An Annotated Image Dataset for Small Apple Fruitlet Detection in Complex Orchard Environments

This study presents a comprehensive dataset comprising 2,517 images of apple fruitlets, each uniformly rescaled to 500×500 pixels to maintain dimensional consistency. Manual annotation was performed using LabelImg software, with bounding box coordinates stored in XML format. To enhance usability across different platforms, we additionally provide annotations in TXT format. Figure 3 illustrates a representative annotation example. During the annotation process, special attention was given to: (1) precise localization of small fruitlets, and (2) accurate annotation of partially occluded targets, where only visible portions were labeled to minimize false positives in subsequent analyses.

本研究构建了一套包含2517张苹果幼果图像的综合性数据集，所有图像均统一缩放至500×500像素以保证维度一致性。标注工作采用LabelImg软件手动完成，边界框（bounding box）坐标以XML格式存储。为提升跨平台适用性，我们额外提供了TXT格式的标注文件。图3展示了一则典型的标注示例。在标注过程中，我们重点关注以下两点：(1) 对小型幼果进行精准定位；(2) 对部分遮挡的目标进行准确标注——仅标注其可见区域，以最大程度降低后续分析中的假阳性率。