Abstract
In order to solve the problem of model accuracy reduction caused by the difficulty of obtaining specific training samples or the insufficient number of samples in the application of existing object detection and recognition model based on deep learning, this article proposes a conditional generative adversarial network model (VSA-CGAN), which integrates the self-attention mechanism of visual perception to optimize the inference of object attention feature maps, so as to learn the global information of the image and the detailed features of the object. It is designed to add conditional features in the generator and the discriminator, associate the specific dimensions of the data with the semantic features, and explicitly indicate the model to generate the corresponding object signature category information, so as to generate the feature representation of the image which is more suitable for the distribution of the original data. The model in this article has completed numerical experiments on several general standard data sets, and compared with several mainstream generative adversarial network models in image data augmentation performance. The experimental results show that the generation model in this article has excellent object simulation ability and strong application prospects.
Highlights
Generative adversarial network (GAN) is an optimized generation model proposed by Goodfellow in 2014 based on the idea of antagonistic competition
Self-Attention Generative Adversarial Networks (SAGAN) can learn the distribution rule of the overall geometric features of the image based on the selfattention module to a certain extent, the self-attention model of SAGAN is not precise enough to learn the distribution of the structural information and geometric features of the object itself, resulting in the poor effect of generating the detailed features of the object in the generated image, the deviation of the geometric distribution between the key structures of the object, and SAGAN still adopts an unsupervised learning method, and the designed attention model has a high demand for the number of training data, which greatly limits the performance and application prospects of the network [7]
On KITTI dataset, because the Convolutional neural network (CNN) classifier has achieved high accuracy, each of the data enhancement method can hardly achieve conspicuous improvement on the performance; on MOD2 dataset, the classification performance of ESA-CGAN enhanced dataset is better than that of duplicate samples and CNN classifier that has been enhanced by affine transformation, which indicates that it has avoided overfitting and that the synthesized images generated by the method proposed in this article prove to be valid
Summary
Generative adversarial network (GAN) is an optimized generation model proposed by Goodfellow in 2014 based on the idea of antagonistic competition. SAGAN can learn the distribution rule of the overall geometric features of the image based on the selfattention module to a certain extent, the self-attention model of SAGAN is not precise enough to learn the distribution of the structural information and geometric features of the object itself, resulting in the poor effect of generating the detailed features of the object in the generated image, the deviation of the geometric distribution between the key structures of the object, and SAGAN still adopts an unsupervised learning method, and the designed attention model has a high demand for the number of training data, which greatly limits the performance and application prospects of the network [7]. We use the recursive nature of LSTM to iteratively optimize the salient features of static images, instead of using LSTM to model the time dependence of sequence data
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
