Website using APIs for Object Detection and Classification Images from Satellite Imagery using with a NoSQL database in the backend

Object Detection of tall objects such as Minerates, Radio Towers (Cellular, TV, Radio, ... etc), roads and tracks, aircrafts, cars is required to be undertaken. Segmentation of areas into built-up areas, water bodies, roads, forests, baren lands, Open shrubs, residential, industrial, commercial, vegetation, agricultural fields etc. You may use two models for achieving the same. Segmented images and Bounded Boxed images are required to be stored as a layer on a copy of the original satellite image.

Inputs:

1. For the projects Datapoints on google earth in the form of kml file can be shared if required.

2. Data freely available can be utilised.

3. DSTL - Kaggle ([login to view URL])

4. Draper Satellite Image Chronology - Kaggle ([login to view URL])

5. Understanding the Amazon from Space - Kaggle ([login to view URL])

6. Ships in Satellite Imagery - Kaggle - From Open California dataset. ([login to view URL])

7. 2D Semantic Labeling - Vaihingen data - This is from a true orthophotographic survey (from a plane). The data is 9cm resolution! ([login to view URL])

8. SAT-4 and SAT-6 airborne datasets - Very high res (1m) whole of US. 65 Terabytes. (DeepSat) 1m/px. (2015). SAT-4: barren land, trees, grassland, other. SAT-6: barren land, trees, grassland, roads, buildings, water bodies. (RGB + NIR)

([login to view URL]~saikat/deepsat/)

9. The Rareplanes Dataset ([login to view URL])

9. Any other dataset deemed fit and available in the open-source or your own.

Pre-Trained Models:

Any open-source pre-trained model may be used to avoid starting from scratch.

Language of programming: Python and/or Jupyter Notebooks. Complete installable files on Ubuntu platform 18.04 or 20.04.

Steps of work:

A complete pipeline of project is required. From data collection stage, Training etc.

Notebook 1 Creating Dataset (Chips) using kml File in case of Google earth imagery as input. Converting the data annotation formats used for testing.

Notebook 2 Dataset enhancement using GANs.

Notebook 3 Training of Deep Neural Network on Dataset. Validation and testing. This should have the facility to use data generated in its own database.

Notebook 4 Input pipeline to detect objects from given images in a directory. The output should be a bounding boxed image as a layer and an image. With a Json file of objects detected and a .csv file.

User should have the option to check which objects he is interested in detecting in the web application. You may follow Yolt-4 ([login to view URL]) or Laika

project ([login to view URL]) or any other algorithm.

Text File describing all steps neatly for installation and demonstration on a standalone PC is required to be given.

Notebook 5 Creation of a text report of objects detected using json file.

The API-based web applications should not be on bootstraps or web dependant. The application should store all the processed images into NoSQL along with detected objects and segmentation areas.

The user should be able to search using his image file name or date of processing of the image.

Any paper or open-source work including data used should be referenced in detail. Any project adjustments may be discussed before implementation. Fast and efficient implementation of open-source code is expected.

Inputs for use:

The inputs i.e. images on which inference should be carried out is on a multiband or single band satellite image.

If satellite image is not of desired resolution provision for enhancing the same need to be provided to user during input stage.

Finally the entire work should be easy to install and work on offline machine.

Python Software Architecture Java Data Mining Machine Learning (ML)

Project ID: #31058770

About the project