基于改进RT-DETR的原棉异性纤维检测方法

A method for detecting foreign fibers in raw cotton based on improved RT-DETR

摘要

为解决在资源受限条件下实现原棉异性纤维检测模型的轻量化部署问题,并为原棉异性纤维的精准检测提供技术支撑,本文提出了一种基于改进RT-DETR的原棉异性纤维检测算法。首先,提出了包含C2F-IEL模块的轻量化主干网络,以降低模型参数量和计算的复杂度,提高模型的部署能力;其次,提出了LDGST模块,提高了对异性纤维的检测精度,使模型更关注与背景颜色相近的异性纤维;最后,引入FSIoU损失函数,通过自适应聚焦不同难度样本的机制提升模型回归精度,增强了对异性纤维的定位精度。结果表明,本文提出的COTTON-DETR模型检测精度和平均精度均值分别为96.22%和95.72%,与基线模型RT-DETR模型相比,检测精度和平均精度均值分别提高了0.7%和2.47%,参数量和性能需求分别降低了37.22%和29.65%。本文提出的COTTON-DETR在保证异性纤维识别精度的同时,显著降低了模型复杂度,为原棉异性纤维的检测提供了技术支撑。

Abstract

To address the challenge of lightweight deployment of a raw-cotton foreign-fiber detection models under resource-constrained conditions and provide technical support for accurate foreign-fiber detection, this paper proposes a raw cotton foreign fiber detection algorithm based on an enhanced RT-DETR architecture. First, a lightweight backbone that incorporates the C2F-IEL module is introduced to reduce model parameter complexity and amputational overhead while improving deployment flexibility. Second, the LDGST module enhances detection accuracy by focusing on foreign fibers with similar background colors. Finally, the FSIoU loss function and adaptive sample-focusing mechanism refine regression precision and boost localization accuracy for foreign fibers. Experiments show that COTTON-DETR achieves a detection accuracy of 96.22% and a mean average precision (mAP) of 95.72%. Compared with the original RT-DETR, it raises detection accuracy and mAP by 0.7% and 2.47%, respectively, while reducing parameters and computational cost by 37.22% and 29.65%. The COTTON-DETR proposed in this paper maintains the accuracy of foreign fiber recognition while significantly reducing model complexity, providing effective technical support for detecting foreign fibers in raw cotton.

关键词

异性纤维 / RT-DETR / 轻量化 / 损失函数 / 目标检测

Key words

foreign fiber / RT-DETR / lightweight / loss / target detection

[1] XIAO C, JI Q Y, ZHANG F C, et al. Effects of various soil water potential thresholds for drip irrigation on soil salinity, seed cotton yield and water productivity of cotton in northwest China[J]. Agricultural Water Management, 2023, 279: 108172.
[2] 李瑜.浅议棉花异性纤维在线检测方法研究[C]//第十六届中国标准化论坛论文集.郑州,2019:971-974.
[3] 任维佳, 杜玉红, 左恒力, 等. 棉花中异性纤维检测图像分割和边缘检测方法研究进展[J].纺织学报,2021,42(12):196-204.
REN W J, DU Y H, ZUO H L, et al. Research progress in image segmentation and edge detection methods for alien fibers detection in cotton [J]. Journal of Textile Research,2021,42 (12):196-204.
[4] 陈亚军, 吴婷荣, 史书伟, 等. 基于光学成像的棉花异性纤维检测方法研究进展[J]. 激光与光电子学进展, 2021, 58(16): 130-146.
CHEN Y J, WU T R, SHI S W, et al. Review of cotton foreign fiber detection method using optical imaging [J]. Laser & Optoelectronics Progress, 2021, 58(16): 130-146.
[5] WANG R, ZHANG Z F, YANG B, et al. Detection and classification of cotton foreign fibers based on polarization imaging and improved YOLOv5[J]. Sensors, 2023, 23(9): 4415.
[6] 张馨, 李道亮, 杨文柱, 等. 高分辨率棉花异性纤维彩色图像的快速分割方法[J]. 农业机械学报, 2011, 42(1): 159-164,192.
ZHANG X, LI D L, YANG W Z, et al. Fast segmentation of high-resolution color images of cotton foreign fibers [J]. Transactions of the Chinese Society for Agricultural Machinery, 2011, 42(1): 159-164,192.
[7] 刘双喜, 王金星, 张菡, 等. 微弱棉花异性纤维图像的多通道小波分割方法研究 [J]. 仪器仪表学报, 2016, 37(S1): 60-66.
LIU S X, WANG J X, ZHANG H, et al. Research on the multi-channel wavelet segmentation method of faint cotton foreign fibers [J]. Chinese Journal of Scientific Instrument, 2016, 37(S1): 60-66.
[8] 林素存, 魏菊, 常帅才. 基于深度学习的毛/粘混纺织物混纺比检测技术 [J]. 毛纺科技, 2024, 52(2): 121-126.
LIN S C, WEI J, CHANG S C. Blending ratio detection of wool/viscose blended fabrics based on deep learning [J]. Wool Textile Journal, 2024, 52(2): 121-126.
[9] LI Q, HAN S K, WANG P, et al. Foreign fiber detecting system based on multispectral technique [J] 2015 International Conference on Optical Instruments and Technology: Optical Systems and Modern Optoelectronic Instruments, 2015, 9618: 96180W.
[10] REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[11] JIANG P Y, ERGU D, LIU F Y, et al. A review of YOLO algorithm developments[J]. Procedia Computer Science, 2022, 199: 1066-1073.
[12] TERVEN J, CORDOVA-ESPARZA D M, ROMERO-GONZALEZ J A. A comprehensive review of YOLO architectures in computer vision: from YOLOv1 to YOLOv8 and YOLO-NAS[J]. Machine Learning and Knowledge Extraction, 2023, 5(4): 1680-1716.
[13] 杜玉红, 董超群, 赵地, 等. 改进 Faster RCNN 模型在棉花异性纤维识别中的应用[J]. 激光与光电子学进展,2020,57(12):124-133.
DU Y H, DONG C Q, ZHAO D, et al. Application of improved Faster RCNN model for foreign fiber identification in cotton [J]. Laser & Optoelectronics Progress, 2020, 57(12): 124-133.
[14] 郭典, 李景彬, 胡立庆, 等. 基于轻量化YOLOv8的疵棉异性纤维检测算法研究[J]. 石河子大学学报(自然科学版),2024,42(6):765-774.
GUO D, LI J B, HU L Q, et al. Research on defective cotton foreign fiber detection algorithm based on lightweight YOLOv8 [J]. Journal of Shihezi University (Natural Science), 2024, 42(6): 765-774.
[15] DAI Z G, CAI B L, LIN Y G, et al. Up-detr: unsupervised pre-training for object detection with transformers[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2021: 1601-1610.
[16] LI F, ZHANG H, XU H Z, et al. Mask dino: towards a unified transformer-based framework for object detection and segmentation[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2023: 3041-3050.
[17] ZHAO Y A, LV W Y, XU S L, et al. DETRs beat YOLOs on real-time object detection[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2024: 16965-16974.
[18] LI X, CAI M, TAN X, et al. An efficient transformer network for detecting multi-scale chicken in complex free-range farming environments via improved RT-DETR[J]. Computers and Electronics in Agriculture, 2024, 224: 109160.
[19] LYU Z M, DONG S J, XIA Z Y, et al. Enhanced real-time detection transformer (RT-DETR) for robotic inspection of underwater bridge pier cracks[J]. Automation in Construction, 2025, 170: 105921.
[20] GONG W K. Lightweight object detection: a study based on YOLOv7 integrated with ShuffleNetv2 and vision transformer [EB/OL]. 2024:arXiv:2403.01736. https://arxiv.org/abs/2403.01736.

基金资助

新疆生产建设兵团科技攻关项目(2019AB014)