基于改进YOLOv7模型的田间麦穗识别与计数

打开文本图片集

中图分类号：S512.1；TP391.4 文献标识码：A 文章编号：2095-5553（2025）12-0077-09

Abstract：Inwheatbreeding，thenumberof spikes isanimportant indicatorforevaluating wheatyield，andtimelyand accurate acquisitionof wheat spikesisof great practicalsignificanceforyield prediction.Therefore，this study proposesanew network model YOLOv7一ASFF-ECA that combines Adaptive Spatial Feature Fusion（ASFF）andEficient Channel Atention（ECA）withYOLOv7.This experimentused alocallcolected wheat earimagedataset，which includedatotalof 3 373 wheat ear images.The experimental results showed that the precision rate of the model reached 97.8% ，the recall rate was 97% ，mean average precision （mAP ）was 98.3% ， F1 value was 97.4% ，average processing time per image was 28.1ms ，andthere were almost no missed detections incounting.Inorder to verifythe accuracyof this model in thefield， thedataset wasdivided into two states based onthe intensityoflight：brightand shadow，andinto smoothand noisy states based onthedegree ofsmoothness.Comparative experiments were conducted with the model and other models respectively. In the shaded state，the precision rate of the model is 95.6% ，the recall rate is 97.5% ，the F1 value is 96.4% ，and the mAP （204号 is 97.3% . In the noisy state，the precision rate is 95.8% ，the recall rate is 97.0% ，the F1 value is 96.3% ，and the mAP is 97.8% . Overallcomparison shows that the YOLv7—ASFF—ECA network model not only ensures high accuracy，but also hasfastdetectionspeedandcalibratedcounting.Inadition，itdemonstrates excellntobustnes indealing ithensewheat densityand severe occlusion in the field，providing new technical support for wheat earrecognition and yield prediction.

Keywords：wheat;wheat ear recognition; adaptive feature fusion; atention mechanism; improved YOLOv7

0 引言

小麦是我国种植面积最广泛的作物之一，及时而准确地检测小麦麦穗的生长情况与麦穗数对有效地支持产量预测至关重要，这对于农产品价格和供应政策的制定和实施具有重大影响1。（剩余14443字）