CN112183579B - Method, medium and system for detecting micro target - Google Patents

Method, medium and system for detecting micro target Download PDF

Info

Publication number
CN112183579B
CN112183579B CN202010905792.2A CN202010905792A CN112183579B CN 112183579 B CN112183579 B CN 112183579B CN 202010905792 A CN202010905792 A CN 202010905792A CN 112183579 B CN112183579 B CN 112183579B
Authority
CN
China
Prior art keywords
target
tiny
network
resnet
micro
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010905792.2A
Other languages
Chinese (zh)
Other versions
CN112183579A (en
Inventor
赵欣洋
叶涛
王玄之
崔鹏
杜巍
赵希洋
朱颖
刘亮
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.)
State Grid Ningxia Electric Power Co Ltd
Original Assignee
State Grid Ningxia Electric Power Co Ltd
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 State Grid Ningxia Electric Power Co Ltd filed Critical State Grid Ningxia Electric Power Co Ltd
Priority to CN202010905792.2A priority Critical patent/CN112183579B/en
Publication of CN112183579A publication Critical patent/CN112183579A/en
Application granted granted Critical
Publication of CN112183579B publication Critical patent/CN112183579B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a method, a medium and a system for detecting a micro target. The method comprises the following steps: inputting an image containing a micro target into a deep learning target detection network, and outputting the micro target extracted from the image, wherein the area of the micro target is smaller than a preset area; inputting the extracted micro target into an amplifying deep learning network, and outputting the micro target amplified by a preset multiple; and inputting the tiny targets amplified by a preset multiple into a deep learning classification network, and outputting the categories of the tiny targets. The invention can effectively improve the detection precision of the micro target and reduce the false detection rate, thereby improving the performance and stability of the whole computer vision system and having better practical application value and economic benefit.

Description

Method, medium and system for detecting micro target
Technical Field
The present invention relates to the field of target detection technologies, and in particular, to a method, medium, and system for detecting a micro target.
Background
In recent years, with the continuous development and breakthrough of deep learning and computer computing power, the deep learning achieves incomparable effects of the traditional algorithm in the fields of computer vision such as classification tasks, target detection, semantic segmentation and the like, and is applied to various industries.
In the field of target detection, currently mainstream application algorithms are: SSD, fasterRcnn, YOLO, etc., which have good effects in practical applications, but these algorithms have a problem that they generally have low detection accuracy when detecting a minute target. The main reason is that firstly, the tiny target contains fewer pixels, the data information carried by the tiny target is less, and secondly, the general rule of deep learning is that the deeper the network depth is, the higher the detection precision is. Therefore, the tiny target can pass through a deeper deep learning network, and less information carried by the tiny target is almost disappeared in a deeper characteristic diagram under the effect of downsampling, so that the detection precision of the tiny target is reduced.
Disclosure of Invention
The embodiment of the invention provides a method, a medium and a system for detecting a micro target, which are used for solving the problem of low precision of detecting the micro target in the prior art.
In a first aspect, a method for detecting a minute object is provided, including: inputting an image containing a micro target into a deep learning target detection network, and outputting the micro target extracted from the image, wherein the area of the micro target is smaller than a preset area; inputting the extracted micro target into an amplifying deep learning network, and outputting the micro target amplified by a preset multiple; and inputting the tiny targets amplified by a preset multiple into a deep learning classification network, and outputting the categories of the tiny targets.
In a second aspect, there is provided a computer readable storage medium having computer program instructions stored thereon; the computer program instructions, when executed by a processor, implement the minute object detection method as described in the embodiment of the first aspect.
In a third aspect, there is provided a minute object detection system comprising: the computer readable storage medium as in the second aspect embodiment.
Therefore, the embodiment of the invention can effectively improve the detection precision of the micro target and reduce the false detection rate, thereby improving the performance and stability of the whole computer vision system and having better practical application value and economic benefit.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a minute object detection method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an enlarged deep learning network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of the partial structure of a residual ResNet network of an embodiment of the present invention;
figure 4 is a schematic diagram of a period shuffler operator in accordance with an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The embodiment of the invention discloses a micro target detection method. As shown in fig. 1, the minute object detection method includes the steps of:
step S1: after an image containing a micro target is input into a deep learning target detection network, the micro target extracted from the image is output.
The deep learning object detection network employs a well-known neural network, such as a YoloV3 object detection network, an SSD object detection network, a FasterRcnn object detection network. It should be appreciated that the images of a typical input deep learning object detection network should all be scaled to a certain size. Therefore, before step S1, the method according to the embodiment of the present invention further includes: the original image is scaled to an image of preset pixels. The preset pixels may be determined based on the selected deep learning object detection network. For example, for a YoloV3 target detection network, the scaled image is a 608×608 (px) image. Specifically, the original image can be read by the immread method of opencv, and then the read original image is scaled to a size of 608×608 by the resize method of opencv.
The area of the micro target in the embodiment of the invention is smaller than the preset area. For example, in the embodiment of the present invention, the preset area is 20×20 (px). It should be appreciated that the tiny target is directed to an image input to the deep learning target detection network and not the original image.
Specifically, the deep learning object detection network may output the width and the height of the image, and may calculate the area according to the product of the width and the height, thereby determining whether it belongs to a minute object.
Step S2: and inputting the extracted micro target into an amplifying deep learning network, and outputting the micro target amplified by a preset multiple.
Specifically, as shown in fig. 2, the structure of the amplifying deep learning network is composed of a tiny target ResNet network structure block and a sub-pixel convolution layer which are connected in sequence.
The micro target ResNet network structure block consists of residual ResNet networks which are cascaded for preset times, a deeper network structure is formed, and micro target depth semantic information can be fully extracted. In the embodiment of the invention, the residual ResNet network is cascaded for 9 times. As shown in fig. 3, the two weight layers (weight layers) in the residual res net network are replaced with a dense densnet network. The residual ResNet network and the dense DenseNet network are well known network structures and are not described in detail herein.
Output result X of residual ResNet network l ＝H l (X l-1 ,w i ,b i )+X l-1 . Wherein X is l-1 Representing the input object of the current residual res net network. H l Representing a function of a dense DenseNet network. w (w) i And b i Parameters representing a dense DenseNet network can be derived from model training in conjunction with instances.
Tiny target feature map F output by tiny target ResNet network structure block l-1 ＝R(X lr W, B). Wherein X is lr Representing the tiny target of the incoming tiny target ResNet network fabric block. R represents a nonlinear function of the tiny target ResNet network structure block, and is generally a Relu nonlinear function. W and B represent parameter weights and bias values of the micro target ResNet network structure block, and can be obtained by model training in combination with examples. The size of the micro target feature map is w×h×c×r 2 . r represents the magnification of the minute object feature map. w×h×c represents the size of the minute object inputting the minute object res net network structural block, w represents the width of the minute object inputting the minute object res net network structural block, h represents the height of the minute object inputting the minute object res net network structural block, and c represents the number of image channels, typically 3, that is, RGB three channels.
By combining the residual ResNet network with the dense DenseNet network, a new network structure block is provided, the residual ResNet network ensures the depth of the network, and the dense DenseNet network ensures that more tiny target information is transferred to the deep feature map.
Because the small target itself carries less information because of less pixels, in order to make the amplified target carry more target characteristic information, the embodiment of the invention amplifies the small target by utilizing sub-pixel convolution. Specifically, the subpixel convolution formula of the subpixel convolution layer is I sr ＝PS(W l ×F l-1 +B l ). Wherein I is sr Representing the output result of the subpixel convolutional layer. F (F) l-1 And a tiny target feature map which represents the output of the tiny target ResNet network structure block. PS represents a period shuffle operator, as shown in FIG. 4, low-pixel feature map is obtained by r 2 The convolution kernels yield r 2 A characteristic diagram, and then r 2 Sub-pixel points in the feature map are sequentially arranged from left to right and from top to bottom to obtain a final amplified image. W (W) l And B l Respectively represent sub-pixel convolutionThe weight and bias values can be obtained by model training in combination with examples. In the embodiment of the invention, the preset multiple is 8 times, and the output tiny target is 8*w i ×8*h i X 3, wherein 3 represents R, G, B channels.
Step S3: and inputting the tiny targets amplified by a preset multiple into a deep learning classification network, and outputting the categories of the tiny targets.
The deep learning classification network adopts a well-known neural network, such as an ideptionv 4 classification network, a vgg classification network and a resnet50 classification network.
In one embodiment of the present invention, the result (p) is output using an accept v4 classification network i ,c i ). Wherein p is i Representing confidence of the ith tiny object, c i The classification result of the i-th minute object is represented. Generally, the classification result is determined by a specific example, and may include, for example, pedestrians and vehicles.
In addition to the above classification result of the micro target, the method according to the embodiment of the present invention may further detect the position information of the micro target, and preferably, the micro target detection method according to the embodiment of the present invention further includes the following steps:
and inputting the image containing the micro target into a deep learning target detection network, and outputting the position information of the micro target.
The deep learning object detection network is described in the foregoing step S1, and will not be described in detail herein. Outputting the position information (x) of all the minute objects through the deep learning object detection network i ,y i ,w i ,h i ). Wherein x is i An abscissa representing the center of the ith minute object, y i An ordinate representing the center of the ith minute object, w i For the width of the ith micro target, h i Is the height of the ith tiny target. w (w) i And h i Can be used to calculate the area of the output object to determine if the object is a tiny object.
Therefore, the tiny target detection method of the embodiment of the invention not only can detect the category of the tiny target, but also can obtain the position information of the tiny targetRest, available (p) i ,c i ,x i ,y i ,w i ,h i ) Representing the combined result containing the minute object category and the location information.
The embodiment of the invention also discloses a computer readable storage medium, wherein the computer readable storage medium is stored with computer program instructions; the computer program instructions, when executed by a processor, implement the minute object detection method as described in the above embodiments.
The embodiment of the invention also discloses a micro target detection system, which comprises: the computer-readable storage medium as in the above embodiments.
In summary, the embodiment of the invention can effectively improve the detection precision of the micro target and reduce the false detection rate, thereby improving the performance and stability of the whole computer vision system and having better practical application value and economic benefit.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.

Claims (9)

1. A method for detecting a minute object, comprising:
inputting an image containing a micro target into a deep learning target detection network, and outputting the micro target extracted from the image, wherein the area of the micro target is smaller than a preset area;
inputting the extracted micro target into an amplifying deep learning network, and outputting the micro target amplified by a preset multiple;
inputting the tiny targets amplified by a preset multiple into a deep learning classification network, and outputting the categories of the tiny targets;
the structure of the amplifying deep learning network consists of tiny target ResNet network structure blocks and sub-pixel convolution layers which are sequentially connected, wherein the tiny target ResNet network structure blocks consist of residual ResNet networks with cascading preset times, two weight layers in the residual ResNet networks are replaced by dense DenseNet networks, the residual ResNet networks ensure the depth of the networks, and the dense DenseNet networks ensure that more tiny target information is transferred to a deep feature map.
2. The minute object detection method according to claim 1, characterized in that: output result X of the residual ResNet network l ＝H l (X l-1 ,w i ,b i )+X l-1 Wherein X is l-1 Representing the input object, H, of the residual ResNet network l A function representing the dense DenseNet network, w i And b i Parameters representing the dense DenseNet network.
3. The minute object detection method according to claim 1, characterized in that: a tiny target feature map F output by the tiny target ResNet network structure block l-1 ＝R(X lr W, B), wherein X lr Representing the tiny targets input into the tiny target ResNet network structure block, R representing a nonlinear function of the tiny target ResNet network structure block, and W and B representing parameter weights and bias values of the tiny target ResNet network structure block.
4. The minute object detection method according to claim 3, wherein: the size of the tiny target feature map is w×h×c×r 2 R represents the magnification of the tiny target feature map, and w×h×c represents the size of the tiny target input to the tiny target ResNet network structure block.
5. The minute object detection method according to claim 1, characterized in that: the subpixel convolution formula of the subpixel convolution layer is I sr ＝PS(W l ×F l-1 +B l ) Wherein I sr Representing the output result of the sub-pixel convolution layer, F l-1 Representing a tiny target feature map output by the tiny target ResNet network structure block, wherein PS represents a periodic shuffling operator, W l And B l Respectively representing the sub-pixel convolution weight and the bias.
6. The minute object detection method according to claim 1, characterized by further comprising:
and inputting the image containing the micro target into a deep learning target detection network, and outputting the position information of the micro target.
7. The minute object detection method according to claim 1 or 6, wherein before the step of inputting the image containing the minute object into the deep learning object detection network, the method further comprises:
scaling the original image to the image of the preset pixels.
8. A computer-readable storage medium, characterized by: the computer readable storage medium has stored thereon computer program instructions; the computer program instructions, when executed by a processor, implement the minute object detection method according to any of claims 1 to 7.
9. A minute object detection system, characterized by comprising: the computer-readable storage medium of claim 8.
CN202010905792.2A 2020-09-01 2020-09-01 Method, medium and system for detecting micro target Active CN112183579B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010905792.2A CN112183579B (en) 2020-09-01 2020-09-01 Method, medium and system for detecting micro target

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010905792.2A CN112183579B (en) 2020-09-01 2020-09-01 Method, medium and system for detecting micro target

Publications (2)

Publication Number Publication Date
CN112183579A CN112183579A (en) 2021-01-05
CN112183579B true CN112183579B (en) 2023-05-30

Family

ID=73924095

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010905792.2A Active CN112183579B (en) 2020-09-01 2020-09-01 Method, medium and system for detecting micro target

Country Status (1)

Country Link
CN (1) CN112183579B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108564097A (en) * 2017-12-05 2018-09-21 华南理工大学 A kind of multiscale target detection method based on depth convolutional neural networks
CN109344821A (en) * 2018-08-30 2019-02-15 西安电子科技大学 Small target detecting method based on Fusion Features and deep learning
CN109635666A (en) * 2018-11-16 2019-04-16 南京航空航天大学 A kind of image object rapid detection method based on deep learning
CN111179212A (en) * 2018-11-10 2020-05-19 杭州凝眸智能科技有限公司 Method for realizing micro target detection chip integrating distillation strategy and deconvolution

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108564097A (en) * 2017-12-05 2018-09-21 华南理工大学 A kind of multiscale target detection method based on depth convolutional neural networks
CN109344821A (en) * 2018-08-30 2019-02-15 西安电子科技大学 Small target detecting method based on Fusion Features and deep learning
CN111179212A (en) * 2018-11-10 2020-05-19 杭州凝眸智能科技有限公司 Method for realizing micro target detection chip integrating distillation strategy and deconvolution
CN109635666A (en) * 2018-11-16 2019-04-16 南京航空航天大学 A kind of image object rapid detection method based on deep learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
双路径反馈网络的图像超分辨重建算法;陶状等;《计算机系统应用》;20200405;第181–186页 *
基于多尺度稠密卷积网络的单图像超分辨率重建;唐家福等;《包装工程》;20200710(第13期);全文 *

Also Published As

Publication number Publication date
CN112183579A (en) 2021-01-05

Similar Documents

Publication Publication Date Title
CN109447994A (en) In conjunction with the remote sensing image segmentation method of complete residual error and Fusion Features
CN114627052A (en) Infrared image air leakage and liquid leakage detection method and system based on deep learning
CN113706581B (en) Target tracking method based on residual channel attention and multi-level classification regression
CN107844743A (en) A kind of image multi-subtitle automatic generation method based on multiple dimensioned layering residual error network
CN111640089A (en) Defect detection method and device based on feature map center point
CN111046917B (en) Object-based enhanced target detection method based on deep neural network
CN106650615A (en) Image processing method and terminal
CN112487915B (en) Pedestrian detection method based on Embedded YOLO algorithm
CN109447897B (en) Real scene image synthesis method and system
CN112364931A (en) Low-sample target detection method based on meta-feature and weight adjustment and network model
CN112418165B (en) Small-size target detection method and device based on improved cascade neural network
CN116310693A (en) Camouflage target detection method based on edge feature fusion and high-order space interaction
CN111611925A (en) Building detection and identification method and device
CN116403127A (en) Unmanned aerial vehicle aerial image target detection method, device and storage medium
CN115311550B (en) Remote sensing image semantic change detection method and device, electronic equipment and storage medium
CN115880495A (en) Ship image target detection method and system under complex environment
CN106600613A (en) Embedded GPU-based improved LBP infrared target detection method
CN116977674A (en) Image matching method, related device, storage medium and program product
CN115170978A (en) Vehicle target detection method and device, electronic equipment and storage medium
CN115792913A (en) Radar echo extrapolation method and system based on time-space network
CN113554656B (en) Optical remote sensing image example segmentation method and device based on graph neural network
CN113592885B (en) SegNet-RS network-based large obstacle contour segmentation method
Zhao et al. ST-YOLOA: a Swin-transformer-based YOLO model with an attention mechanism for SAR ship detection under complex background
CN111340139B (en) Method and device for judging complexity of image content
CN112183579B (en) Method, medium and system for detecting micro target

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