基于深度学习的冬枣智能分选技术与装备研究
Research on intelligent sorting technology and equipment for winter jujube based on deep learning
王伟辉 1李阳 1徐慧群 1张建军1
作者信息
- 1. 青岛理工大学机械与汽车工程学院,山东青岛 266520
- 折叠
摘要
针对冬枣分选难、分选精度低等问题,设计一种基于深度学习的冬枣智能分选机,搭建卷积神经网络模型,对冬枣准确地识别分类;冬枣360°滚动至图像采集区域,采用颜色空间变换、二值处理、图像分割等方法提取冬枣目标,消除背景对检测效率的影响;由Arduino控制8路继电器通断完成冬枣的分选.试验结果表明,搭建的冬枣缺陷检测模型平均判别准确率达98.12％,其中畸形果的判别准确率100％、病果99.6％、优质果98.8％、裂果97.8％,每组平均测试时间149.32 ms;当传送速度为0.5 m/s时,检测准确率达98.01％.研究对冬枣分选工作有重要意义.
Abstract
In response to the problems of difficult sorting and low sorting accuracy of winter jujube,a winter jujube quality intelligent sorting machine was designed based on deep learning,the convolutional neural network model was built to accurately identify and classify the winter jujube;the winter jujube is rolled 360°to the image acquisition area,color space transformation,binary processing,image segmentation and other methods were adopted to extract the target of winter jujube to eliminate the influence of background on the detection efficiency;Arduino controls ON/OFF of eight relays to complete the sorting of winter jujube.The test results show that the average discriminating accuracy of the constructed winter jujube defect detection model reaches 98.12％,among which the discriminating accuracy of deformed fruit reaches 100％,99.6％ for diseased fruit,98.8％ for high-quality fruit,97.8％ for cracked fruit,and the average test time of each group reaches 149.32 ms;when the transfer speed is 0.5 m/s,the detection accuracy reaches 98.01％.The research is of significant importance for the winter jujube sorting work.
关键词
冬枣/深度学习/机器视觉/品质分级/表面缺陷Key words
winter jujube/deep learning/machine vision/quality grading/surface defect引用本文复制引用
基金项目
山东省农机装备研发创新计划项目(2018YF015)
出版年
2024
国家科技期刊平台
NSTL
万方数据