恶劣环境下图像算法数据增强方法

Data augmentation in harsh environments for computer vision algorithms

为解决恶劣环境下数据采集难度较大、数据匮乏导致模型性能受限的问题,提出一种基于风格迁移的数据增强方法,用于增加恶劣环境下的样本数量,提升模型在恶劣环境下的鲁棒性。建立包含22500张图片的数据集,使用卷积神经网络进行图片去重,进行手工标注,用于进行分类模型的训练。设计正常环境和恶劣环境对比实验,验证提出的数据增强方法效果,实验结果表明,该方法可以有效提升分类模型在恶劣环境下的鲁棒性。

In some harsh environments,data acquisition is difficult,and the lack of data has a great impact on model perfor-mance.A data augmentation method based on style transfer was proposed to increase the number of samples in harsh environments and improve the robustness of the model.The established dataset contained 22500 manually-labeled images,where a con-volutional neural network was used for image deduplication.The classification model was trained on this dataset to verify the proposed data augmentation method.The contrast experiment between normal environment and harsh environment was designed to verify the effect of the proposed method.Experimental results show that this method can effectively improve the robustness of the classification model in harsh environments.