Abstract
We propose a simple dense convolutional neural network model (SDenseNet) to solve the problem of low accuracy and poor performance in classification of navel orange with large external differences. Specifically, AlexNet is adopted as the backbone network, through the introduction of batch normalization(BN) and reduction initial size of the convolution kernel to accelerate convergence of the network model. Besides, we design a new feature reuse structure to promote the connection between layers, and modify the full connection layer by using global pooling to reduce training parameters. The experimental results demonstrate that our proposed SDenseNet significantly improves the performance with 99.33% accuracy on self-built navel orange dataset, outperforming the classic models of LeNet, AlexNet, SqueezeNet and ResNet with significant improvements of 5.33%, 1.55% and 3.55% respectively. The research results provide a new solution for external quality automatic detection and classification of navel orange.
Export citation and abstract BibTeX RIS
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.