Aiming at the problem of insufficient labeled data in the medical field, a Named Entity Recognition model that introduces counterfactual mechanism to enhance vocabulary is proposed in this article. With the idea of Semi-Supervised Learning, the model aims at a small amount of labeled data, builds a counterfactual vocabulary generator that captures more dependencies to enhance medical data by introducing and improving the counterfactual mechanism in the structural causal model. Further more, a vocabulary information fusion recognizer is constructed to verify the effectiveness of the data. The recognizer integrates character feature embedding, vocabulary information feature embedding in training data, and position feature embedding. While achieving medical vocabulary enhancement, it also solves the problem of inaccurate entity recognition and improves the accuracy of entity recognition. Through comparative and ablation experiments, it is shown that the named entity recognition model with counterfactual mechanism in this paper achieved an F1 score of 84.67% and 86.15% on the CCKS2019 and CCKS2020 datasets, respectively, which were 0.22%–4.57% and 0.315–5.04% higher than other related models, and 3.82% and 3.86% higher than traditional counterfactual generators, respectively, proving the effectiveness of the model.

While iris recognition boasts a notably high recognition rate, unimodal biometrics suffer from vulnerabilities such as environmental factors and spoofing attacks. Particularly in remote or mobile settings with minimal constraints, the efficacy of iris recognition diminishes significantly. To address this challenge, leveraging periocular biometric features, situated proximately to the iris and known for their high discriminative power, presents a promising solution. By integrating iris and periocular features through bimodal fusion recognition, this issue can be effectively tackled. This paper introduces a novel approach for accurate and adaptive fusion recognition termed MultipleFusionNet, which incorporates iris and periocular depth features. Embracing the concept of channel attention and channel grouping attention, an automatic weight generation network is devised to dynamically assign weights to iris and periocular features through network learning. These weighted features, derived from convolutional neural networks, enable precise fusion and subsequently enhance recognition accuracy. The fusion module of this network model serves as a versatile deep feature fusion component, seamlessly integrable into any CNN backbone network, and characterized by its lightweight and straightforward implementation. Experimental validation is conducted using the CASIA–Iris–Distance dataset for remote scenarios and the CASIA–Iris–Lamp dataset for close-range illumination variations, both provided by the Chinese Academy of Sciences. Results from diverse comparative and distance measurement experiments demonstrate the superiority of the proposed fusion model. Achieving a peak accuracy of 99.56% and a minimal Equal Error Rate (EER) of 0.0027, measured by cosine distance, the model outperforms unimodal approaches and other related feature fusion methods. Furthermore, in terms of computational complexity, the proposed model exhibits efficiency, with a parameter count and computational workload approximately 1.5% lower than twice that of unimodal methods, and only 1% higher than standard fusion methods. These findings underscore the model's favorable balance of computational efficiency and performance prowess.

In the medical field, entity recognition can extract valuable information from the text of large-scale electronic medical records. Due to the lack of features for locating entity boundaries and incomplete semantic information extraction, the implementation of Chinese Named Entity Recognition(NER) is more difficult. In this paper, a model combining Multi-Feature Embedding and Multi-Net-work Fusion model (MFE-MNF) is proposed. The model embeds multi-granularity features, i.e. characters, words, radicals and external knowledge, extends the feature representation of characters and defines the entity boundary. The feature vectors are input respectively into the two paths of Bi-directional Long Short-Term Memory (BiLSTM) and adaptive graph convolution network to capture comprehensively and deeply the context semantic information and global semantic information, and alleviate the problem of incomplete semantic information extraction. The experimental results on CCKS2019 and CCKS2020 datasets show that compared with the traditional entity recognition model, the proposed model can extract entities accurately and effectively.

In order to tackle limitations of current iris segmentation methods based on deep learning, such as an enormous amount of parameters, intensive computation and excessive storage space, a lightweight and efficient iris segmentation network is proposed in this article. Based on the classical semantic segmentation network U-net, the proposed approach designs a dual-path fusion network model to integrate deep semantic information and rich shallow context information at multiple levels. Our model uses the depth-wise separable convolution for feature extraction and introduces a novel attention mechanism, which strengthens the capability of extracting significant features as well as the segmentation capability of the network. Experiments on four public datasets reveal that the proposed approach can raise the MIoU and F1 scores by 15% and 9% on average compared with traditional methods, respectively, and 1.5% and 2.5% on average compared with the classical semantic segmentation method U-net and other relevant methods. Compared with the U-net, the proposed approach reduces about 80%, 90% and 99% in terms of computation, parameters and storage, respectively, and the average run time up to 0.02 s. Our approach not only exhibits a good performance, but also is simpler in terms of computation, parameters and storage compared with existing classical semantic segmentation methods.

Iris recognition is a very important biometric technology. Given sufficient labeled data, iris recognition algorithms combined with deep learning have achieved excellent performance. With limited training samples, however, over-fitting often occurs and affects recognition performance if deep learning methods are directly used for training. The learning problem with insufficient samples may be solved by using few-shot learning methods. In this paper, we propose an attention meta-transfer learning (AttentionMTL) approach for iris recognition through an improved attention network model. Experiments on the publicly available datasets show that AttentionMTL has achieved the highest accuracy of 99.95% and obtained higher accuracy (up to 6%) than conventional MTL method and other related approaches.

Abstract Traditional iris recognition has high recognition accuracy and low misrecognition rate. However, in the case of mobile terminal or distance, the image resolution and image quality decrease, and the recognition rate also decreases. To solve the above problems, this article is based on deep learning technology, on the basis of single mode state recognition, from different levels of multimodal integration, the iris and the eyes in the score level fusion recognition research, put forward the adaptive dynamic weighted score fusion method, to determine the weighing values can adaptive algorithm of the modal, without artificial specified, dynamic weighting algorithm more flexible, stronger applicability. Experimental results of casIA-Iris-LAMP and CasIA-Iris-Distance Iris database in Chinese Academy of Sciences show that the proposed fusion algorithm has higher recognition accuracy and better recognition performance than the single mode recognition algorithm and the traditional fractional fusion method, which proves the effectiveness of the algorithm.

Abstract Periocular recognition technology is a biometric recognition technology widely used in identity verification. Because of its high precision, high ease of use and high security, Periocular recognition has a broad application prospect and scientific research value. In order to solve the problem of angular rotation of eyes in practical application, this paper proposes a deep learning periocular recognition method based on multi-angle data augmentation. The method is to rotate the original data set from small angle to large angle, so that the amount of data is expanded to 7 times of the original, and the diversity of data is increased at the same time. The InceptionV3 network and MobileNetV2 lightweight network are used for experimental verification respectively, and good results are obtained from multi-angle tests, indicating that the proposed method can improve the generalization ability of the model and has good robustness.

数学思维和数学素养不仅体现在数学学科,也体现在计算机和其它学科中,因此培养学生的数学思维能力和数学核心素养正在越来越被重视.通过学校的教育教学活动,提升学生的数学核心素养,也是在学生迈入社会的时候,必须经历的事情.由此看出,学生的思维能力和数学核心素养尤其重要.本文主要讲述如何探讨逻辑思维能力培养提升学生数学核心素养,让学生适应当今社会的发展,具备终身学习的品格和能力.

Iris recognition is a popular research field in the biometrics, and it plays an important role in automatic recognition. Given sufficient training data, some deep learning‐based approaches have achieved good performance on iris recognition. However, when the training data are limited, overfitting may occur. To address this issue, in this paper, we proposed a few‐shot learning approach for iris recognition, based on model‐agnostic meta‐learning (MAML). To our best knowledge, we are the first to apply few‐shot learning for iris recognition. Our experiments on the benchmark datasets have demonstrated that the proposed approach can achieve higher performance than the original MAML, and it is competitive to deep learning‐based approaches.

为研究单机随机排序问题的近似,论文对WSEPT规则在期望值角度给以新的更深入的分析.该分析对开始期限及完工期限模型均适用,之后,通过由WSEPT的期望值与最优适应性策略排序的期望值的关系来修正上下界.从而给出WSEPT规则下的近似程度分析.

In order to detect UAV in real time, the paper choose to use a dynamic detection method based on two consecutive inter-frame differences method to extract the region of interest. The position of the target appeared on the image was obtained by the method of two consecutive inter-frame difference, and the UAV was detected by the trained SVM classifier. UAV could be detected quickly and accurately in complex background and in different position and angle circumstances. Compared to the traditional HOG + SVM sliding window detection method, the experimental results show that the detecting speed with the methods is obviously improved when the recognition accuracy is invariable.

Aiming at the low accuracy of traditional brain tumor detection, a three-dimensional brain tumor detection method based on deep learning was proposed. Firstly, the magnetic resonance images of different modal brain tumors were fused to obtain the three-dimensional features of brain tumor focus under different modalities. Then, an instance normalization layer was added between the convolution layer and the pooling layer to improve the convergence speed of the network and relieve the problem of overfitting. And the loss function was improved, the weighted loss function was used to enhance the feature learning of the focus area. Finally, the problem of more focuses in the false positive brain tumor was solved combining with the post-processing method. The experimental results show that the proposed brain tumor detection method can effectively detect the tumor focuses. The Dice coefficient, sensitivity and specificity of the three evaluation indexes reach 0.926 7, 0.928 1 and 0.997 7 respectively. The three indicators improve 4.6%, 3.96% and 0.04% compared with the 2D detection network, and improve 13.2%, 10.42% and 0.12% compared with the initial single modal brain.

用线性规划的优先策略寻求两类NP难的随机排序问题的近似解法,并分别给出了其近似度.

UAVs (unmanned aerial vehicles) have been widely used in many fields, where they need to be detected and controlled. Small-sample UAV recognition requires an effective detecting and recognition method. When identifying a UAV target using the backward propagation (BP) neural network, fully connected neurons of BP neural network and the high-dimensional input features will generate too many weights for training, induce complex network structure, and poor recognition performance. In this paper, a novel recognition method based on non-negative matrix factorization (NMF) with sparseness constraint feature dimension reduction and BP neural network is proposed for the above difficulties. The Edgeboxes are used for candidate regions and Log-Gabor features are extracted in candidate target regions. In order to avoid the complexity of the matrix operation with the high-dimensional Log-Gabor features, preprocessing for feature reduction by downsampling is adopted, which makes the NMF fast and the feature discriminative. The classifier is trained by neural network with the feature of dimension reduction. The experimental results show that the method is better than the traditional methods of dimension reduction, such as PCA (principal component analysis), FLD (Fisher linear discrimination), LPP (locality preserving projection), and KLPP (kernel locality preserving projection), and can identify the UAV target quickly and accurately.

为了优化环境污染监测,需要对环境污染目标区域进行有效识别,提出基于深度学习和无人机技术的环境污染目标区域识别方法,采用无人机机载空间扫描方法进行图像采集,对采集的环境图像进行污染区域的边缘轮廓检测和图像分割处理,采用深度学习算法对环境污染区域的图像进行自适应分块标记和识别,实现对环境污染区域的图像检测和三维区域识别.采用该方法进行环境污染区域的目标识别准确性较好,对环境污染区域分块识别的精度较高,较强.

Abstract In order to detect non-cooperative target UAV quickly and accurately, a novel method of UAV detection method based on graph theory and HOG-FLD feature fusion is presented in this paper. In order to avoid the time-consuming full search, the candidate areas of the UAV are obtained through the selective search of the image segmentation and the similarity, and the features are extracted through the method of gradient orientation histogram fusion FLD linear to train the SVM classifier with generalization ability to identify the UAV. The method can detect the UAV quickly and accurately under complicated background and circumstances of various position and angle. Compared with the sliding window method based on image segmentation and HOG+SVM, the experimental results show that the speed of this method has been obviously improved with the same recognition accuracy.

为提高场景不断变化下车牌定位的准确性,提出应用深度学习定位的方法.基于最小生成树原理将图像转换成图论中的图,结合相似度进行选择搜索获取车牌候选区域.利用车牌长宽比条件对候选区域进行筛选,以减少数据的不平衡性.通过带平滑项的交叉熵损失函数以及Bootstrapping技术训练深度网络,并用该模型从候选区域中识别出车牌区域.实验表明,该方法在候选区域提取、网络训练阶段以及定位准确率方面均优于其他算法,测试集中准确率为97.6％.

A novel pedestrian detection method based on maximum entropy threshold and random forest classifier is presented in order to detect effectively pedestrian in infrared images.The entropy of target and background is calculated respectively, and the sum of entropy of target and background is maximized to obtain the maximum information between target and background distributions in the image.According to the rule of maximum entropy, the segmentation threshold value is selected to get interest regions.The morphological and experiential methods are used to further eliminate noise effect.The interest region is transformed to eigenvector by wavelet transform.Random forest classifier is applied to distinguish pedestrians from non-pedestrian candidates.Experimental results in different infrared images show that the method presented has high correct detection rate and low false detection rate with fast calculation speed.

The performance of ear recognition is influenced by pose variation. For the similar position of ear and profile face, a multimodal recognition method is proposed based on the feature fusion of ear and profile face information. A model for ear and profile face feature fusion and recognition is built. The Log-Gabor features of ear and profile face are first extracted separately, and two features are integrated into a combined feature after two Log-Gabor features are standardized. Then combined feature is mapped to kernel space to fuse further, and acquired stronger discriminant feature for classification by kernel Fisher discriminant analysis (KFDA). The minimum distance classifier is finally used in recognition. Experimental results on the profile face database of Notre Dame University show that the fused method improves the recognition rate of pose variation, and the performance of multimodal recognition is better than unimodal recognition using either ear or profile face alone. The method of ear and profile face feature fusion and recognition is effective and robust for the pose variation.