基于轻量化YOLOX算法的东北虎检测

基于轻量化YOLOX算法的东北虎检测

以大型哺乳动物东北虎（<italic>Panthera tigris altaica</italic>）为例，通过使用圈养东北虎监测影像数据集（ATRW），采用YOLOX算法对东北虎进行目标检测研究，检测速度为87.59张/s，0.50阈值准确率<inline-formula><alternatives><math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1"><mo stretchy="false">...

Full description

Saved in:
Bibliographic Details
Main Authors: 于文华, 谢永华
Format: Article
Language:zho
Published: Editorial Department of Chinese Journal of Wildlife 2023-11-01
Series:野生动物学报
Subjects:
YOLOX算法
目标检测
轻量化
东北虎
Online Access:http://ysdw.nefu.edu.cn/thesisDetails#10.12375/ysdwxb.20230405
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:以大型哺乳动物东北虎（<italic>Panthera tigris altaica</italic>）为例，通过使用圈养东北虎监测影像数据集（ATRW），采用YOLOX算法对东北虎进行目标检测研究，检测速度为87.59张/s，0.50阈值准确率<inline-formula><alternatives><math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1"><mo stretchy="false">(</mo><mi>m</mi><mi>A</mi><msub><mrow><mi>P</mi></mrow><mrow><mn mathvariant="normal">0.50</mn></mrow></msub><mo stretchy="false">)</mo></math><graphic specific-use="big" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/29316FC3-CBDB-4584-956E-5B0CDFD28C88-M001.jpg"><?fx-imagestate width="12.10733318" height="3.89466691"?></graphic><graphic specific-use="small" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/29316FC3-CBDB-4584-956E-5B0CDFD28C88-M001c.jpg"><?fx-imagestate width="12.10733318" height="3.89466691"?></graphic></alternatives></inline-formula>为97.32%，0.75阈值准确率（<inline-formula><alternatives><math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M2"><mi>m</mi><mi>A</mi><msub><mrow><mi>P</mi></mrow><mrow><mn mathvariant="normal">0.75</mn></mrow></msub><mo stretchy="false">)</mo></math><graphic specific-use="big" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/29316FC3-CBDB-4584-956E-5B0CDFD28C88-M002.jpg"><?fx-imagestate width="11.00666618" height="3.89466691"?></graphic><graphic specific-use="small" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/29316FC3-CBDB-4584-956E-5B0CDFD28C88-M002c.jpg"><?fx-imagestate width="11.00666618" height="3.89466691"?></graphic></alternatives></inline-formula>为75.10%，模型总参数量为8.938×10<sup>6</sup>。通过筛选无锚框算法，对选出的YOLOX算法进行轻量化、添加注意力机制及网络损失函数的优化，优化后的算法检测速度提升1.74张/s，<inline-formula><alternatives><math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M3"><mi>m</mi><mi>A</mi><msub><mrow><mi>P</mi></mrow><mrow><mn mathvariant="normal">0.50</mn></mrow></msub></math><graphic specific-use="big" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/29316FC3-CBDB-4584-956E-5B0CDFD28C88-M003.jpg"><?fx-imagestate width="9.65200043" height="3.21733332"?></graphic><graphic specific-use="small" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/29316FC3-CBDB-4584-956E-5B0CDFD28C88-M003c.jpg"><?fx-imagestate width="9.65200043" height="3.21733332"?></graphic></alternatives></inline-formula>准确率提升1.02个百分点，<inline-formula><alternatives><math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M4"><mi>m</mi><mi>A</mi><msub><mrow><mi>P</mi></mrow><mrow><mn mathvariant="normal">0.75</mn></mrow></msub></math><graphic specific-use="big" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/29316FC3-CBDB-4584-956E-5B0CDFD28C88-M004.jpg"><?fx-imagestate width="9.65200043" height="3.21733332"?></graphic><graphic specific-use="small" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/29316FC3-CBDB-4584-956E-5B0CDFD28C88-M004c.jpg"><?fx-imagestate width="9.65200043" height="3.21733332"?></graphic></alternatives></inline-formula>准确率提升1.53个百分点，模型的总参数量减少18.47%。算法改进后，在提升识别准确率的同时，有效降低了检测算法依托硬件的需求，为东北虎的野外行为研究、保护生物多样性及东北虎的野外相关数据收集提供了检测算法。
ISSN:2310-1490

Similar Items