Optimizing Hyperspectral Anomaly Detection with a Scalable Quantum Snow Geese Non-Local Neural Network Algorithm

Hyperspectral Anomaly Detection (HAD) focuses on identifying unusual objects or patterns in hyperspectral imagery (HSI) that differ from the normal background or surrounding pixels. In the past two decades, anomaly detection (AD) has become essential in hyperspectral image analysis, offering a novel approach to differentiate targets from the background without requiring prior information. Current methods face multiple drawbacks, either by neglecting the problem of low resolution or by depending on super-resolution models that aim to reconstruct the entire image, which can result in ineffective detection of abnormal pixels. To address these challenges, a Scalable Quantum Snow Geese Non-local Neural Network Algorithm (ASQSGNon-LNNet) is introduced for hyperspectral anomaly detection, aimed at achieving high accuracy. In this approach, input images are sourced from the ABU-urban dataset. First, the features of these input images are extracted using the Memory Efficient Vision Transformer (ME-ViT) method. Subsequently, the extracted features undergo Hyperspectral Super Resolution using the Deform-Mamba Network. After that, anomaly detection is conducted using a Scalable Quantum Non-local Neural Network (SQNonL-NNet), optimized by the Snow Geese Algorithm (SGA), leading to automated hyperspectral anomaly detection with improved efficiency and accuracy. The performance of the proposed ASQSGNon-LNNet is assessed using the ABU-urban dataset, achieving an accuracy of 99.8% and surpassing existing methods. The results of this technique reveal unusual patterns or objects in hyperspectral images by analyzing their spectral signatures and enhancing accuracy in detecting hidden or subtle anomalies in remote sensing, environmental monitoring, and military applications. This technology provides greater accuracy than traditional methods.