My ongoing research focuses on defect detection in solar panels. By utilizing multiple modalities, my research aims to address the complexities of identifying various types of defects, including cracks, hotspots, delamination, corrosion, and other structural anomalies that affect the performance and longevity of solar panels.
During my undergraduate studies, I worked on data mining and tuning conventional machine learning algorithms to predict some target class with the highest accuracy. This also led me to creating an application that could automate this process to a certain extent. Afterwards, I shifted my focus from text based datasets to image based ones, working on enabling low powered systems to make certain estimations based on color space.
Profile: Google Scholar
[ResearchGate] [IEEEXplore] [Code] [Video]
This work proposes a generic model with low computational overhead, employing GrabCut, k-Means Clustering and kNN which when implemented on a Raspberry Pi 3 Model B can predict the ripeness of any fruit in well under a second with 85% accuracy.
[DSpace]
This work involves metadata of around 5,000 apps, comprising over 170,000 reviews scraped using a custom script to analyze patterns associated with specific ratings. Moreover, a success model was defined based on specific features and predicted with 85% accuracy.