Satellite image classification using HOG and DAISY Feature Descriptors

Advancements in remote sensing techniques provide crucial information for a variety of applications, including landscape changes, land cover categorization, enhanced weather forecasting, and climate observation. With the recent breakthroughs that have made satellites both more powerful and cheaper to deploy, the volume of high-resolution satellite images collected has increased exponentially. Manual classification techniques of these high-resolution satellite images are too inaccurate and inefficient to handle the challenge. Hence, automation of satellite image classification has become a crucial part of the field of remote sensing. The major purpose of this work is to evaluate the ability of a feature extraction model to identify satellite pictures utilizing different approaches. This study presents actual data demonstrating that the Bag of Feature (BoF) approach outperforms conventional feature extraction algorithms for multilabel scene classification. We employ local feature descriptors such as DAISY features for local feature extraction and classification, and global feature descriptors such as the histogram of oriented gradients (HOG) for global feature extraction and classification, using the Bag of Features technique. The bag of feature encoding is augmented by the Mini-Batch K-Means method to decrease the complications of the feature encoding technique. A method for pooling is applied to combine DAISY and HOG data relevant to each image. Finally, we perform a 10-fold cross-validation experiment with a support vector machine (SVM) classifier and classify our results on a dataset of 21 scene categories. The hybrid model attained an accuracy of 81.76% and the KNN model attained an accuracy of 67.14%. The study demonstrates how this categorization method may be used to identify changes in land use and land cover, as well as aid in the improvement of satellite classification.