Augment dataset

本研究中收集的初始图像包括桃子在各种条件下的表现。然而，现实世界的果园除了天气和光线之外，还存在许多影响水果外观的因素。考虑到果园中不平坦的地面可能会导致未来机器人作期间拍摄角度不同，因此对 Pascal VOC 格式注释文件附带的原始数据集图像进行了数据增强。采用高斯噪声、随机亮度调整、图像旋转和水平翻转等技术来增强数据集。在增强过程中，至少对每个原始图像应用了一种方法。增强后，生成了 10,350 张新图像。在增强过程中，数据集中的标签文件信息会根据目标位置的变化自动更新。然后将新生成的标注文件转换回 YOLO 格式，以满足后续可行性实验模型验证的要求。The initial images collected in this study include photographs of peaches under diverse conditions. However, real-world orchards encompass numerous additional factors that affect fruit appearance beyond weather and lighting. Considering that uneven ground in orchards may result in varying shooting angles during future robotic operations, data augmentation was therefore conducted on the original dataset images paired with Pascal VOC format annotation files. Techniques including Gaussian noise injection, random brightness adjustment, image rotation, and horizontal flipping were employed to augment the dataset. At least one of these augmentation techniques was applied to each original image during the process. Following augmentation, a total of 10,350 new images were generated. During the augmentation process, the information in the dataset's label files was automatically updated in accordance with changes in target positions. The newly generated annotation files were then converted back to the YOLO format to meet the requirements for subsequent model validation in feasibility experiments.