Deep Hybrid Model to Classify Satellite Image Using Vegetation Indices Feature Set

A variety of information about the region of interest can be found in satellite images through multispectral and panchromatic elements. For several purposes, including mapping of vegetation and crop, accurate classification of satellite imagery is necessary. Hybrid classifiers, providing techniques to combine and integrate many classifiers, have proven significant efficiency in classification of satellite images in recent years. In the way, this paper attempts to propose a satellite image classification model. The satellite image is first pre-processed by turning it into a grayscale image, applying a histogram to sharpen the image's edges, and eliminating any noise. Then the suitable vegetation index feature set is extracted from the preprocessed image, which includes EVI, NDVI, WDRVI and ENDVI. Finally, a hybrid model is proposed to classify the image, which is the combination of Improved CNN and LSTM that trained with the feature set. Lastly, the suggested work's efficiency is verified with regard to of sensitivity, accuracy, and specificity compared to the traditional methods.