农田杂草分割多光谱-RGB融合图像数据集

数据采集： 数据集通过农田环境的多光谱成像设备与可见光摄像设备协同采集，同时采集对应的地面实拍 RGB 图像。通过严格校准与配准流程，保证多模态图像的空间对应关系。 数据清洗与标注： 所有图像由专业人员完成像素级杂草区域标注，标注内容包含杂草、作物、土壤等类别。通过质量审核机制过滤模糊、不清晰或不合格样本，提高标注准确性。 算法构建流程： （1）通过独立的 RGB 与多光谱分支提取纹理、光谱等多维度特征； （2）利用特征融合模块（包括加权融合、注意力机制或 Transformer/Mamba 融合结构）合成互补信息； （3）采用多尺度解码器或上采样结构，实现杂草区域的像素级分割； （4）模型训练过程中使用交叉熵、Dice loss 等组合损失函数，提高杂草分割精度。 模型应用： 输出可用于杂草检测、区域统计、精准施药路径生成及农田精细化管理，为相关决策提供技术依据。Data Collection: The dataset is collaboratively collected using multispectral imaging equipment and visible light camera devices in farmland environments, with corresponding on-site ground RGB images captured simultaneously. A strict calibration and registration process is implemented to ensure the spatial correspondence of multimodal images. Data Cleaning and Annotation: All images are annotated with pixel-level weed regions by professional personnel, covering categories such as weeds, crops, soil and others. A quality review mechanism is adopted to filter blurry, unclear or unqualified samples, thereby improving annotation accuracy. Algorithm Construction Pipeline: (1) Extract multi-dimensional features such as texture and spectrum via separate RGB and multispectral branches; (2) Synthesize complementary information using feature fusion modules, including weighted fusion, attention mechanisms, or Transformer/Mamba-based fusion architectures; (3) Adopt multi-scale decoders or upsampling structures to achieve pixel-level segmentation of weed regions; (4) Use combined loss functions such as cross-entropy and Dice loss during model training to enhance the accuracy of weed segmentation. Model Application: The output can be applied to weed detection, regional statistics, precise pesticide application path generation and refined farmland management, providing technical basis for relevant decision-making.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于农田杂草分割的多光谱-RGB融合图像数据集，包含1404张图像，通过多光谱和可见光设备协同采集，并经过专业像素级标注。其特点在于融合多模态信息以提升杂草识别精度，主要应用于精准农业、智能管理和科研教育等领域。