Research On Key Techniques Of Radioactive Sources Localization In Coded-Aperture Imaging

Coded-aperture imaging is an important technique for the search and positioning of radioactive sources in the field of nuclear safety.It has high angular resolution and detection sensitivity,and is suitable for accurate reconstruction of the spatial distribution of radiation hotspots.It has been widely used in nuclear accident emergency,decommissioning and decontamination of nuclear facilities,and homeland security.However,this technique still has the following problems:（1）The detection energy range is wide in nuclear safety applications so that a coded-aperture imager needs a thick collimator to ensure that the projection data has sufficient contrast,but it will introduce the collimation effect at the same time.The number of Compton scattered events increases with the increment of incident photon energy.These problems will make the accurate positioning of radiation hotspots in a wide energy range challenging.（2）The dual-plane geometric structure formed by the collimator and the detector severely limits its field-of-view（FOV）.Not only does it affect the detection region,but the cyclic characteristics of the collimator are more likely to cause the pseudo reconstructed hotspots due to the sources within the partially coded field-of-view（PCFOV）.This partial coding effect is the inherent problem of the coded-aperture imaging technique and will affect the accurate localization of radioactive sources.（3）A large number of measurements at different angles are required for searching for radioactive sources in a broad region due to the imagers’limited FOV.Not only will it take a long measurement time and lead the operator to receive more radiation,but also the results will be interfered by the pseudo hotspots caused by the partial coding effect.Compton imaging technique has a large FOV but it cannot be applied to the low-energy photons and has low angular resolution.Therefore,it is of great significance to develop methods with large FOV and high-precision for the source localization in coded-aperture imaging.This thesis researches the image reconstruction algorithms,neural network methods and panoramic imaging methods.The main research results are as follows:（1）The maximum-likelihood expectation-maximization（MLEM）reconstruction algorithm using multi-energy windows is established.An imager prototype is constructed based on bismuth germanate.The imaging energy range is divided into 4energy windows according to its crystal characteristics.The multi-energy system response matrices（SRM）are calculated using Monte Carlo simulations to reconstruct the corresponding energy windows.5 point sources with the same energy,5 point sources with different energies and 2 ring sources with different energies were simulated,and five pitchblende samples with different sizes were experimented with to verify the algorithm.The results show that the algorithm can reconstruct the point sources with average positioning biases along the X and Y axes are less than 1.25 mm.And the average reconstruction ratio between the ring sources is 0.84.It can also accurately reconstruct the different sizes of the pitchblende samples.It demonstrates that this algorithm has higher localization accuracy and image quality than the results using a single SRM,and with similar performance to the photopeak window method.Compared with the photopeak window method that requires nuclides identification and calculation of the system response matrix of all nuclides,this algorithm can reduce the requirements for the energy resolution of the detector and greatly save the calculation cost,while accurately positioning radiation hotspots with a wide energy range.（2）A neural network method for the identification and localization of the radioactive sources within the PCFOV is established.The localization of the source region is converted to an image recognition problem.The source within the PCFOV can be identified and localized in real-time by combining the identification result of the source region and the hotspot information extracted from the reconstructed image using the cross-correlation method.The result of the simulated test set shows that the method can identify the source within the PCFOV with the F1 value up to 0.997,and the average positioning errors along the X and Y axes are 15.0 mm and 16.4 mm,respectively.In addition,four additional test sets with different count levels,background noises,source energies and the number of sources were simulated to evaluate the generalization performance of this method.The results show that except for the test set with the energy of 122 ke V has a 0.812 F1 value,the method can identify the source’s region with the F1 value above 0.95 and the coefficient of determination（）of the localization results close to 1 for all other test sets.This method also shows strong robustness to the statistical noise.A ¹³⁷Cs source is placed at 23 positions to collect projection data with different count levels in the experiment.The results show that all the source positions can be accurately identified and localized at 180×10³ count level,and only 5 positions are incorrectly identified even at 18×10³ count level.It demonstrated the feasibility and robustness of the model in real world scenarios.In addition,an extended-FOV MLEM reconstruction algorithm is established for the complex radiation scenes where both the fully coded field-of-view（FCFOV）and PCFOV have radioactive sources,and the radiation scenes with multiple sources are constructed with both simulation and experimental data.The results show that the existing algorithms incorrectly reconstruct the radioactive source within the PCFOV,and the extended-FOV MLEM algorithm can accurately reconstruct the radioactive sources of both the FCFOV and PCFOV at the same time.Both of the above methods effectively extend the localization and reconstruction region of radioactive sources to the PCFOV,and solve the problem of the limited FOV and interference by the pseudo hotspot caused by the partial coding effect.（3）A panoramic imaging method combined with the imaging system and the robot is established to aim at the application scenario of searching for radioactive sources in a broad region.This method controls the robot to collect radiation data and optical images at different angles,and reconstructs radiation images using the extended-FOV MLEM algorithm.And the panoramic imaging can be achieved.A ¹³⁷Cs source and a pitchblende sample are placed at different angles in the experiment for verification.The results demonstrate that compared with the conventional measurement methods,this method can not only avoid the interference caused by the partial coding effect,but also it can rapidly and accurately realize panoramic imaging in the horizontal 2πFOV,and vertical and diagonal directions including the PCFOV with fewer imaging times.It can reduce the radiation exposure of operators and provide a new solution for large FOV imaging for nuclear safety applications.The main innovations in this work are as follows:（1）An MLEM reconstruction algorithm using multi-energy windows is established.Compared with the photopeak window method,it can achieve accurate localization of radiation hotspots in a wide energy range with less calculation cost.（2）A neural network method for identification and localization of radioactive sources within the PCFOV and an extended-FOV MLEM reconstruction algorithm are established,which solves the inherent problem of the coded aperture imaging being interfered by the pseudo hotspot caused by the partial coding effect.（3）A panoramic imaging method combined with the imaging system and the robot is established,which can avoid the interference caused by the partial coding effect,and realize the rapid and accurate panoramic imaging of the horizontal 2πfield of view,vertical and diagonal directions including PCFOV.It provides a new large-FOV imaging solution that replaces people with a robot in nuclear safety applications.