Omdena collaboration projects(Part-2): My Omdena experience
Introduction
This article contains a summary of the latest Omdena project I and other collaborators participated in. It explains the problem, collaboration structure, contribution and knowledge gained.
Project name: Developing an AI Model to Identify School Locations in Sudan from Satellite Imagery.
The Problem
Accurate data about school locations is critical to Giga, a joint initiative by UNICEF and ITU aimed at connecting the unconnected schools in the world to the internet. This will help bridge the digital divide in the world. However, for many countries, school locations records are often inaccurate, incomplete, or non-existent. Traditional methods of the field visit and mapping of the school locations are not only heavily expensive but some schools are located in remote, inaccessible, and insecurity-prone areas.
Therefore, the mission of this AI project is to develop a Deep Learning model(s) to accurately and comprehensively identify school locations in Sudan from high-resolution satellite imagery.
Dataset description: Schools and non school satelite images of sudan
Link to the project details: https://omdena.com/projects/omdenalorexgiga/
Collaboration/solution structure
Divided into six tasks
data preparation: involved the data labelling task where the dataset provided were annotated and labelled with a tool called label box. It was used to determine the region of interests.
data inferencing: This involved some EDA tasks with python and powerbi. we also discovered school patterns and important information in the dataset.
data preprocessing: this involved data augmentation. Then the labelled images were exported to COCO format. tfRecords was also used to store augmented images.
modelling: This involved building the ML model to classify the images. Using VGG, Custom Unet, YOLO v5, Efficient Det modelling
testing: a relatively new field was explored. The ML code for the pipeline was tested.
super resolution: GF GANs was used to restore the low quality satelite image by enhancing colors and removing noise.
My contributions.
I contributed in the data inferencing task by performing EDA, also in the super resolution task which involved using GF GANs.
Knowledge gained:
I learnt labelling concepts, also in modelling tasks. and across all the task i participated in.
Link to project presentation by collaborators: https://www.linkedin.com/posts/oluyede-segun-adedeji-jr-a5550b167_ai-activity-6892838081477644288-zQrQ
Conclusion
I have learnt a lot from this project which took place over eight weeks. Building scalable real life AI solutions and model was exciting. I recommend any AI enthusiast to join Omdena
Link to join omdena: https://lnkd.in/dyHkSsrH
WRITER: OLUYEDE SEGUN . A(jr)
linkedin profile: https://www.linkedin.com/in/oluyede-segun-adedeji-jr-a5550b167/
twitter profile: https://twitter.com/oluyedejun1
TAGS: #AI #ComputerVision #machinelearning #COCO #YOLO #omdena
