Effectivity of super resolution convolutional neural network for the enhancement of land cover classification from medium resolution satellite images

TL;DR

This paper demonstrates that using Super-Resolution Convolutional Neural Networks (SRCNN) to enhance medium-resolution satellite images significantly improves land cover classification accuracy, reducing misclassification compared to traditional interpolation methods.

Contribution

The study provides empirical evidence that SRCNN-based resolution enhancement outperforms bilinear and bicubic interpolation for satellite image analysis.

Findings

SRCNN significantly improves land cover classification accuracy.

SRCNN reduces pixel misclassification compared to traditional methods.

Enhanced resolution leads to better forest change detection.

Abstract

In the modern world, satellite images play a key role in forest management and degradation monitoring. For a precise quantification of forest land cover changes, the availability of spatially fine resolution data is a necessity. Since 1972, NASAs LANDSAT Satellites are providing terrestrial images covering every corner of the earth, which have been proved to be a highly useful resource for terrestrial change analysis and have been used in numerous other sectors. However, freely accessible satellite images are, generally, of medium to low resolution which is a major hindrance to the precision of the analysis. Hence, we performed a comprehensive study to prove our point that, enhancement of resolution by Super-Resolution Convolutional Neural Network (SRCNN) will lessen the chance of misclassification of pixels, even under the established recognition methods. We tested the method on original LANDSAT-7 images of different regions of Sundarbans and their upscaled versions which were produced by bilinear interpolation, bicubic interpolation, and SRCNN respectively and it was discovered that SRCNN outperforms the others by a significant amount.