Synthetic Aperture Radar Image Fusion Using Firefly Algorithm: A Comparative Study

Abstract

Picture fusion of SAR is an important process in improving remote sensing images in all weather conditions by merging complimentary information of several SAR images. This paper compares three fusion techniques: Fuzzy Logic based fusion, Firefly Algorithm based optimised fusion, and DL based fusion, and evaluates the performances of the three fusion techniques on three datasets of SAR (Sentinel-1 C-band, ALOS PALSAR L-band and UAVSAR quad-polarimetric). As the primary recommended approach to automatically optimise fusion parameters to maximise the degree of similarity of the structure, information preservation and edge clarity, the Firefly Algorithm, a nature-inspired metaheuristic optimisation methodology is examined. The experimental evaluation is carried out on six metrics of performance which are complementary in nature. They include the following: Entropy, Spatial Frequency, Average Gradient, Mutual Information, SSIM and PSNR. The results show that Firefly Algorithm performs the highest overall fusion performance with the highest Spatial Frequency (36.52), Average Gradient (15.71), SSIM (0.601), PSNR (15.91 dB) and composite quality score (57.63). Therefore it is ideal for preserving the structure and retaining edges details. Fuzzy Logic has competitively evaluated results with the highest Entropy (7.647), Mutual Information (2.603) and the fastest execution time (0.1807 seconds). DL requires a high quality and quantity of training data, which leads to less effective metrics but can be enhanced with bigger datasets. It has been investigated through Firefly Algorithm that the fitness is improved in a stable manner after 5th iteration which proves the efficient optimisation for the application of SAR image fusion. Keywords: Synthetic Aperture Radar (SAR), Image Fusion, Firefly Algorithm (FA), Fuzzy Logic, Deep Learning, Optimization, Performance Metrics, Remote Sensing