A Method for Augmenting Camouflage Data Based on Texture Space Scrambling

Data plays a crucial role in training deep learning models. To address the issue of insufficient sample size in the natural camouflage dataset, a data augmentation method based on texture space scrambling is proposed to effectively expand the sample size. Firstly, dynamic scrambling scales are selected based on the number of target pixels, and then the target area is divided into several texture blocks according to the scrambling scale. Finally, the spatial arrangement order of these texture blocks is scrambled. The experiments show that the data generated by the data augmentation method based on texture space scrambling has a similar data distribution to the original dataset. By training on the augmented dataset, the detection performance of the model has improved by 2.2%. Compared with traditional data augmentation methods, the proposed method achieves better results.