YOLOv11-WMF:基于频域增强与多尺度注意力的棉桃检测算法OA
针对复杂棉田环境中棉花检测存在的尺度差异大、遮挡严重、背景干扰强及小目标漏检误检等问题,该文以 YOLOv11为基础,提出改进模型 YOLOv11-WMF.该模型通过引入 WaveCNet 模块增强边缘细节特征、嵌入 MSCA 模块聚合多尺度特征、采用 C3K2_Former 结构融合局部与上下文信息,提升检测性能.实验表明,在自建棉田数据集上,模型 mAP@0.5 达 83.9%,较基线提升 2.5 个百分点,参数量控制在 2.46×106,可有效解决检测痛点,为棉花智能采摘机器人视觉感知提供技术支撑.
In response to the problems of large scale differences,severe occlusion,strong background interference,and small target missed detections and false detections in cotton detection in complex cotton field environments,this paper proposes an improved model YOLOv11-WMF based on YOLOv11.This model enhances detection performance by introducing WaveCNet module to enhance edge detail features,embedding MSCA module to aggregate multi-scale features,and using C3K2_former structure to fuse local and contextual information.Experiments have shown that on a self built cotton field dataset,the model mAP@0.5 Reaching 83.9%,an increase of 2.5 percentage points from the baseline,the parameter quantity is controlled at 2.46×106,which can effectively solve the detection difficulty and provide technical support for the visual perception of cotton intelligent picking robots.
燕子豪;徐家梁;赵自新
西安交通大学 仪器科学与技术学院,西安 710001西安交通大学 仪器科学与技术学院,西安 710001西安交通大学 仪器科学与技术学院,西安 710001
农业科技
YOLOv11棉花检测小目标检测频域特征增强多尺度注意力机制
YOLOv11cotton detectionsmall target detectionfrequency-domain feature enhancementmulti-scale attention mechanism
《智慧农业导刊》 2026 (18)
35-40,6
评论