Satellite Image Classification by Hybrid Optimization Assisted Hybrid Classifier

Satellite image classification is the most important process for the automated analysis and identification of patterns in satellite information. This becomes challenging because the noises in satellite images will lead to misclassification and hence decrease the classification accuracy. Hence, need more study on attaining accurate classification outcomes. This study introduces a new satellite image classification model following the steps of pre-processing, image augmentation, extraction of features, feature selection, and classification. Initially, the noise is removed by the nonlocal means denoising (NLM) method. Afterwards, to enhance system performance, an image augmentation process takes place. From the noise-free (pre-processed) and augmented images, normalized difference vegetation index (NDVI), Kauth–Thomas tasseled cap (KTC), normalized difference indices (NDIs), and enhanced vegetation index (EVI)-based features are extracted. From these extracted features, more suitable features are chosen via an improved chi-square-based feature selection process. Subsequently, chosen features are subjected to the classification process. Here, we follow the hybrid classification process that combines the improved long short-term memory (LSTM) and optimized deep belief network (DBN). The weight tuning of the DBN classifier is done by the new training algorithm termed as chimp position updated bald eagle search (CPUBES) algorithm. This optimal training process ensures betterment in the classification outcome. Finally, the effectiveness of the CPUBES-based hybrid classifier (CPUBES-HC) is assessed by evaluating its performance over the conventional methods. The results show that the CPUBES model has an accuracy of 93.59%, making it superior to more conventional methods such as [Formula: see text]%, [Formula: see text]%, [Formula: see text]%, [Formula: see text]%, modified [Formula: see text]%, and [Formula: see text]%, respectively. In contrast to SVM, DT, DBN, and LSTM.