Research On The Image Detection Algorithm Of X-ray Security Inspection Based On Deep Learning

With the continuous development of science and technology,in the face of increasing security threats,people have realized the importance of security inspection,and X-ray security inspection technology has also received widespread attention.However,in the actual scene,due to the diversity of item types,the particularity of imaging angle and the limitations of detection algorithm,the detection accuracy is not high.In order to solve the related problems,this paper proposes the research of X-ray security image detection algorithm based on deep learning.The main work and research contents are as follows:(1)For the object detection model based on anchor free,this paper selects Faster R-CNN as the benchmark model to improve it.First,in order to expand the receptive field of the feature map and effectively extract the regional features of the object with shape distortion in the feature map,this paper proposes the Malformed Attention Module(MAM).Secondly,the Swin Transformer block is used to connect the corresponding backbone output feature layer to strengthen the exchange of output feature map information,which can refine the semantic information of the backbone network output feature layer.Finally,the MAM Faster R-CNN model is formed.In this paper,the MAM Faster R-CNN model is evaluated on the Hi Xray and OPIXray datasets,and the results show that the model has superior performance in detection accuracy.At the same time,ablation experiments are carried out on the Hi Xray dataset.The results indicate that the MAM Faster R-CNN model outperforms the baseline model Faser R-CNN in terms of detection accuracy.(2)For the object detection model not based on anchor free,this paper selects VFNet as the baseline model to improve it.First,the Large Kernel Attention(LKA)Module is connected to the corresponding backbone output feature layer,and the adaptive feature selection of the self-attention module is used to better focus on the effective feature information in the feature map.Secondly,for the Neck part,this paper replaces the feature pyramid network(FPN)with the path aggregation network(PAN),and adds the Conv-MLP block in the bottom-up feature fusion part on the right side of the PAN network to reduce the loss of some underlying details.Finally,the HA VFNet model is formed.In this paper,the HA VFNet model is evaluated on the Hi Xray and OPIXray datasets.The results show that the HA VFNet model not only improves the accuracy,but also reduces the number of parameters and computation compared with MAM Faster R-CNN.At the same time,ablation experiments are carried out on the Hi XRay dataset,and the results show that the HA VFNet model has excellent detection performance compared to the baseline model(VFNet).