To address the problems of missing detection and low detection accuracy of the existing helmet wearing detection algorithms for small and crowded targets detection, this paper proposes a helmet wearing detection method based on improved YOLOv5 and transfer learning. First, different from the default priori frame that is not suitable for the task, we use the K-means algorithm to cluster the suitable priori frame size for the detection task. Then, in the back of the feature extraction network, we introduce a spatial channel mixed attention module to strengthen the learning of relevant weights and suppress the weights of irrelevant backgrounds, respectively. Further, we improve the judgment metric of the non-maximum-suppression (NMS) algorithm in the post-processing stage of YOLOv5 to reduce the phenomenon of false deletion and missing of prediction boxes. After that, the proposed network is trained based on the strategy of transfer learning, which can overcome the scarcity of limited existing data sets and improve the generalization ability of the model. Finally, we build a cascade judgment framework for helmet wearing deployed in visual sensor networks. The experimental results show that our proposed method improves the average accuracy (IOU=0.5) to 93.6%, which is 5% higher than the original model in the helmet wearing data set. The proposed model also outperforms other state-of-the-art algorithms by obviously improving the accuracy of helmet wearing detection in the construction scenarios.

The medical named entity recognition aims to automatically identify and classify medical entities in electronic medical records, which plays a very important role in downstream tasks such as information retrieval and knowledge graph. Existing methods usually ignore the dependencies between entities. To address this, this paper proposes a gated attention unit-based model for Chinese medical named entity recognition. First, the proposed model uses the pre-training model MC-BERT to capture contextual information. Then, it uses the cross-attention and gated attention unit to enhance the interaction between entity query and contextual semantics, and further extract the dependency and correlation between entities. Finally, the proposed model uses the matching algorithm of bipartite graph to calculate the loss. This paper conducted experiments on three datasets, including the CMeEE, CMQNN, and MSRA. The experimental results show that the F₁ values of the proposed model on the three datasets are 70.74%, 96.92%, and 95.53%, respectively, which outperforms other related models, demonstrating the effectiveness of the proposed model in Chinese medical named entity recognition task.

In intrusion detection, while certain unsupervised models can detect unknown attack types using thresholds, they usually fall to effectively utilize the identified traffic patterns to capture the similarities and differences across traffic flows. As a result, unknown attack traffics are frequently misclassified as normal traffic. To address this issue, this paper proposes a Multi-scale Contrastive Embedding Enhanced Adaptive Intrusion Detection Model (MCE-IDM). It employs hierarchical contrastive learning to integrate known attack types with their associated data features, generating embeddings that are subsequently combined with the original data to train an unsupervised model.Furthermore, a lightweight gradient boosting machine is used for feature selection, which significantly reduces the time complexity during this phase compared to previous models. Experimental results on multiple datasets demonstrate that the proposed model not only exhibits stable performance but also improves the Matthews Correlation Coefficient (MCC) by 15.78 percentage points over baseline models on a highly imbalanced subset of data.The proposed method also consistently achieves competitive results across other subsets.

The use of Web security scanners for scanning and probing serves as a primary way of cyber reconnaissance in the early stages of Web attacks. Existing Web intrusion detection methods primarily focus on addressing the binary classification problem between known attack flows and normal flows, which have issues such as being easily bypassed and an inability to detect unknown attacks. To address these, this paper proposes a method for detecting Web intrusion attacks from the perspective of Web security scanner scanning. Firstly, this paper collects scanning data from various Web security scanners through experiments, and analyzes the similarity of the scanning data. Secondly, a Web intrusion detection scheme based on scanning data is proposed, which includes a feature extraction method and a convolutional gated recurrent network model. Finally, comparative experiments are conducted on the collected dataset, and the results show that the Web intrusion detection scheme proposed in this paper achieves superior attack detection performance. Specifically, it achieves a detection precision of 99.87% and an F₁-score of 98.99% for known Web security scanners, and a detection precision of 92.98% and an F₁-score of 95.71% for unknown Web security scanners.

Unsupervised domain adaptation aims to transfer knowledge from labeled data in the source domain to the unlabeled target domain. The unsupervised domain adaptation method based on domain alignment learns domain invariant features by minimizing the difference in cross domain feature distribution, ignoring the learning of discriminative features in the target domain, resulting in poor discriminability of the learned domain invariant features in the target domain, confusion of different types of features, and performance decrease of the model. To address the problem of poor discriminative power of domain invariant features in the target domain during domain alignment, this paper proposes a deep Subdomain Adaptive method based on domain invariant Discriminative Feature learning (DFSA). The deep subdomain adaptive method is combined with the constraints of minimizing class confusion to increase the inter-class differences of domain invariant features, and consistency regularization is added to reduce the intra-class differences of domain invariant features. By doing so, domain invariant discriminative features are mined to better preserve the discriminative power of domain invariant features in the target domain and improve domain adaptability. The effectiveness of DFSA was validated on the Office-31 and Office-Home benchmark datasets, and the results showed that DFSA effectively improves domain adaptation performance, with an average accuracy improvement of approximately 0.79 percentage points and 3.51 percentage points, respectively, when compared with existing deep subdomain adaptation methods on the two benchmark datasets.