融合多尺度特征与目标注意力的YOLOv5s棉花叶片病虫害检测模型建立及评价

打开文本图片集

中图分类号 TP391.4;S436.629 文献标志码A 文章编号 1007-7731（2026）05-0113-05

DOI号 10.16377/j.cnki.issn1007-7731.2026.05.026

Establishment and evaluation of YOLOv5s cotton leaf disease and pest detection model integratingmulti-scalefeaturesand targetattention

Shang Peng12En Dakai 1，2 Jing Xiaojie1Liu Zhaojie1Zhang Weidong1Zhang Xiao 1，2 （2 （ 1 College of Information Engineering， Tarim University，Alar 84330o， China; ²Key Laboratory of Tarim Oasis Agriculture and Education，Tarim University，Alar 8433Oo， China）

AbstractInresponse to the need to improve theeficiencyof cotondisease and pest detection，this paper proposed an enhanced detection model named YOLOv5sMBT，based on YOLOv5s，for identifying cotton diseases and pests inleaf images.Themodel incorporates 3 key improvements over theoriginalYOLOv5s framework：amulti-scale featureextraction network（Multi-scale）was constructed to enhance feature extraction capabilities;anda Transformer atention mechanism was integrated between the feature extraction network andtheneck network，combined with the C3 module to form C3TR，thereby improving themodel’satention to target features;a BiFPN（Bidirectional feature pyramid network）structure was introduced to efciently fuse shalowanddeep features.Adataset of2179 leaf images covering 4 common coton diseases and pests （Nesidiocoris tenuis，spider mite，wilting，aphid） wasused for validation. The dataset was split into training，validation，and test sets in a ratio.Experimental results showed that the YOLOv5sMBT model achieved a mean average precision （ mAP ）of 0.838，outperforming the original model （ mAP of 0.799）.This study provides a reference for the intelligent detection of cotton diseases and pests.

Keywordsdiseasesand pests detection; multi-scale feature extraction network; cotton leaf; intelligent deection

棉花作为主要经济作物之一，在新疆地区种植面积较大。（剩余5839字）