Satellite image classification using Resnet_12 architecture

Author Reg. ID: TIJER_101517

Published Paper Id: TIJER2303194

Published In: Volume 10 Issue 3, March-2023

Abstract: There has been a rise in interest in the use of satellite image analysis for a range of purposes, including mineralogy, forestry, agriculture, military operations, mapping, urban planning, ocean surveillance, and disaster management. However, conventional algorithms for object detection and classification are not precise or dependable enough to tackle this problem. To automate this task, deep learning algorithms, specifically convolutional neural networks, have shown potential., it has achieved success in image understanding. The Hyperspectral images have been classified into four distinct categories, namely Cloudy, Water, Deserts, and green area. This classification has been performed on the satellite images used in the study. The process of manually classifying satellite images using image interpretation methods is a time-consuming task that requires the expertise of field professionals. Therefore, our research focuses on creating an effective automated system for satellite image classification. We plan to achieve this by utilizing a Deep Convolutional Neural Network (DCNN) model, specifically the ResNet-12 framework with 12 layers. The ResNet-12 model utilizes skip connections that merge the input with the convolutional layer's output to tackle the issue of vanishing and exploding gradients often encountered in traditional CNN models. Efficiency of all the models were measured using the metrics Accuracy & Precision. Resnet-12 model got highest accuracy of 97.3% maintaining the Precision same as other models.

Keywords: Resnet-12, Hyperspectral images, Skip connections, Deep Learning, CNN.

Downloads: 000220

Page No: 471-478

Country: viziyanagram, Andhra Pradesh, India

Research Area: Science and Technology

Published Paper URL: https://tijer.org/TIJER/viewpaperforall?paper=TIJER2303194

Published Paper PDF: https://tijer.org/TIJER/papers/TIJER2303194

Click Here to Download This Article

Article Preview