Research On Dangerous Goods Safety Inspection Detection Algorithm Based On YOLOv4

X-ray machines for security check are widely used in public transportation,logistics,customs and major security occasions to check contraband and restricted goods in luggage and parcels.Security staff need within a limited time to examine what is mixed and disorderly,dense and forms of luggage,while mature security technology improved the overall quality of security,but in the process of security image processing or rely on security operators for artificial screen calibration,it is not able to do auxiliary security personnel for contraband automatic marking system.Therefore,it is of great significance to realize the automatic identification of dangerous goods in security check.In this paper,the deep learning target detection algorithm is applied to the detection of dangerous goods in security inspection images.The convolutional neural network is used to identify and classify the detection targets,so as to realize the real-time detection of dangerous goods in security inspection.The YOLOV4 model was selected as the basic model for research and improvement.Based on the real-time performance of detection,the emphasis was placed on improving the accuracy of detection.The feature extraction ability of the target algorithm is improved by increasing the feature Receptive Field(Receptive Field Block,RFB).The void convolution captures the position of small targets in the global Field of view and reduces the rate of missed detection of small targets.Through the cascade connection between different levels,the deep features and the shallow features are further combined and integrated to fully learn the background features and contraband features,so as to accelerate the convergence of the model.The(Convolution Block Attention Module(CBAM)Attention mechanism is used to enhance the overall data.The features of each channel are weighted to learn the features of different channels.The original feature vectors are replaced by the filtered and weighted feature vectors for residual fusion.Reduced redundant features and effectively improved contraband detection accuracy under background interference.Finally,through comparative experiments,it is found that the Mean Average Precision(MAP)of the improved YOLOV4 model is 10% higher than that of the original model,in which the small target AP is significantly improved,and the lighter’s Average Precision(AP)is 30%higher than that of the original model.It is proved that the improved model is effective and has a certain application value.