Research & Projects

Project 1

A Comparative Study of Fuzzy Machine Learning Techniques for avalanche debris detection in the Indian Himalayas

Project 2

Optimized Machine Learning Model for Natural Rubber Crop Mapping with Optical Satellite Data

Rubber (Hevea brasiliensis) is a tropical cash crop cultivated for latex production. The trees are typically grown in plantations and are long-lived, perennial, and tall. The majority of the plantation area is owned by smallholders in most of the rubber-producing countries. Monitoring the extent and age of rubber plantations are done through traditional methods in most countries and is a difficult process. The issues associated with monitoring traditional practices can be overcome by using remote sensing techniques. A challenge in the use of remote sensing for rubber crop mapping is the similarity of the rubber crop with forest cover. In this study, satellite image-derived spectral indices like Normalised Difference Vegetation Index (NDVI) and Normalised Difference Moisture Index (NDMI) along with machine learning are used to overcome the challenge of misclassification. Random Forest, Decision Tree, Support Vector Machine, and Gradient Boosting classifiers are used for binary classification of rubber and forest in the parts of Malappuram District. The study produces high accuracy of 81% for the Random Forest classifier and efficiently classifies Rubber and Forest.

Project 3

Quantum Computing enhanced Machine Learning for Satellite Image Classification

Project 4

Experimental Study of Raster and Vector in-memory processing

Apache Sedona (previously GeoSpark) is an in-memory cluster computing system that aids the processing of large-scale spatial data, which is an extension of Apache Spark to support spatial data types, indexing, operations, and partitioning techniques. Apache Sedona framework effectively handles big geodata and can be used for various location-based services. K Nearest Neighbor Query (kNN Query) Sedona offers an exciting opportunity to cluster point layers in Apache. Three nearest vaccination centers of a location are found using kNN Query in order to understand the vector handling capacity of the framework. Similarly, raster processing is done using various spectral indices. Spectral indices derived from remotely sensed imagery helps in quantitative and qualitative analysis of land use and land cover. These indices are widely used in remote sensing applications using different satellite and airborne sensors. Landsat, Sentinel, Resourcesat, and other freely available satellites with high spatial and temporal resolution enable cost-effective extraction of time series data consistently. With the wider use of the internet, the scientific community today is more inclined towards open-source software that has wide community support from developers, engineers, and scientists. In the present study, the Apache Sedona framework is used to calculate various spectral indices like NDVI, NDWI, etc. from Landsat 8 satellite data and is compared with traditional methods

Project 5

GIS Approach for Rubber Crop Yield Modeling using Machine Learning: A Case Study in Kerala

Rubber is an important cash crop in the tropics. It is exponentially expanding in both optimal and sub-optimal climatic zones in tropical countries. Kerala, the leading producer of rubber in India, has seen rapid growth in its plantation area. The production of rubber is highly dependent on various parameters like temperature, rainfall, humidity, etc. A project is done to model the relationship between rubber crop production and various parameters like area of production, rainfall and temperature as well as various indices like NDVI, NDWI, etc. The rubber production and cultivation area of Kerala are obtained from Farm Guide publication and Agricultural Report published by the Government of Kerala and various research papers. Rainfall and temperature data is obtained from IMD Pune. Various machine learning algorithms such as RandomForestRegressor, GradientBoostingRegressor, DecisionTreeRegressor, and Support Vector Machine is implemented in the data and the R2 for each algorithm is found out. GradientBoostingRegressor is found out to have the highest accuracy with 0.9391, followed by DecisionTreeRegressor (0.9052), RandomForestRegressor (0.9024), and Support Vector Machine (-0.0005).

Page updated

Google Sites

Report abuse