摘要
针对传统语义图像分割复杂背景下的苹果目标鲁棒性较差的问题,提出一种基于深度学习的全卷积网络图像分割方法。方法首先利用全卷积网络和人工标注的标准结果进行网络的训练和验证,然后利用训练好的全卷积网络进行网络的实验与测试。基于苹果样本集的实验结果表明:全卷积网络模型能够在提高苹果目标分割精度的同时有效降低错分风险,而且在消耗近似时间的前提下,平均召回率均达到90%以上,平均绝对误差的范围控制在1%以内。
In view of the poor robustness of the apple target in the complex background of the traditional semantic image segmentation, a full convolution network image segmentation method based on depth learning is proposed. First, the network is trained and verified by the standard results of full convolution network and manual annotation. Then, the experiment and test of the network are carried out by the trained full convolution network. The experimental results of apple samples show that the full convolution network model can effectively reduce the accuracy of the fruit target segmentation and reduce the error risk effectively, and the average recall rate of apples is above 90% under the approximate time of consumption, and the average absolute error is equal. Range control is within 1%.
作者
朱悦云
Zhu Yueyun(School of Mechanical and Automotive Engineering,Kaifeng University,Kaifeng Henan 475000,China)
出处
《信息与电脑》
2018年第19期18-19,共2页
Information & Computer
关键词
苹果图像分割
全卷积网络
深度学习
apple image segmentation
fully convolution network
deep learning