中圖分類號(hào)：TP391.41 文獻(xiàn)標(biāo)志碼：A DOI: 10.16157/j.issn.0258-7998.256995 中文引用格式： 把奉昕，燕振剛，陳蕾. SEC-YOLO:一種高效的中草藥復(fù)雜背景識(shí)別模型[J]. 電子技術(shù)應(yīng)用，2026，52(6)：97-105. 英文引用格式： Ba Fengxin，Yan Zhengang，Chen Lei. SEC-YOLO: an efficient complex background recognition model for Chinese herbal medicine[J]. Application of Electronic Technique，2026，52(6)：97-105.
SEC-YOLO： an efficient complex background recognition model for Chinese herbal medicine
Ba Fengxin，Yan Zhengang，Chen Lei
College of Information Science and Technology， Gansu Agricultural University
Abstract： To address the challenges of complex backgrounds, occlusion, and illumination variations in Chinese herbal medicine image recognition, this paper constructs a large-scale dataset containing over 20 000 images of 76 herbal categories and applies various data augmentation techniques to enhance model generalization. Based on YOLO11n, an improved model named SEC-YOLO is proposed, which incorporates the C3k2_Star module, ECA_SR attention mechanism, and C2PSA_CGLU module to improve recognition performance under complex conditions. Furthermore, the feature fusion structure and detection head are optimized to enhance the detection of small and overlapping herbs. The improved model achieves a mean average precision (mAP) of 98.0%, with a weight size of 3.7 MB, 4.7 GFLOPs, and 1.80M parameters. Compared with YOLO11n, the weight size is reduced by 30.1%, FLOPs by 28.7%, and parameters by 31.0%. Both detection accuracy and speed meet the requirements of real-time detection. Experimental results demonstrate that SEC-YOLO achieves high accuracy and real-time recognition while maintaining lightweight characteristics, providing strong support for the automation of Chinese herbal medicine identification.
Key words : Chinese herbal medicine recognition in complex backgrounds；YOLO11；StarNet；channel spatial attention module C2PSA_CGLU；ECA_SR attention mechanism
引言
中草藥是中國(guó)傳統(tǒng)中醫(yī)的特有藥物，在人類的健康中占據(jù)著重要的位置，同時(shí)在國(guó)際上也有著極其強(qiáng)大的威信力[1]。在傳統(tǒng)的中藥識(shí)別過程中，主要依靠人力以及自身經(jīng)驗(yàn)對(duì)中草藥加以識(shí)別，此種識(shí)別方式效率低下并且容易因?yàn)槿藶樵虺霈F(xiàn)誤檢等情況。尤其在野外采集或復(fù)雜背景（如多類藥材堆疊、遮擋、光照變化、相似環(huán)境等）下，識(shí)別難度進(jìn)一步加大，嚴(yán)重制約了數(shù)字中藥資源管理及其產(chǎn)業(yè)化發(fā)展。而人工智能、圖像處理和模式識(shí)別領(lǐng)域的進(jìn)步可以拓展和改進(jìn)中醫(yī)藥技術(shù)的實(shí)踐[2]。近年來(lái)，隨著深度學(xué)習(xí)在目標(biāo)檢測(cè)領(lǐng)域的迅猛發(fā)展，YOLO（You Only Look Once）系列算法因其端到端、一階段、檢測(cè)速度快等優(yōu)勢(shì)，廣泛應(yīng)用于農(nóng)業(yè)、交通、醫(yī)療等圖像識(shí)別場(chǎng)景。其中，YOLO11作為該系列的最新演進(jìn)版本，在保持模型輕量化的同時(shí)，引入了改進(jìn)的C3k2模塊、ECA_SR注意力機(jī)制以及解耦檢測(cè)頭等結(jié)構(gòu)，進(jìn)一步提升了模型的識(shí)別精度與泛化能力。
