Floating Waste and Aquatic Vegetation Image Dataset from Turbid Rivers

This dataset contains images of floating waste and river vegetation collected directly from turbid river environments in Banjarmasin, Indonesia. The dataset was developed to represent the visual characteristics of rivers with high water turbidity and complex environmental conditions, including light reflections, vegetation shadows, surface ripples, varying illumination, and objects with similar visual characteristics. The dataset contains two object classes: floating_waste and river_vegetation. The floating_waste class represents waste materials found floating on the river surface, including plastic, plastic bottles, styrofoam, and other lightweight materials. The river_vegetation class represents natural vegetation observed on or near the river surface. The images were collected directly from river environments during daytime, between approximately 09:00 and 15:00 WITA, using smartphone cameras. During image acquisition, the camera was positioned at approximately 160 cm above the ground, approximately 3 m from the river, and at an angle of approximately 45° toward the water surface. The collected data consisted of both photographs and video recordings, with video recordings subsequently converted into image frames. The initial data collection resulted in 1,578 images. The images were subsequently processed through resizing and cropping into 640 × 640 pixel images, resulting in a final dataset of 3,156 images. Manual object annotation was performed using the bounding box method through Roboflow. Each annotated object was assigned to one of the two classes, floating_waste or river_vegetation. The dataset is organized in a YOLO-compatible format, consisting of image files and corresponding text annotation files containing class identifiers and bounding box coordinates. A YAML configuration file is provided to define the dataset structure, class names, and data locations. The final dataset was divided into training, validation, and testing subsets using a 70:20:10 ratio, consisting of 2,209 training images, 631 validation images, and 316 testing images. The dataset can be used for computer vision and object detection research, particularly for floating waste and river vegetation detection in visually challenging turbid river environments.本数据集采集自印度尼西亚马辰（Banjarmasin）的浑浊河流环境，涵盖漂浮垃圾与河流植被的图像样本。数据集旨在还原高水浊度、环境复杂度较高的河流视觉特征，涵盖光照反射、植被阴影、水面波纹、多变光照以及视觉特征相似的干扰物体等典型场景。 该数据集包含两类目标类别：漂浮垃圾（floating_waste）与河流植被（river_vegetation）。其中漂浮垃圾类指漂浮于河面的废弃物，包括塑料、塑料瓶、泡沫塑料及其他轻质废弃材料；河流植被类指分布于河面或河畔的自然水生/滨水植被。 所有图像均于白天实地采集，采集时段为印尼中部时间（WITA）09:00至15:00，采用智能手机摄像头完成拍摄。拍摄参数设置为：相机距地面高度约160厘米，距河面约3米，以约45°的俯角朝向水面。采集数据包含静态照片与视频录像两类，后续将视频素材转换为独立图像帧。 初始采集阶段共获得1578张原始图像，经尺寸调整与裁剪统一为640×640像素的标准化图像后，最终数据集规模扩充至3156张。标注工作采用边界框（bounding box）标注方法，通过Roboflow平台完成人工标注，每个标注目标被归类为漂浮垃圾或河流植被两类之一。 本数据集采用适配YOLO的格式进行组织，包含图像文件与对应的文本标注文件，后者存储有类别标识符与边界框坐标信息。同时配套提供YAML格式的配置文件，用于定义数据集结构、类别名称与数据存储路径。 最终数据集按70:20:10的比例划分为训练集、验证集与测试集，分别包含2209张训练图像、631张验证图像与316张测试图像。本数据集可用于计算机视觉与目标检测相关研究，尤其适用于视觉挑战较强的浑浊河流环境中的漂浮垃圾与河流植被检测任务。