Utilizing Computational Spectral Imaging Technology to Promote Innovation in Non-Destructive Testing in the Industry 4.0 Era

Abstract

In the framework of Industry 4.0, the requirement for high accuracy and efficiency in non-destructive testing (NDT) technology continues to rise. Traditional methods can be inaccurate and slow in complex situations. This paper introduces a new inspection solution based on computational spectral imaging (CSI). This method builds a multispectral imaging system to collect high-dimensional data with defect detection and imaging quality improved by feature extraction and spectral merging. The experiments demonstrate noticeable performance of the proposed method for typical industrial material inspections with root mean square error (RMSE) of 2.16×10⁻², peak signal-to-noise ratio (PSNR) of 35.42 dB, and structural similarity (SSIM) of 0.987. The single-sample inspection time is reduced to approximately 62% of the original based on traditional 3D-CNN (three dimensional convolutional neural networks), with an accuracy above 95% at a noise σ of 0.065. The Kappa consistency coefficient for production line verification is also 0.924, with research showing that this system provides significant advantages for improved detection accuracy and level of automation - offering a practical solution for high-quality, low-cost intelligent detection in some manner or another in the Industry 4.0 environment.