Scholarlink

<p>To avoid the impact of hard samples on the training of the Flying Bird Object Detection model (FBOD model), the Self-Paced Learning strategy with Easy Sample Prior Based on Confidence (SPL-ESP-BC) is proposed. First, the loss-based Minimizer Function is improved and a confidence-based one is introduced, making it more suitable for one-category object detection. Second, to enable early-stage judgment of easy and hard samples using SPL, an SPL strategy with Easy Sample Prior (ESP) is put forward. The FBOD model is initially trained with easy samples via the standard strategy and then continues with all samples using SPL. Combining ESP and the confidence-based Minimizer Function, the SPL-ESP-BC strategy is formed. Training the FBOD model with this strategy helps it learn flying bird object features in surveillance videos from easy to hard. Experimental results show that compared to the non-distinguishing standard training strategy, the Average Precision at 50% Intersection-over-Union (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4795_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="37" /> </InlineMediaObject> <EquationSource Format="TEX">\(\text {AP}_{50}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>AP</mtext> <mn>50</mn> </msub> </math></EquationSource> </InlineEquation>) of the FBOD model trained by SPL-ESP-BC increases by 2.0% on the FBD-SV-TSS dataset and 1.0% on the FBD-SV-2024 dataset. It also outperforms other loss-based SPL strategies in comprehensive detection performance. This project is publicly available at <a href="https://github.com/Ziwei89/FBOD-BSPL">https://github.com/Ziwei89/FBOD-BSPL</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Self-paced learning strategy with easy sample prior based on confidence for the flying bird object detection model training

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Zi-Wei Sun,
  • Ze-Xi Hua,
  • Heng-Chao Li,
  • Yan Li

摘要

To avoid the impact of hard samples on the training of the Flying Bird Object Detection model (FBOD model), the Self-Paced Learning strategy with Easy Sample Prior Based on Confidence (SPL-ESP-BC) is proposed. First, the loss-based Minimizer Function is improved and a confidence-based one is introduced, making it more suitable for one-category object detection. Second, to enable early-stage judgment of easy and hard samples using SPL, an SPL strategy with Easy Sample Prior (ESP) is put forward. The FBOD model is initially trained with easy samples via the standard strategy and then continues with all samples using SPL. Combining ESP and the confidence-based Minimizer Function, the SPL-ESP-BC strategy is formed. Training the FBOD model with this strategy helps it learn flying bird object features in surveillance videos from easy to hard. Experimental results show that compared to the non-distinguishing standard training strategy, the Average Precision at 50% Intersection-over-Union ( \(\text {AP}_{50}\) AP 50 ) of the FBOD model trained by SPL-ESP-BC increases by 2.0% on the FBD-SV-TSS dataset and 1.0% on the FBD-SV-2024 dataset. It also outperforms other loss-based SPL strategies in comprehensive detection performance. This project is publicly available at https://github.com/Ziwei89/FBOD-BSPL.