Research On Target Detection Of Nuclear Waste Image In Nuclear Radiation Environment Based On Deep Learning

The rapid development of China’s nuclear energy industry has resulted in a large amount of nuclear waste generated every year,and the correct disposal of nuclear waste has attracted more and more attention from the public.In a complex nuclear radiation environment,due to the influence of various radioactive elements,humans cannot directly enter the work,and nuclear emergency robots can replace humans to enter the nuclear radiation area,complete the transmission of on-site conditions and perform nuclear emergency tasks.The effective detection of nuclear waste through vision in the nuclear radiation environment is the first step for the nuclear emergency robot to work normally.Therefore,from the perspective of the deep learning target detection model,this paper effectively detects the nuclear waste image in the nuclear radiation environment To assist the nuclear emergency robot to complete the task.This article first analyzes the research status of nuclear emergency robots,nuclear waste image target detection and deep learning,then preprocesses nuclear waste image samples and makes data sets,and then studies and improves the deep learning target detection model suitable for this topic.To obtain a deep learning target detection model with a smaller model and a faster real-time detection speed to achieve the detection of nuclear waste images in a nuclear radiation environment.The main tasks are as follows:In the aspect of nuclear waste image preprocessing,firstly,the noise category of radioactive particles on the image under nuclear radiation environment was studied,and the corresponding filtering algorithm was used to perform noise removal experiments on the nuclear waste image,and the median filter algorithm with better effect was selected Filter the nuclear waste image;then use the histogram equalization algorithm to improve the contrast of the nuclear waste image,improve the quality of the nuclear waste image,and complete the nuclear waste image preprocessing.Finally,a nuclear waste image data set was produced.The collected nuclear waste image was first subjected to histogram equalization and median filter preprocessing.Then,the generated anti-network technology and image processing technology were applied to the pre-processed nuclear waste image data set.The expansion makes the nuclear waste image data set sufficient in the deep learning model training to ensure the training effect of the deep learning target detection model.In the aspect of nuclear waste image target detection,select the regression-based deep learning target detection model with faster detection speed for experiments,and comprehensively compare the detection speed,accuracy and model size of Retina Net,YOLO V3,Tiny-yolo v3,SSD and other models.Select the SSD target detection model with better detection speed and detection accuracy to improve the experiment.Considering that the SSD model is too large and the detection speed is not prominent,this paper improves the SSD model based on the lightweight Mobile Net V2 model with fast speed,high classification accuracy,and small model size.The Mobile Net V2 network structure is used as the basic network part of the SSD model.The nuclear waste image information learned by the Mobile Net V2 network structure is sent to the target detection part of the SSD model to perform target detection on the nuclear waste image.Finally,a mobile computer platform is built for experimental verification,and a test system for nuclear waste image detection is designed.The experimental results on the nuclear waste data set show that the model size is compressed to 1/5 of the original while the model detection accuracy is guaranteed,the video detection speed is increased from 14.3fps to 34.4fps,the m AP reaches 89.8%,and the lightweight SSD The real-time detection speed of the model is fast,the precision is high,and the size of the model is small,which can meet the nuclear waste detection requirements of the mobile end of the nuclear emergency robot.The nuclear waste image target detection model based on deep learning can assist the nuclear emergency robot to detect nuclear waste in the nuclear radiation environment in real time and accurately on the mobile terminal,reduce the work pressure of the rescuer in the nuclear radiation environment,and reduce the radiation radiation environment to the body of the rescuer Impact,improve the efficiency of rescuers,and reduce safety costs,this method has good research significance and has certain engineering application value.