows:
w＝σ·s·edw，h＝σ·s·edh
where s is the step size, σ is the empirical scaling factor, and dw and dh are the mapping of (w, h). The value [0,1000] is converted into [ -1,1] through the formula, the value space is reduced, and the learning target is easier.
In the concrete designUsing a sub-network NsTo perform shape prediction, which is a two-channel 1 × 1 convolutional layer containing dw and dh values.
Due to the dynamic random size, the shape prediction has higher recall rate, and extremely wide or extremely high objects can be predicted better.
(4) According to the determined anchor frame with the anchor branch, the anchor frame and the FSAF branch are processed in parallel, and a more accurate anchor frame is selected;
the FSAF anchorless branch and the anchored branch guided and positioned by the anchor frame are processed in parallel, and two anchorless convolution layers (conv) are additionally added on the basis of the FSAF anchorless branch and the branch based on the anchor frame, so that the anchorless branch and the anchored branch work jointly, and the optimal anchor frame is selected.
(5) And classifying and positioning the target by utilizing the cascade detector according to the determined more accurate anchor frame.
By utilizing a Cascade R-CNN Cascade method, each detector model is trained by cascading a plurality of detector modules, IOU thresholds which are continuously improved in three stages are set, an anchor frame output by a detector in the previous stage is used as input of pooling in the next stage, and a better detection effect is obtained through the continuously improved thresholds.
As shown in FIG. 2, the first stage is backbone network processing, and then using the abstraction header network (N)0) A preliminary prediction is generated. The second phase is candidate area extraction network subnetwork (N)1) For the extraction head network (N)0) Generated prediction result (Bbox)0) And (6) processing. Third and fourth stages replace the region of interest sub-network with N on the basis of repeating the second stage2And N3. Equation (1) is the loss equation for the cascaded detector. The main purpose of the cascade is to train for the next stage to find a higher positive sample of the IOU by adjusting the bounding box. The performance of the detector is improved by the ever-increasing IOU threshold.
L(xt,g)＝Lcls(ht(xt),yt)+λ[yt≥1]Lloc(ft(xt,bt),g) (1)
Wherein, bt＝ft-1(xt-1,bt-1) G is xtThe basic truth object of, lambda trade-off coefficient, [ ·]Indicating function, ytIs xtThe label of (1).
Example two
In one or more embodiments, disclosed is an improved object detection system based on a Grid R-CNN model, comprising:
a module for obtaining a target image to be detected, processing the image complexity of the image, and then distributing the image to different feature extraction networks for extracting feature maps;
means for determining a location of an anchor box using grid-guided positioning for the extracted feature map;
means for dynamically determining an anchor frame with anchor branches in conjunction with shape prediction based on anchor frame position;
a module for processing the anchor frame with the anchor branch and the FSAF branch in parallel to obtain a more accurate anchor frame;
and a module for classifying and positioning the target by using the cascade detector based on the obtained anchor frame.
It should be noted that the specific working process of the above module is the same as the method disclosed in the first embodiment, and is not described again.
EXAMPLE III
In one or more embodiments, a terminal device is disclosed, which includes a server including a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements an object detection method based on a Grid R-CNN model improvement in embodiment one when executing the program. For brevity, no further description is provided herein.
It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general purpose processors, digital signal processors DSP, application specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and so on. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory may include both read-only memory and random access memory, and may provide instructions and data to the processor, and a portion of the memory may also include non-volatile random access memory. For example, the memory may also store device type information.
In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in a processor or instructions in the form of software.
The Grid R-CNN model-based improved target detection method in the first embodiment may be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The software modules may be located in ram, flash, rom, prom, or eprom, registers, among other storage media as is well known in the art. The storage medium is located in a memory, and a processor reads information in the memory and completes the steps of the method in combination with hardware of the processor. To avoid repetition, it is not described in detail here.
Those of ordinary skill in the art will appreciate that the various illustrative elements, i.e., algorithm steps, described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
Example four
In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, the instructions being adapted to be loaded by a processor of a terminal device and implementing the Grid R-CNN model-based improved object detection method described in the first embodiment.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.

Claims (9)

1. A Grid R-CNN model-based improved target detection method is characterized by comprising the following steps:
acquiring a target image to be detected, performing image complexity processing on the image, and then distributing the image to different feature extraction networks to extract feature maps;
for the extracted feature map, determining the position of an anchor frame by using grid-guided positioning;
dynamically determining the shape of the anchor frame by combining shape prediction based on the position of the anchor frame, and determining the anchor frame with the anchor branch by the position and the shape;
processing the anchor frame with the anchor branches and the FSAF branches in parallel to obtain a more accurate anchor frame;
and based on the obtained anchor frame, classifying and positioning the target by utilizing a cascade detector.
2. The Grid R-CNN model-based improved target detection method as claimed in claim 1, wherein the image complexity processing is performed on the image, and the specific process comprises:
calculating the complexity of the pictures to be detected, determining the number of network layers for feature extraction according to the threshold range to which the complexity of the pictures belongs, and inputting the pictures with different complexities into feature extraction networks with different numbers of layers.
3. The improved object detection method based on the Grid R-CNN model as claimed in claim 1, wherein the Grid-guided localization is used to determine the position of the anchor frame by the following specific procedures:
and predicting the positions of the predefined grid points by using the full convolution network, and selecting one point with the maximum probability value from the generated grid points as the position location of the prediction anchor frame.
4. The improved target detection method based on the Grid R-CNN model as claimed in claim 1, wherein the anchor frame with anchor branches is dynamically determined based on the anchor frame position in combination with shape prediction, specifically comprising:
dynamically determining the shape of an anchor frame by using a GA-RPN shape prediction method; the size of the anchor frame is dynamically variable.
5. The improved target detection method based on the Grid R-CNN model as claimed in claim 1, wherein processing the anchor frame with the anchor branch and the FSAF branch in parallel specifically comprises:
and additionally adding two non-anchored convolution layers on the basis of the anchored branch, wherein the two non-anchored convolution layers are respectively used for classifying the branch and predicting regression.
6. The improved target detection method based on the Grid R-CNN model as claimed in claim 1, wherein based on the obtained anchor frame, the classification and positioning of the target are performed by using a cascade detector, the specific process is as follows:
by utilizing a Cascade R-CNN Cascade method, each detector model is trained by cascading a plurality of detector modules, IOU thresholds which are continuously improved in three stages are set, an anchor frame output by a detector in the previous stage is used as input of pooling in the next stage, and a better detection effect is obtained through the continuously improved thresholds.
7. An improved target detection system based on a Grid R-CNN model is characterized by comprising:
a module for obtaining a target image to be detected, processing the image complexity of the image, and then distributing the image to different feature extraction networks for extracting feature maps;
means for determining a location of an anchor box using grid-guided positioning for the extracted feature map;
means for dynamically determining an anchor frame with anchor branches in conjunction with shape prediction based on anchor frame position;
a module for processing the anchor frame with the anchor branch and the FSAF branch in parallel to obtain a more accurate anchor frame;
and a module for classifying and positioning the target by using the cascade detector based on the obtained anchor frame.
8. A terminal device comprising a processor and a computer-readable storage medium, the processor being configured to implement instructions; the computer-readable storage medium storing a plurality of instructions adapted to be loaded by a processor and to perform the method for object detection based on the Grid R-CNN model refinement of any of claims 1-6.
9. A computer-readable storage medium having stored therein a plurality of instructions, wherein the instructions are adapted to be loaded by a processor of a terminal device and to perform the Grid R-CNN model-based improved object detection method according to any one of claims 1 to 6.
CN202011343605.2A 2020-11-25 2020-11-25 Improved target detection method and system based on Grid R-CNN model Pending CN113011415A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011343605.2A CN113011415A (en) 2020-11-25 2020-11-25 Improved target detection method and system based on Grid R-CNN model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011343605.2A CN113011415A (en) 2020-11-25 2020-11-25 Improved target detection method and system based on Grid R-CNN model

Publications (1)

Publication Number Publication Date
CN113011415A true CN113011415A (en) 2021-06-22

Family

ID=76383213

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011343605.2A Pending CN113011415A (en) 2020-11-25 2020-11-25 Improved target detection method and system based on Grid R-CNN model

Country Status (1)

Country Link
CN (1) CN113011415A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113435302A (en) * 2021-06-23 2021-09-24 中国农业大学 GridR-CNN-based hydroponic lettuce seedling state detection method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200151564A1 (en) * 2018-11-12 2020-05-14 Honda Motor Co., Ltd. System and method for multi-agent reinforcement learning with periodic parameter sharing
CN111178196A (en) * 2019-12-19 2020-05-19 东软集团股份有限公司 Method, device and equipment for cell classification
CN111950612A (en) * 2020-07-30 2020-11-17 中国科学院大学 FPN-based weak and small target detection method for fusion factor

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200151564A1 (en) * 2018-11-12 2020-05-14 Honda Motor Co., Ltd. System and method for multi-agent reinforcement learning with periodic parameter sharing
CN111178196A (en) * 2019-12-19 2020-05-19 东软集团股份有限公司 Method, device and equipment for cell classification
CN111950612A (en) * 2020-07-30 2020-11-17 中国科学院大学 FPN-based weak and small target detection method for fusion factor

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
CHENCHEN ZHU ET AL.: "Feature Selective Anchor-Free Module for Single-Shot Object Detection", 《ARXIV:1903.00621V1 [CS.CV]》 *
JIAQI WANG ET AL.: "Region Proposal by Guided Anchoring", 《ARXIV:1901.03278V1 [CS.CV]》 *
XIN LU ET AL.: "Grid R-CNN", 《ARXIV:1811.12030V1 [CS.CV]》 *
ZHAOWEI CAI ET AL.: "Cascade R-CNN: Delving into High Quality Object Detection", 《ARXIV:1712.00726V1 [CS.CV]》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113435302A (en) * 2021-06-23 2021-09-24 中国农业大学 GridR-CNN-based hydroponic lettuce seedling state detection method
CN113435302B (en) * 2021-06-23 2023-10-17 中国农业大学 Hydroponic lettuce seedling state detection method based on GridR-CNN

Similar Documents

Publication Publication Date Title
CN110400332B (en) Target detection tracking method and device and computer equipment
CN110796048B (en) Ship target real-time detection method based on deep neural network
CN110781756A (en) Urban road extraction method and device based on remote sensing image
CN111753682B (en) Hoisting area dynamic monitoring method based on target detection algorithm
CN111627050B (en) Training method and device for target tracking model
WO2023193401A1 (en) Point cloud detection model training method and apparatus, electronic device, and storage medium
CN111640089A (en) Defect detection method and device based on feature map center point
CN114897779A (en) Cervical cytology image abnormal area positioning method and device based on fusion attention
CN112419202B (en) Automatic wild animal image recognition system based on big data and deep learning
WO2023116632A1 (en) Video instance segmentation method and apparatus based on spatio-temporal memory information
CN113487610B (en) Herpes image recognition method and device, computer equipment and storage medium
CN113807361B (en) Neural network, target detection method, neural network training method and related products
CN113095418A (en) Target detection method and system
CN113592060A (en) Neural network optimization method and device
CN115187530A (en) Method, device, terminal and medium for identifying ultrasonic automatic breast full-volume image
CN112348116A (en) Target detection method and device using spatial context and computer equipment
CN111523463A (en) Target tracking method and training method based on matching-regression network
CN115578616A (en) Training method, segmentation method and device of multi-scale object instance segmentation model
CN116486288A (en) Aerial target counting and detecting method based on lightweight density estimation network
CN115187786A (en) Rotation-based CenterNet2 target detection method
CN113554656B (en) Optical remote sensing image example segmentation method and device based on graph neural network
CN113763412A (en) Image processing method and device, electronic equipment and computer readable storage medium
CN113011415A (en) Improved target detection method and system based on Grid R-CNN model
CN113256683A (en) Target tracking method and related equipment
CN111914809A (en) Target object positioning method, image processing method, device and computer equipment

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