CN111881984A - Target detection method and device based on deep learning - Google Patents

Target detection method and device based on deep learning Download PDF

Info

Publication number
CN111881984A
CN111881984A CN202010753252.7A CN202010753252A CN111881984A CN 111881984 A CN111881984 A CN 111881984A CN 202010753252 A CN202010753252 A CN 202010753252A CN 111881984 A CN111881984 A CN 111881984A
Authority
CN
China
Prior art keywords
target
data
sampling
image
detected
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
CN202010753252.7A
Other languages
Chinese (zh)
Inventor
赵文超
张樯
李斌
张蛟淏
侯棋文
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.)
Beijing Institute of Environmental Features
Original Assignee
Beijing Institute of Environmental Features
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 Beijing Institute of Environmental Features filed Critical Beijing Institute of Environmental Features
Priority to CN202010753252.7A priority Critical patent/CN111881984A/en
Publication of CN111881984A publication Critical patent/CN111881984A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06F18/211—Selection of the most significant subset of features
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Abstract

The invention discloses a target identification detection method and a target identification detection device based on deep learning, wherein the method comprises the following steps: a data sampling step, a feature extraction step and a target detection step; the data sampling step comprises: carrying out down-sampling processing on the acquired original image of the target to be detected for multiple times to obtain multi-sampling data; the feature extraction step includes: extracting deep features and abstract image features of the multiple times of sampling data respectively, and extracting image features for target detection; the target detection step includes: and identifying the deep features and the abstract image features which are sampled for multiple times through a neural network, and determining the detection result of the target to be detected in the original image. The invention can realize high detection rate and low false scene rate in long-distance air small target detection.

Description

Target detection method and device based on deep learning
Technical Field
The invention relates to the technical field of imaging, in particular to a target identification detection method and device based on deep learning.
Background
The remote target detection is often required in the fields of national defense safety, terrain observation, civil aviation guarantee and the like, and especially in the field of air defense, a longer early warning distance means that an air defense system has longer response time. However, long-distance objects often exhibit small object characteristics on images, small number of pixels, unobvious features, and are easily mixed with the background. Therefore, the false alarm rate is high and the detection rate is low during detection. This makes small target detection always a research focus in the field of target detection.
Existing target detection algorithms are largely classified into two categories, two-phase algorithms and single-phase algorithms. The two-stage algorithm firstly extracts candidate regions according to the target position, and then carries out image classification in each candidate region. The method needs to scan the picture twice, has complicated steps and low calculation speed, and cannot meet the real-time requirement. The single-stage algorithm is an end-to-end algorithm based on regression, the method utilizes a convolutional neural network to predict the position and the type of a target at the same time, the calculation speed is high, the method can be used in engineering application with high real-time requirement, but the precision is low when dense targets and small targets are predicted.
Disclosure of Invention
The invention aims to solve the technical problem of providing a target identification detection method and a target identification detection device based on deep learning, and realizing high detection rate and low false alarm rate in long-distance air small target detection.
The invention discloses a target detection method based on deep learning, which comprises the following steps: a data sampling step, a feature extraction step and a target detection step;
the data sampling step comprises: carrying out down-sampling processing on the acquired original image of the target to be detected for multiple times to obtain multi-sampling data;
the feature extraction step includes: extracting deep features and abstract image features of the multiple times of sampling data respectively, and extracting image features for target detection;
the target detection step includes: and identifying the deep features and the abstract image features which are sampled for multiple times through a neural network, and determining the detection result of the target to be detected in the original image.
Preferably, the method previously comprises: a target data set is acquired and constructed, the data set including bird image data and/or drone image data.
Preferably, the down-sampling processing the acquired original image of the object to be detected for a plurality of times includes:
and sampling the target image of the original image in different scales for multiple times, wherein the step length of each sampling is N unit pixels, and N is a positive integer.
Preferably, the extracting deep features and abstract image features from the multiple times of sampling data respectively comprises:
carrying out residual error processing on sampling data obtained by each downsampling, and extracting the characteristics of the target to be detected for position identification;
and fusing the characteristics of the target to be detected sampled currently and the characteristics of the target to be detected sampled at the previous time, and taking the fused characteristics as deep characteristics and abstract image characteristics.
Preferably, the residual processing of the sample data comprises:
and respectively passing the sampling data through a 3 × 3 convolution layer, a 1 × 1 convolution layer and a residual layer, wherein the residual layer is used for avoiding gradient disappearance when the network depth of the sampling data is increased.
Preferably, after acquiring and building the target data set, the method further comprises:
for each cell of different scale characteristic graphs in the data set, three boundary boxes are predicted by using the prior boxes, a plurality of values are predicted for each boundary box, and the central coordinate of the target and the prediction result of the width and the height of the target are obtained according to the predicted values.
Preferably, the obtaining of the prediction results of the center coordinates of the object and the width and height of the object according to the plurality of predicted values includes:
predict 4 values for each bounding box, denoted t respectivelyx,ty,tw,th；
Obtaining the target center coordinates (x, y) and the width w and height h prediction result b of the target according to the following formulax,by,bw,bh。
bx＝σ(tx)+cx
by＝σ(ty)+cy
Figure BDA0002610690300000031
Figure BDA0002610690300000032
Where σ is a function of t, txAnd tyIs the predicted coordinate offset in the x and y directions; t is twAnd thIs a scaled value of width and height;
cxand cyIs the coordinate of the upper left corner of the cell in which the target is located, PwAnd PyIs the width and height of the prior box.
Preferably, the original image is downsampled by a step size of 2 five times, and the sampled data obtained by the downsampling 5 times are respectively subjected to residual processing 1 time, residual processing 4 times, residual processing 8 times and residual processing 4 times.
Preferably, the target data set comprises: positive samples and negative samples, the positive samples being correctly identified samples and the negative samples being incorrectly identified samples.
In a second aspect, the present invention further provides a target detection apparatus based on deep learning, including: the system comprises a data sampling module, a feature extraction module and a target detection module;
the data sampling module: performing multiple downsampling processing on an acquired original image of a target to be detected to obtain multiple sampled data;
the feature extraction module: extracting deep layer characteristics and abstract image characteristics of the multiple times of sampling data respectively, and extracting image characteristics for target detection;
the target detection module: and identifying deep features and abstract image features sampled for multiple times through a neural network, and determining a detection result of the target to be detected in the original image.
Compared with the prior art, the invention has the following advantages:
the method firstly carries out targeted image data acquisition and labeling on various unmanned aerial vehicles. And then, processing the label of the unmanned aerial vehicle data by using a clustering algorithm, and selecting a prior frame suitable for the unmanned aerial vehicle target. And then carrying out data enhancement on the unmanned aerial vehicle data. And inputting the enhanced data into a neural network for training, wherein the trained model can complete the target detection of the unmanned aerial vehicle.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the example serve to explain the principles of the invention and not to limit the invention.
FIG. 1 is a flowchart of a deep learning-based target detection method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a target detection network structure according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a deep learning-based target detection apparatus according to an embodiment of the present invention;
FIG. 4 is a block diagram of a computing device according to another embodiment of the present application;
fig. 5 is a diagram of a computer-readable storage medium structure according to another embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
The steps illustrated in the flow charts of the figures may be performed in a computer system such as a set of computer-executable instructions. Also, while a logical order is shown in the flow diagrams, in some cases, the steps shown or described may be performed in an order different than here.
Example one
As shown in fig. 1, an embodiment of the present invention provides a target detection method based on deep learning, including: a data sampling step, a feature extraction step and a target detection step;
s101, the data sampling step comprises: carrying out down-sampling processing on the acquired original image of the target to be detected for multiple times to obtain multi-sampling data;
s102, the characteristic extraction step comprises the following steps: extracting deep features and abstract image features of the multiple times of sampling data respectively, and extracting image features for target detection;
s103, the target detection step comprises: and identifying the deep features and the abstract image features which are sampled for multiple times through a neural network, and determining the detection result of the target to be detected in the original image.
The embodiment of the invention carries out down sampling on the input image for multiple times, detects the targets with different scales on the subsequent sampling, and utilizes the deep-layer characteristics and the abstract image characteristics to ensure that the targets with multiple scales can be detected, and simultaneously, the small target detection has abundant enough image characteristics, thereby realizing high detection rate and low false scene rate in the long-distance air small target detection.
In the embodiment of the invention, the method comprises the following steps: a target data set is acquired and constructed, the data set including bird image data and/or drone image data.
The embodiment of the invention firstly acquires and builds a small target data set, wherein the target data set mainly comprises images of birds and various unmanned aerial vehicles at a distance of about two kilometers.
In the embodiment of the invention, a camera with a focal length of 800mm and a resolution of 1920 x 1080 is selected for data acquisition. The pixel size of the camera is 3 μm. The size of mainstream small consumer grade drones and birds is 200mm to 500 mm.
The image distance can be calculated according to the following gaussian imaging formula when the object is two kilometers.
Figure BDA0002610690300000051
Wherein f is focal length, u is object distance, v is image distance, the size of the imaged object can be obtained according to the image distance and the size of the object, and the number of pixels corresponding to the object can be obtained by dividing the imaged size by the size of the pixel. The calculated number of pixels of the target aimed at by the embodiment of the invention is mostly distributed below 50 pixels by 50 pixels, and the proportion of the pixels in the whole image is less than 0.15%, so that the target is considered as a small target. During collection, the unmanned aerial vehicle and birds in different weather, different backgrounds and different flight postures are shot in a long distance (2 kilometers).
In the embodiment of the invention, the processing of downsampling the acquired original image of the target to be detected for multiple times comprises the following steps:
and sampling the target image of the original image in different scales for multiple times, wherein the step length of each sampling is N unit pixels, and N is a positive integer.
In the embodiment of the present invention, the extracting deep features and abstract image features from the multiple sampling data respectively includes:
carrying out residual error processing on sampling data obtained by each downsampling, and extracting the characteristics of the target to be detected for position identification;
and fusing the characteristics of the target to be detected sampled currently and the characteristics of the target to be detected sampled at the previous time, and taking the fused characteristics as deep characteristics and abstract image characteristics.
In the embodiment of the present invention, the processing of the residual error of the sample data includes:
and respectively passing the sampling data through a 3 × 3 convolution layer, a 1 × 1 convolution layer and a residual layer, wherein the residual layer is used for avoiding gradient disappearance when the network depth of the sampling data is increased.
In the embodiment of the present invention, after the target data set is collected and built, the method further includes:
for each cell of different scale characteristic graphs in the data set, three boundary boxes are predicted by using the prior boxes, a plurality of values are predicted for each boundary box, and the central coordinate of the target and the prediction result of the width and the height of the target are obtained according to the predicted values.
In the embodiment of the present invention, obtaining the prediction results of the center coordinate of the target and the width and height of the target according to the plurality of predicted values includes:
predict 4 values for each bounding box, denoted t respectivelyx,ty,tw,th；
Obtaining the target center coordinates (x, y) and the width w and height h prediction result b of the target according to the following formulax,by,bw,bh。
bx＝σ(tx)+cx
by＝σ(ty)+cy
Figure BDA0002610690300000061
Figure BDA0002610690300000071
Where σ is a function of t, txAnd tyIs the predicted coordinate offset in the x and y directions; t is twAnd thIs a scaled value of width and height;
cxand cyIs the coordinate of the upper left corner of the cell in which the target is located, PwAnd PyIs the width and height of the prior box.
In the embodiment of the invention, the sigma is fitted by a deep learning network.
In the embodiment of the present invention, the original image is downsampled by a step length of 2 five times, and sampling data obtained by downsampling 5 times are respectively subjected to residual error processing 1 time, residual error processing 4 times, residual error processing 8 times, and residual error processing 4 times.
The network structure of the embodiment of the invention is improved on a dark net53 network structure of an end-to-end single-stage algorithm to detect a long-distance small aerial target, as shown in fig. 2, a feature extraction part comprises five times of down sampling with the step size of 2, a residual error module is added after each down sampling to extract deep features and abstract image features, and each residual error module comprises a 3 x 3 convolution layer, a 1 x 1 convolution layer and a residual error layer. The convolution layer 1 x 1 is used to reduce the number of convolution kernel channels to reduce the parameters generated by the network, and the residual layer is used to avoid the gradient disappearance when the network depth is increased. In this network, 1 residual module is connected after the first down-sampling. After the second down-sampling, 2 residual modules are added on the basis of the original darknet53 structure, and the two residual modules are changed into 4 residual modules, so that more position information of the target is extracted. And when sampling is carried out for the third, fourth and fifth times, respectively connecting 8 residual modules, 8 residual modules and 4 residual modules. In the detection, one detection scale is added in the second downsampling, namely, the detection is carried out on four scales. And when the down sampling is performed twice, the number of target image pixels is large, and the small target detection is facilitated. The features used for detection are obtained by up-sampling the features of the high layer to be fused with the features of the previous layer. Therefore, on the detection scale corresponding to the lower layer, the utilized image features have the position information of the lower layer and the high-layer semantic information of the higher layer, so that the target detection is more accurate.
During training, 3 prior frames are preset for each detection scale, and four scales are needed so that 12 sizes of prior frames are needed in total.
For each cell of the feature map with different scales, the neural network predicts three bounding boxes by means of the prior box, 4 values are predicted for each bounding box, and the four values are respectively marked as tx,ty,tw,th. Correcting to obtain final target center coordinates (x, y) and width w and height h of the target according to the following formulaMeasurement result bx,by,bw,bh。
bx＝σ(tx)+cx
by＝σ(ty)+cy
Figure BDA0002610690300000081
Figure BDA0002610690300000082
Where σ is a function of t, txAnd tyIs the predicted coordinate offset twAnd thIs the scaled value of the scale. c. CxAnd cyIs the coordinate of the upper left corner of the cell in which the target is located, PwAnd PyIs the width and height of the prior box.
In an embodiment of the present invention, the target data set includes: positive samples and negative samples, the positive samples being correctly identified samples and the negative samples being incorrectly identified samples.
Example two
As shown in fig. 2, an embodiment of the present invention further provides a deep learning-based target detection apparatus, including: the system comprises a data sampling module, a feature extraction module and a target detection module;
the data sampling module: performing multiple downsampling processing on an acquired original image of a target to be detected to obtain multiple sampled data;
the feature extraction module: extracting deep layer characteristics and abstract image characteristics of the multiple times of sampling data respectively, and extracting image characteristics for target detection;
the target detection module: and identifying deep features and abstract image features sampled for multiple times through a neural network, and determining a detection result of the target to be detected in the original image.
EXAMPLE III
Embodiments also provide a computing device, referring to fig. 4, comprising a memory 1120, a processor 1110 and a computer program stored in said memory 1120 and executable by said processor 1110, the computer program being stored in a space 1130 for program code in the memory 1120, the computer program, when executed by the processor 1110, implementing the method steps 1131 for performing any of the methods according to the invention.
The embodiment of the application also provides a computer readable storage medium. Referring to fig. 5, the computer readable storage medium comprises a storage unit for program code provided with a program 1131' for performing the steps of the method according to the invention, which program is executed by a processor.
Example four
The specific application object of the embodiment of the invention is the aerial small target detection, and the software code of the aerial small target detection is realized by python programming.
In the embodiment, aiming at the requirement of the small air target, the detection of the small air target based on deep learning and the real-time detection and identification of the small air target are realized, and the method is described in detail as follows:
(1) collecting small target data:
the small aerial targets of the present embodiment are primarily directed to a variety of consumer-grade drones and birds. During collection, the unmanned aerial vehicle and birds in different weather, different backgrounds and different flight postures are shot in a long distance (2 kilometers). 10000 small target data are finally obtained, coordinates, positions and categories of targets in the image are labeled, 8000 are used as a training set, and 2000 are used as a test set. As the long-distance image is difficult to obtain, the image needs to be stretched, turned over, adjusted in brightness, oversampled and the like for data enhancement during training.
(2) Training and testing
In this embodiment, the model is trained and tested on the upper computer, the upper computer is configured with a video card NVIDIA GeForceGTX 1080, the memory is 32g, and the CPU is Intel core i 7.
And inputting the data into a neural network for training. During training, the training batches were set to 150 times, the learning rate was 0.001, and the attenuation was 0.0005. After training is finished, the obtained model can be used for small target detection.
During training, 3 prior frames are preset for each detection scale, four scales are provided, and therefore 12 sizes of prior frames are needed in total, and the prior frames are respectively as follows: (1 × 2), (2 × 3), (4 × 5), (7 × 8), (12 × 7), (10 × 11), (18 × 9), (14 × 12), (12 × 15), (18 × 20), (28 × 24), (51 × 59).
More in the small target detection algorithm, attention is paid to target detection conditions under different pixels, so that recall rate and accuracy rate are selected as detection indexes. The recall is defined as follows:
truepositives (tp): positive samples are correctly identified as positive samples; representing the correctly identified target in target detection;
FalseNeegactive (FN): the positive samples are misidentified as negative samples; in object detection it is indicated that the object is misidentified as background.
Recall: and the recall rate represents the proportion of the identified real target in the result returned by the algorithm to the total target of the class. I.e., the proportion of all positive sample samples in the test set that are correctly identified as positive samples. The formula is as follows:
Recall＝TP/(TP+FN)
precision: and the accuracy rate represents the proportion of the identified real target in the result returned by the algorithm to all the identified targets. I.e. the proportion of the identified samples that are correctly identified as positive samples. The formula is as follows:
Rrecision＝TP/(TP+FP)
the recall and accuracy statistics for the tests are shown in table 1 below:
TABLE 1
Figure BDA0002610690300000101
It will be understood by those of ordinary skill in the art that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. In a hardware implementation, the division between functional modules/units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be performed by several physical components in cooperation. Some or all of the components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art.

Claims (10)

1. A target detection method based on deep learning is characterized by comprising the following steps: a data sampling step, a feature extraction step and a target detection step;
the data sampling step comprises: carrying out down-sampling processing on the acquired original image of the target to be detected for multiple times to obtain multi-sampling data;
the feature extraction step includes: extracting deep features and abstract image features of the multiple times of sampling data respectively, and extracting image features for target detection;
the target detection step includes: and identifying the deep features and the abstract image features which are sampled for multiple times through a neural network, and determining the detection result of the target to be detected in the original image.
2. The object detection method according to claim 1, characterized in that the method previously comprises: a target data set is acquired and constructed, the data set including bird image data and/or drone image data.
3. The object detection method according to claim 1 or 2, wherein the down-sampling processing the acquired original image of the object to be detected for a plurality of times includes:
and sampling the target image of the original image in different scales for multiple times, wherein the step length of each sampling is N unit pixels, and N is a positive integer.
4. The object detection method according to claim 1 or 2, wherein the extracting deep features and abstract image features from the plurality of times of sampled data respectively comprises:
carrying out residual error processing on sampling data obtained by each downsampling, and extracting the characteristics of the target to be detected for position identification;
and fusing the characteristics of the target to be detected sampled currently and the characteristics of the target to be detected sampled at the previous time, and taking the fused characteristics as deep characteristics and abstract image characteristics.
5. The object detection method of claim 4, wherein the residual processing of the sampled data comprises:
and respectively passing the sampling data through a 3 × 3 convolution layer, a 1 × 1 convolution layer and a residual layer, wherein the residual layer is used for avoiding gradient disappearance when the network depth of the sampling data is increased.
6. The target detection method of claim 2, further comprising, after collecting and building a target data set:
for each cell of different scale characteristic graphs in the data set, three boundary boxes are predicted by using the prior boxes, a plurality of values are predicted for each boundary box, and the central coordinate of the target and the prediction result of the width and the height of the target are obtained according to the predicted values.
7. The object detection method of claim 6, wherein obtaining the prediction of the center coordinates of the object and the width and height of the object based on the plurality of predicted values comprises:
predict 4 values for each bounding box, denoted t respectivelyx、ty、tw、th；
Obtaining the target center coordinates (x, y) and the width w and height h prediction result b of the target according to the following formulax、by、bw、bh。
bx＝σ(tx)+cx
by＝σ(ty)+cy
Figure FDA0002610690290000021
Figure FDA0002610690290000022
Where σ is a function of t, txAnd tyIs the predicted coordinate offset in the x and y directions; t is twAnd thIs a scaled value of width and height;
cxand cyIs the coordinate of the upper left corner of the cell in which the target is located, PwAnd PyIs the width and height of the prior box.
8. The object detection method according to claim 4, wherein the original image is downsampled five times with a step size of 2, and sample data obtained by the 5 downsampling is respectively subjected to 1 residual processing, 4 residual processing, 8 residual processing, and 4 residual processing.
9. The identification method of claim 2, wherein the target data set comprises: positive samples and negative samples, the positive samples being correctly identified samples and the negative samples being incorrectly identified samples.
10. An object detection device based on deep learning, characterized by comprising: the system comprises a data sampling module, a feature extraction module and a target detection module;
the data sampling module: performing multiple downsampling processing on an acquired original image of a target to be detected to obtain multiple sampled data;
the feature extraction module: extracting deep layer characteristics and abstract image characteristics of the multiple times of sampling data respectively, and extracting image characteristics for target detection;
the target detection module: and identifying deep features and abstract image features sampled for multiple times through a neural network, and determining a detection result of the target to be detected in the original image.
CN202010753252.7A 2020-07-30 2020-07-30 Target detection method and device based on deep learning Pending CN111881984A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010753252.7A CN111881984A (en) 2020-07-30 2020-07-30 Target detection method and device based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010753252.7A CN111881984A (en) 2020-07-30 2020-07-30 Target detection method and device based on deep learning

Publications (1)

Publication Number Publication Date
CN111881984A true CN111881984A (en) 2020-11-03

Family

ID=73205815

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010753252.7A Pending CN111881984A (en) 2020-07-30 2020-07-30 Target detection method and device based on deep learning

Country Status (1)

Country Link
CN (1) CN111881984A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112465794A (en) * 2020-12-10 2021-03-09 无锡卡尔曼导航技术有限公司 Golf ball detection method based on YOLOv4 and embedded platform
CN112508924A (en) * 2020-12-15 2021-03-16 桂林电子科技大学 Small target detection and identification method, device, system and storage medium
CN113300986A (en) * 2021-04-17 2021-08-24 湖南红船科技有限公司 Unmanned aerial vehicle image transmission signal and hotspot signal identification method, medium and computer equipment

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796046A (en) * 2019-10-17 2020-02-14 武汉科技大学 Intelligent steel slag detection method and system based on convolutional neural network
CN110889399A (en) * 2019-12-23 2020-03-17 北京航天泰坦科技股份有限公司 High-resolution remote sensing image weak and small target detection method based on deep learning
CN111127438A (en) * 2019-12-25 2020-05-08 昆明理工大学 Pulmonary CT image junction detection method based on improved YOLO algorithm
CN111444809A (en) * 2020-03-23 2020-07-24 华南理工大学 Power transmission line abnormal target detection method based on improved YO L Ov3

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796046A (en) * 2019-10-17 2020-02-14 武汉科技大学 Intelligent steel slag detection method and system based on convolutional neural network
CN110889399A (en) * 2019-12-23 2020-03-17 北京航天泰坦科技股份有限公司 High-resolution remote sensing image weak and small target detection method based on deep learning
CN111127438A (en) * 2019-12-25 2020-05-08 昆明理工大学 Pulmonary CT image junction detection method based on improved YOLO algorithm
CN111444809A (en) * 2020-03-23 2020-07-24 华南理工大学 Power transmission line abnormal target detection method based on improved YO L Ov3

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
JOSEPH REDMON, ALI FARHADI: "YOLO9000: Better, Faster, Stronger", COMPUTER VISION AND PATTERN RECOGNITION, pages 1 - 9 *
JOSEPH REDMON, ALI FARHADI: "YOLOv3: An Incremental Improvement", 《COMPUTER VISION AND PATTERN RECOGNITION》, pages 1 - 6 *
YUANYUAN HU等: "Object Detection of UAV for Anti-UAV Based on Improved YOLO v3", 《2019 CHINESE CONTROL CONFERENCE (CCC)》, pages 8386 - 8390 *
李斌;张彩霞;杨阳;张文生;: "复杂场景下深度表示的无人机目标检测算法", 计算机工程与应用, no. 15, pages 118 - 123 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112465794A (en) * 2020-12-10 2021-03-09 无锡卡尔曼导航技术有限公司 Golf ball detection method based on YOLOv4 and embedded platform
CN112508924A (en) * 2020-12-15 2021-03-16 桂林电子科技大学 Small target detection and identification method, device, system and storage medium
CN112508924B (en) * 2020-12-15 2022-09-23 桂林电子科技大学 Small target detection and identification method, device, system and storage medium
CN113300986A (en) * 2021-04-17 2021-08-24 湖南红船科技有限公司 Unmanned aerial vehicle image transmission signal and hotspot signal identification method, medium and computer equipment

Similar Documents

Publication Publication Date Title
WO2022100470A1 (en) Systems and methods for target detection
US20150054824A1 (en) Object detection method, object detection device, and image pickup device
CN107273832B (en) License plate recognition method and system based on integral channel characteristics and convolutional neural network
CN111881984A (en) Target detection method and device based on deep learning
CN111222395A (en) Target detection method and device and electronic equipment
CN110147714B (en) Unmanned aerial vehicle-based coal mine goaf crack identification method and detection system
CN110659601B (en) Depth full convolution network remote sensing image dense vehicle detection method based on central point
CN113723377A (en) Traffic sign detection method based on LD-SSD network
CN112364843A (en) Plug-in aerial image target positioning detection method, system and equipment
CN108509826B (en) Road identification method and system for remote sensing image
CN115861756A (en) Earth background small target identification method based on cascade combination network
CN116258940A (en) Small target detection method for multi-scale features and self-adaptive weights
Dousai et al. Detecting humans in search and rescue operations based on ensemble learning
CN112464933B (en) Intelligent identification method for weak and small target through foundation staring infrared imaging
KR20200145174A (en) System and method for recognizing license plates
CN112288701A (en) Intelligent traffic image detection method
CN111881982A (en) Unmanned aerial vehicle target identification method
CN112288702A (en) Road image detection method based on Internet of vehicles
CN113012215A (en) Method, system and equipment for space positioning
CN115035429A (en) Aerial photography target detection method based on composite backbone network and multiple measuring heads
CN115601538A (en) Target detection method, target detector, target detection system, and medium
Wu et al. Research on Asphalt Pavement Disease Detection Based on Improved YOLOv5s
CN114332814A (en) Parking frame identification method and device, electronic equipment and storage medium
CN115485746A (en) License plate character recognition method, device, equipment and storage medium
CN113326891A (en) Method, system and device for detecting small target object

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