Drone-Based Shrub and Grass Monitoring Dataset
Description
This dataset contains drone-captured aerial images of shrubs, grass, and surrounding vegetation collected from the campus area of Daffodil International University, located in Savar, Dhaka, Bangladesh, Asia. The data was acquired using low-altitude drone footage and later converted into high-resolution image frames to support computer vision and deep learning research. The primary objective of this dataset is to enable analysis of environmental impacts on shrub and grass life cycles using deep learning techniques. The images capture natural variations in vegetation structure, density, color, and growth patterns influenced by environmental factors such as sunlight exposure, soil conditions, seasonal changes, and human activity. The dataset is suitable for a wide range of tasks, including vegetation classification, object detection, segmentation, lifecycle analysis, and environmental monitoring. It can be used to train and evaluate deep learning models such as convolutional neural networks for understanding plant health, growth stages, and stress indicators from aerial imagery. All images were extracted from drone videos to ensure diversity in viewing angles, scales, and spatial context. Minimal preprocessing was applied to preserve real-world conditions and visual complexity. This makes the dataset particularly useful for developing robust models that can generalize to real environmental scenarios. This dataset is intended for academic research, environmental studies, and machine learning experimentation. Researchers can use it to explore sustainable environmental monitoring approaches, assess vegetation dynamics, and develop data-driven solutions for mitigating environmental impacts on shrub and grass ecosystems.
Files
Steps to reproduce
The dataset was developed using aerial imagery collected from green areas within the campus of Daffodil International University, located in Savar, Dhaka, Bangladesh. The site was selected due to its diverse presence of shrubs, grass, and mixed vegetation exposed to natural and human-influenced environmental conditions. Data acquisition was performed using a consumer-grade unmanned aerial vehicle equipped with an RGB camera. Drone flights were conducted during daylight hours under stable weather conditions to ensure uniform lighting and minimize motion artifacts. Low-altitude flights were used to capture fine-grained vegetation details. Both stationary hovering and slow, linear flight paths were employed to introduce variation in scale, angle, and spatial context. Video footage was recorded in high definition and later converted into still images for analysis. Frame extraction was carried out at fixed time intervals using open-source video processing tools such as FFmpeg and OpenCV. This approach reduced redundancy while preserving sufficient visual diversity across samples. Following extraction, images were manually inspected to remove blurred frames, excessive shadows, or non-vegetation scenes. Minimal preprocessing was applied, limited to resolution normalization and format standardization, in order to retain real-world environmental characteristics. The final dataset was organized into structured directories based on vegetation type and capture conditions. When available, metadata such as capture location and acquisition time was retained to support downstream analysis. This workflow can be reproduced by selecting a vegetation-rich area, capturing low-altitude drone video under consistent lighting, extracting frames at regular intervals, applying quality control, and organizing the images for deep learning experiments.
Institutions
- Daffodil International University