Enhancing Remote Sensing Image Classification using Horse Herd Optimization with Deep Transfer Learning Model

The classification of aerial image is very important in the remote sensing community, and several studies are performed in recent years. In these studies, the majority of them concentrate on classifying an image as to one semantic label, but in real life, an aerial image can frequently connect to several labels, for example, multi-object level labels. In addition, a widespread image of existing objects in providing high resolution aerial images offers further in-depth knowledge of the area of study. Therefore, aerial image multi-label classification gains an appeal-improving attention. In this aspect, this research develops an Automated Remote Sensing Image Classification by employing Horse Herd Optimization with Deep Transfer Learning (ARSIC-HHODTL) model. The goal of the ARSIC-HHODTL model is to exploit a hyperparameter-tuned DL model for the recognition of aerial images into distinct classes. At the initial step, the ARSIC-HHODTL technique uses a bilateral filtering method for noise removal. Next, the features in the preprocessed images can be extracted utilizing the EfficientNet-B7 method. For the classification of aerial images, a Long Short-Term Memory (LSTM) approach has been executed. Finally, the HHO method can boost the performance of the LSTM method. The promising performance of the ARSIC-HHODTL technique has been validated on standard datasets. The simulation values exhibit a superior accuracy output of 94.65% over the existing techniques.