Bee Image Object Detection

The dataset was created for bee object detection based on images. Videos were taken at the entrance of 25 beehives in three apiaries in San Jose, Cupertino, and Gilroy in CA, USA. The videos were taken above the landing pad of different beehives. The camera was placed at a distinct angle to provide a clear view of the hive entrance.

The images were saved one frame per second from videos. The annotation platform Label Studio was selected to annotate bees in each image due to the friendly user interface and high quality. The below criteria was followed in the labeling process. First, at least 50% of the bee's body must be visible. Second, the image cannot be too blurry. After tagging each bee with a rectangle box in the annotation tool, output label files with Yolo labeling format were generated for each image. The output label files contained one set of bounding-box (BBox) coordinates for each bee in the image. If there were multiple objects in the image, there would be one line for one object in the label file. It recorded the object ID, X-axis center, Y-axis center, BBox width, and height with normalized image size from 0 to 1.

本数据集旨在基于图像进行蜜蜂目标检测，其创建旨在捕捉蜜蜂在加州圣何塞、库比蒂诺和吉尔罗伊三个蜂场25个蜂箱入口的视频。摄像机置于不同蜂箱的着陆平台上方，以独特的角度安装，以确保对蜂箱入口的清晰视图。视频以每秒一帧的速度进行捕捉，并将图像保存。鉴于用户界面友好且标注质量上乘，选用了Label Studio标注平台对每帧图像中的蜜蜂进行标注。在标注过程中遵循以下标准：首先，蜜蜂的身体至少需有50%可见；其次，图像不得过于模糊。在标注工具中对每只蜜蜂进行矩形框标注后，为每张图像生成了Yolo标注格式的输出标签文件。输出标签文件包含图像中每只蜜蜂的边界框（BBox）坐标集。若图像中存在多个物体，则在标签文件中为每个物体占据一行。记录了物体ID、X轴中心、Y轴中心、BBox宽度和高度，均以归一化图像尺寸（0至1）表示。