Tuning CNN Hyperparameters for Satellite Image Segmentation: A Grid Search Strategy for Enhanced Performance

• Author(s): Aditya Kinnori ; Aniket Tripathi
• Paper ID: 1709000
• Page: 164-173
• Published Date: 31-12-2020
• Published In: Iconic Research And Engineering Journals
• Publisher: IRE Journals
• e-ISSN: 2456-8880
• Volume/Issue: Volume 4 Issue 6 December-2020

Abstract

Applications that rely on satellite data such as land use classification, environmental monitoring, and urban planning—can significantly benefit from accurate satellite image segmentation. Convolutional Neural Networks (CNNs) are especially effective for this task, as they can learn complex spatial feature hierarchies from high-resolution imagery. However, their performance is highly dependent on the careful tuning of hyperparameters such as learning rate, batch size, filter size, dropout rate, and the number of convolutional layers. Poorly chosen hyperparameters can lead to underperforming models that fail to generalize to new data. This study explores how grid search-based hyperparameter tuning can optimize CNN performance for satellite image segmentation. By systematically evaluating different CNN configurations on a high-resolution satellite dataset, the research identifies optimal parameter settings that enhance segmentation quality. Key metrics such as Intersection over Union (IoU), F1-score, and pixel accuracy were used to assess each configuration’s effectiveness. Results show that even minor adjustments in hyperparameters can lead to significant improvements in segmentation accuracy. The grid search method not only helped eliminate weak configurations early but also led to more robust models. Importantly, the findings highlight the value of domain-specific tuning over relying on generic or default settings. This paper presents a replicable approach for practitioners looking to enhance segmentation accuracy in satellite image analysis, supporting use cases like agricultural monitoring, disaster assessment, and climate research.

Keywords

Satellite Image Segmentation, CNN Hyperparameter Tuning, Grid Search Optimization, Remote Sensing Applications, Deep Learning in Geospatial Analysis

Citations

IRE Journals:
Aditya Kinnori , Aniket Tripathi "Tuning CNN Hyperparameters for Satellite Image Segmentation: A Grid Search Strategy for Enhanced Performance" Iconic Research And Engineering Journals Volume 4 Issue 6 2020 Page 164-173

IEEE:
Aditya Kinnori , Aniket Tripathi "Tuning CNN Hyperparameters for Satellite Image Segmentation: A Grid Search Strategy for Enhanced Performance" Iconic Research And Engineering Journals, 4(6)