• Simplifying Low-Light Image Enhancement Networks with Relative Loss Functions

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-10-08

    摘要:Image enhancement is a common technique used to mitigate issues such as severe noise, low brightness, low contrast, and color deviation in low-light images. However, providing an optimal high-light image as a reference for low-light image enhancement tasks is impossible, which makes the learning process more difficult than other image processing tasks. As a result, although several low-light image enhancement methods have been proposed, most of them are either too complex or insufficient in addressing all the issues in low-light images. In this paper, to make the learning easier in low-light image enhancement, we introduce FLW-Net (Fast and LightWeight Network) and two relative loss functions. Specifically, we first recognize the challenges of the need for a large receptive field to obtain global contrast and the lack of an absolute reference, which limits the simplification of network structures in this task. Then, we propose an efficient global feature information extraction component and two loss functions based on relative information to overcome these challenges. Finally, we conducted comparative experiments to demonstrate the effectiveness of the proposed method, and the results confirm that the proposed method can significantly reduce the complexity of supervised low-light image enhancement networks while improving processing effect.

  • 空气质量预测的深度学习模型研究与评估

    分类: 环境科学技术及资源科学技术 >> 环境科学技术及资源科学技术其他学科 分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-09-22

    摘要:目的 及时和准确的空气质量预测数据对于环境管理至关重要,尤其是在空气重污染期间,预测数据可以为政府生态环境管理部门应对污染状况、精准地调配社会资源的决策提供数据支撑。
    方法 笔者研发的基于深度学习的空气质量预测模型AirNet6,可以兼顾准确性和实时性,实现臭氧、二氧化硫、一氧化碳等因子的7天甚至更长时间的空气质量预测。
    结果 与传统的化学模型演算不同,本模型使用时空图卷积网络(STGCN),能捕获历史监测数据、天气预测数据、社会活动等数据的规律,在2分钟内完成一百多个点位未来168小时数据的预测。
    结论 实验表明,AirNet6模型在速度、节能和准确度上,比传统的化学模型及时间序列AI模型均有明显进步。

  • Adaptively hybrid fractal image coding

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-09-06

    摘要:In this study, an adaptively hybrid method was proposed to improve the performance of fractal coding methods. First, we found that the range blocks with large variances (RBLVs) play a crucial role in degrading decoded images, and the effect of the remaining range blocks with small variances (RBSVs) can be ignored. Second, RBLVs were designed to be encoded in an extended domain block pool (EDBP), and the remaining RBSVs were encoded with the no-search fractal encoding method. Moreover, an effective method to adaptively divide the range blocks into the above two categories was proposed. Finally, four fractal coding methods were adopted to assess the performance of the proposed method. Experimental results show that, compared with the previous methods, the proposed method can achieve better-decoded image quality with fewer bits per pixel and fewer computations.

  • Fractal Decoded Image Quality Prediction

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-08-25

    摘要:To predict fractal decoded image quality more efficiently, an effective decoded image quality prediction method was proposed in this study. In fractal encoding process, the dynamic range of the linear correlation coefficients (LCCs) between range blocks and their best-matched domain blocks was greatly extended by several outliers which increased uncertainty and resulted in reduced prediction accuracy. To remove the interference of outliers, we introduced the effective minimum and maximum of LCCs, which provided the effective bottom and top limits of the actual percentage of accumulated collage error (EBL-APACE and ETL-APACE), respectively. Further, when EBL-APACE reached a large percentage, the average collage error (ACER) can be estimated, and the decoded image quality can be predicted directly.
    Experimental results show that compared with the previous method, the proposed method can provide higher prediction accuracy with fewer computations.

  • 基于改进GPT模型的文本生成研究

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-07-01

    摘要:[目的] 本研究旨在提出一种基于词和词性的联合文本生成模型,以提高生成文本的质量。
    [方法] 该模型由两个预训练的文本生成模型组成,一个是基于词的模型,另一个是基于词性的模型。此外,本文还提出并使用了BERT模型对进行二分类任务,以判断文本生成效果。
    [结果] 在三个数据集上的实验结果表明,与传统的GPT模型相比,GPT-WP模型生成文本的质量有明显提升。
    [局限] BERT模型在二分类任务中参数较大,大规模数据训练下评价效果差,本文提出的模型在数据量较小的场景下表现较好,大规模数据表现差异缩小。
    [结论] GPT-WP模型在本文提出的评价方法下表明其能够有效地提高生成文本的质量,对于自然语言生成任务的改进和评估提供了参考。

  • 基于GAN+XGBoost+LR的个性化推荐方法

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-05-16

    摘要:目的 解决长尾商品的推荐中存在的样本数据偏少,现有协同过滤法计算量大,性能难以满足需求的问题
    方法 出了基于GAN+XGBoost+LR的解决方案,通过协同过滤寻找初始样本数据,利用GAN生成更多样本数据训练XGBoost+LR模型,并针对不同模型的特点指定针对性的训练策略。
    结果 该方案在兼顾性能和精确度要求下,可以提高推荐模型的鲁棒性。
    局限 XGBoost模型承担自动化特征工程能力有限。
    结论 基于GAN+XGBoost+LR的个性化推荐方法可以提高长尾商品的推荐的有鲁棒性。

  • 基于ChatGPT的用户图书评分偏好预测研究

    分类: 计算机科学 >> 计算机应用技术 分类: 图书馆学、情报学 >> 情报学 提交时间: 2023-05-12

    摘要:目的/意义:随着以ChatGPT为代表大语言模型技术的不断发展与变革,使得许多领域的经典场景都重新焕发出新的机会。同时,越来越多的学者开始关注如何将大语言模型的智能化能力与技术应用到现有的场景,并分析这些技术带来的挑战和机遇。 方法/过程:本文以ChatGPT为建模对象,首次将大语言模型技术引入用户图书评分偏好预测这一图情领域的典型应用场景,并落地实践。通过构建基于ChatGPT的用户图书评分预测模型(CUBR, ChatGPT-based model for User Book Rating Prediction),来探索大语言模型技术在图书推荐领域实践和落地的可行性。同时,本文基于图书评分任务的不同评估方案与现有经典推荐模型进行对比,探讨并给出了CUBR在用户图书评分预测场景的优势与劣势,并分析了后续大语言模型在图书推荐其他场景可能的研究机会点。 结果/结论:本文实验研究表明,(1)CUBR模型在现有用户图书评分偏好预测任务上能够取得不错的推荐效果,特别是单样本(One-shot)这类待推荐目标信息较少的情况下,其表现接近或超过当前经典推荐算法,且泛化能力较强,较适用于冷启动推荐场景。(2)随着单个用户提示样本内容的增加(如从One-shot到Ten-shot),CUBR的预估效果会有显著的提升,说明CUBR具备不错的实时上下文学习能力。 局限:本文研究场景仅限于用户图书评分偏好理解与推荐,未来将尝试在更多的图情场景应用和改造现有大语言模型技术,并获得更好的实践效果。

  • An Improved YOLOv5-Based Method for UAV Object Detection

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-03-23

    摘要:Object detection based on unmanned aerial vehicle (UAV) images is very challenging. The multi-scale size and high density of objects in the UAV view bring great difficulties. To fully address this issue to unleash the potential of UAV applications, the YOLOv5-STD model is proposed. First, add one more head to locate extremely small object detection by shallow image features; second, use the attention mechanism to optimize the backbone by the transformer; third, use SPD-Conv to avoid the loss of fine-grained image feature information. At the last, sufficient experiments on the dataset VisDrone 2022 have proven that the model has good performance, compared with the basic model, the improved model has an average improvement of about 7% in mAP@.5 metrics, and the ablation experiments have verified that its improvement skills have a positive effect on the model. This paper can help developers and researchers get a better experience in the analysis and processing of unmanned aerial vehicle images.

  • 人工智能技术在国有企业数字化转型中的机遇、挑战与实施路线

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-21


  • 基于B/S架构的期刊业务办公自动化研究

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》


  • 基于ACO算法在可重构扫描网络中搜索最优测试链路的应用

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》


  • 基于本体的轴类零件外圆工序尺寸参数的自动生成方法

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》


  • 一种非均匀图滤波器组的设计方法

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》


  • 一种改进雅可比算法的频域临界采样图滤波器组

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》


  • 一种模糊互模拟的局部算法

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》


  • 一种用于交通预测的自适应时空图神经网络

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》


  • 基于彩色QR码的信息无损提取隐藏算法

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》


  • 基于稀疏优化的异常分布检测方法

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-15 合作期刊: 《桂林电子科技大学学报》

    据集的平均 FPR 95分别降低了15.02%和15.41%。

  • Comprehensive evaluation of gene sequence encoding methods in deep learning

    分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-09

    摘要:Background: The prediction of genomic structure has become a hot spot in genome research. At present, the prediction method based on deep learning is more effective and accurate than other machine learning algorithms. Since gene sequence data cannot directly enter the deep learning model, the original data need to be encoded and converted into numerical features before model prediction. As a result, different encoding methods may affect final accuracy.Methods: In order to explore the performance of different encoding methods, we compared ten strategies in six deep learning models. We also compared the performance of all methods on independent datasets and models from our laboratory. For all models, we used their original parameters.Results: Dummy encoding, hash encoding, and one-hot encoding perform best in various models. In addition, dummy encoding and one-hot encoding are the best for processing RNA data, while hash encoding is superior to other methods for processing promoter data. Also, when processing part- or full-sequence data, the performance of dummy encoding, hash encoding, and one-hot encoding is similar. Besides that, in sisRNA datasets and prediction models of Arabidopsis and rice, dummy encoding and one-hot encoding achieve higher prediction accuracy.Conclusions: We conclude that the best encoding method varies when the data set changes. One-hot encoding, dummy encoding, and hash encoding are the three best methods for six models. This study fills the gap on sequence encoding methods in deep learning and can provide a valuable reference for the community.

  • CropCircDB: a comprehensive circular RNA resource for crops in response to abiotic stress

    分类: 生物学 >> 生物学其他学科 分类: 计算机科学 >> 计算机应用技术 提交时间: 2023-02-09

    摘要:Circular RNA (circRNAs) may mediate mRNA expression as miRNA sponge. Since the community has paid more attention on circRNAs, a lot of circRNA databases have been developed for plant. However, a comprehensive collection of circRNAs in crop response to abiotic stress is still lacking. In this work, we applied a big-data approach to take full advantage of large-scale sequencing data, and developed a rich circRNA resource: CropCircDB for maize and rice, later extending to incorporate more crop species. We also designed a metric: stress detections score, which is specifically for detecting circRNAs under stress condition. In summary, we systematically investigated 244 and 288 RNASeq samples for maize and rice, respectively, and found 38 785 circRNAs in maize, and 63 048 circRNAs in rice. This resource not only supports user-friendly JBrowser to visualize genome easily, but also provides elegant view of circRNA structures and dynamic profiles of circRNA expression in all samples. Together, this database will host all predicted and validated crop circRNAs response to abiotic stress.