A structured-light 3D hyperspectral imaging system with a multilayer perceptron reflectance correction model

The combination of three-dimensional (3D) imaging and hyperspectral imaging has been demonstrated to provide significant information about the scene, thereby facilitating enhanced perception and comprehension of the underlying structure of the real world. In response to the inadequate utilisation of the physical potential of 3D geometric information in current multimodal data fusion research, a structured-light three-dimensional hyperspectral imaging system (STHS) and a hyperspectral data correction method based on 3D geometry were proposed, and their application value in the analysis of plant physiological parameters was systematically explored. The system integrates structured-light 3D reconstruction, hyperspectral imaging, and illumination modules, achieving sequential acquisition and fusion of 3D and hyperspectral data within the 473–900 nm spectral range. It demonstrates a topographic accuracy of 0.04 mm at 800 mm working distance, 7 nm spectral resolution, and 0.98 Structural Similarity Index (SSIM) registration accuracy. An innovative reflectance correction model based on a Multilayer Perceptron-based Reflectance Correction (MLP-RC) was developed, which physically correlates geometric features with spectral response and reduces spatial non-uniformity to 1.1%. Compared with Lambertian correction and white-board correction, the standard deviation obtained using the MLP-RC model is significantly smaller than those of the other two methods, demonstrating its superior capability in restoringreflectance consistency. Leveraging the corrected data, we achieved three-dimensional visualization of chlorophyll distribution in plant leaves, revealing heterogeneous distribution patterns within complex structures. Destructive biochemical experiments were conducted on multiple plant species to validate the reliability of the system for estimating chlorophyll content. This geometry-constrained spectral correction substantially improves the accuracy of plant phenotype analysis and provides a novel technical solution for precision agriculture monitoring.