CN111582367A - Small metal threat detection method - Google Patents
Small metal threat detection method Download PDFInfo
- Publication number
- CN111582367A CN111582367A CN202010376167.3A CN202010376167A CN111582367A CN 111582367 A CN111582367 A CN 111582367A CN 202010376167 A CN202010376167 A CN 202010376167A CN 111582367 A CN111582367 A CN 111582367A
- Authority
- CN
- China
- Prior art keywords
- image
- small metal
- detection method
- neural network
- ray
- 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
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
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- 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
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)
- Analysing Materials By The Use Of Radiation (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a detection method of small metal threats, which comprises the following steps: carrying out X-ray inspection on an object to be inspected to obtain an X-ray image; preprocessing the X-ray image to obtain a preprocessed X-ray image; density sampling is carried out by a small window; classifying the detected object by using a classifier trained on a preset deep learning model based on a convolutional neural network, calculating a confidence score of a detected image, and judging a small metal threat under the condition that the confidence is greater than a specific threshold value; the detection method can improve the detection performance of the SMT, and has magnitude order improvement compared with the traditional neural network detection.
Description
Technical Field
The invention relates to a related application of detecting threats in an X-ray image by utilizing a convolutional neural network, in particular to a method for detecting small metal threats.
Background
Nowadays, security detection is more and more common in our daily life, and meanwhile, the security detection also draws high attention of all social circles. In important places such as airports, stations, subways, institutions and the like all over the world, safety workers detect whether people carry dangerous articles such as firearms, ammunitions, flammability, explosive, toxic radioactivity and the like through safety infrastructure equipment so as to ensure the personal and property safety of the people, prevent lawless persons or terrorist attacks and maintain peace and stability of social security.
In some special scenarios, such as shipping containers, Small Metal Threats (SMT) currently rely heavily on statistical risk analysis, intelligence reports and visual inspection of X-ray images by security personnel. These methods are very slow and unreliable in the case of difficult tasks, the objects to be detected are usually very small in volume, and in a very complicated and confusing context, it is necessary to detect objects that may be below 50 pixels in an image containing more than 200 ten thousand pixels, so these methods are not accurate, and in the detection of freight containers, the scene in the image is often larger and more complicated, there is almost no limitation on the arrangement of the goods, and in addition to some dense obscurations, it is difficult for general security personnel to see the threat hidden in the legitimate goods with the naked eyes, so that there is still a possibility of being threatened by illegal smuggling or terrorist attacks.
Prior related art scheme one
The manual inspection of the X-ray security image is a common security inspection method at present, and is commonly used in security inspection places such as stations and airports. However, in the image of the freight container, the imaging size of the small metal is small, for example, the typical value of the pixel in the image of 2600 × 850 pixels is only 0.1%, so that the security inspector can hardly distinguish the image during the inspection process, and the inspection time is increased. The first prior related technical scheme has the following disadvantages: when facing a large cargo box of 10 meters in length such as a container, if the cargo is placed in a complicated or dense manner and is shielded, small metal objects are difficult to distinguish, and the metal threat in qualified cargo is not found at all.
Prior related art scheme two
People use a feature extraction and clustering method to detect small metals in container images, such as a BoW bag-of-words model, and the method is commonly used in the fields of computer vision, information retrieval and the like. The bag-of-words model is that firstly the picture is subjected to feature extraction through SIFT, SURF and other modes, then a dictionary is constructed by using k-means clustering and other methods, and new input data is classified through histogram statistics. The second prior art has the following disadvantages: the method is developed aiming at natural images and the like, the transparency, the noise level, the disordered and inclined visual angles of the method are obviously different from those of the X-ray images, and the detection is directly applied to the X-ray images and has poor identification effect.
The prior related technical scheme is three:
the detection is carried out based on the X-ray cargo images of the shape and the texture, the scheme divides the X-ray cargo images into 22 categories, such as categories of tire grains and the like, training is carried out on a training set of the categories, and 78% of the images can be correctly classified and recognized. The third prior art scheme has the following disadvantages: because the direction of the small metal is not limited, and the characteristics of the small metal are not obvious, the special characteristic textures cannot be extracted, the recognition degree is not high, and the method is only suitable for recognizing large objects with obvious rules and characteristics.
At present, there are some patent reports on the use of neural networks to detect articles.
Chinese patent CN108303747A discloses an inspection apparatus and a method of detecting a firearm. And carrying out X-ray inspection on the inspected object to obtain a transmission image. A plurality of candidate regions in the transmission image are determined using a trained firearm detection neural network. Classifying the plurality of candidate regions with the firearm detection neural network to determine whether a firearm is included in the transmission image. By the aid of the scheme, whether the gun is contained in the container/vehicle or not can be determined more accurately. The disclosed embodiment is only provided with a Convolutional Neural Network (CNN) as an embodiment, and the learning process of the convolutional neural network includes two links of calculating output (forward propagation) and adjusting parameter (backward propagation). A network is obtained by learning a sample image, namely, the network is optimized after multi-layer convolution, excitation function, pool sampling, full connection and error calculation, so that the prediction error of the network under the current sample image is ensured to be minimum, namely, the model is considered to be the optimal model. But no more detailed study was performed.
Chinese patent CN107871122A discloses a security inspection detection method, device, system and electronic equipment, wherein the method includes obtaining an X-ray image collected by an X-ray machine in a security inspection machine received by a security inspection terminal, and preprocessing the X-ray image to obtain a preprocessed X-ray image; extracting article characteristics of the corresponding object to be detected in the preprocessed X-ray image according to a preset deep learning model, wherein the preset deep learning model comprises a deep learning model based on a convolutional neural network; identifying the characteristics of the articles by using a classifier trained based on a preset deep learning model to generate an identification result corresponding to the object to be detected; and sending the identification result of the object to be detected to the security inspection terminal so that the security inspection terminal displays the identification result. The technical scheme provided by the embodiment of the invention realizes automatic identification and detection of contraband, effectively ensures the accuracy of contraband identification and prevents potential safety hazards while improving the identification efficiency. Carrying out smooth denoising on the acquired X-ray image by adopting a neighborhood averaging method to obtain a smooth denoised X-ray image; and enhancing the edge information of the smoothed and denoised image by adopting a histogram equalization method to obtain a preprocessed X-ray image.
Chinese patent CN108734183A discloses an inspection apparatus and an inspection method. The container to be inspected is X-ray scanned to obtain a transmission image, then a first vector describing the partial transmission image is generated from the transmission image using a convolutional neural network, and a word vector is generated from the textual description of the container cargo as a second vector using a cyclic neural network. And integrating the first vector and the second vector to obtain a third vector expressing the transmission image and the text description. And judging the category of the goods in the container based on the third vector. According to the embodiment of the disclosure, the rough category of the target goods can be preliminarily judged, and further judgment of a diagraph is facilitated. The biggest difference of the invention is that word vectors generated by the word description are added.
Chinese patent CN110488368A discloses a contraband identification method and device based on a dual-energy X-ray security inspection machine, the method includes: acquiring a multichannel image set with marking information, wherein the multichannel image set comprises an HLS image, an equivalent atomic number image and an X-ray received energy image, and the marking information comprises position information and category information of contraband; inputting the multichannel image set into a convolutional neural network for training; and identifying the image to be detected by using the trained convolutional neural network, and outputting the position and the category of the contraband.
In summary, the above patent performs optimization training on the X-ray image by using the convolutional neural network method, so as to reduce the influence of factors such as external environment and better obtain the classification and detection results, and greatly improve the work efficiency and the accuracy of detecting forbidden articles and the like in the field of security detection. The patents mainly detect and identify the common forbidden dangerous goods in daily life such as guns, knives and the like, and meanwhile, the training data set is relatively sufficient and a great number of cases are supported. However, in some special cases, such as in some terrorist attacks abroad, some small metal objects can detonate explosives, which are more destructive and destructive, and meanwhile, when facing a freight container with a huge space and being in a very concealed state, the small metal objects often cannot be detected well, so that the large metal objects have great potential safety hazards. The invention mainly aims at detecting the small metal threat, realizes the extraction and training of the characteristics by using a deep learning method, has relatively fewer existing small metal threat data sets, and enlarges the data sets by data enhancement and the like when only a small amount of data is available. The innovation point is as follows: the use of a two-channel input in a convolutional network and an improvement in natural image processing makes the network model more suitable for processing such X-ray images with translucency; the innovation point is two: the existing SMT example is projected by utilizing multiplicative characteristics, so that a new data set is obtained, and the training of subsequent data is facilitated.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the method aims at the problems that the detection of the X-ray image is inaccurate and the detection time is long when a manual check is adopted in a related scheme. And detecting and classifying by adopting a feature extraction and clustering method aiming at the second related scheme, wherein the scheme does not process the X-ray image, and optimizes the class features of different images. And a detection method based on shape and texture is adopted for the third related scheme, and when the third related scheme faces Small Metal Threats (SMT), the imaging size is not considered to be small, the direction is not restricted, and special texture information is difficult to extract for classification. This approach does not enable accurate detection of Small Metal Threats (SMT).
The method for detecting the threats in the X-ray images by using the convolutional neural network effectively solves the problem that the threats cannot be accurately and respectively identified, is short in time consumption and strong in computing power, can meet the requirement of more inspection images, and can obtain a better data set training effect by expanding data by a learning representation method.
Specifically, the invention provides a method for detecting a small metal threat, which comprises the following steps: carrying out X-ray inspection on an object to be inspected to obtain an X-ray image; preprocessing the X-ray image to obtain a preprocessed X-ray image; density sampling is carried out by a small window; classifying the detected object by using a classifier trained on a preset deep learning model based on a convolutional neural network, calculating a confidence score of a detected image, and judging a small metal threat under the condition that the confidence is greater than a specific threshold value; and sending the detection result of the detected object to a security inspection terminal.
In addition, by mapping the normalized mean window score to pixel values at each location, in addition to computing the confidence score for the detected image, a thermal imaging map is generated during the classification process by which the detected small metal threats are localized.
The preprocessing of the X-ray image comprises: and the image is converted into a gray image, noise interference is reduced, logarithmic transformation is introduced, and the expression information of the image is improved. The deep learning model based on the convolutional neural network is obtained through training of a certain amount of sample data, and the sample data comprises existing related commercial flow image data and enhancement data. The enhancement data is based on the semi-transparent nature of the X-ray image, synthesizing a physically accurate image by projecting small metal threat instances into existing commercial stream images, while scaling and flipping the image intensity, resulting in diversified enhancement data.
The small metal threat example comprises the following operation steps: cutting out a patch containing a single small metal threat instance from a full-size image, manually performing pixel segmentation of the small metal threat instance to obtain a small metal threat binary mask, and performing background correction by dividing the cut-out color patch by the average intensity of pixels except the small metal threat binary mask.
According to the deep learning model based on the convolutional neural network, the CNN type is trained from the beginning by using a MatConvNet library, the MatConvNet is a MATLAB toolbox of the CNN in computer vision, the toolbox is simple to use, supports a complex model of a large data set, and provides calculation modules with filtering, such as linear convolution, pooling and the like. The network architecture is based on Simnyan and Zisserman (the neural network structure proposed by Simnyan and Zisserman can process large-scale image recognition, and the network depth is pushed to 16 to 19 layers, so that the detection performance is obviously improved, and the research and exploration of deep networks in the computer vision field are promoted), and a 19-layer CNN network structure is adopted, and 19 layers comprise 16CONV and 3 FC. Or a CNN network structure with 11 layers is adopted, wherein the 11 layers comprise 8CONV and 3 FC. In the process of training the neural network, the image input adopts dual-channel image input or gray scale image input.
The present invention proposes a new type of Small Metal Threat (SMT) detection technique, since threats are rare in commercial flows (SoC) and acquisition of phased images, data is difficult and time consuming to acquire. The scheme firstly expands by synthesizing data to solve the problem of few training data samples, performs convolutional neural network training on the basis of obtaining a large amount of data, then locates the detected SMT (or the source of a false positive signal) to perform classification decision, finally provides the result of automatically detecting the SMT in a full-size image, and performs performance evaluation on all types of backgrounds.
The overall structure of the invention is shown in figure 1.
The experimental steps of the invention are as follows: and enhancing a data set, processing data, building a CNN network structure and evaluating the performance.
Data set enhancement: the method comprises the steps of firstly collecting the existing related business flow (SOC) image data, including some industrial equipment and the like. We first crop out a patch containing a single SMT instance from a full size image. Pixel segmentation of the SMT instance is performed manually, resulting in an SMT binary mask. Background correction is performed by dividing the clipped color patch by the average intensity of the pixels except for the SMT binary mask. If an unrelated object or structure is present in the patch (e.g., part of other SMT or supporting structures), the corresponding pixel is also ignored during background correction. The SMT instance can then be projected into another X-ray image by intensity multiplication. Due to the semi-transparent characteristic of the X-ray image, the same SMT instance is projected to different image results to generate distinct appearances, a physically accurate image is synthesized by projecting the threat to the SoC image, and meanwhile, the data set is enhanced by zooming and image overturning of the image intensity, so that the data set is more diversified.
Data processing: during training, the data set is divided into a training set and a testing set, and the training data and the testing data cannot be overlapped. Preprocessing and logarithmic transformation are performed before classification, and the logarithmic transformation can stretch low gray values with narrow range and compress high gray values with wide range. Can be used to expand dark pixel values in an image while compressing light pixel values. This transformation typically detects the covert editor by security during visual inspection, and logarithmic conversion of the image is key to improving performance.
CNN network architecture: the CNN type of the present invention is ab initio Trained (TFS) using MatConvNet libraries, this architecture is based on simony and Zisserman, with multiple convolutional layers (CONV) with 3 x 3 filters of the neural network layer stacked between Max-Pooling layers (Max-Pooling), and adding three fully connected layer (FC) layers, making up a 19-layer (16CONV +3FC) CNN network structure altogether. The image input adopts a dual-channel image input (TFS-B), namely, a raw intensity and a logarithmic transformation intensity, and finally, the classification probability is output through a softmax loss function. Batch normalization (determining the mean and variance of the input distribution) is used for network regularization and to speed up the training speed, subtracting the average image computed in the training set from each input image. In addition, the horizontal and vertical directions of the image are also randomly flipped during training.
Performance evaluation: the experimental results were verified by the tested data set. The resulting test results reflect the performance index of the network structure and the accuracy of detecting Small Metal Threats (SMT).
Has the advantages that:
compared with the prior art, the Small Metal Threat (SMT) in X-ray is detected based on the convolutional neural network, and the data are trained to optimize class characteristics through a representation learning method. When 90% of the SMT composition detected in the detection result is hidden in the commercial flow image, a false alarm of less than 6% is reported. Compared with the traditional histogram statistical information retrieval method, the method is improved by one order of magnitude.
For the related patents, the data set can be expanded through multiplicative characteristics of X-rays, the characteristic information of the object is accurately extracted, and the condition of under-fitting caused by insufficient data is reduced.
Drawings
For a better understanding of the present invention, reference will now be made in detail to the following drawings, in which:
FIG. 1 is an overall block diagram of the present invention;
FIG. 2 is a graph of the effect of logarithmic transformation on X-ray images of bolt cutters;
fig. 3 is a TFS network configuration diagram;
FIG. 4 is a comparison graph of SMT detection performance for each method;
FIG. 5 is a comparison graph of non-SMT detection;
FIG. 6 is a diagram of an example of SMT detection of CNN-19-TFS-B;
FIG. 7 is a comparison graph of SMT testing.
Detailed Description
The present invention will be described in further detail with reference to specific examples. It is to be understood that these embodiments are provided to illustrate the general principles, major features and advantages of the present invention, and the present invention is not limited in scope by the following embodiments. The implementation conditions used in the examples can be further adjusted according to specific requirements, and the implementation conditions not indicated are generally the conditions in routine experiments.
The first step is to acquire relevant image data: during data enhancement we used the R60 orbit scan of a Rapiscan Eagle R scanner to obtain benign images (no SMT) of the job, which was equipped with a 6MV linac power supply and was able to obtain X-ray images with a resolution of 6mm/pixel-1 in the horizontal direction. While the image is a 16-bit grayscale image, the image size varies between 1290 × 850 and 2570 × 850 pixels, respectively, for containers 20 feet and 40 feet long. In practice, we randomly collected sample data from commercial stream (SoC) images, which may be empty samples (which account for about 20% of the entire data set), and may also contain commercial cargo containers, heavy machinery, industrial equipment, trays of household goods and bulk materials, and so on. Based on multiplicative property of X-ray transmission image formation, different appearance information is generated by projecting the same SMT example into different images, the SMT example is used as a new data set, and data enhancement is carried out on the images in a rotating mode, a zooming mode and the like.
And a second step of image data processing: in the data processing process, due to the semitransparent property of an X-ray image, preprocessing is firstly carried out, the X-ray image is converted into a gray image, noise interference is reduced, logarithmic transformation is introduced, and expression information of the image is improved. The detection of SMT in X-ray cargo images is a binary classification task, benign images (without SMT) are assigned to negative categories, SMT images (at least one SMT) are assigned to positive categories, density sampling is performed in a small window, classification is performed and confidence is given, and the classification prediction result is obtained by comparing the classification with a threshold T.
The third step is to train a convolutional neural network: in the process of training a neural network, the network architecture is based on simony and Zisserman, a CNN network structure with 19 layers (16CONV +3FC) is adopted, and structural variants with 11 layers (8CONV +3FC) are also discussed, and three configurations are performed on the two structures, as shown in fig. 3: grayscale image input (TFS-A), dual channel image input (TFS-B), then the original function and the logarithmically converted input are assigned to the first fully connected layer FC (TFS-C) with separated functions and then connected to the branches of the network (without right sharing). Batch normalization is used for network regularization and accelerates the training speed, and an overfitting phenomenon in the training process is prevented. The weight decay and momentum are fixed at 10-4 and 0.9 respectively. The learning rate decreased from 10-3 to 10-6 during the 30 sessions of training. In addition to the CNN model of TFS, a pre-training (PT) network was trained, which features were extracted from the FC1 and FC2 layers of the VGG-VD-19 model, whose structure was very similar to the 19-layer CNN-TFS, trained on ImageNet (natural photographic image dataset), and classified using a random forest classifier, which has fewer parameters and strong classification ability than other classifiers and is commonly used in engineering tasks. The input image is resized to 224 x 224 and the grayscale channel is replicated twice in the third dimension to match the desired RGB format. In addition to calculating confidence scores for the inspection images, a thermal imaging map may also be generated during the classification process by mapping the normalized average window score to pixel values at each location. These visualizations can generally locate detected SMT (or sources of false positive signals) to classify the classification decision, while also providing subsequent work (e.g., physical inspection) for security personnel.
Bag of words (BoW) function was also evaluated in network training experiments, including oriented basic image features (oBIF) and visual word Pyramid Histograms (PHOW). Where BIF is a fixed geometric feature and PHOW is an extended form of dense SIFT (scale invariant feature transform) multi-scale variation.
The fourth step is performance evaluation: after the neural network is built and trained, performance comparison of test classification is carried out next. The Titan X GPU from Invita corporation was used in the training process, with an average processing time of 3.5 seconds per image, which is also significantly less than the time for security personnel to inspect the cargo box. As can be seen from fig. 4, the bag-of-words model has poor detection effect, wherein the output after PHOW logarithmic transformation obtains the best AUC and H measures. The pre-trained (PT) CNN model does not behave as the bag-of-words model. The CNN trained from zero (TFS) has a good detection effect, the CNN-19-TFS-B model used in the method obtains the best detection result, and meanwhile, the logarithmic transformation plays an important role in improving the performance. In FIG. 4, AUC is the area under the ROC curve, and FPR90 is the false positive rate at 90% detection rate. The thermographic visual image of fig. 5 shows the detection without SMT, and the red signal (1.0) indicates FPR90, illustrating the best detection. FIG. 6 is a thermal imaging view showing the detection in the presence of SMT, and the detection can be prepared by classifying and identifying SMT at various positions by using CNN-19-TFS-B, and the positions of SMT in red areas. In general, the result of the comparison between the scheme of the patent and the bag of words and the pre-training (PT) network method is shown in FIG. 7, which shows that the AUC in the CNN-TFS method is 0.97, the H-measure is 0.78, the method has good detection and classification effects, and the performance of the method is obviously superior to that of other evaluation methods.
The present invention has been described in detail, and the detailed description is only for the purpose of facilitating understanding the method and the core concept of the present invention, which is intended to enable those skilled in the art to understand the contents of the present invention and to implement the present invention, and not to limit the scope of the present invention. All equivalent changes and modifications made according to the spirit of the present invention should be covered within the protection scope of the present invention.
Claims (10)
1. A method for detecting a small metal threat, comprising: carrying out X-ray inspection on an object to be inspected to obtain an X-ray image; preprocessing the X-ray image to obtain a preprocessed X-ray image; density sampling is carried out by a small window; classifying the detected object by using a classifier trained on a preset deep learning model based on a convolutional neural network, calculating a confidence score of a detected image, and judging a small metal threat under the condition that the confidence is greater than a specific threshold value; and sending the detection result of the detected object to a security inspection terminal.
2. The detection method according to claim 1, characterized in that: in addition to calculating confidence scores for the detected images, a thermographic map is generated during the classification process by mapping the normalized mean window scores to pixel values at each location, through which the detected small metal threats are localized.
3. The detection method according to claim 1, characterized in that: the preprocessing of the X-ray image comprises: and the image is converted into a gray image, noise interference is reduced, logarithmic transformation is introduced, and the expression information of the image is improved.
4. The detection method according to claim 1, characterized in that: the deep learning model based on the convolutional neural network is obtained through training of a certain amount of sample data, and the sample data comprises existing related commercial flow image data and enhancement data.
5. The detection method according to claim 4, characterized in that: the enhancement data is based on the semi-transparent nature of the X-ray image, synthesizing a physically accurate image by projecting small metal threat instances into existing commercial stream images, while scaling and flipping the image intensity, resulting in diversified enhancement data.
6. The detection method according to claim 5, characterized in that: the small metal threat example comprises the following operation steps: cutting out a patch containing a single small metal threat instance from a full-size image, manually performing pixel segmentation of the small metal threat instance to obtain a small metal threat binary mask, and performing background correction by dividing the cut-out color patch by the average intensity of pixels except the small metal threat binary mask.
7. The detection method according to claim 1, characterized in that: the deep learning model based on the convolutional neural network has a CNN type which is trained from scratch by using a MatConvNet library.
8. The detection method according to claim 1, characterized in that: the network architecture of the deep learning model based on the convolutional neural network adopts a 19-layer CNN network structure, wherein the 19 layers comprise 16CONV and 3 FC.
9. The detection method according to claim 1, characterized in that: the network architecture of the deep learning model based on the convolutional neural network adopts a CNN network structure with 11 layers, wherein the 11 layers comprise 8CON and 3 FC.
10. The detection method according to claim 1, characterized in that: according to the deep learning model based on the convolutional neural network, in the process of training the neural network, image input adopts dual-channel image input or gray image input.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010376167.3A CN111582367A (en) | 2020-05-07 | 2020-05-07 | Small metal threat detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010376167.3A CN111582367A (en) | 2020-05-07 | 2020-05-07 | Small metal threat detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111582367A true CN111582367A (en) | 2020-08-25 |
Family
ID=72112539
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010376167.3A Pending CN111582367A (en) | 2020-05-07 | 2020-05-07 | Small metal threat detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111582367A (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103975233A (en) * | 2012-02-06 | 2014-08-06 | 株式会社日立高新技术 | X-ray inspection device, inspection method, and x-ray detector |
| CN107530040A (en) * | 2015-04-01 | 2018-01-02 | 株式会社日立制作所 | X ray CT device, restructing operation device and X ray CT image generating method |
| CN109948565A (en) * | 2019-03-26 | 2019-06-28 | 浙江啄云智能科技有限公司 | A kind of not unpacking detection method of the contraband for postal industry |
| CN110163179A (en) * | 2019-05-29 | 2019-08-23 | 浙江啄云智能科技有限公司 | A kind of contraband detecting recognition methods, system, equipment and its storage medium based on deep learning |
| CN110155572A (en) * | 2019-06-25 | 2019-08-23 | 杭州电子科技大学 | A kind of intelligence community garbage classification system and method |
| CN110176010A (en) * | 2019-05-24 | 2019-08-27 | 上海联影医疗科技有限公司 | A kind of image detecting method, device, equipment and storage medium |
| CN110533051A (en) * | 2019-08-02 | 2019-12-03 | 中国民航大学 | Contraband automatic testing method in X-ray safety check image based on convolutional neural networks |
| CN110543857A (en) * | 2019-09-05 | 2019-12-06 | 安徽启新明智科技有限公司 | Contraband identification method, device and system based on image analysis and storage medium |
-
2020
- 2020-05-07 CN CN202010376167.3A patent/CN111582367A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103975233A (en) * | 2012-02-06 | 2014-08-06 | 株式会社日立高新技术 | X-ray inspection device, inspection method, and x-ray detector |
| CN107530040A (en) * | 2015-04-01 | 2018-01-02 | 株式会社日立制作所 | X ray CT device, restructing operation device and X ray CT image generating method |
| CN109948565A (en) * | 2019-03-26 | 2019-06-28 | 浙江啄云智能科技有限公司 | A kind of not unpacking detection method of the contraband for postal industry |
| CN110176010A (en) * | 2019-05-24 | 2019-08-27 | 上海联影医疗科技有限公司 | A kind of image detecting method, device, equipment and storage medium |
| CN110163179A (en) * | 2019-05-29 | 2019-08-23 | 浙江啄云智能科技有限公司 | A kind of contraband detecting recognition methods, system, equipment and its storage medium based on deep learning |
| CN110155572A (en) * | 2019-06-25 | 2019-08-23 | 杭州电子科技大学 | A kind of intelligence community garbage classification system and method |
| CN110533051A (en) * | 2019-08-02 | 2019-12-03 | 中国民航大学 | Contraband automatic testing method in X-ray safety check image based on convolutional neural networks |
| CN110543857A (en) * | 2019-09-05 | 2019-12-06 | 安徽启新明智科技有限公司 | Contraband identification method, device and system based on image analysis and storage medium |
Non-Patent Citations (4)
| Title |
|---|
| THOMAS W. ROGERS等: "《Automated X-ray Image Analysis for Cargo Security》", 《ARXIV》 * |
| 唐小川 等: "《基于交互作用的文本分类特征选择算法》", 《计算机应用》 * |
| 王宇 等: "《基于计算机视觉的X射线图像异物分类研究》", 《液晶与显示》 * |
| 高红霞 等: "《微焦点X射线图像乘性加性混合噪声的去除》", 《光学精密工程》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3349050B1 (en) | Inspection devices and methods for detecting a firearm | |
| US10013615B2 (en) | Inspection methods and devices | |
| Rogers et al. | Automated x-ray image analysis for cargo security: Critical review and future promise | |
| CN110956225B (en) | Contraband detection method and system, computing device and storage medium | |
| CN107145898B (en) | Radiographic image classification method based on neural network | |
| US10509979B2 (en) | Inspection methods and systems | |
| Rogers et al. | A deep learning framework for the automated inspection of complex dual-energy x-ray cargo imagery | |
| Bhowmik et al. | The good, the bad and the ugly: Evaluating convolutional neural networks for prohibited item detection using real and synthetically composited X-ray imagery | |
| CN110765990A (en) | Intelligent article detection method and system, computing device and storage medium | |
| Jaccard et al. | Automated detection of smuggled high-risk security threats using deep learning | |
| CN115439693A (en) | Training method of target recognition network model, electronic device and program product | |
| CN109726660A (en) | A kind of remote sensing images ship identification method | |
| Wang et al. | Fire situation detection method for unmanned fire-fighting vessel based on coordinate attention structure-based deep learning network | |
| CN110992324B (en) | Intelligent dangerous goods detection method and system based on X-ray image | |
| CN117274623A (en) | Identification method and system for intelligent retrieval and identification of forbidden articles | |
| CN116912519A (en) | Contraband detection method integrating image color features | |
| EP4086812A1 (en) | Object identification system and method | |
| CN116543308A (en) | Landslide detection early warning model and early warning method based on multi-model fusion | |
| CN111582367A (en) | Small metal threat detection method | |
| CN112633286B (en) | Intelligent security inspection system based on similarity rate and recognition probability of dangerous goods | |
| Liu et al. | A Lightweight Dangerous Liquid Detection Method Based on Depthwise Separable Convolution for X‐Ray Security Inspection | |
| Lyu et al. | X-ray image classification for parcel inspection in high-speed sorting line | |
| Pumpong et al. | Building detection in airports through remote sensing image using YOLOv3 with Jet Saliency map | |
| Deena et al. | Prohibited Object Detection using YOLOv7 on X-Ray Images of Airport Luggages (OPIXray) | |
| Nasim et al. | Incremental Instance Segmentation for Cluttered Baggage Threat Detection |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200825 |
|
| WD01 | Invention patent application deemed withdrawn after publication |