A New Improved Convolutional Neural Network Flower Image Recognition Model
Authors:
- 1 December 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 3110-3117
Abstract
In order to improve the accuracy of the flower image recognition, a convolutional neural network (A-LDCNN) model based on attention mechanism and LD-loss (Linear Discriminant Loss Function) is proposed. Unlike traditional CNN (Convolutional Neural Networks), A-LDCNN uses the VGG-16 network pre-trained by ImageNet to perform feature learning on preprocessed flower images. The attention feature is constructed by fusing the local features of the multiple intermediate convolution layers with the global features of the fully connected layer and using it as the final classification feature. LDA (Latent Dirichlet Allocation) is introduced into the model to construct a new loss function LD-loss, which participates in the training of CNN to minimize the feature distance in class and maximize the feature distance between classes, and to solve the problem of Inter-class similarity and intra-class difference in flower image classification. Classification experiments show that the accuracy of A-LDCNN is 87.6%, which is higher than other traditional networks and can realize the accurate recognition of flower images under natural conditions.Keywords
This publication has 4 references indexed in Scilit:
- Pest identification via deep residual learning in complex backgroundComputers and Electronics in Agriculture, 2017
- Efficient Object Detection and Segmentation for Fine-Grained Recognition2013
- Automated Flower Classification over a Large Number of Classes2008
- A Visual Vocabulary for Flower Classification2006