Analyzing Convolutional Neural Networks as feature extractors for Video Regression
Siddhant Jain, Rushikesh Ghotekar, Amit Dawande, Aarti Pawar, Anup Ingle, Vijay M. Marathe · 2023
This literature study and experimentation aim to study the current best performing architectures in the domain of image classification. In particular, we investigate the approaches for characterising the Convolutional Neural Network (CNN) architectures for extracting features from frames for our task of video regression. Weather Forecasting and wind speed are important factors for many economic, business and management sectors. Generalizability was successfully demonstrated by the network by accurately predicting based on recorded wind speeds based from other flags in different geographical locations and a controlled wind tunnel test. The accurate measurement of wind speed helps in assisting the prediction of weather fore-cast conditions. In this review, we undertake a comprehensive comparison of various CNN architectures and present a novel and accurate prediction model for wind speed prediction. We specifically explore existing methodology under image classification architectures, data sources, pre-processing methods, for extraction of features, and the variety of evaluation metrics and approaches. The relevance of extracting characteristics is highlighted in this review from individual frames to capture a better representation of information, including the difficulties that the researchers confront during the procedure and the research gaps that exist in this subject.