沈阳农业大学学报2026,Vol.57Issue(5):96-103,8.DOI:10.3969/j.issn.1000-1700.2026.05.009
基于改进YOLOv12的松材线虫形态学特征检测
Morphological Feature Detection of Bursaphelenchus xylophilus Based on Improved YOLOv12
摘要
Abstract
[Objective]Bursaphelenchus xylophilus is an important pathogen causing pine wilt disease,and its microscopic morphological identification still depends largely on expert experience.Conventional fluorescence PCR and manual microscopic identification have limitations such as complex sample processing,high personnel dependence,long detection cycles,and insufficient suitability for large-scale rapid screening.To improve the automation and accuracy of key morphological structure detection,an improved YOLOv12 microscopic image detection model integrating a lightweight backbone network and cross-scale feature enhancement was proposed.[Methods]The microscopic images collected from the Jilin Academy of Forestry Sciences were used to construct a dataset containing 200 original images,which was augmented to 1 200 images through cropping,flipping,rotation,brightness disturbance,and noise disturbance.The dataset was divided into training,validation,and testing sets at a ratio of 7.0:1.5:1.5.Based on YOLOv12,ShuffleNetV2 was introduced to replace the original backbone network to reduce parameter redundancy and computational complexity.Meanwhile,a Cross-Scale Feature Fusion Module(CCFM)was designed to strengthen semantic interaction and detail compensation among features at different scales,thereby improving the recognition of small morphological structures such as the vulval flap,spicule,tail tip,median esophageal bulb,and stylet.[Results]Comparative experiments showed that the improved model achieved a precision of 94.4%,an mAP of 93.2%,and a recall of 88.5%,with 25.3 M parameters.In the ablation experiment,the YOLOv12+ShuffleNetV2+CCFM combination achieved an accuracy of 95.51%,a precision of 94.44%,a recall of 88.50%,a F1-score of 0.93,a mAP of 93.20%,and an inference speed of 25.7 images·s-1.Compared with the original YOLOv12,the mAP and recall increased by 7.0% and 6.4%,respectively,indicating the complementary benefits of ShuffleNetV2 and CCFM.[Conclusion]The proposed improved YOLOv12 model improves the detection accuracy of microscopic morphological features while maintaining high inference efficiency,and it can provide technical support for rapid diagnosis,morphology-assisted identification,and intelligent monitoring of agricultural and forestry pests.关键词
松材线虫/目标检测/ShuffleNetV2/CCFM/显微图像识别Key words
Bursaphelenchus xylophilus/object detection/ShuffleNetV2/CCFM/microscopic image recognition分类
农业科技引用本文复制引用
王宇飞,姜岚峰,宋少忠,安晓峰,于凯威..基于改进YOLOv12的松材线虫形态学特征检测[J].沈阳农业大学学报,2026,57(5):96-103,8.基金项目
吉林省科技厅农业重点研发项目(20250202036NC) (20250202036NC)