Metaheuristic Optimization Based Deep Learning Model for Multispectral Image Classification

Abstract

Abstract Multispectral image classification has received significant attention among research communities and academicians. Owing to the difficulties (spatial, spectral, dynamic data sources, and temporal discrepancies) that exist in the online and time-series multispectral image investigation, there is a high incidence probability in dissimilarities of spectral bands from the input stream that degrades the classifier results. Recently, several artificial intelligence (AI) models can be used for the extraction of prominent features. Besides, deep learning (DL) methods become more familiar and gained interest in the remote sensing community for the classification of multispectral and hyperspectral images. With this motivation, this paper presents an automated parameter tuned deep learning enabled multispectral image classification (AHPTDL-MSIC) technique. The proposed AHPTDL-MSIC technique aims to categorize the different class labels of the multispectral images. Besides, the AHPTDL-MSIC technique applies multi-level discrete wavelet transform (DWT) based image decomposition technique. Moreover, the EfficientNet technique is applied as a feature extractor to generate a collection of features. Furthermore, the chaotic satin bowerbird optimization (CSBO) algorithm with kernel extreme learning machine (KELM) model is applied for the classification process. The application of CSBO algorithm helps to appropriately tune the class labels of the KELM model. In order to ensure the enhanced performance of the AHPTDL-MSIC technique, a wide range of simulations take place using the Madurai LISS IV multispectral images and the results are examined under several aspects. The extensive comparative study highlighted the better performance of the AHPTDL-MSIC technique over the recent methods.