Adaptive mayfly optimization-based denoising for enhanced edge and structure preservation in remote sensing images

Abstract Remote Sensing (RS) imagery has greatly benefited from advances in spatial and spectral resolution, but it remains susceptible to sensor limitations, atmospheric interference, and transmission errors. With traditional denoising methods, these distortions lead to over-smoothing and obscure edge and texture details. To address this, the proposed Adaptive Mayfly Optimization Algorithm (AMOA) tunes the parameters of three spatial filters such as median, bilateral and Modified Decision-Based Unsymmetric Trimmed Median Filter (MDBUTMF), based on each image’s specific noise characteristics. Conventional Particle Swarm Optimization (PSO) and Mayfly Optimization Algorithms (MOA) rely on fixed search strategies. The proposed AMOA introduces dynamic control parameters such as the Attraction Coefficient ( β ) and Decay Rate ( γ ), to improve convergence and balance exploration and exploitation. By adaptively adjusting the kernel size ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>T</mml:mi> </mml:math> ), spatial diameter ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>d</mml:mi> </mml:math> ), intensity similarity ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>σ</mml:mi> </mml:mrow> <mml:mrow> <mml:mi>c</mml:mi> </mml:mrow> </mml:msub> </mml:math> ) and spatial proximity ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>σ</mml:mi> </mml:mrow> <mml:mrow> <mml:mi>s</mml:mi> </mml:mrow> </mml:msub> </mml:math> ), the framework further enhances the denoising while preserving the image’s key features. Validation was conducted on three diverse RS datasets - LANDSAT 9, UCMerced and WHU-RS19. Across both quantitative and qualitative evaluations using PSNR, SSIM, MSE, VIF and GMSD, AMOA consistently outperforms PSO and MOA. In particular, the MDBUTMF-AMOA experiments achieved a peak PSNR of 42.41 dB and a minimum GMSD of 0.04, alongside a pixel-level accuracy of 99.97% and a reconstruction error rate of just 0.0216%, demonstrating near-perfect denoising performance. This indicates superior structural preservation and visual quality. These findings indicate superior structural preservation and visual quality, demonstrating AMOA’s potential for downstream tasks such as classification, feature extraction. and object detection.