CN110473145B - Detection method and device for passive terahertz image fusion ultrahigh resolution reconstruction - Google Patents
Detection method and device for passive terahertz image fusion ultrahigh resolution reconstruction Download PDFInfo
- Publication number
- CN110473145B CN110473145B CN201910754243.7A CN201910754243A CN110473145B CN 110473145 B CN110473145 B CN 110473145B CN 201910754243 A CN201910754243 A CN 201910754243A CN 110473145 B CN110473145 B CN 110473145B
- Authority
- CN
- China
- Prior art keywords
- model
- super
- image
- resolution
- resolution reconstruction
- 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
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 227
- 230000004927 fusion Effects 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims description 81
- 238000000034 method Methods 0.000 claims description 40
- 238000012545 processing Methods 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 6
- 238000007689 inspection Methods 0.000 abstract description 10
- 230000008569 process Effects 0.000 description 17
- 238000005516 engineering process Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000003384 imaging method Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000003672 processing method Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000003631 expected effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
The application provides a detection method for passive terahertz image fusion ultrahigh resolution reconstruction, which comprises the following steps: acquiring a passive terahertz image to be detected; inputting the passive terahertz image into a pre-optimized super-resolution reconstruction model for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image; the super-resolution reconstruction model is a model constructed based on a generative countermeasure network ESPCN; inputting the high-resolution image into a pre-trained object detection model for object detection to obtain an object detection result; the object detection model is a model constructed based on a detection algorithm model SSD. The super-resolution reconstruction model constructed based on the generation type countermeasure network ESPCN is fused into an object detection mode based on a passive terahertz image, so that the detection speed is increased, the detection accuracy is improved, and the requirements of human body security inspection on speed and accuracy are well met.
Description
Technical Field
The invention relates to the technical field of object detection, in particular to a detection method and a detection device for passive terahertz image fusion ultrahigh resolution reconstruction.
Background
Terahertz waves have a wavelength of 30 μm to 3mm, and can penetrate through paper, plastic, cloth and other substances, so that articles hidden under clothes of a human body can be found, and therefore terahertz waves are often used for human body security inspection. The terahertz wave imaging technology of the human body can be divided into an active type and a passive type.
Because the passive mode forms an image by using the terahertz waves emitted by the human body, the object is judged by shielding or absorbing the emitted terahertz waves by the object carried by the human body reflected in the image. This mode does not have any radiation to the human body and also does not require a terahertz wave transmitting device. Therefore, the passive terahertz wave imaging technology is more widely applied to security inspection.
At that time, the resolution of the formed terahertz wave image is low because the intensity of the terahertz wave emitted by the human body is weak. In the prior art, a traditional image processing method is adopted, a terahertz wave image mode is improved, and then object detection is carried out. However, the conventional image processing methods and object detection methods cannot meet the requirements of human body security inspection in both accuracy and speed.
Disclosure of Invention
Based on the defects of the prior art, the invention provides a detection method and a detection device for passive terahertz image fusion ultrahigh resolution reconstruction, and aims to solve the problem that the good requirements of human body security inspection cannot be met in both detection accuracy and speed in the prior art by performing object detection after a traditional image processing mode.
In order to achieve the purpose, the invention provides the following technical scheme:
the invention provides a detection method for fusion ultrahigh resolution reconstruction of a passive terahertz image, which comprises the following steps:
acquiring a passive terahertz image to be detected;
inputting the passive terahertz image into a pre-optimized super-resolution reconstruction model for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image; the super-resolution reconstruction model is a model constructed based on a generative countermeasure network ESPCN;
inputting the high-resolution image into a pre-trained object detection model for object detection to obtain an object detection result; the object detection model is a model constructed based on a detection algorithm model SSD.
Optionally, in the detection method, the method for optimizing the super-resolution reconstruction model includes:
processing the format of the passive terahertz image training sample into a first format; the first format is an image format before pixels of the image are rearranged by the generative countermeasure network ESPCN; the passive terahertz image training sample comprises an object required to be detected;
inputting the processed passive terahertz image training sample into the super-resolution reconstruction model for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image training sample; wherein the high-resolution image is used as a training sample of the object detection model; the super-resolution reconstruction model adopts an activation function Tanh;
and adjusting parameters in the super-resolution reconstruction model according to a detection result obtained by detecting the high-resolution image by the object detection model.
Optionally, in the above detection method, the training method of the object detection model includes:
inputting a high-resolution image corresponding to the passive terahertz image training sample obtained by performing super-resolution reconstruction on the super-resolution reconstruction model into the object detection model for object detection to obtain a detection result; the object detection model is obtained by adjusting the number of network layers of the detection algorithm model SSD;
and adjusting the parameters of the object detection model according to the detection result.
Optionally, in the foregoing detection method, after the adjusting parameters of the object detection model according to the detection result, the method includes:
and returning the passive terahertz image training samples corresponding to the detection results of the object requiring detection and the object requiring detection errors to the super-resolution reconstruction model for super-resolution reconstruction again after the parameters are adjusted, so as to train the object detection model after the parameters are adjusted by using the passive terahertz image training samples again.
The invention provides a detection device for passive terahertz image fusion ultrahigh resolution reconstruction, which comprises:
the acquisition unit is used for acquiring a passive terahertz image to be detected;
the first input unit is used for inputting the passive terahertz image into a super-resolution reconstruction model which is optimized in advance to perform super-resolution reconstruction so as to obtain a high-resolution image corresponding to the passive terahertz image; the super-resolution reconstruction model is a model constructed based on a generative countermeasure network ESPCN;
the second input unit is used for inputting the high-resolution image into a pre-trained object detection model for object detection to obtain an object detection result; the object detection model is a model constructed based on a detection algorithm model SSD.
Optionally, in the above apparatus, an optimization unit is further included; the optimization unit comprises:
the processing unit is used for processing the format of the passive terahertz image training sample into a first format; the first format is an image format before pixels of the image are rearranged by the generative countermeasure network ESPCN; the passive terahertz image training sample comprises an object required to be detected;
the third input unit is used for inputting the processed passive terahertz image training sample into the super-resolution reconstruction model for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image training sample; wherein the high-resolution image is used as a training sample of the object detection model; the super-resolution reconstruction model adopts an activation function Tanh;
and the first adjusting unit is used for adjusting parameters in the super-resolution reconstruction model according to a detection result obtained by detecting the high-resolution image by the object detection model.
Optionally, in the above apparatus, further comprising a training unit; the training unit comprises:
the fourth input unit is used for inputting the high-resolution image corresponding to the passive terahertz image training sample obtained by performing super-resolution reconstruction by the super-resolution reconstruction model into the object detection model for object detection to obtain a detection result; the object detection model is obtained by adjusting the number of network layers of the detection algorithm model SSD;
and the second adjusting unit is used for adjusting the parameters of the object detection model according to the detection result.
Optionally, in the above apparatus, the training unit further includes:
and the return unit is used for returning the passive terahertz image training samples corresponding to the detection results of the object requiring detection and the object requiring detection errors to the super-resolution reconstruction model for super-resolution reconstruction after the parameters are adjusted, so that the passive terahertz image training samples are used again for training the object detection model after the parameters are adjusted.
The detection method and the device for the passive terahertz image fusion ultrahigh resolution reconstruction provided by the invention fuse a pre-optimized super-resolution reconstruction model constructed based on the generation countermeasure network ESPCN and a pre-trained object model constructed based on the detection algorithm model SSD. And performing super-resolution reconstruction on the passive terahertz image to be detected through a super-resolution reconstruction model to obtain a high-resolution image corresponding to the passive terahertz image. Therefore, the passive terahertz image is efficiently processed into a high-resolution image. Then, the object detection model is used for carrying out object detection on the high-resolution image, and an accurate detection result can be quickly obtained. Thereby not only providing the speed of detection, but also improving the accuracy of detection. Thereby being capable of well meeting the requirement of human body security inspection.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a schematic flowchart of a detection method for passive terahertz image fusion ultrahigh resolution reconstruction according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating a method for performing optimization of a super-resolution reconstruction model and training of an object detection model in combination according to another embodiment of the present invention;
fig. 3 is a schematic structural diagram of a detection apparatus for passive terahertz image fusion ultrahigh resolution reconstruction according to another embodiment of the present invention;
fig. 4 is a schematic structural diagram of a detection apparatus for passive terahertz image fusion ultrahigh resolution reconstruction according to another embodiment of the present invention;
fig. 5 is a schematic structural diagram of a detection apparatus for passive terahertz image fusion ultrahigh resolution reconstruction according to another embodiment of the present invention.
Detailed Description
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 only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In this application, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising one of 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
The embodiment of the invention provides a detection method for fusion of passive terahertz images and ultrahigh resolution reconstruction, which aims to solve the problem that the requirement of good human body security check on detection accuracy and speed cannot be met in the prior art by a mode of carrying out object detection after a traditional image processing mode is adopted.
The embodiment of the invention provides a detection method for fusion ultrahigh resolution reconstruction of a passive terahertz image, which comprises the following steps of:
s101, obtaining a passive terahertz image to be detected.
The passive terahertz image refers to an image obtained by a passive terahertz imaging technology. Terahertz imaging techniques are classified into passive and active. The active mode is that a terahertz transmitting device transmits terahertz waves with specific frequency to a human body, and corresponding images are obtained through reflection and scattering of the terahertz waves by the human body and a carrying object. The passive type is the terahertz waves emitted by the human body, and the corresponding images are obtained by shielding or absorbing the emitted terahertz waves by the object carried by the human body. Since there is radiation to the human body passively and there is no need to configure a special terahertz wave emitting device, it is more widely applied to human body security inspection. Therefore, the embodiment of the present invention is directed to a picture obtained by a passive terahertz wave imaging technology.
S102, inputting the passive terahertz image into a super-resolution reconstruction model optimized in advance to perform super-resolution reconstruction, and obtaining a high-resolution image corresponding to the passive terahertz image.
The super-resolution reconstruction model is constructed based on the generative countermeasure network ESPCN.
It should be noted that, because the intensity of the terahertz wave emitted by the human body is weak, the resolution of the obtained passive terahertz image is low. And too low resolution of the image can seriously affect the accuracy of object detection. Therefore, the super-resolution reconstruction method adopts the pre-constructed and optimized super-resolution reconstruction model to perform super-resolution reconstruction on the low-resolution passive terahertz image to obtain the high-resolution image, and then performs object detection through the high-resolution image so as to enable the detection result to be more accurate.
The super-resolution reconstruction of the image refers to a process of reconstructing a low-resolution image through a series of image processing and machine learning algorithms to obtain a high-resolution image. The main principle is to extract information with strong correlation and complementarity from a plurality of images and comprehensively utilize the information so as to improve the resolution of the images. Because of the dependence on the machine learning algorithm and the computer running capability, compared with the traditional mode, the mode of processing the image has higher speed and better effect.
In the embodiment of the invention, the adopted super-resolution reconstruction model is constructed based on the generative countermeasure network ESPCN. The generative countermeasure network ESPCN is a network model that is specially applied to super-resolution reconstruction of images. The generative countermeasure network ESPCN directly carries out calculation convolution on the low-resolution image to obtain a high-resolution image. The super-resolution reconstruction network models SRCNN and DRCN, etc. are obtained by upsampling and interpolating the low-resolution image to obtain the same size as the high-resolution image, and then are used as network input, which means that the convolution operation is performed at a higher resolution. This reduces efficiency compared to computing the convolution directly on a low resolution image. Therefore, the super-resolution model is constructed by the generation type countermeasure network ESPCN with higher efficiency, so that the requirement of security inspection on speed can be better met.
It should be noted that step S103 is executed after the super-resolution is performed in step S102 to obtain a high-resolution image.
S103, inputting the high-resolution image into a pre-trained object detection model for object detection to obtain an object detection result.
The object detection model is a model constructed based on a detection algorithm model SSD.
It should be noted that the detection algorithm model SSD is a network model based on machine learning technology, and has a greater speed advantage compared with other detection algorithms, so that it is a detection algorithm model widely used in the detection field.
The object detection model of the embodiment of the invention is constructed based on the detection algorithm model SSD, and the object detection model is trained, so that the trained object detection model can be used for quickly detecting the object of the high-resolution image, and the accurate detection result can be quickly obtained.
The detection method for the passive terahertz image fusion ultrahigh resolution reconstruction provided by the invention fuses a pre-optimized super-resolution reconstruction model constructed based on the generative countermeasure network ESPCN and a pre-trained object model constructed based on the detection algorithm model SSD. And performing super-resolution reconstruction on the passive terahertz image to be detected through a super-resolution reconstruction model to obtain a high-resolution image corresponding to the passive terahertz image. Therefore, the passive terahertz image is efficiently processed into a high-resolution image. Then, the object detection model is used for carrying out object detection on the high-resolution image, and an accurate detection result can be quickly obtained. The speed of detection is provided, and the accuracy of detection is improved. Thereby being capable of well meeting the requirement of human body security inspection.
The detection method for the passive terahertz image fusion ultrahigh resolution reconstruction is completed by the mutual matching of a super-resolution reconstruction model and an object detection model. Moreover, the object detection can be realized through the super-resolution reconstruction model and the object detection model only after the super-resolution reconstruction model is optimized and the object detection model is trained. Therefore, the super-resolution reconstruction model and the object detection model can be better matched to work. Another embodiment of the present invention provides a method for performing optimization of a super-resolution reconstruction model and training of an object detection model in combination, as shown in fig. 2, including:
s201, processing the format of the passive terahertz image training sample into a first format.
The first format is an image format before the generation countermeasure network ESPCN rearranges the pixels of the image. The passive terahertz image training sample refers to a passive terahertz image as a training sample. The passive terahertz image training sample comprises an object to be detected, namely the passive terahertz image training sample comprises one or more known objects, and the one or more known objects are used for detecting the objects accurately according to a detection result after the detection is carried out by the object detection model, so that whether the training of the object detection model is completed or not is determined, and the objects are defined as the objects to be detected.
It should be noted that the process of processing the image by the generative countermeasure network ESPCN mainly includes two processes. The first process is as follows: processing a low-resolution image into a multi-channel image; the second process is as follows: and rearranging the same pixel points in the multi-channel image into a small area, and rearranging all the small areas to form a high-resolution image. Therefore, in order to avoid the need to rearrange the pixels during training, the images may be processed into an image format before being rearranged into a high-resolution image before being input into the model.
S202, inputting the processed passive terahertz image training sample into a super-resolution reconstruction model for super-resolution reconstruction, and obtaining a high-resolution image corresponding to the passive terahertz image training sample.
The super-resolution reconstruction model in the embodiment of the invention adopts an activation function Tanh, and the loss function adopts a mean square error.
After the passive terahertz image training sample is processed, the passive terahertz image training sample with low resolution can be input into the constructed super-resolution reconstruction model for resolution reconstruction, so that a high-resolution image corresponding to the passive terahertz image training sample is obtained.
It should be noted that the obtained high-resolution image corresponding to the passive terahertz image training sample will be used as a training sample of the object detection model. Therefore, step S103 is performed to obtain a high resolution image corresponding to the passive terahertz image training sample in step S102.
S203, inputting a high-resolution image corresponding to the passive terahertz image training sample obtained by performing super-resolution reconstruction on the super-resolution reconstruction model into the object detection model for object detection to obtain a detection result.
The object detection model is obtained by adjusting the number of network layers of the detection algorithm model SSD. That is to say, the object detection model constructed in the embodiment of the present invention is obtained by adjusting the number of network layers of the existing detection algorithm model SSD to some extent.
Specifically, the detection algorithm model SSD is mainly divided into a basic network layer and a feature extraction layer. The basic network layer comprises a plurality of convolution layers and a pooling layer, and shallow low-dimensional features to deep high-dimensional features of the image are sequentially extracted. The feature extraction layer extracts six layers of network layers from the basic layer at intervals, and the output results of the six layers of network layers are convoluted by the two convolution layers again to carry out frame regression and object classification, so that the target object is detected. Since the passive terahertz image features are relatively few, the deep high-dimensional image features required are few in number compared with the shallow low-dimensional image features. Therefore, in the embodiment of the present invention, in the feature extraction, the network layer that outputs the deep high-dimensional image is replaced with the network layer that outputs the shallow low-dimensional image in the base layer, from the input result that six layers are extracted at intervals from the base layer. Because the extraction of the shallow low-dimensional features is more than that of the deep high-dimensional features, the detection speed of the object detection model can be greatly improved.
S204, adjusting parameters in the super-resolution reconstruction model and adjusting parameters of the object detection model according to a detection result obtained by detecting the high-resolution image by the object detection model.
Because the object detection model and the super-resolution reconstruction model work with each other, the object detection model and the super-resolution reconstruction model are optimized and trained simultaneously, and parameters are adjusted simultaneously. Therefore, the object can be detected by the object detection model, and a high-resolution image is obtained by processing the super-resolution reconstruction model, so that the method is more suitable for detecting the object detection model.
Therefore, after the detection result is obtained, parameters of the super-resolution reconstruction model and the object detection model need to be adjusted correspondingly according to the detection. Therefore, the training of the object detection model is realized, and the optimization of the super-resolution reconstruction model is also completed.
S205, returning the passive terahertz image training samples corresponding to the detection results of the undetected object requiring detection and the detection result of the object requiring detection error to the super-resolution reconstruction model for super-resolution reconstruction again after adjusting the parameters.
The object detection model fails to detect the object required to be detected or detects the object error required to be detected, which indicates that the training of the object detection model and the optimization of the super-resolution reconstruction model do not reach the expected effect, and after the two models are adjusted, in order to verify the adjusted effect and continue to adjust parameters, the passive terahertz image training sample corresponding to the detection result in which the object required to be detected and the object error required to be detected are not detected needs to be returned to the super-resolution reconstruction model after the parameters are adjusted for carrying out the super-resolution reconstruction again, so that the passive terahertz image training sample is used again for training the object detection model after the parameters are adjusted. I.e., after performing step S204, it may return to step S201.
According to the method for optimizing the super-resolution reconstruction model and training the object detection model, provided by the embodiment of the invention, the super-resolution reconstruction model is used for carrying out super-resolution reconstruction on the passive terahertz image training sample to obtain a high-resolution image, and the high-resolution image is input into the object detection model for training to obtain a detection result. And then correspondingly adjusting parameters of the super-resolution reconstruction model and the object detection model according to the detection result. Therefore, the combined adjustment of the super-resolution reconstruction model and the object detection model is realized, and the final matching effect of the two models can be optimal.
Another embodiment of the present invention provides a detection apparatus for passive terahertz image fusion ultrahigh resolution reconstruction, as shown in fig. 3, including:
an acquiring unit 301, configured to acquire a passive terahertz image to be detected.
It should be noted that, the specific working process of the obtaining unit 301 may refer to step S101 in the foregoing method embodiment accordingly, and details are not described here again.
A first input unit 302, configured to input the passive terahertz image into a pre-optimized super-resolution reconstruction model for super-resolution reconstruction, so as to obtain a high-resolution image corresponding to the passive terahertz image.
The super-resolution reconstruction model is a model constructed based on a generative countermeasure network ESPCN.
It should be noted that, the specific working process of the first input unit 302 may refer to step S102 in the foregoing method embodiment accordingly, and details are not described here again.
A second input unit 303, configured to input the high-resolution image into a pre-trained object detection model for object detection, so as to obtain an object detection result.
The object detection model is a model constructed based on a detection algorithm model SSD.
It should be noted that, the specific working process of the second input unit 303 may refer to step S103 in the foregoing method embodiment accordingly, which is not described herein again.
The passive terahertz image fusion ultrahigh resolution reconstruction detection device provided by the embodiment of the invention fuses a pre-optimized super-resolution reconstruction model constructed based on a generative countermeasure network ESPCN and a pre-trained object model constructed based on a detection algorithm model SSD. And inputting the passive terahertz image to be detected acquired by the acquisition unit into a super-resolution reconstruction model through a first input unit for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image. Therefore, the passive terahertz image is efficiently processed into a high-resolution image. Then, the second input unit inputs the high-resolution image into the object detection model for object detection, so that an accurate detection result can be obtained quickly. Therefore, the detection speed is increased, and the detection accuracy is also improved. Thereby being capable of well meeting the requirement of human body security inspection.
Optionally, in another embodiment of the present invention, the detection apparatus for passive terahertz image fusion ultrahigh resolution reconstruction further includes an optimization unit. As shown in fig. 4, the optimization unit includes:
the processing unit 401 is configured to process a format of the passive terahertz image training sample into a first format.
The first format is an image format before pixels of the image are rearranged by the generative countermeasure network ESPCN. The passive terahertz image training sample contains an object required to be detected.
It should be noted that, the specific working process of the processing unit 401 may refer to step S201 in the foregoing method embodiment accordingly, and is not described herein again.
A third input unit 402, configured to input the processed passive terahertz image training sample into the super-resolution reconstruction model for super-resolution reconstruction, so as to obtain a high-resolution image corresponding to the passive terahertz image training sample.
Wherein the high-resolution image is used as a training sample of the object detection model; the super-resolution reconstruction model adopts an activation function Tanh.
It should be noted that, the specific working process of the third input unit 402 may refer to step S202 in the above method embodiment accordingly, and is not described herein again.
A first adjusting unit 403, configured to adjust parameters in the super-resolution reconstruction model according to a detection result obtained by detecting the high-resolution image by the object detection model.
It should be noted that, the specific working process of the first adjusting unit 403 may refer to step S204 in the foregoing method embodiment accordingly, and details are not described herein again.
Optionally, in another embodiment of the present invention, the detection apparatus for fusion of passive terahertz images and ultrahigh resolution reconstruction may further include a training unit. As shown in fig. 5, the training unit includes:
a fourth input unit 501, configured to input the high-resolution image corresponding to the passive terahertz image training sample obtained by performing super-resolution reconstruction on the super-resolution reconstruction model into the object detection model for object detection, so as to obtain a detection result.
The object detection model is obtained by adjusting the number of network layers of the detection algorithm model SSD.
It should be noted that, the specific working process of the fourth input unit 501 may refer to step S203 in the foregoing method embodiment accordingly, and details are not repeated here.
A second adjusting unit 502, configured to adjust parameters of the object detection model according to the detection result.
It should be noted that, the specific working process of the second adjusting unit 502 may refer to step S204 in the foregoing method embodiment accordingly, and is not described herein again.
Optionally, in another embodiment of the present invention, the training unit further includes:
and the return unit is used for returning the passive terahertz image training samples corresponding to the detection results of the object requiring detection but not detected and the object requiring detection error to the super-resolution reconstruction model for super-resolution reconstruction again after the parameters are adjusted, so that the passive terahertz image training samples are used again for training the object detection model after the parameters are adjusted.
Those of skill would further appreciate that the various illustrative components and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the components and steps of the various examples have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the technical solution. 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 invention.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (6)
1. A detection method for passive terahertz image fusion ultrahigh resolution reconstruction is characterized by comprising the following steps:
acquiring a passive terahertz image to be detected;
inputting the passive terahertz image into a pre-optimized super-resolution reconstruction model for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image; the super-resolution reconstruction model is a model constructed based on a generative countermeasure network ESPCN;
inputting the high-resolution image into a pre-trained object detection model for object detection to obtain an object detection result; the object detection model is a model constructed based on a detection algorithm model SSD;
the optimization method of the super-resolution reconstruction model comprises the following steps:
processing the format of the passive terahertz image training sample into a first format; the first format is an image format before pixels of the image are rearranged by the generative countermeasure network ESPCN; the passive terahertz image training sample comprises an object required to be detected;
inputting the processed passive terahertz image training sample into the super-resolution reconstruction model for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image training sample; wherein the high-resolution image is used as a training sample of the object detection model; the super-resolution reconstruction model adopts an activation function Tanh;
and adjusting parameters in the super-resolution reconstruction model according to a detection result obtained by detecting the high-resolution image by the object detection model, wherein the object detection model and the super-resolution reconstruction model work mutually.
2. The method of claim 1, wherein the training method of the object detection model comprises:
inputting a high-resolution image corresponding to the passive terahertz image training sample obtained by performing super-resolution reconstruction on the super-resolution reconstruction model into the object detection model for object detection to obtain a detection result; the object detection model is obtained by adjusting the number of network layers of the detection algorithm model SSD;
and adjusting the parameters of the object detection model according to the detection result.
3. The method of claim 2, wherein the adjusting the parameters of the object detection model according to the detection result comprises:
and returning the passive terahertz image training samples corresponding to the detection results of the object requiring detection and the object requiring detection, which are not detected, to the super-resolution reconstruction model for super-resolution reconstruction after the parameters are adjusted, so as to train the object detection model after the parameters are adjusted by using the passive terahertz image training samples again.
4. The utility model provides a detection device that passive form terahertz image fuses ultrahigh resolution and rebuilds which characterized in that includes:
the acquisition unit is used for acquiring a passive terahertz image to be detected;
the first input unit is used for inputting the passive terahertz image into a pre-optimized super-resolution reconstruction model for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image; the super-resolution reconstruction model is a model constructed based on a generative countermeasure network ESPCN;
the second input unit is used for inputting the high-resolution image into a pre-trained object detection model for object detection to obtain an object detection result; the object detection model is a model constructed based on a detection algorithm model SSD;
the optimization unit comprises:
the processing unit is used for processing the format of the passive terahertz image training sample into a first format; the first format is an image format before pixels of the image are rearranged by the generative countermeasure network ESPCN; the passive terahertz image training sample comprises an object required to be detected;
the third input unit is used for inputting the processed passive terahertz image training sample into the super-resolution reconstruction model for super-resolution reconstruction to obtain a high-resolution image corresponding to the passive terahertz image training sample; wherein the high-resolution image is used as a training sample of the object detection model; the super-resolution reconstruction model adopts an activation function Tanh;
and the first adjusting unit is used for adjusting parameters in the super-resolution reconstruction model according to a detection result obtained by detecting the high-resolution image by the object detection model, and the object detection model and the super-resolution reconstruction model work mutually.
5. The apparatus of claim 4, further comprising a training unit, the training unit comprising:
the fourth input unit is used for inputting the high-resolution image corresponding to the passive terahertz image training sample obtained by performing super-resolution reconstruction on the super-resolution reconstruction model into the object detection model for object detection to obtain a detection result; the object detection model is obtained by adjusting the number of network layers of the detection algorithm model SSD;
and the second adjusting unit is used for adjusting the parameters of the object detection model according to the detection result.
6. The apparatus of claim 4, wherein the training unit further comprises:
and the return unit is used for returning the passive terahertz image training samples corresponding to the detection results of the object requiring detection and the object requiring detection errors to the super-resolution reconstruction model for super-resolution reconstruction after the parameters are adjusted, so that the passive terahertz image training samples are used again for training the object detection model after the parameters are adjusted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910754243.7A CN110473145B (en) | 2019-08-15 | 2019-08-15 | Detection method and device for passive terahertz image fusion ultrahigh resolution reconstruction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910754243.7A CN110473145B (en) | 2019-08-15 | 2019-08-15 | Detection method and device for passive terahertz image fusion ultrahigh resolution reconstruction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110473145A CN110473145A (en) | 2019-11-19 |
| CN110473145B true CN110473145B (en) | 2023-03-14 |
Family
ID=68510147
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910754243.7A Active CN110473145B (en) | 2019-08-15 | 2019-08-15 | Detection method and device for passive terahertz image fusion ultrahigh resolution reconstruction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110473145B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113706383A (en) * | 2021-08-30 | 2021-11-26 | 上海亨临光电科技有限公司 | Super-resolution method, system and device for terahertz image |
| CN114037609B (en) * | 2021-10-21 | 2024-04-19 | 中国科学院沈阳自动化研究所 | Terahertz image super-resolution algorithm based on learning terahertz imaging inverse process |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10386650B2 (en) * | 2016-10-22 | 2019-08-20 | Massachusetts Institute Of Technology | Methods and apparatus for high resolution imaging with reflectors at staggered depths beneath sample |
| CN108665509A (en) * | 2018-05-10 | 2018-10-16 | 广东工业大学 | A kind of ultra-resolution ratio reconstructing method, device, equipment and readable storage medium storing program for executing |
-
2019
- 2019-08-15 CN CN201910754243.7A patent/CN110473145B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110473145A (en) | 2019-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Deep learning–based fully automated pavement crack detection on 3D asphalt surfaces with an improved CrackNet | |
| CN106796650B (en) | Multi-stage Activity determination | |
| CN107229918A (en) | A kind of SAR image object detection method based on full convolutional neural networks | |
| CN110473145B (en) | Detection method and device for passive terahertz image fusion ultrahigh resolution reconstruction | |
| CN103955922B (en) | Method for detecting flaws of printed fabric based on Gabor filter | |
| CN102708373B (en) | Method and device for classifying remote images by integrating space information and spectral information | |
| Huynh-The et al. | Accurate LPI radar waveform recognition with CWD-TFA for deep convolutional network | |
| CN102592280B (en) | Hyperspectral image anomaly detection method using multi-window feature analysis | |
| CN104143185A (en) | Blemish zone detecting method | |
| CN103578091A (en) | Infrared polarization image fusion method based on Laplacian pyramid | |
| CN103712560A (en) | Part detection method, system and device based on information fusion of multiple sensors | |
| CN104392245B (en) | The method for recognizing road surface types and device of Multi-sensor Fusion | |
| Prasad et al. | Mobile plant species classification: a low computational aproach | |
| CN110268441A (en) | The method for obtaining the 3D model data of multiple components of object | |
| CN105809649A (en) | Variation multi-scale decomposing based SAR image and visible light image integration method | |
| CN113792623B (en) | Security check CT target object identification method and device | |
| CN112699952B (en) | Train fault image amplification method and system based on deep learning | |
| TWI731689B (en) | Tomography method, system and apparatus based on time-domain spectroscopy | |
| Di Angelo et al. | A computational method for bilateral symmetry recognition in asymmetrically scanned human faces | |
| CN107622476A (en) | Image Super-resolution processing method based on generative probabilistic model | |
| Thiyaneswaran et al. | Iris Recognition using Left and Right Iris Feature of the Human Eye for Biometric Security System | |
| JP7462119B2 (en) | Three-dimensional imaging method and apparatus, and three-dimensional imaging device | |
| CN103530876B (en) | A kind of infrared image analysis method based on Fourier transform | |
| Hu et al. | Predicting high-fidelity human body models from impaired point clouds | |
| CN105241813A (en) | Compressive sampling photoacoustic microscopic imaging method and apparatus |
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 |