完美论文网提供代写毕业论文、代写本科论文、代写硕士论文、论文发表及查重服务!QQ:505676555

电话:18796337551
当前位置:完美论文网信息系统 → 文章正文

X光安检中危险品的多视角检测研究

作者:完美论文网  来源:www.wmlunwen.com  发布时间:2019/10/10 8:58:27  

摘要:交通运输安检主要通过安检人员根据安检扫描仪获取的X光图像判断行李中是否存在危险品。但如果危险品被其它物品遮挡或行李内部混乱,则人眼很难分辨出来,检测精度会有所降低,从而对公共安全造成威胁。论文研究的多视角自动检测方法可以有效地提高检测的精度和速度,大大减少安检人员的工作量,对维护公共安全具有十分重要的意义。

本课题以危险品单视角X光检测为基础,针对两种多视角检测算法检测精度低、检测速度慢以及平均耗时高等缺点,通过大量对比实验验证改进的两种多视角检测方法的优越性。本课题的主要研究工作如下:

第一,基于Q学习算法的X光多视角主动视觉安检方法。针对该方法检测精度低和检测速度慢等缺点,基于Q学习算法提出采用状态回溯的启发式Q学习算法估计最佳视角,引入代价函数和启发函数,提高学习效率,加快Q学习的收敛速度。通过对比实验可以得出改进的主动视觉安检方法能大大提高检测精度和速度,相比于以Q-Learning为基础的主动视觉方法,检测剃刀刀片所得精确率和召回率之间的加权平均值 值提高了2.51%,速度提高了17.39%,检测手枪所得的 值提高了9.60%,速度提高了12.45%。

第二,基于匹配和跟踪算法的多视角自动检测方法。针对该方法计算量大、检测精度不高等缺点,在结构估计部分做出改进,首先利用主成分分析方法对SIFT特征降维,保留关键特征,有效降低计算量,提高匹配效率;然后采用基于采样优化的随机抽样一致算法消除误匹配点,提高特征匹配的精度。通过对比实验可以得出,基于匹配和跟踪算法的多视角自动检测方法可以提高检测精度并且平均耗时较少,相比于改进之前的方法,检测剃刀刀片所得的值提高了2.69%,平均耗时缩短了20.21s,检测手枪所得的值提高了3.10%,平均耗时缩短了21.87s。

Traffic security inspection mainlydetermines whether threat objects exit in luggage by X-ray images obtained bysecurity scanner. However, if threat objects are covered by other objects orluggage is confused, it will be difficult for the human eye to distinguish, andthe detection accuracy will be reduced, and even threatening to public safety.Multi-view automatic detection methods studied in this paper can effectivelyimprove the accuracy and speed of detection, and it will be greatly reduce theworkload of security inspectors, and is very important to maintain publicsafety.

This topic is based on the single-viewX-ray detection of threat objects. In order to overcome the shortcomings of lowdetection accuracy, slow detection speed and high average time-consuming of thetwo multi-view detection algorithms, a large number of comparative experimentswere conducted to verify the advantages of the two improved multi-viewdetection methods. The main research work of this topic is as follows:

First, X-ray multi-view active visionsecurity inspection method based on Q-learning algorithm. In order to solve theproblems of the poor detection accuracy and slow detection speed of thismethod, a heuristically accelerated state backtracking Q-Learning algorithmbased on the Q-Learning algorithm is proposed to estimate the next best view.Cost function and heuristic function are introduced to improve the learningefficiency and speed up the convergence rate of Q-learning. Through thecontrast experiments, it can be concluded that the improved active visionsecurity inspection method can greatly improve the detection accuracy andspeed. Compared with the active vision approach based on Q-Learning, theweighted average value of   between theprecision and recall of detecting razor blades is increased by 2.51% and thedetection speed is increased by 17.39%, while the   of detecting handguns is increased by 9.60%,and the detection speed is increased by 12.45%.

Second, multi-view automatic detectionmethod based on matching and tracking algorithm. In order to overcome theshortcomings of large computation and low detection accuracy of this method,the structure estimation part is improved. Firstly, the principal componentanalysis method is used to reduce the dimension of SIFT features, retain keyfeatures, effectively reduce computation and improve matching efficiency.Secondly, the random sampling consensus algorithm based on samplingoptimization is used to eliminate mismatching points and improve the accuracyof feature matching. By comparing experiments, it can be concluded that themulti-view automatic detection method based on matching and tracking algorithmcan improve the detection accuracy and the average time-consuming is less.Compared with the method before improvement, the   of detecting razor blades is increased by2.69%, the average time-consuming is shortened by 20.21s, while the   of detecting handguns is increased by 3.10%,and the average time-consuming is shortened by 21.87s.

关键词:X光安检;多视角检测;Q学习;主动视觉;匹配和跟踪

X-ray security inspection; multiple viewdetection; Q-Learning; active vision; matching and tracking

联系方式

客服QQ 505676555
客服热线18796337551
网站地址 www.wmlunwen.com
郑重承诺 原创,包修改,包通过!
完美论文网真诚欢迎新老客户的光临与惠顾!