基于YOLO的钢缆表面损坏检测

为了解决检测钢缆表面损坏时检测设备资源有限、时间过长等问题,将深度学习的先进技术以及卷积神经网络(CNN)应用于钢缆表面损坏检测.提出了一种基于YOLO的缺陷检测网络模型,将GhostNet融入主干网络,并基于ShuffleNet及注意力机制提出了新的特征提取模块(ShuffleC3),再对Head部分进行剪枝改进.实验结果表明,改进后网络相比基线YOLOv5s的平均精度提高1.1%,参数量和计算量分别降低了43.4%和31%,模型大小减少了42.3%.可以在降低网络计算成本的同时,保持较高的识别精确度,更好地满足了对钢缆材料表面损坏检测的要求.     

改进AOD-Net的道路交通图像去雾算法

针对现有图像去雾算法在处理道路交通图像时无法同时兼顾去雾效果和实时性的问题,本文以快速一体化网络去雾算法(AOD-Net)为基础进行改进.首先,在AOD-Net中添加SE通道注意力,以自适应的方式分配通道权重,关注重要特征;其次,引入金字塔池化模块,扩大网络的感受野,并融合不同尺度特征,更好地捕捉图像信息;最后,使用复合损失函数同时关注图像像素信息和结构纹理信息.实验结果表明,改进后的AOD-Net算法对道路交通图像去雾后的峰值信噪比提升了2.52 dB,结构相似度达到了91.2%,算法复杂度和去雾耗时略微增加,但仍满足实时要求.     

跨层协同注意和通道分组注意的细粒度图像分类

细粒度图像分类的主要挑战在于类间的高度相似性和类内的差异性. 现有的研究多数基于深层的特征而忽略了浅层细节信息, 然而深层的语义特征由于多次卷积和池化操作往往会丢失大量的细节信息. 为了更好地整合浅层和深层的信息, 提出了基于跨层协同注意和通道分组注意的细粒度图像分类方法. 首先, 通过ResNet50加载预训练模型作为骨干网络提取特征, 由最后3个阶段提取的特征以3个分支的形式输出, 每一个分支的特征通过跨层的方式与其余两个分支的特征计算协同注意并交互融合, 其中最后一个阶段的特征经过通道分组注意模块以增强语义特征的学习能力. 模型训练可以高效地以端到端的方式在没有边界框和注释的情况下进行训练, 实验结果表明, 该算法在3个常用细粒度图像数据集CUB-200-2011、Stanford Cars和FGVC-Aircraft上的准确率分别达到了89.5%、94.8%和94.7%.     

Protein–Protein Interaction Prediction for Targeted Protein Degradation

Protein–protein interactions (PPIs) play a fundamental role in various biological functions; thus, detecting PPI sites is essential for understanding diseases and developing new drugs. PPI prediction is of particular relevance for the development of drugs employing targeted protein degradation, as their efficacy relies on the formation of a stable ternary complex involving two proteins. However, experimental methods to detect PPI sites are both costly and time-intensive. In recent years, machine learning-based methods have been developed as screening tools. While they are computationally more efficient than traditional docking methods and thus allow rapid execution, these tools have so far primarily been based on sequence information, and they are therefore limited in their ability to address spatial requirements. In addition, they have to date not been applied to targeted protein degradation. Here, we present a new deep learning architecture based on the concept of graph representation learning that can predict interaction sites and interactions of proteins based on their surface representations. We demonstrate that our model reaches state-of-the-art performance using AUROC scores on the established MaSIF dataset. We furthermore introduce a new dataset with more diverse protein interactions and show that our model generalizes well to this new data. These generalization capabilities allow our model to predict the PPIs relevant for targeted protein degradation, which we show by demonstrating the high accuracy of our model for PPI prediction on the available ternary complex data. Our results suggest that PPI prediction models can be a valuable tool for screening protein pairs while developing new drugs for targeted protein degradation.     

Development of Deep-Learning-Based Single-Molecule Localization Image Analysis

Recent developments in super-resolution fluorescence microscopic techniques (SRM) have allowed for nanoscale imaging that greatly facilitates our understanding of nanostructures. However, the performance of single-molecule localization microscopy (SMLM) is significantly restricted by the image analysis method, as the final super-resolution image is reconstructed from identified localizations through computational analysis. With recent advancements in deep learning, many researchers have employed deep learning-based algorithms to analyze SMLM image data. This review discusses recent developments in deep-learning-based SMLM image analysis, including the limitations of existing fitting algorithms and how the quality of SMLM images can be improved through deep learning. Finally, we address possible future applications of deep learning methods for SMLM imaging.     

Identification of Spliceogenic Variants beyond Canonical GT-AG Splice Sites in Hereditary Cancer Genes

Pathogenic/likely pathogenic variants in susceptibility genes that interrupt RNA splicing are a well-documented mechanism of hereditary cancer syndromes development. However, if RNA studies are not performed, most of the variants beyond the canonical GT-AG splice site are characterized as variants of uncertain significance (VUS). To decrease the VUS burden, we have bioinformatically evaluated all novel VUS detected in 732 consecutive patients tested in the routine genetic counseling process. Twelve VUS that were predicted to cause splicing defects were selected for mRNA analysis. Here, we report a functional characterization of 12 variants located beyond the first two intronic nucleotides using RNAseq in APC, ATM, FH, LZTR1, MSH6, PALB2, RAD51C, and TP53 genes. Based on the analysis of mRNA, we have successfully reclassified 50% of investigated variants. 25% of variants were downgraded to likely benign, whereas 25% were upgraded to likely pathogenic leading to improved clinical management of the patient and the family members.     

基于知识蒸馏与改进ViT网络的花卉图像细粒度分类

由于自然条件下拍摄的花卉图像背景复杂,而且其存在类内差异性大和类间相似性高的问题,现有主流方法仅依靠卷积模块提取花卉的局部特征难以实现准确的细粒度分类。针对上述问题,本文提出了1种高精度、轻量化的花卉分类方法(ConvTrans-ResMLP),通过结合Transformer模块和残差MLP(multi-layer perceptron)模块实现对花卉图像的全局特征提取,并在Transformer模块中加入卷积计算使得模型仍保留提取局部特征的能力;同时,为了进一步将花卉分类模型部署到边缘设备中,本研究基于知识蒸馏技术实现对模型的压缩与优化。实验结果表明,本文所提出的方法在Oxford 17、Oxford 102和自制的Flowers 32数据集上的准确率分别达98.62%、97.61%和98.40%;知识蒸馏后本文的轻量化模型的大小约为原来的1/18,而准确率仅下降2%左右。因此,本研究能较好地提升边缘设备下花卉细粒度分类的效率,对促进花卉培育的自动化发展具有切实意义。     

基于LSDANet的手机芯片屏蔽壳表面缺陷检测方法

为了解决手机芯片屏蔽壳表面白印缺陷微小、尺度各异等因素影响检测快速性和准确性的问题,本文提出一种基于长短连接通路和双注意力网络(long short link and double attention network, LSDANet)的手机芯片屏蔽壳表面缺陷检测方法。首先,通过构建基于编码和解码的语义分割模型和利用长短距离连接通路,提高网络模型对尺度各异缺陷的特征提取能力。其次,分别设计基于通道和空间的注意力机制,增大5—10 pixel尺寸的白印缺陷在空间和通道上的特征权重。最后,融合双注意力机制和长短距离连接通路分割模型,构建LSDANet缺陷检测网络,应用于手机芯片屏蔽壳表面缺陷检测。实验数据表明,LSDANet网络能够达到96.21%的平均像素精度、66.13%的平均交并比和39.03的每秒检测帧数,相比多种语义分割算法均具有更高的检测精度和速度。     

融合注意力线性特征多样化的DR分级模型

糖尿病视网膜病变(diabetic retinopathy, DR)是目前人类的主要致盲疾病之一。针对DR数据集中样本类间差异小和类分布不均衡等制约分级性能提高的问题,本文提出一种融合注意力线性特征多样化(fusion of attention linear feature diversification, FALFD)的分级算法。该算法首先用改进的Res2Net残差网络作为模型骨干来增大感受野,进一步提高网络捕捉特征信息的能力;其次引入自适应特征多样化模块(adaptive feature diversification module, AFDM)对眼底图像可分辨的微小病理特征进行识别,获得具有高语义信息的局部特征,避免单一特征区域的限制,进而提高分级准确度;再后利用双线性注意力融合模块(bilinear attention fusion module, BAFM)增加可判别区域特征的网络权重占比;最后采用正则化焦点损失(focal loss, FL)进一步提升算法的分类性能。在IDRID数据集上,灵敏度和特异性分别为94.20%和97.05%,二次加权系数为87.83%;在APTO...     

一种新型的无人机牧草分割网络—LMS-DeeplabV3+

目前草原环境复杂、牧草分散且与背景颜色差异小,无法实现高效精准的分割,因此本文提出了一种新型的轻量化多尺度DeeplabV3+网络(lightweight and multi-scale DeeplabV3+network, LMS-DeeplabV3+)。该网络以DeeplabV3+为基础网络,首先选用轻量级的MobilenetV2作为骨干网络用于初步特征提取,并为了适应牧草分割任务做了网络配置上的调整;其次在加强特征提取模块和解码模块中均使用深度可分离卷积代替普通卷积以轻量化网络;此外利用密集空洞空间金字塔池化(dense atrous spatial pyramid pooling, DASPP)模块捕获更大的感受野,加强各特征之间的交互;又引入卷积注意力机制(convolutional block attention module, CBAM)重分配权重加强特征提取。实验证明,提出的新网络与原始网络相比平均交并比(mean intersection over union,mIOU)提升了8.06个百分点、平均像素精度(mean pixel accuracy,mPA)提升了6.7...