基于改进YOLOv4的苹果检测与果径估测方法  

Apple detection and fruit diameter estimation method based on improved YOLOv4

作　　者：岳琳茜 李文宽 杨晓峰 李海芳 杨其晟 YUE Linxi;LI Wenkuan;YANG Xiaofeng;LI Haifang;YANG Qisheng(College of Information and Computer,Taiyuan University of Technology,Taiyuan 030600,China)

机构地区：[1]太原理工大学信息与计算机学院,太原036000

出　　处：《激光杂志》2022年第2期58-65,共8页Laser Journal

基　　金：国家自然科学基金(No.61976150);省重点研发计划项目(No.201803D31038);山西省晋中市科技重点研发计划(No.Y192006)。

摘　　要：为实现自然场景下苹果快速、精确检测与果实生长状态监测,提出基于改进的YOLOv4苹果检测模型与果径估测方法。针对自然场景下背景杂乱、目标较小等问题,在特征提取网络中引入注意力机制使模型更关注于果实区域,并在路径聚合网络中使用DO-Conv卷积以丰富网络提取的特征信息,提高模型检测性能。将检测后的目标区域进行CIELAB颜色空间分量下的阈值分割,利用RGB-D深度信息进行尺寸转换,实现果径估测。实验结果表明,模型检测精度达91.5%。果径估测平均绝对误差为1.91 mm,均方根误差(RMSE)为2.17 mm,总体分级准确率在90%以上,可为田间苹果分级与生长状态监测提供参考。In order to realize fast and accurate apple detection and fruit growth status monitoring in natural scenes, an improved YOLOv4 apple detection model and fruit diameter estimation method are proposed. Aiming at the problems of cluttered background and small targets in natural scenes, the attention mechanism is introduced in the feature extraction network to make the model pay more attention to the fruit area, and DO-Conv convolution is used in the path aggregation network to enrich the feature information extracted by the network. Improve model checking performance. The detected target area is subjected to threshold segmentation under CIELAB color space components, and RGB-D depth information is used for size conversion to realize fruit diameter estimation. Experimental results show that the model detection accuracy reaches 91.5%. The average absolute error of fruit diameter estimation is 1.91 mm, the root mean square error(RMSE) is 2.17 mm, and the overall classification accuracy is above 90%, which can provide a reference for field apple grading and growth status monitoring.

关 键 词：YOLOv4 苹果检测 注意力机制 果径测量 苹果分级 

分 类 号：TN209[电子电信—物理电子学]