| X-ray image security inspection for contraband is a critical technology to protect personal and property security,which is widely used in many public places including airports,stations,customs and so on.Recently,with the rapid development of deep learning in the field of image processing,the intelligent detection of X-ray contraband images has made enormous progress.However,when applied to practical scenes,there are many challenges such as complex background noise interference,target overlapping occlusion,and various types of objects to face.Based on the convolutional neural network,many optimization strategies such as attention mechanism,reverse bottleneck layer design and color enhancement have been employed to improve the X-ray contraband detection methods.The proposed methods have been trained and tested on the public datasets,indicating their excellent competitiveness compared with other advanced methods.In addition,the practice of visual application demonstration interface design has also been carried out.The main contributions are as follows:(1)To resolve the problems of high overlap,heavy occlusion and complex background interference in the X-ray luggage image security detection,an improved YOLOv5 network model has been proposed by introducing the weighted boxes fusion algorithm and Convolutional Block Attention Module for X-ray prohibited items detection.The convolutional block attention module has been introduced in the Neck to enhance the extraction of deep important features and suppress background interference of X-ray prohibited items features.The Mixup data augmentation strategy in the training process has been employed to simulate the detection scene with high overlap and heavy occlusion items to strengthen the model’s learning ability for complex samples.In the testing process,the weighted boxes fusion algorithm is used to optimize the redundant prediction boxes to enhance its prediction accuracy ability.The proposed model has been tested on the datasets SIXray,Hi Xray and OPIXray,the results show that the proposed model has high accuracy and robustness.(2)To tackle the problems of the miss and false detection due to position and angle changing,a model improved by inverted bottleneck and light convolution block attention module for the X-ray prohibited items detection has been proposed.The inverted bottleneck design has been introduced in the backbone to emphasize the detailed features and improve the model to cope with the large-angle change problem.The light convolution block attention module has been designed to suppress background interference and reduce model parameter.The Gaussian Error Linear Unit activation function and improved loss function are used to enhance the nonlinear expression ability.The proposed model is trained and tested on the public datasets.The results show that the proposed method can effectively solve the problem of angel change in X-ray luggage,improving the learning ability of difficult samples.(3)To resolve the problems of monotonous color and the difficult detection of the small objects.An X-ray prohibited items detection network has been proposed by introducing the color enhancement algorithm,improved feature pyramid structure,and re-parameterized convolution layers.The proposed method enhances the image color and augments the dataset to improve the training effect.The improvement of feature pyramid structure is used to enhance the recognition performance of small-sized objects.The structural re-parameterized convolution is employed to enhance the feature sensitiveness and coding/decoding ability.In addition,the weighted coefficient of classification loss function is optimized and the punishment of classification task is increased,enhancing the performance of network classification.The proposed model is trained and tested on three public datasets,including CLCXray.The results show the proposed method can effectively cope with the difficult detection including small objects and monotone color objects like liquid containers. |
PDF Full Text Request