基于机器视觉的轻量化芒果果面缺陷检测  

Light weight detection of mango surface defects based on machine vision

作　　者：聂衍文 杨佳晨 文慧心 高路 徐建 NIE Yan-wen;YANG Jia-chen;WEN Hui-xin;GAO Lu;XU Jian(School of Electrical and Electronic Engineering,Wuhan Polytechnic University,Wuhan,Hubei 430023,China)

机构地区：[1]武汉轻工大学电气与电子工程学院,湖北武汉430023

出　　处：《食品与机械》2023年第3期91-95,240,共6页Food and Machinery

基　　金：湖北省科技厅基金项目(编号:CXYH2019000535)。

摘　　要：目的:降低自动化芒果分级检测设备的制造成本。方法:对比了3种常用的目标检测算法在芒果缺陷检测中的效果,并以轻量化、移动设备可移植性为目标,提出了一种基于YOLOv5的轻量化芒果果面缺陷检测算法。结果:试验算法在满足芒果表面缺陷检测要求的前提下,相较于原算法可使参数量减少45.9%,浮点运算次数减少46.7%,权重文件大小减小45.2%。结论:试验算法有效降低了对部署设备的性能需求,在降低芒果分级检测设备的制作成本方面具有潜在价值。Objective:Reduce the manufacturing cost of automated mango grading equipment.Methods:The effects of three commonly detection algorithms for mango defect detection were compared,and a defect detection algorithm based on YOLOv5 for mango surface was proposed for the light weight design to work on mobile devices.Results:Compared with the original algorithm,the experimental algorithm can reduce the number of parameters by 45.9%,the number of floating point operations by 46.7%,and the weight file size by 45.2%under the premise of meeting the requirements for mango surface defect detection.Conclusion:the experimental algorithm effectively reduces the performance requirements for deployment equipment,and has potential value in reducing the manufacturing cost of mango grading detection equipment.

关 键 词：芒果 果面检测 神经网络 轻量化 机器视觉 图像识别 

分 类 号：S226.5[农业科学—农业机械化工程] TP391.41[农业科学—农业工程]