CN111783750A

CN111783750A - Data processing method and device

Info

Publication number: CN111783750A
Application number: CN202010824770.3A
Authority: CN
Inventors: 张毅; 刘昊
Original assignee: Lanzhou University
Current assignee: Lanzhou University
Priority date: 2020-08-17
Filing date: 2020-08-17
Publication date: 2020-10-16
Anticipated expiration: 2040-08-17
Also published as: CN111783750B

Abstract

The application discloses a data processing method and a data processing device, wherein signal data generated by incident particles through a GEM (phase change memory) detector are acquired; carrying out dimensionality reduction processing on the signal data; and (5) performing track reconstruction on the data subjected to the dimension reduction processing by using a neural network. The difficulty of a track reconstruction algorithm can be simplified, and the imaging quality is finally improved.

Description

Data processing method and device

Technical Field

The application relates to the field of particle track reconstruction, in particular to a data processing method.

Background

The nondestructive imaging technology can adopt a neutron imaging mode, avoids the influence of the radioactivity of the detected object, and can form a high-quality image on the strong radiation substance.

In the related art, when a plurality of protons or X-rays generate signals in a single electronic clock cycle, a reconstruction module needs to judge and process the signals in advance, and the reconstruction algorithm has low efficiency due to huge data volume.

Disclosure of Invention

The application mainly aims to provide a data processing method to solve the technical problem that the reconstruction algorithm of the track reconstruction algorithm is low in efficiency.

In order to achieve the above object, in a first aspect, the present application provides a data processing method.

The data processing method comprises the following steps: acquiring signal data generated by incident particles through a GEM detector; performing dimensionality reduction processing on the signal data; and performing track reconstruction on the data subjected to the dimensionality reduction by using a neural network.

In some embodiments, the performing the dimensionality reduction on the signal data comprises: performing dimensionality reduction processing on the signal data to obtain initial signal data; and screening the data type contained in the initial signal data to obtain signal data of a preset data type.

In some embodiments, the dimension reduction processing on the signal data includes selecting a data type included in the signal data to obtain signal data of a preset data type; and performing dimensionality reduction processing on the signal data of the preset data type.

In some embodiments, the performing the trajectory reconstruction on the dimensionality reduced data by using the neural network further comprises: and sending the reconstructed particle track information to an upper computer for storage.

In a second aspect, the present application further provides a data processing apparatus, including an obtaining unit configured to obtain signal data generated by an incident particle via a GEM detector; a processing unit configured to perform a dimension reduction process on the signal data; and the reconstruction unit is configured to perform track reconstruction on the data subjected to the dimensionality reduction processing by utilizing a neural network.

In some embodiments, the processing unit includes a first processing module configured to perform a dimension reduction process on the signal data to obtain initial signal data; and the first selection module is configured to select the data type contained in the initial signal data to obtain signal data of a preset data type.

In some embodiments, the processing unit includes a second selection module configured to select a data type included in the signal data to obtain signal data of a preset data type; and the second processing module is configured to perform dimension reduction processing on the signal data of the preset data type.

In some embodiments, the apparatus further comprises a sending unit configured to send the reconstructed particle trajectory information to a host computer for storage.

In a third aspect, the present application provides an electronic device comprising one or more processors; storage means having one or more programs stored thereon which, when executed by the one or more processors, cause the one or more processors to carry out the method according to any one of the first aspects.

In a fourth aspect, the present application provides a computer storage medium having a computer program stored thereon, wherein the program when executed by a processor implements a method as in any of the embodiments of the first aspect

In the embodiment of the application, a data processing method and a data processing device are provided, wherein signal data generated by incident particles through a GEM (phase change memory) detector are acquired; carrying out dimensionality reduction processing on the signal data; and (5) performing track reconstruction on the data subjected to the dimension reduction processing by using a neural network. By reducing the dimension of the data, the difficulty of a track reconstruction algorithm can be simplified, and the imaging quality is finally improved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this application, serve to provide a further understanding of the application and to enable other features, objects, and advantages of the application to be more apparent. The drawings and their description illustrate the embodiments of the invention and do not limit it. In the drawings:

fig. 1 is a flowchart of a data processing method provided according to an embodiment of the present application;

fig. 2 is a selection flowchart in a data processing method according to an embodiment of the present application.

Fig. 3 is a diagram illustrating an effect of the degree of conformity after track reconstruction in a data processing method according to an embodiment of the present application.

Fig. 4 is a schematic diagram of obtaining an experimental data set in a data processing method according to an embodiment of the present application.

Fig. 5 is a block diagram of a data processing apparatus according to an embodiment of the present application.

Detailed Description

In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only partial embodiments of the present application, but not all 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 application.

It should be noted that the terms "first," "second," and the like in the description and claims of this application and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It should be understood that the data so used may be interchanged under appropriate circumstances such that embodiments of the application described herein may be used. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

Moreover, some of the above terms may be used to indicate other meanings besides the orientation or positional relationship, for example, the term "on" may also be used to indicate some kind of attachment or connection relationship in some cases. The specific meaning of these terms in this application will be understood by those of ordinary skill in the art as appropriate.

In addition, the term "plurality" shall mean two as well as more than two.

It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.

FIG. 1 shows an embodiment of a data processing method, comprising:

and step 101, acquiring signal data generated by the incident particles through a GEM detector.

And 102, performing dimensionality reduction on the signal data.

In this embodiment, the dimension reduction process may perform dimension reduction on the signal data by using a sparse encoder. The normalization may be performed by taking at least one period of data as a data set; and training the untrained sparse automatic encoder based on the normalized data set, the preset learning rate, the preset maximum iteration number, the preset network structure and the data compression to obtain the trained sparse automatic encoder. The sparse autoencoder may be trained from at least one cycle of data from the output of a GEM detector (gas electron Multiplier). For example, 30000 cycles of data can be used to train a sparse self-encoder, before training a sparse self-encoding network, the input data set of the network needs to be normalized to a range of (0.1,0.9), the learning rate of the training is 0.0001, the maximum iteration number is 1000, the structure of the trained network is 400-40-400, and the data compression ratio is 10: 1. After 973 periods of network training, the iteration is stopped, and the loss function is converged.

In some optional implementations of this embodiment, the performing the dimension reduction processing on the signal data includes: performing dimensionality reduction processing on the signal data to obtain initial signal data; and screening the data types contained in the initial signal data to obtain the signal data of the preset data types.

In this embodiment, the signal data includes signal data of at least one data type, such as a noise data type, a single particle data type, and/or a plurality of particle data types.

The signal data of at least one data type may be selected (for example, event data meeting requirements is selected and captured by using a selection algorithm), so as to obtain single particle data used for reconstructing particle track information, where the signal data of a preset data type is the single particle data. In the screening process, the data types of the signal data can be classified through a network model (such as a classification algorithm model), and then the classified signal data is screened to obtain single particle data.

In some optional implementations of this embodiment, the performing the dimension reduction processing on the signal data includes: selecting data types contained in the signal data to obtain signal data of preset data types; and performing dimension reduction processing on the signal data of the preset data type.

In this embodiment, the signal data may be first selected (for example, event data meeting requirements are selected and captured by using a selection algorithm), so as to obtain single particle data before dimensionality reduction, and then the single particle data is subjected to dimensionality reduction, so as to obtain signal data for track reconstruction.

In the screening process, data types of the signal data can be classified through a network model (for example, a neural network model such as a classification algorithm model), and then the classified signal data is screened to obtain single particle data.

In the classification, a large number of events can be used as network model input data for classification, the output of the network is three nodes of a label of the data, the label a of the noise data is 100, the single particle data b is 010, and the multiple particle data c is 001. The classification algorithm is trained, and the classification of the data inputted in series is determined by using the classification algorithm, and when the data is determined to be a-type data, the identifier 0 is outputted, when the data is determined to be b-type, the identifier 1 is outputted, and when the data is determined to be c-type, the identifier 2 is outputted. And then entering a selection process.

Referring to fig. 2, fig. 2 shows a selection flowchart in the data processing method provided in the present embodiment, in fig. 2, Buffer: and shifting and caching the coded data and the replaced data, and transmitting the data to the Workspace according to the judgment result of the event capture queue.

Workspace: and carrying out a reconstruction algorithm once after receiving data transmitted by the Buffer once, and then emptying.

Event capture queue: the input end inputs the judgment result of the classification algorithm, 0 only contains noise, 1 contains data of one event, and 2 contains data of two events.

When there is: (taking the example of capturing an 8-cycle event)

The symbol 0111111110 indicates that an 8-cycle event was captured, 8 sets of data in 1 corresponding buffer were passed to the workpace, and 1 of the events in the queue was cleared to 0, and the next cycle was followed by a continued shift to the right.

Simple process: identification area:

the information was obtained by identifying 8 cycles and making it possible to transmit data >

At this time, 8-18 groups of data in the Buffer are transmitted into the region for Workspace pair, and an enabling is given to the reconstruction algorithm, and meanwhile, the 1 on the right side is set to be 0. This completes the capture of an event data and data reconstruction.

Other cycles were similar, and 8 cycles of data were captured by 0111111110

9 cycles of data were captured by 01111111110

At 011111111110, 10 cycles of data were captured

At 0111111111110 capture 11 cycles of data

At 01111111111110 capture 12 cycles of data

011111111111110 captures 13 cycles of data

0111111111111110 captures 14 cycles of data

Other periods are similar.

The signal data can be classified through a classification algorithm, the periodic data suitable for the work of a reconstruction algorithm are screened from the data collected by a detector through a screening method, and the complex data caused by the fact that excessive periodic data of null noise and multiple particles are overlapped and collected by the detector are filtered. The gamma signal doped in the data is removed by the limitation of the cycle number.

And 103, performing track reconstruction on the data subjected to the dimensionality reduction by using a neural network.

In the embodiment, the neural network can be trained in a preset mode in advance, the neural network obtained through training can be used for track reconstruction of events with different duration periods, the reconstruction algorithm is reliable in operation, and good position reconstruction can be achieved. The trajectory reconstruction algorithm can be implemented using Verilog HDL. When the neural network is trained to carry out track reconstruction, the input data respectively lasts for 8-20 cycles by using the data of the events, the original data is not directly used during training, but the original data passes through the sparse self-encoder which is trained, and the output of the hidden layer of the sparse self-encoder is used as the input of the neural network (for example, a feedforward network). When the period of a certain event is less than 20, the result of dimensionality reduction of the free period is replaced by 0, and the output of the feedforward network is the position of the incident particle in the simulation. Fig. 3 shows a graph of the effect of the degree of conformity after the reconstruction of the track in the data processing method.

In some optional implementation manners of this embodiment, in this embodiment, after performing trajectory reconstruction on the data subjected to the dimensionality reduction processing by using the neural network, sending the reconstructed particle trajectory information to an upper computer for storage.

In this embodiment, after completing the track reconstruction, the method may further include: and correcting the reconstructed track.

Because of the accuracy limitation of simulation software, simulation data cannot be completely developed instead of real data. The analog data and the scale data of the detector can be mixed in a certain proportion to be used as a data set for correction. Because the simulation data and the real experiment have errors but limited ranges, the training is suspended and the network is saved when the mean square error of the feedforward network is reduced to about 2, and then the network is continuously trained by a data set mixed by the simulation data and the calibration data. This speeds up the training of the network. The procedure for obtaining the experimental data set is shown in fig. 4:

a1 mm thick epoxy plate of the same shape (30 mm) as the sensitive area of the detector was prepared, with 101 mm wide slits spaced 1.5mm apart. Two additional unslit epoxy panels were used to mask them as in figure 4 a.

The baffle is placed on top of the sensitive area of the detector, covered with two non-slit baffles and left with a desired slit. Am241 is placed over the slit. The irradiation was performed for 1 hour at each slit and the data collected by the detector was collected with the slit middle position as the incident position, as shown in fig. 4 b.

Based on the acquired trajectory data of 10 incident positions, the data are directly used for training the algorithm, and the training defect is not representative of the data set. Since the detector has translational invariance, translation is then used for data enhancement. The incident position of the particle and the corresponding track data in the above data are simultaneously translated to the left or the right (ensuring that the incident position and the range of the data are still inside the detector), so that the track data of a large number of incident positions are obtained

After the experimental data acquisition is finished and the data enhancement is carried out, the simulation data and the experimental data are processed according to the following steps of 2: 8, and continuing to train the feedforward network until the loss function converges. Therefore, the development of an algorithm level is completed, when the working condition of the detector changes, the calibration is carried out again, the above work is repeated, and the sparse self-encoder does not need to be trained again.

In this embodiment, after the event screening and proton trace reconstruction are completed, the reconstructed result is uploaded to a computer by using a communication interface protocol (for example, USB, Ethernet), and the data is used for storage or imaging.

FIG. 5 shows an embodiment of a data processing apparatus comprising: an acquisition unit 501 configured to acquire signal data generated by the incoming particles via the GEM detector;

a processing unit 502 configured to perform a dimension reduction process on the signal data;

a reconstruction unit 503 configured to perform a trajectory reconstruction on the dimension-reduced data by using a neural network.

As an optional implementation manner of this embodiment, the processing unit includes a first processing module, configured to perform dimension reduction processing on the signal data to obtain initial signal data; and the first selection module is configured to select the data type contained in the initial signal data to obtain signal data of a preset data type.

As an optional implementation manner of this embodiment, the processing unit includes a second selection module configured to select a data type included in the signal data to obtain signal data of a preset data type; and the second processing module is configured to perform dimension reduction processing on the signal data of the preset data type.

As an optional implementation manner of this embodiment, the apparatus further includes: and the sending unit is configured to send the reconstructed particle track information to an upper computer for storage.

An embodiment of an electronic device comprises: one or more processors; a storage device having one or more programs stored thereon which, when executed by the one or more processors, cause the one or more processors to implement a method as in any one of the embodiments of the data processing method.

An embodiment of the computer-readable medium comprises, the storage medium storing a computer program which, when executed by a processor, implements the method as described in any of the data processing method embodiments.

In this embodiment, the program included in the data processing method can be transplanted to the FPGA platform on the FPGA platform, a large number of combinational logics are used to complete the operation part in the algorithm, the data transmitted from the detector is analyzed and processed by using the powerful parallel capability of the FPGA, and the gamma signal and the proton overlapping proton or other particle signal are filtered out by the selection algorithm through the period information.

The data processing method can be realized by adopting an industrial-grade FPGA (field programmable gate array), the main frequency can reach 1GHz, the processing speed can catch up with the signal transmission speed of the GEM detector, the defect that the traditional reconstruction algorithm depends on the simulation precision is overcome, and the online reconstruction is realized.

The ADC inputs data to the FPGA in a period of every 25ns, the FPGA performs coding operation at a corresponding rate, then the compressed data is stored in the on-chip ROM, the storage amount is data of an event, namely 30 periods, the data of a full period stored enter the next module for operation, and then the operation result is transmitted to the computer through the USB communication protocol. Parameters used by the algorithm are stored in the ROM in design, a large number of vector dot product operations need to be carried out in the operations, the algorithm parameters stored in the ROM are continuously called, and the operation of the excitation function in the algorithm is completed in a lookup table mode.

Because FPGA is difficult to realize floating point number operation, the fixed point number operation realized by FPGA mainly needs to fix the algorithm realized by PC.

And (3) fixed point formation: the algorithm obtained by the PC terminal through a program is a result of floating-point number operation, and the obtained parameters comprise tens of decimal places, so that the feasibility and the accuracy of the algorithm are ensured. These tens of decimal points need to be represented by an appropriate fixed point number, and a certain calculation accuracy is lost to make hardware implementation possible.

Counting number: the decimal point is expressed by appointing a binary digit number, then representing the positive and negative of the numerical value by the sign bit at the first digit, and then appointing the position of the decimal point. For example, fix dt (1.16.12) represents a 16-bit fixed point number with a 1-bit sign bit and a 12-bit decimal bit. FIxdt (0.32,16) represents a 32-bit fixed point number, including 16-bit decimal places and an unsigned bit.

And (3) selecting a proper fixed point number for representing the parameters obtained by the algorithm, and losing certain precision. And (3) performing fixed point number simulation by using software, comparing floating point number operation results with fixed point number operation results in various formats, and selecting a proper fixed point number format. Firstly, the fixed point number of 32 has high precision and small operation error with a floating point number, but the 32-bit operation has higher difficulty in realizing the FPGA, the 12-bit fixed point number has higher calculation loss, and the algorithm significance is lost. The final decision uses the parameters of the algorithm in the form of the expression of fixdt (1,18,15), since the parameters of the algorithm are all in the range of (-1,1), and therefore only 2 integer bits are used to maximize the use of hardware resources.

A vector dot product module: and (3) completing a required vector point multiplication module by using a multiply accumulator, respectively transmitting each dimension of the vector and the corresponding parameter thereof, calculating the product of the each dimension and accumulating the product. The parameters are stored in a file independently, when the detection change needs to be redeveloped, only the parameters in the file need to be modified, and the algorithm framework does not need to be modified. And finally, outputting the calculation result.

Implementation of the excitation function: the excitation function is used in the algorithm, wherein the self-coding module and the feedforward network need to use two different Sigmoid functions in a chain, which are respectively:

obviously, the excitation function needs a large number of exponential operations and division operations. The method can be realized directly on the FPGA level, can replace an excitation function by using a lookup table method, and has the route that a function value in a certain definition domain of the excitation function is calculated on a computer, the function value is written on a ROM of the FPGA point to point, and the function value output of the point is directly found from the ROM when calculation is needed.

And carrying out algorithm development on the Matlab platform, and reserving corresponding parameters of the track reconstruction algorithm.

Hardware development is carried out on a Quartus II platform, Verilog HDL is used for realizing the algorithm steps, the developed codes are downloaded into FLASH through a USB port and an embedded lead of a computer, and after a development board is powered on, an FPGA reads program codes from the FLASH and starts to run the codes. And the ADC transmits data to the FPGA, and the FPGA directly transmits the information of the reconstructed particles obtained by calculation to a computer after executing algorithm operation.

Compared with the traditional reconstruction algorithm, the track reconstruction algorithm in the data processing method and the device provided by the embodiment of the application can obtain higher accuracy, so that a detector with a larger sensitive area or position measurement with higher accuracy can be realized. The on-line processing of the track signal of the detector is realized, the step of transmitting, recording and processing a large amount of data is omitted, and the complicated data processing process is saved. The same data acquisition system realizes more functions and can be applied to the field of complex requirements. The track reconstruction is carried out by using a specific algorithm and is verified on a computer, so that the hardware calculation loss is reduced to the maximum extent, and the real-time reconstruction of the track signal is realized on a hardware level, so that the imaging with higher quality can be obtained when the method is used for detecting imaging under the same working condition.

The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method of data processing, comprising:

acquiring signal data generated by incident particles through a GEM detector;

performing dimensionality reduction processing on the signal data;

and performing track reconstruction on the data subjected to the dimensionality reduction by using a neural network.

2. The data processing method of claim 1, wherein performing dimensionality reduction processing on the signal data comprises:

performing dimensionality reduction processing on the signal data to obtain initial signal data;

and screening the data type contained in the initial signal data to obtain signal data of a preset data type.

3. The data processing method of claim 1, wherein performing dimensionality reduction processing on the signal data comprises:

selecting data types contained in the signal data to obtain signal data of preset data types;

and performing dimensionality reduction processing on the signal data of the preset data type.

4. The data processing method of claim 1, wherein the performing the trajectory reconstruction of the dimensionality reduced data with the neural network further comprises:

and sending the reconstructed particle track information to an upper computer for storage.

5. A data processing apparatus comprising:

an acquisition unit configured to acquire signal data generated by the incoming particles via the GEM detector;

a processing unit configured to perform a dimension reduction process on the signal data;

and the reconstruction unit is configured to perform track reconstruction on the data subjected to the dimensionality reduction processing by utilizing a neural network.

6. The data processing apparatus according to claim 5, wherein the processing unit comprises:

the first processing module is configured to perform dimensionality reduction processing on the signal data to obtain initial signal data;

and the first selection module is configured to select the data type contained in the initial signal data to obtain signal data of a preset data type.

7. The data processing apparatus according to claim 5, wherein the processing unit comprises:

the second selection module is configured to select data types contained in the signal data to obtain signal data of preset data types;

and the second processing module is configured to perform dimension reduction processing on the signal data of the preset data type.

8. The data processing apparatus of claim 5, wherein the apparatus further comprises:

and the sending unit is configured to send the reconstructed particle track information to an upper computer for storage.

9. An electronic device comprising one or more processors;

a storage device having one or more programs stored thereon,

when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-4.

10. A computer storage medium having a computer program stored thereon, wherein the program when executed by a processor implements the method of any one of claims 1-4.