基于改進(jìn)YOLOv4的實(shí)時(shí)目標(biāo)檢測(cè)方法研究

打開(kāi)文本圖片集

摘要：為提升實(shí)時(shí)目標(biāo)檢測(cè)的準(zhǔn)確性和穩(wěn)健性，該文采用增強(qiáng)特征融合技術(shù)、網(wǎng)絡(luò)架構(gòu)技術(shù)、損失函數(shù)技術(shù)等對(duì)YOLOv4算法進(jìn)行優(yōu)化。結(jié)果表明，改良后的YOLOv4算法在多變環(huán)境下對(duì)小型目標(biāo)檢測(cè)表現(xiàn)出色，展現(xiàn)了其實(shí)用性和穩(wěn)定性，為廣泛應(yīng)用奠定了堅(jiān)實(shí)基礎(chǔ)。

關(guān)鍵詞：實(shí)時(shí)目標(biāo)檢測(cè)；YOLOv4；特征融合；GIoU損失函數(shù)

doi：10.3969/J.ISSN.1672-7274.2024.09.006

中圖分類號(hào)：TP 31                 文獻(xiàn)標(biāo)志碼：B            文章編碼：1672-7274（2024）09-00-04

Research on Real-time Object Detection Method Based on Improved YOLOv4

LU Jianheng

（College of Artificial Intelligence， Guangzhou Huashang College， Guangzhou 511300， China）

Abstract： To enhance the accuracy and robustness of real-time object detection， this paper optimizes the YOLOv4 algorithm by employing enhanced feature fusion technology， network architecture technology， loss function technology， and other strategies. The results demonstrate that the improved YOLOv4 algorithm exhibits excellent performance in detecting small objects in diverse environments， showcasing its practicality and stability， and laying a solid foundation for its widespread application.

Keywords： real-time object detection; YOLOv4; feature fusion; GIoU loss function

0   引言

在人工智能與視覺(jué)科技領(lǐng)域，實(shí)時(shí)目標(biāo)檢測(cè)技術(shù)日益凸顯其重要性，廣泛應(yīng)用于自動(dòng)駕駛、視頻監(jiān)控、無(wú)人機(jī)導(dǎo)航等多個(gè)領(lǐng)域。（剩余4785字）