Research On Special Object Recognition Of X Ray Image Based On Neural Network

In recent years,image recognition and target detection methods have become one of the significant directions in the field of machine learning research,traditional method of target recognition and new target detection method appearing at present are very different in recognition accuracy and speed.This project is from an actual project of logistics enterprise in Erdos,Inner Mongolia.The project requires automatic detect whether specified special goods exists in the image of item after passing x ray security inspection machine,such as common mobile hard drives and knives and similar item,and it has high recognition rate and accuracy.Based on this problem,this paper adopts BP Neural Network image recognition algorithm,target detection of Fast Region Convolutional Neural Network,combination of pixel level segmentation and Mask Region Convolutional Neural Network of target detection.The neural network using standard gradient descent algorithm has slow convergence rate and it is easy to fall into local minimum values,this paper proposes adaptive momentum method to train network.Pre-processing and extract the feature of sample,and then training optimized BP neural network.Experimental results show that optimized BP neural network improves convergence speed of network and reduces overall error effectively.In order to achieve automatic extraction of target object features,further improve the accuracy and speed of target recognition and detection,this topic introduces regionally selected convolutional neural network to identify and detect x ray images of special objects.Target detection of region convolutional neural network extracts features through convolutional layer inside the network,it no need to manually design and extract features in advance,compared with traditional recognition algorithm,its detection accuracy and training speed has leap improvement.Finally,this paper combines pixel-level segmentation method and region-convolutional neural network to detect targets in x ray image,it fundamentally solved the problem of positional deviation of bounding box of region-convolutional neural network target detection,and it have richer expression on target position and target outline.By implementing the above identified recognition and detection algorithm,compare and analysis the experimental results,it improve the recognition accuracy and speed effectively.