Improving the Accuracy of One-Shot Detectors for Small Objects in X-ray Images

In this paper, we address the problem of detecting small objects on high-quality X-ray imagesusing deep neural networks. We propose to implement the two-stage approach, in which, firstly, input image issplit into partially overlapping blocks to make small objects more discriminative for detection. Secondly, the small blocks are fed into conventional single-shot detectors. These detectors are trained using the blocks of the training images extracted by the same procedure.Two datasets of X-ray images from the customs inspection complex are examined in the experimental study. It was shown thatthe proposed algorithm with data augmentationleads tomore precise results when compared to the conventional technique:ourmethod outperforms the traditional approach by 5.4 - 25.7% depending on the type of used backbone convolutional neural network.