Agriculture Land Analysis System Using GIS And Machine Learning

Project Title

Agriculture Land Analysis System Using GIS And Machine Learning

Project Area of Specialization Artificial IntelligenceProject Summary

Agriculture land segmentation is critical in the fields of agricultural insurance and digital agriculture, and it is becoming increasingly significant. Precision smallholder farmland segmentation and mapping are required to correctly accomplish insured crop area and disaster loss estimation.  One of the most significant advantages are categorizing detecting trees, farms, crops, agricultural land, developed land, vacant plots, and so on is the examination of agricultural land on a large scale. Agricultural yield, farming area, and annual agricultural production are all computed. This will also help the government manage the illegal housing society.  Deep learning technology has proven its worth by outperforming state-of-the-art alternatives in a variety of disciplines. We use Google Earth Pro for taking pictures. After taking pictures we annotate and label them with specific classes. Then through deep learning, we train a model of CNN-based instance segmentation model. That learn different features from annotated satellite data sets. A deep learning  Prediction model is deployed using a Web-based front-end application.

Project Objectives

A web-based application for analyzing any agricultural land in the country. Training a CNN-based instance segmentation model for the prediction of different land types like Agri land, commercial land, empty plot, residential land, etc. Satellite imaginary data gathering and image annotations. Detailed analysis system for yearly yield estimation. The necessary deep learning basics to provide an understanding of the Instance segmentation task. It establishes the concepts of neural networks, image recognition via convolutional neural networks (CNN), semantic segmentation via fully convolutional neural networks (FCN), and object detection. Building on top of these concepts, we presents the functionality and state of the art of deep learning instance segmentation.

Project Implementation Method

Our project agricultural land analysis system is based on several steps, which are mentioned below:

1. Satellite images accessibility: We use google earth pro satellite imaginary dataset, scraped using automation built based on python. We made an auto GUI controlling program for automatically scraping specific coordinates of current and past satellite images. 

2. Satellite image annotations: For training, the deep learning segmentation model annotated images are required. We annotated satellite images with a polygon for specific objects like agricultural land, residential land, trees farms, and road. The polygons are saved using coco JSON annotation format.

3. Instance segmentation model: We train the instance segmentation model based on Resnet-101 architecture using annotated images. Model is built based on Keras algorithm in python, trained using adam optimizer, and save as hdf5 format.

4. Web Application: We built flask based front-end web application for an interactive user interface. The trained model is deployed in the backend connected with frontend inputs as flask API. Front-end web applications provide access to input geo-coordinates of any area in square and polygon format. The web application also shows google API-based interactive maps and overlay detected segmentation class categories. 

Benefits of the Project

One of the most significant advantages are categorizing detecting trees, farms, crops, agricultural land, developed land, vacant plots, and so on is the examination of agricultural land on a large scale. Agricultural yield, farming area, and annual agricultural production are all computed. This will also help the government manage the illegal housing society.

Technical Details of Final Deliverable

Our project agricultural land analysis system is a web-based frontend application 

Accessible from anywhere using the project website. The overall project deliverables include following items:

1. Deployed frontend web application on a server accessible through URL. Application is based on python flask interface.

2. Keras hdf5 saved segmentation model deployed on backend server connected to frontend using flask API.

Final Deliverable of the Project Software SystemCore Industry AgricultureOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Zero Hunger, Good Health and Well-Being for People, Sustainable Cities and Communities, Responsible Consumption and Production, Climate ActionRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	68450
Google Maps Api	Equipment	1	18500	18500
Amazon EC2 Server(gpu t4)	Equipment	1	27750	27750
Website domain and hosting (hostinger)	Equipment	1	22200	22200