Semi-supervised learning is a sound measure to relieve the strict demand of abundant annotated datasets, especially for challenging multi-organ segmentation . However, most existing SSL methods predict pixels in a single image independently, ignoring the relations among images and categories. In this paper, we propose a two-stage Dual Contrastive Learning Network for semi-supervised MoS, which utilizes global and local contrastive learning to strengthen the relations among images and classes. Concretely, in Stage 1, we develop a similarity-guided global contrastive learning to explore the implicit continuity and similarity among images and learn global context. Then, in Stage 2, we present an organ-aware local contrastive learning to further attract the class representations. To ease the computation burden, we introduce a mask center computation algorithm to compress the category representations for local contrastive learning. Experiments conducted on the public 2017 ACDC dataset and an in-house RC-OARs dataset has demonstrated the superior performance of our method.

Facial expression recognition (FER) has attracted intensive attention due to its critical role in various computer vision tasks. However, existing FER approaches suffer from either noisy annotations or expression ambiguity (high inter-class and low intra-class similarity), limiting the FER performance. To this end, we propose a robust end-to-end collaborative learning based transformer for FER (CL-TransFER) in this paper. Specifically, CL-TransFER co-trains a CNN feature extractor and a transformer feature extractor jointly to extract both rich local semantic features as well as global structural information from facial images. By enforcing the consensus between the predictions of two extractors, the CL-TransFER could suppress the influence of noisy annotations. To further tackle the expression ambiguity problem, we design a simple yet efficient self-supervised masked reconstruction (SSMR) task to pre-train the transformer feature extractor of CL-TransFER. This enhances the model's capability of learning fine-grained discriminative representations. Extensive experiments on three popular benchmarks have demonstrated the effectiveness and superiority of our method.

Cavities in a weld seriously affect the airtightness of the chip, which makes chip inspection a crucial step in intelligent manufacturing. In recent years, deep learning-based defect inspection models have shown significant advantages in reducing human errors. However, due to the scarcity of defective data, deep learning-based models are susceptible to overfitting. Moreover, the multiscale and uneven grayscale distribution of cavities further compound the challenges faced by these models. To address these issues, we develop a chip inspection system based on a multiscale subarea attention network (MSANet) for cavity defect detection. In the system, the segment anything model is embedded to interactively segment the weld. Furthermore, to circumvent the overfitting problem, a large-scale cavity dataset is built by splitting the segmented weld into multiple patches. Notably, a novel MSANet is proposed to precisely segment the varying cavities, and a source-to-destination Dijkstra algorithm is designed to assess the chip quality. The experimental results demonstrate that our chip inspection system achieves a 99.24% F1-score and 99.26% AUC.

Universal multi-source domain adaptation (UniMDA) aims to transfer the knowledge from multiple labeled source domains to an unlabeled target domain without constraints on the label space. Due to its inherent domain shift (different data distributions) and class shift (unknown target classes), UniMDA stands as an extremely challenging task. However, existing solutions mainly focus on excavating image features to detect unknown samples, ignoring the abundant information contained in the textual semantics. In this paper, we propose a Semantic-aware Adaptive Prompt Learning method based on Contrastive Language Image Pretraining (SAP-CLIP) for UniMDA classification tasks. Concretely, we utilize the CLIP with learnable prompts to leverage textual information of both class semantics and domain representations, thus helping the model detect unknown samples and tackle domain shifts. Besides, we propose a novel margin loss with a dynamic scoring function to enlarge the margin distance between known and unknown sample sets, facilitating a more precise classification. Experiment results on three benchmarks confirm the state-of-the-art performance of our method.

Radiotherapy is one of the leading treatments for cancer. To accelerate the implementation of radiotherapy in clinic, various deep learning-based methods have been developed for automatic dose prediction. However, the effectiveness of these methods heavily relies on the availability of a substantial amount of data with labels, i.e. the dose distribution maps, which cost dosimetrists considerable time and effort to acquire. For cancers of low-incidence, such as cervical cancer, it is often a luxury to collect an adequate amount of labeled data to train a well-performing deep learning (DL) model. To mitigate this problem, in this paper, we resort to the unsupervised domain adaptation (UDA) strategy to achieve accurate dose prediction for cervical cancer (target domain) by leveraging the well-labeled high-incidence rectal cancer (source domain). Specifically, we introduce the cross-attention mechanism to learn the domain-invariant features and develop a cross-attention transformer-based encoder to align the two different cancer domains. Meanwhile, to preserve the target-specific knowledge, we employ multiple domain classifiers to enforce the network to extract more discriminative target features. In addition, we employ two independent convolutional neural network (CNN) decoders to compensate for the lack of spatial inductive bias in the pure transformer and generate accurate dose maps for both domains. Furthermore, to enhance the performance, two additional losses, i.e. a knowledge distillation loss (KDL) and a domain classification loss (DCL), are incorporated to transfer the domain-invariant features while preserving domain-specific information. Experimental results on a rectal cancer dataset and a cervical cancer dataset have demonstrated that our method achieves the best quantitative results with [Formula: see text], [Formula: see text], and HI of 1.446, 1.231, and 0.082, respectively, and outperforms other methods in terms of qualitative assessment.

为解决扣件数据集不平衡问题,引入代价敏感策略对卷积神经网络算法进行改进,并以此检测断裂、丢失的缺陷扣件.该算法借鉴AdaBoost算法的思路,在训练过程中对整体误差函数中每个样本分配不同的权重,并依据先前模型的错误率不断地加以调整,使算法关注各个类别中的难学习样本,并对调整后的权重按类别进行归一化处理,以增大小类样本的关注度.分别在高速铁路无砟轨道和普速铁路有砟轨道2个扣件数据集上进行对照试验验证算法的有效性.引入G-mean作为评价指标平衡不同类别的召回率.结果表明:将改进后算法应用于高速铁路无砟和普速铁路有砟轨道的扣件数据集,改进后算法的G-mean值比原算法分别提高10％和25％以上;比传统的扣件识别方法分别提高13％和39％以上.

Real‐world classification often encounters a problem called class imbalance. When the data of some classes are redundant than that of other classes, traditional classifiers usually bias their decision boundaries to the redundant majority classes. Most proposed cost‐sensitive strategies often ignore the hard‐to‐learn examples or have a large amount of hyper‐parameters. This article proposes an adaptive learning cost‐sensitive convolutional neural network to solve this problem. During the training process, the proposed method embeds a class‐dependent cost to each class in the global error, making the decision boundary bias to the minority classes. Meanwhile, a distribution weight is assigned to each example to enhance the learning of the hard‐to‐learn examples. Both the class‐dependent costs and distribution weights are learnt automatically in the net. This cost‐sensitive approach makes the algorithm focus on the examples in the minority classes as well as the hard‐to‐learn examples in each class. Besides, this approach can be applied to both binary and multi‐class image classification problems without any modification. Experiments are conducted on four image classification datasets to evaluate this algorithm. The experimental results show that the proposed method achieves better performance than the baseline algorithms and some other algorithms.

高速铁路线路中的扣件故障会影响高速铁路列车的平顺性,甚至威胁到行车安全.针对扣件视觉检测在复杂背景和光照多变的条件下检测精度低的问题,提出一种基于局部区域特征的缺陷扣件检测算法.从原始图像中截取包括扣件和钢轨的感兴趣区域并对该区域图像进行预处理,接着对预处理后的图像提取极小值区域(二值化),然后基于二值图和模板匹配算法实现扣件区域的精确定位,最后提取扣件区域内上下弹条的局部区域特征并送入决策树完成扣件图像的分类,不同线路的真实图像用于试验.试验证明了算法的有效性,其中漏检率为0.6％,误检率为0.32％,可以满足工程应用需求.

为了实现高铁缺陷扣件的准确、快速和自动化检测,提出一种基于图像处理技术的高铁扣件自适应视觉检测算法.针对高铁扣件图像的特性,使用改进的LBP (local binary pattern)算子提取扣件的显著特征;在扣件特征图的基础上,采用模板匹配算法得到扣件区域在原始图中的精确位置,进而得到扣件子图并用扣件的位置信息校验定位结果;以相邻两个扣件子图的差值作为判断依据,如果差值大于预设的阈值,相应的扣件则被判断为缺陷扣件.将该检测算法应用于高铁工务部门提供的真实扣件图.研究结果表明:本文提出的自适应扣件检测算法在雨天的表现最差,检出率为96％,误检率为0.50％;在晴天的表现最好,检出率为100％,误检率为0.22％;在不同天气、光照、环境下的综合检出率为99％,综合误检率为0.33％.

Ensemble learning is an excellent method for imbalance classification. However, the existing ensemble methods often ignore noise in the dataset, which may reduce the accuracy of classifier. In this paper, we propose a density-based undersampling algorithm (DBU) and integrate it with AdaBoost (DBUBo ost) to improve the classification performance. The major contribution of this paper is the development of an undersampling strategy for dealing with both noise and class imbalance problem. We first divide the examples from each class into three categories: useful examples, noise and potentially useful examples. Then we introduce a similarity coefficient to distinguish the examples from each category. Through a selection mechanism based on similarity coefficients, we retain the useful examples and remove the noisy examples. To demonstrate the effectiveness, we compare our DBUBoost with four ensemble methods and three anti-noise methods. The experiments were conducted on 9 KEEL datasets and their noise-modified datasets. Experimental results have shown that our DBUBoost performs better than other state-of-the-art methods.

Imbalance classification is an interesting issue in machine learning and data mining. In recent years, many related algorithms have been proposed to solve such an issue. Among them, under-sampling is an effective and timesaving data pre-processing method, which balances the dataset by removing some examples from the majority class. However, these proposed under-sampling methods often lose some useful information or ignore noise in the datasets, which will result in the performance degradation. This paper proposes a density-based under-sampling algorithm (DBU) to solve these two problems. In feature space, similar examples are close to each other and noisy example is far from other examples belonging to the same class. Thus the similar examples have a high density while the noisy example has a low density. DBU uses the local density peaks to represent the whole majority class, so that it can retain the useful information and eliminate the noisy examples automatically. To evaluate our algorithm, experiments are conducted on 15 two-class imbalanced datasets. Experimental results show that DBU achieves the better results than other under-sampling methods.

Traditional image features are not able to effectively represent railway fasteners under varied illumination and conditions. We propose the line local binary pattern encoding method that considers the relationship between the center point and its upper and lower neighborhoods. The method can effectively represent the key components of fasteners. In comparison with several state-of-the-art methods, the proposed method has good performance on detecting the completely missing and partly missing fasteners on real data sets, especially when the illumination and background are not ideal.

铁路扣件是固定轨道的连接件,扣件丢失或断裂直接影响铁路运输的安全。扣件运行环境复杂,采集的图像与背景差异性较小,难以自动识别。针对扣件图像自动化识别提出了一种新的图像局部二元模式编码算法。用高斯分布进行随机采样得到邻域点,利用随机点对的差分关系得到二元模式编码,称为高斯采样二元模式编码GSLBP (Gaussian sampling local binary pattern)。为了避免噪声影响,利用随机点邻域像素值之和代替随机点的值,通过比较像素值之和得到编码,最后利用卡方距离对图像进行分类。该算法更加准确的反映了图像局部纹理信息,图像差异化信息更加明显。在铁路扣件图像上进行了实验,比较了各种LBP编码方法与提出的方法的分类结果,结果表明提出的方法具有更好的分类结果。

为了提高铁路扣件基于视觉的自动化检测精度,提出了一种改进的LBP(local binary pattern)编码算法.该方法为了避免基本LBP对噪声敏感问题,根据不同邻域的不同噪声强度,结合测量误差服从高斯分布的原则,计算邻域内像素均值和偏差;根据偏差大小,自动设置阈值,实现自适应噪声抑制.为了避免基本LBP表达邻域差分关系不完整的缺陷,提出了利用邻域内随机采样的方式得到采样点对,通过比较随机点对的差分关系得到LBP编码.对在晴天、阴天、雨天等不同天气条件下的铁路扣件图像上进行实验,并与原始以及其他改进LBP进行比较.结果表明,本文的算法具有更高的检测准确率,晴天提高了3.32％,阴天提高了3.27％,雨天提高了4.10％,能够满足铁路扣件自动化检测的需要.

The original local binary pattern (LBP) operator or LBP variants adopt the difference between the neighboring pixels and the center pixel to describe the pixel that does not consider the relationship between the neighboring pixels. The block region characteristics of an image are determined by the relationship between neighboring pixels, not just the neighboring pixels and the center pixel. In this letter, a new local neighborhood encoding method is proposed, which we call random sampling LBP (RSLBP). Based on the distribution of the image difference signal, point pairs are randomly selected in the local neighborhood, and LBP encoding is carried out after comparing the sums of pixels neighboring the random point. Image local difference is more obvious and noise resistance is better. By comparing the classification results of LBP and LBP variants with the proposed method, we show that the proposed method achieves better classification performance on the standard images library and the real fastener images, and the performance gain is significant when the noise level is high.

针对现有扣件定位方法对不同光照条件下的铁路扣件图像适用性差、定位不准确的问题,提出一种鲁棒的、准确的铁路扣件定位检测算法.首先利用人眼视敏度特性来确定Rank变换窗口的大小,并通过改进Rank变换增强图像边缘特征;然后基于改进的LSD(Line Segment Detector)方法在增强后的图像中搜寻垫板与轨枕接触的边界线,并利用先验知识实现扣件的准确定位.实验表明,该方法定位准确且具有很好的适应性和鲁棒性.

The randomly operating of truck near the entrance of highway toll gate has become an sever problem in the expressway management.On based of the analyzing of the vehicle prior knowledge,the vehicle parameter extracting model and a tracking algorithm are designed in this paper.After the prior knowledge is applied in the modeling process of this algorithm,the model matching method is also applied in a real-time image sequences to track the vehicle.Depend on the method mentioned before,the continuously real-time model updating and consecutively vehicles tracking are achieved in this paper.The algorithm implementation process which is tested by a real experiment is showed in this paper.The experiment results show that the algorithm,which can avoid the environment impacts to the matching result,also has a certain degree of ability to recognize a shielded-plate vehicle.