Inspecting Contraband Hidden In Milk Powder Using Deep Convolutional Neural Networks

Smugglers often hide contraband in milk cans, so the task of checking the cans is very heavy for the customs. This makes the automated detection of milk cans become very important. Because inconspicuous entrainment objects are relatively small and traditional methods are insensitive to small entrainment, it is difficult to detect the entrainment of milk powder. In this paper, we propose a method using convolutional neural networks (CNNs) for detecting contraband goods hidden in milk powder on customs X-ray images. This method can segment the original image and mark the local area. Specifically, we first divide the images into blocks and rotate them at many angles. Then we mark whether these blocks contain entrained objects. This not only increases the number of samples but improves the adaptability of the CNNs model. Finally, we modified the CaffeNet for this purpose. In experiments, the proposed method detection accuracy reaches 83.3% and the recall is 100% with seconds' computation time, which show a great improvement over other methods based on low level image processing techniques. Excellent results show that this method can greatly help the customs officers to check the milk powder.