Transformer architecture based on mutual attention for image-anomaly detection

Image-anomaly detection, which is widely used in industrial fields. Previous studies that attempted to address this problem often trained convolutional neural network-based models(e.g., autoencoders and generative adversarial networks) to reconstruct covered parts of input images and calculate the difference between the input and reconstructed images. However, convolutional operations are effective at extracting local features, making it difficult to identify larger image anomalies. Method To this end, we propose a transformer architecture based on mutual attention for image-anomaly separation. This architecture can capture long-term dependencies and fuse local and global features to facilitate better image-anomaly detection. Result Our method was extensively evaluated on several benchmarks, and experimental results showed that it improved the detection capability by 3.1% and localization capability by 1.0% compared with state-of-the-art reconstruction-based methods.

Budget analysis of mesoscale available potential energy in a heavy rainfall event over the eastern Tibetan Plateau

Using model simulated data,the distribution characteristics,genesis,and impacts on precipitation of available potential energy(APE)are analyzed for a heavy rainfall event that took place over the eastern Tibetan Plateau during 10–11 July 2018.Results show that APE was mainly distributed below 4 km and within 8–14 km.The APE distribution in the upper level had a better correspondence with precipitation.Northwestern cold advection and evaporation of falling raindrops were primary factors leading to positive anomalies of APE in the lower level,while positive anomalies of APE in the upper level were caused by a combination of thermal disturbances driven by latent heat and potential temperature perturbations resulting from the orography of the Tibetan Plateau.Budget analysis of APE indicated that APE fluxes and conversion between APE and kinetic energy(KE)were the main source and sink terms.Meridional fluxes of APE and conversion of KE to APE fed the dissipation of APE in the lower level.Vertical motion enhanced by conversion of APE to KE in the upper level was the major factor that promoted precipitation evolution.A positive feedback between APE and vertical motion in the upper level generated a powerful correlation between them.Conversion of KE to APE lasted longer in the lower level,which weakened vertical motion;whereas,northwestern cold advection brought an enhanced trend to the APE,resulting in a weak correlation between APE and vertical motion.

Image copy-move forgery passive detection based on improved PCNN and self-selected sub-images

Image forgery detection remains a challenging problem.For the most common copy-move forgery detection,the robustness and accuracy of existing methods can still be further improved.To the best of our knowledge,we are the first to propose an image copy-move forgery passive detection method by combining the improved pulse coupled neural network(PCNN)and the self-selected sub-images.Our method has the following steps:First,contour detection is performed on the input color image,and bounding boxes are drawn to frame the contours to form suspected forgery sub-images.Second,by improving PCNN to perform feature extraction of sub-images,the feature invariance of rotation,scaling,noise adding,and so on can be achieved.Finally,the dual feature matching is used to match the features and locate the forgery regions.What’s more,the self-selected sub-images can quickly obtain suspected forgery sub-images and lessen the workload of feature extraction,and the improved PCNN can extract image features with high robustness.Through experiments on the standard image forgery datasets CoMoFoD and CASIA,it is effectively verified that the robustness score and accuracy of proposed method are much higher than the current best method,which is a more efficient image copy-move forgery passive detection method.

A privacy-enhancing scheme against contextual knowledge-based attacks in location-based services

1 Introduction and main contributions Location-based services are springing up around us,while leakages of users'privacy are inevitable during services.Even worse,adversaries may analyze intercepted service data,and extract more privacy like health and property.Therefore,privacy preservation is an indispensable guarantee on LBS security.Among the previous approaches to privacy preservation,k-anonymity-based ones have drawn much research attention[1-3].However,some privacy concern will be aroused if these schemes are adopted directly.For instance,Ut issues a query"Find the nearest hotel around me"in such an area as Fig.1(privacy profile k=4).DLS algorithm[2]constructs anonymity set A because these four cells have similar probabilities of being queried in the past.However,experienced adversaries can exclude some cells if they have learned rich contextual knowledge(side information)from historical data,such as features of each cell and LBS users.