期刊论文

Research of image recognition of camellia oleifera fruit based on improved convolutional auto-encoder

中文

林业工程学报

2096-1359

2019

4

3

118-124

10.13360/j.issn.2096-1359.2019.03.018

湖南省科技计划重点研发项目(2016NK2142) 湖南省高校科技创新团队支持计划(2014207)。

本校为第一机构

为进一步提高油茶果识别的速度与精度,提出一种基于改进卷积自编码机神经网络的油茶果图像识别方法。该方法以非对称分解卷积核提高训练速度,以输出端重构输入端的直连训练方式减少信息损失;利用改进的自编码机训练不同颜色通道的图像,随后在共享层中利用金字塔池化算法融合高阶特征,利用softmax算法构建分类器。结合selective-search算法生成候选区域,并对其进行筛选,利用训练好的网络设计油茶果果实图像检测器验证算法的实用性。改进算法100次迭代所需时间为166 s,平均识别准确率为90.4%;改进模型收敛性能较好,学习能力强,经过10次迭代之后即可达到85%的识别准确率,在样本数目为200的小数据集...

In order to further improve the speed and accuracy of the camellia oleifera recognition,an image recognition method based on convolutional auto-encoder was proposed to solve the problems of long convergence time and large number of model parameters in the traditional methods. The method used asymmetric decomposed convolution kernels to improve network structure,then increasing the training speed,and reduced information loss by reconstructing the input with output. Compared with the traditional method,the improvement of the convolution kernel structure helped to reduce the computational difficu...