Akademik Veri Yönetim Sistemi
Chaos, Solitons and Fractals, cilt.208, 2026 (SCI-Expanded, Scopus)
Spiking Neural Networks (SNNs), known as third-generation networks, associate biological neural structures with brain-like computations. In recent years, memristive synapse mechanisms have been adapted to SNN structures to meet the STDP learning rule. Memristive Spiking Neural Networks (MSNN) structures are developed and successfully applied to many problems. These MSNN structures generally include neuron models with low biological meaningfulness. The effects of using neuron models with high biological meaningfulness in MSNN structures on network performance have not been fully revealed in the literature yet. Therefore, this study aims to compare the performances of MSNNs created by using neuron models with high biological meaningfulness and Voltage ThrEshold Adaptive Memristor (VTEAM) memristor structures through an image restoration problem. For this purpose, three different basic MSNN structures are designed, each consisting of 25 × 5 neurons and using FitzHugh–Nagumo, Hindmarsh–Rose, and Morris–Lecar neuron models. Nine different memristive-based network structures are obtained by diversifying these three basic MSNNs with Joglekar, Biolek, and Prodromakis window functions. An image restoration problem is handled using each network. In this context the noisy induced images are restored with these MSNNs by using STDP rule and WTA algorithm. All performance outputs of these MSNNs such as their memductance responses, settling times or error comparisons are presented in detailed graphs. The differences between noisy input images and the restored-output images are determined by using Structural Similarity Index Masure (SSIM) for a deterministic comparison. According to these performance comparisons, two critical results are inferred as follows: i) As the biological meaningfulness of the neuron model increases, MSNN reaches the true restored output images faster, and ii) the change of the window function in the memristor element also critically affects the MSNN performance.