Abstract:The small-sample task is a current challenge in the field of deep learning, due to the huge annotation cost and the inherent limitations of targets, such as the acquisition of rare animal and plant images. Data augmentation is an effective method to solve the semantic sparseness and overfitting of deep convolution neural network in small-sample classification, but its effectiveness remains to be improved. We propose a Wasserstein GAN with confidence loss (WGAN-CL) to implement the expansion of small-sample plant dataset. Firstly, a shallower GAN's structure is designed to adapt to less plant data. Meanwhile, shortcut-stream connections are brought into the basic network to enlarge the solution space of the model without producing additional training parameters. Secondly, the Wasserstein distance combined with confidence loss is used for optimizing the model. Experiments demonstrate that the Wasserstein distance with gradient penalty guarantees the stability of model training and the diversity of outputs. And the sample screening strategy based on confidence loss can ensure that the generated image is close to the real image in semantic features, which is critical for subsequent image classification. To verify the effectiveness of the WGAN-CL in plant small-sample augmentation, 2000 flower images of 5 categories in the "Flowers" dataset are utilized as training samples, while 2000 augmented images are employed for model training as well to improve the performance of a classical classifier. WGAN-CL has a significant performance improvement over state-of-the-art technologies, i.e., a 2.2% improvement in recall and a 2% improvement in F1-score. Experiments on the "Plant Leaves" dataset also achieved excellent results demonstrating that WGAN-CL can be migrated to other tasks. WGAN-CL uses less computational resources while considering both effectiveness and robustness, proved the practicality of our model.

Data Augmentation Generated by Generative Adversarial Network for Small Sample Datasets Clustering

Data Augmentation Based on Generative Adversarial Network with Mixed Attention Mechanism

Study Of Restrained Network Structures For Wasserstein Generative Adversarial Networks (Wgans) On Numeric Data Augmentation

Data Augmentation Computing Model Based on Generative Adversarial Network

Self-Paced Video Data Augmentation by Generative Adversarial Networks with Insufficient Samples.

Gaussian Mixture Model and Double-Weighted Deep Neural Networks for Data Augmentation Soft Sensing

Cluster-based Generative Adversarial Network Imbalanced Data Generation Method

Ensemble Data Augmentation for Imbalanced Fault Diagnosis.

WGAN-CL: A Wasserstein GAN with confidence loss for small-sample augmentation

Toward Understanding Generative Data Augmentation

Gaussian Mixture Conditional Tabular Generative Adversarial Network for Data Imbalance Problem

Integrating data augmentation and hybrid feature selection for small sample credit risk assessment with high dimensionality

SSGAN: A Semantic Similarity-Based GAN for Small-Sample Image Augmentation

DOPING: Generative Data Augmentation for Unsupervised Anomaly Detection with GAN

An improved generative adversarial network to oversample imbalanced datasets

Balanced Self-Paced Learning for Generative Adversarial Clustering Network.

Conditional Wasserstein GAN-based Oversampling of Tabular Data for Imbalanced Learning

Numeric Data Augmentation Using Structural Constraint Wasserstein Generative Adversarial Networks

A new imbalanced data oversampling method based on Bootstrap method and Wasserstein Generative Adversarial Network

Time series data augmentation method of small sample based on optimized generative adversarial network

On Data Augmentation for GAN Training