Anusha Manur

This study focuses on developing a solar forecasting model to address the challenge of daily and seasonal variation in solar irradiance and load usage in rural solar home systems.

• The forecasting methodology accounts for variation in weather data as well as battery utilization, leading to more optimal energy usage.
• Performed analysis of multivariate time series field data from real homes power and weather data leading to normalized RMSE of 0.11 in the predictions.

This project uses machine learning techniques to better predict, analyze and track land use and land cover changes in the Indian subcontinent using satellite remote sensing data from Indian Space Research Organization (ISRO) to develop a thematic map depicting land cover.

• Developed a machine learning model to predict land cover change using satellite images for segmentation and classification of terrain using deep learning techniques using OpenCV, Keras, and Python.
• Collaborated with ISRO scientists to optimize model with 20% higher accuracy than statistical methods, reported analytical insights, and designed a dashboard for easy processing and visualization of large-scale, high-dimensional data.
• Confirmed hypothesis of land change due to seasonal variation and wrote Python scripts to automate data cleaning tasks (report) (code).

The aim of this work is to show how a convolutional neural network can be applied to remote sensing multispectral images to achieve segmentation. This exploration uses National Remote Sensing Centre, India (NRSC), obtained remotely sensed data.

• The ROIs (regions of interest) selected, comprise of four classes i.e. water bodies, land, forest, and snow
• Unsupervied techniques like KMeans were used for segmentation but gave poor results.
• Various design choices of the CNN architecture were analyzed. An accuracy of 92% for segmentation using labelled input data was achieved which shows that CNNs are a viable tool for solving segmentation tasks in the area of remote sensing (report).

The project involved developing android applications for remote sensing applicatons.

• The application MapCam, was mainly used to enable control of a remote drone by a user and act as a sensor and data acquisition entity for aerial mapping by sensor data logging, capturing high resolution images using a smartphone, geotagging and appending camera calibration information to the images.
• MapCam was developed as a cost efficient solution to incorporate sensor and image acquisition features to a drone, using Android Studio with Java.
• Led a cross functional team of 3, which was responsible for complete lifecycle development of the application, from initial requirement gathering to design, coding, testing, documentation and implementation (presentation).