Download Report IEEE DataPort : MH-Weed16: An Indian Crop/Weed Dataset for Computer Vision Tasks in Precision Farming - 2025
JPG, PNG, JPEG by Sayali Shinde, Dr.Vahida Attar
Information
Format: JPG, PNG, JPEG Publisher: IEEE DataPort Publication Date of the Electronic Edition: 10/27/2025
?
ISBN: 10.21227/bx9z-8c25
$15
$3Discount Coupon
Delivery time: Maximum 24 hours
Description
MH-Weed16 is a region-specific crop–weed image dataset developed to advance computer vision research in precision farming, with a particular focus on selective weed management in Indian agricultural systems. Weeds pose a major threat to crop productivity by competing for vital resources and reducing overall yield. Conventional weed-control practices—such as manual removal or indiscriminate herbicide spraying—are labour-intensive, costly, and environmentally unsustainable. While recent advances in deep learning and computer vision have opened pathways for automated and site-specific weed detection, the progress of such technologies is constrained by a lack of representative and high-quality field datasets originating from Indian farms.The MH-Weed16 dataset comprises images of 16 prevalent weed species found in crop fields across Maharashtra, India, collected during the 2023–2024 growing seasons. The dataset captures real-world agricultural variability, including diverse illumination conditions, multiple crop and weed growth stages, heterogeneous soil backgrounds, varying weed densities, occlusions, and high inter-class visual similarity. It includes expert-validated bounding-box annotations for weed instances, supplemented with UAV-based aerial imagery to support macro-scale crop–weed mapping. All images were acquired under natural field conditions, ensuring authenticity for developing robust machine learning models for weed classification, object detection, and density estimation. MH‐Weed16 dataset offers several unique advantages including:1.High-Quality Data and Comprehensive coverage with regional relevance:MH‐Weed16 provides comprehensive coverage of 16 weed species from real fields of Maharashtra region. This regional focus plays important role for addressing local agronomic challenges. Furthermore MH‐Weed16 is enriched with detailed morphology of weeds across multiple stages of growth, useful for both robust model training and fine-grained analysis.2.Expert Validated Annotations:Images are annotated under guidance of subject matter experts, from agriculture universities institutes, and research centres. Bounding boxes are used for precisely annotating different weed species, which ensures reliable ground truth for both object detection and accurate weed area estimation.3.Standardized Data Acquisition Protocols:All images are captured from a consistent top-down view at a fixed camera height and angle. Customized ground vehicle was driven with depth camera mounted on it within real fields of Soybean. This standardization minimizes perspective distortions and enhances reproducibility, which is critical when comparing models and benchmarking performance. Also UAV was employed to facilitate rapid scanning across dense canopies and to acquire data in the areas which are inaccessible otherwise.4.Integration of RGB and RGB-D Modality:The MH‐Weed16 dataset incorporates both RGB (Red-Green-Blue) and RGB-D (Red-Green-Blue-Depth) modalities. The RGB modality captures high-quality color and texture information, which is essential for identifying visual features such as leaf color, shape, and texture. On the other hand, the RGB-D modality adds a valuable depth dimension, offering geometric insights like plant height, spatial arrangement, and the relative distance of objects in the field. By utilizing both appearance and structural information, this approach improves the robustness of computer vision models in challenging environments, such as those with varying lighting, shadows, and occlusions.5.Diverse Natural Lighting and atmospheric Conditions:Unlike many datasets that are limited to controlled or uniform lighting, MH-Weed16 includes images captured under diverse natural lighting conditions, such as shadows, color shifts, and varying atmospheric conditions (e.g., rainy, sunny). This variability provides valuable training data to enhance model adaptability in real-field environments6.Weed Species Coverage:The dataset encompasses weed species that are commonly found across multiple crop fields in Maharashtra region. This broad coverage makes the dataset highly representative and ensures that the developed models are applicable to a wide range of Indian agricultural contexts.7.Weed Area Estimation:The combination of high-resolution top-down images and precise annotations facilitates accurate weed area estimation, a crucial metric for optimizing targeted herbicide applications and improving weed management practices
$15
$3Discount Coupon
Delivery time: Maximum 24 hours