Skip to content

Object Detection on Satellite Imaging for Sustainable Water Harvesting Placements in Maasai Region

Roshan Taneja & Yuvraj Taneja

Sacred Heart Preparatory School, Atherton

Background and Challenge

  • What is the Maasai tribe?
  • What problem are they facing?
  • Impact of water harvesting on the community

note: * The Maasai tribe is a semi-nomadic ethnic group in Northern Tanzania and Kenya. * They face severe water scarcity due to unreliable sources exacerbated by climate change. * Implementing water harvesting has significantly improved their daily lives, reducing water collection time and enhancing public health.

Goal and Methodology

  • What is a boma?
  • Our approach

note: * A boma is a homestead with an outer barrier, shelters, and a smaller circle to hold cattle, housing 10-50 people. * We aim to map boma locations using satellite images and machine learning for optimal water harvesting unit placements.

Data Collection for Detection

  • 2000 photos collected and processed
  • 2000 non-boma landscape photos for confirmation

note: * Photos were cropped, labeled, and multiplied to enhance the dataset. * Ensuring a balanced dataset between boma and non-boma images is crucial for accuracy.

Training the Model

  • Google Earth Engine for plotting bomas
  • Training process

note: * We used the Google Earth Engine to recognize and plot bomas over a set area. * The model was trained with augmented data to improve detection accuracy.

Results and Accuracy

  • Model accuracy and findings

note: * Initial accuracy was 30% with 2000 boma photos. * Increased to 93% after augmenting data to 8000 boma photos. * Model marked coordinates with over 80% confidence in a 250 sq. mile test area.

Conclusions

  • Impact and future applications

note: * AI-driven water resource planning can significantly improve water accessibility for communities like the Maasai. * Future expansion of this approach can aid in sustainable water management across various regions.

Challenges in the Monduli District

  • Rainy and dry seasons' impact on water sources like Nanja Dam

note: * Nanja Dam, a crucial water source, dries up completely by mid-July, posing a severe challenge for the Maasai community. * The need for sustainable water solutions is critical for their livelihood.

Day in the Life Before and After Water Harvesting Units

  • Activities and improvements in daily life

note: * Water collection time reduced from 9 to 2-3 hours. * Increased time for economic, social, and agricultural activities. * Significant improvement in public health and education.

Techniques Considered for Computer Vision

  • YOLOv7, OpenCV, TensorFlow
  • Pros and cons of each

note: * YOLOv7 is optimized for speed and accuracy but has integration challenges. * OpenCV is accessible and compatible but resource-intensive. * TensorFlow is highly customizable but complex.

Data Augmentation and Model Training

  • Enhanced dataset and training layers

note: * Resized images, convolution, and pooling layers are essential for accurate detection. * Enhanced dataset with superimposed boma photos significantly improved model accuracy.

Mapping and Validation

  • Green dots indicating Maasai living units

note: * The map shows precise locations for water unit placements. * Point density calculations help in strategic deployment for maximum benefit.

Key Observations

  • Living patterns and strategic placement

note: * Dense communities around geological formations and highways. * AI analysis reveals optimal locations for water units, benefiting large populations.

Conclusion and Future Work

  • Expanding AI applications for water resource management

note: * Leveraging AI for water planning ensures sustainable solutions. * The approach can be scaled to other regions facing similar challenges, promoting efficient resource allocation.