A novel model for hyper spectral image enhancement and classification: PCA, MBAO and CNN integration

• Published in: Optical and quantum electronics 2024-03, Vol.56 (3), Article 473
• Main Authors: Lalitha, V., Latha, B.
• Format: Article
• Language: English
• Subjects: Adaptive filters; Algorithms; Artificial neural networks; Characterization and Evaluation of Materials; Classification; Computer Communication Networks; Data collection; Electrical Engineering; Hyperspectral imaging; Image classification; Image contrast; Image enhancement; Image quality; Lasers; Optical Devices; Optics; Photonics; Physics; Physics and Astronomy; Preprocessing; Principal components analysis; Remote sensing; Spatial resolution
• ISSN: 0306-8919; 1572-817X
• DOI: 10.1007/s11082-023-06101-z

Hyperspectral images (HSI) are contiguous band images commonly used in remote sensing applications. Over the past decades, significant advancements have been made in HSI enhancement and classification. However, challenges such as computational complexity, low spatial resolution, overfitting, noisy images, misclassification, and slow convergence speed persist. So the novel HSI enhancement and classification model is proposed to address and overcome these issues. The data are collected from the different HSI datasets namely Pavia University (PU), Indian Pines (IP), Salinas Valley (SV), Houston University (HU) and Kennedy Space Center (KSC) for preprocessing the collected data. The collected data undergoes preprocessing, which includes noise reduction, geometric correction, and radiometric calibration operations to improve image quality. To reduce the high dimensionality of the preprocessed image, Principal Component Analysis (PCA) is employed. Additionally, the Mutation Boosted Aquila Optimization (MBAO) algorithm is applied to enhance the visual quality of the image. The image enhancement process relies on three types of enhancement parameters: histogram equalization, contrast stretching, and adaptive filtering. Finally, the enhanced image is classified using a Convolutional Neural Network (CNN) architecture, with weight updates performed iteratively using the Stochastic Gradient Descent (SGD) model to minimize the loss function. A comparative analysis is conducted to evaluate the superiority of the proposed model. The experiments of the proposed model achieves an F1-score of 98.2%, kappa statistics of 98.3%, SSIM of 0.99, PSNR of 38.81 dB, RMSE of 1.97, and UQI of 0.85. These findings indicate that the proposed model outperforms other existing HSI enhancement and classification models.