Abstract Pancreas segmentation has great significance in computer‐aided diagnosis of pancreatic diseases. The small size of the pancreas, high variability in shape, and blurred edges make the task of pancreas segmentation challenging. A new model called SEY‐Net is proposed to solve the above problems, which is a one‐stage model with multi‐inputs. SEY‐Net is composed of three main components. Firstly, the edge information extraction (EIE) module is designed to improve the segmentation accuracy of the pancreas boundary. Then, the SE_ResNet50 is selected as the encoder's backbone to fit the size of the pancreas. Finally, the dual cross‐attention is integrated into the skip connection to better focus on the variable shape of the pancreas. The experimental results shows that the proposed method has better performance and outperforms the other existing state‐of‐the‐art pancreas segmentation methods.

Abstract In current era, the intelligent development of traditional Chinese medicine (TCM) has attracted more and more attention. As the main carrier of clinical medication, formulas use synergies of active substances to enhance efficacy and reduce side effects. Related studies show that there is a nonlinear relationship between the efficacy of formulas and herbs. Deep learning is an effective technique for fitting nonlinear relationships. However, it is not good for using deep learning model directly due to ignoring the characteristics of formulas. In this paper, we propose a detached feature extraction approach (TCM2Vec) based on deep learning for better feature extraction and efficacy prediction. We build two detached encoders, one of it uses cross-feature-based unsupervised pre-training model (FMh2v) to extract the relationship features of herbal medicines for initializing, while the other one simulates multi-dimensional characteristics of medicines by normal distribution. Then we integrate relationships and medicinal characteristics for deep feature extraction. We processed 31,114 unlabeled formulas for pre-training and two classification tasks in-domain for predicting and fine-tuning. One of tasks is multi-classed with 1036 formulas, other one is multi-labelled with 1,723 formulas. For labelled formulas, different feature extraction models based on detached encoder are trained to predict efficacy. Compared with the no pre-training, CBOW and BERT baseline models, FMh2v leads to performance gains. Moreover, the detached encoder offers large positive effects in different models which for efficacy prediction, where ACC increased by 5.80% on average and F1 increased by 12.06% on average. Overall, the proposed feature extraction is an effective method for obtaining characteristic representation of TCM formulas, and provides reference for the adaptability of artificial intelligence technology in the domain of TCM.

In recent years, fatigue driving has been a serious threat to the traffic safety, which makes the research of fatigue detection a hotspot field. Research on fatigue recognition has a great significance to improve the traffic safety. However, the existing fatigue detection methods still have room for improvement in detection accuracy and efficiency. In order to detect whether the driver has fatigue driving, this paper proposes a fatigue state recognition algorithm. The method first uses MTCNN (multitask convolutional neural network) to detect human face, and then DLIB (an open-source software library) is used to locate facial key points to extract the fatigue feature vector of each frame. The fatigue feature vectors of multiple frames are spliced into a temporal feature sequence and sent to the LSTM (long short-term memory) network to obtain a final fatigue feature value. Experiments show that compared with other methods, the fatigue state recognition algorithm proposed in this paper has achieved better results in accuracy. The average accuracy of the proposed method in detecting key points of the face is as high as 93%, and the running time is less than half of the ordinary DLIB method.

With the large-scale application of cloud storage, how to ensure cloud data integrity has become an important issue. Although many methods have been proposed, they still have their limitations. This paper improves some defects of the previous methods and proposes an efficient cloud data integrity verification scheme based on blockchain. In this paper, we proposed a lattice signature algorithm to resist quantum computing and introduced cuckoo filter to simplify the computational overhead of the user verification phase. Finally, the decentralized blockchain network is introduced to replace traditional centralized audit to publicize and authenticate the verification results, which improves the transparency and the security of this scheme. Security analysis shows that our scheme can resist malicious attacks and experimental results show that our scheme has high efficiency, especially in the user verification phase.

In recent years, the wide applications of the wireless sensor networks have achieved great success. However, the security is a critical issue in many scenarios ranging from covert military operations to the organization of the social unrest. Because the traditional encrypting methods are easy to arouse suspicion, an adaptive audio steganography method is proposed. The method is based on interval and variable low bit coding, which can be applied to covert wireless communication. The interval for embedding secret messages into the audio file and the threshold in variable low bit coding are used for selecting the embedding location and embedding bits adaptively; thus the embedding capacity and the embedding rate are variable. Experimental results demonstrate that the proposed method has better performance in embedding rate and invisibility than other audio steganography methods.