Topic of Research Seminar: Towards Automatic weather Classification Using DCNNs

Abstract: Remote sensing (RS) technology has led to the widespread availability of a substantial volume of satellite imaging data. To ensure the effective implementation of the RS in practical situations, it is necessary to develop efficient and scalable solutions that can be applied in various interdisciplinary fields. To achieve the objective of rapid analysis and accurate categorization in the RS imaging, deep Convolution Neural Networks (CNNs) are commonly employed. This study presents a distinctive residual network called ResNet101, which is based on snapshot images. The network includes fully connected layers (FC-1024), dropout layers, a dense layer, and data augmentation techniques. The inter-class similarity problem is addressed by implementing architectural improvements, whereas imbalanced classes are tackled through the use of data augmentation. The ResNet101 model utilizes the demanding Large-Scale Cloud photos Dataset for Meteorology Research (LSCIDMR), which consists of 10 classes including numerous high-resolution photos. The model’s objective is to accurately categorize these images into their appropriate classes. Our developed model surpasses the performance of several recently reported deep learning algorithms in terms of Precision, Accuracy, and F1 scores.

Subject field of Topic: Machine Learning/Deep Learning

Name of Speaker: Mattia Tun Nabi

Professional Rank of Speaker: MS Student

University Email of Speaker: [email protected]

Affiliation of Speaker: School of Mechanical and Manufacturing Engineering (SMME), NUST

Date and Venue: 30 July 2024, 1430 – 1530, SMME Classroom, School of Mechanical and Manufacturing Engineering, NUST Islamabad